exchange rate regime choice: some determinants and ...30052513/chowdhury-exchange-2012a.pdf ·...
TRANSCRIPT
i
Exchange Rate Regime Choice: Some Determinants and Consequences
by
Mohammad Tarequl Hasan ChowdhuryBSS (CU), MSS (CU), MSc (UMB)
Submitted in fulfilment of the requirements for the degree of
Doctor of Philosophy
Deakin UniversityJuly, 2012
i
AcknowledgementsAll praise is due to Allah
First of all, I would like to extend my heartfelt gratitude to my principal supervisor,
Associate Professor Mehmet Ali Ulubasoglu, for his excellent supervision and
invaluable suggestions throughout the duration of my PhD study. His
encouragement and support have been instrumental in the timely completion of my
thesis. My sincere thanks also go to my Associate Supervisors, Dr Prasad S.
Bhattacharya and Dr Debdulal Mallick, for their guidance and helpful comments. I
greatly acknowledge the compassion and ‘open-door’ policy of my supervisory
team.
I would like to thank Deakin University for providing me with a Deakin University
International Research Scholarship (DUIRS) to pursue my PhD study. Thanks also
to the faculty members of the School of Accounting, Economics and Finance. My
interactions with them on different occasions have been very helpful. Thanks go to
all school administrative staff for making my study at Deakin enjoyable. I would
like to thank my friends and fellow PhD students in the school, especially Ranajit,
Syed, Habib, Reza, Anshu, Muttakin, Suresh, Anzum, Zohid and Jabbar, for their
support.
I acknowledge the generosity of the University of Chittagong, my employer in
Bangladesh, for granting me study leave to pursue my PhD at Deakin. I would also
like to extend my gratitude to my teachers and colleagues at the Department of
Economics, University of Chittagong, Bangladesh.
My parents have always been a great source of inspiration to me. Their prayers,
affection and encouragement have been the main source of my success throughout
my life. I am grateful to them. I also thank my brothers and sister for their love and
support. My father-in-law, Professor Shah Muhammad Shafiqullah, has been very
keen about my PhD, as were my other in-laws, which is much appreciated.
Finally, I show my appreciation to my wife, Salma, herself a PhD student, for her
love, constant support and understanding during the last three years, and my
daughter, Nazeefa, who has been eagerly waiting for her father to be ‘free’. Without
their support, it would have been difficult for me to complete this thesis.
ii
Dedication
This thesis is dedicated to my parents
Badrul Alam Chowdhury
and
Gulnahar Begum
iii
Contents
Acknowledgements ...............................................................................................i
Dedication ...............................................................................................................ii
Abstract...................................................................................................................vi
Chapter 1: Introduction.....................................................................................1
1.1 Introduction ...............................................................................................1
1.2 A brief historical overview of exchange rate regime choice ..................6
1.2.1 Classical Gold Standard .....................................................................6
1.2.2 Inter-war arrangements .....................................................................8
1.2.3 The Bretton Woods system.................................................................9
1.2.4 Post–Bretton Woods era...................................................................10
1.3 A brief account of the determinants and consequences of exchange rate regime choice....................................................................................11
1.4 Thesis overview........................................................................................14
References ................................................................................................................18
Tables and Figures ..................................................................................................23
Chapter 2: The Role of Capital Account Openness and Financial Development in Exchange Rate Regime Choice......................................27
2.1 Introduction .............................................................................................27
2.2 Relevant literature and motivation ........................................................31
2.3 Data...........................................................................................................34
2.4 Estimation strategy.................................................................................35
2.4.1 Model specification...........................................................................35
2.4.2 Estimation method ...........................................................................36
2.5 Results......................................................................................................38
2.5.1 Descriptive statistics...............................................................................38
2.5.2 Graphical analysis .............................................................................39
2.5.3 Results for the dynamic panel estimation .......................................40
2.5.4 Results for the fixed or random effect estimation ..........................43
2.6 Conclusion................................................................................................44
References ................................................................................................................46
Tables and Figures ..................................................................................................50
Appendix A.1: Variable descriptions and data sources.......................................58
iv
Chapter 3: An Empirical Inquiry into the Role of Product Diversification in Exchange Rate Regime Choice ..................................59
3.1 Introduction ............................................................................................59
3.2 Theoretical underpinnings......................................................................64
3.2.1 Real external shocks, diversification, and exchange rate regime choice ..................................................................................................65
3.2.2 Financial development, diversification, and exchange rate regime choice ..................................................................................................66
3.2.3 Rent-seeking, diversification, and exchange rate regime choice...68
3.3 Data..........................................................................................................69
3.4 Empirical analysis ..................................................................................70
3.4.1 Model specification...........................................................................70
3.4.2 Estimation issues...............................................................................74
3.4.2.1 Instrumental variables estimation...........................................75
3.4.2.2 Identification through heteroskedasticity ...............................79
3.5 Estimation results ...................................................................................80
3.5.1 OLS, instrumental variables, and identification through heteroskedasticity .............................................................................80
3.5.2 The marginal effect of concentration on regime choice, conditional on the moderator variables...............................................................85
3.6 Conclusion................................................................................................88
References ................................................................................................................90
Tables and Figures ..................................................................................................95
Chapter 4: Exchange Rate Regime and Fiscal Discipline: The Joint Role of Trade Openness and Central Bank Independence ...............106
4.1 Introduction ..........................................................................................106
4.2 Literature review..................................................................................111
4.2.1 Theoretical literature .....................................................................111
4.2.1.1 Trade openness, exchange rate regime and fiscal policy.....114
4.2.1.2 Central bank independence, exchange rate regime and fiscal policy..........................................................................................115
4.2.2 Empirical literature........................................................................116
4.3 Data and methodology .........................................................................118
4.3.1 Data..................................................................................................118
v
4.3.2 Model specification.........................................................................120
4.3.3 Estimation method .........................................................................122
4.3.3.1 Pooled OLS ..............................................................................123
4.3.3.2 Identification and IV estimation............................................124
4.3.3.3 Robustness analysis..................................................................126
4.4 Results and discussion...........................................................................126
4.4.1 Descriptive statistics .......................................................................127
4.4.2 Overall budget balance ..................................................................129
4.4.2.1 Endogeneity correction ...........................................................133
4.4.2.2 Robustness check.....................................................................134
4.4.2.3 Marginal effect of exchange rate regime choice on overall budget balance..........................................................................135
4.4.3 Primary budget balance and cash surplus ...................................136
4.4.3.1 Primary budget balance .........................................................137
4.4.3.2 Robustness check......................................................................138
4.4.3.3 Marginal effects of regime choice on primary budget balance......................................................................................................138
4.4.3.4 Cash surplus ............................................................................138
4.4.3.5 Robustness check......................................................................139
4.4.3.6 Marginal effects of regime choice on cash surplus................139
4.4.4 Government expenditure ................................................................140
4.4.4.1 Marginal effects of regime choice on government expenditure....................................................................................................141
4.5 Conclusion.............................................................................................142
References ..............................................................................................................145
Tables and Figures ................................................................................................151
Chapter 5: Conclusion....................................................................................166
vi
AbstractExchange rate regime choice and the macroeconomic consequences of that choice
have been among the most controversial and important issues in open economy
macroeconomics. This thesis, comprising three essays, analyses three previously
unexplored determinants and the consequences of de facto, or actual, exchange rate
regime choice during the post–Bretton Woods period.
The first essay explores the role of capital account openness and financial sector
health as possible determinants of de facto exchange rate regime choice. Further, for
the first time, the essay examines the persistence of regime choice over time. Due
importance is given to regime choices by different country groups, such as low-
income, middle-income and high-income countries. The essay uses the Reinhart-
Rogoff de facto exchange rate classification and a panel dataset covering the post–
Bretton Woods period of 1971 to 2007, for a large number of countries. The
empirical methodology includes dynamic panel data models with system GMM, and
a static model with fixed effects and random effects estimation techniques. The
essay finds that regime choice is relatively persistent over time for low-income and
high-income countries, while such persistence does not exist for middle-income
countries. The study also finds robust evidence that capital account openness is
associated with fixed exchange rate regimes, particularly for low- and middle-
income countries. The analysis also provides some evidence that financial
development is another important determinant of regime choice. It is found that
countries with underdeveloped financial sectors are not good candidates for flexible
exchange rate policies. Therefore, low-income countries generally appear to be the
reluctant floaters. The effect of another measure of financial sector health—namely,
financial sector fragility—on regime choice remains largely insignificant.
vii
The second essay investigates whether product diversification influences the
exchange rate regime choice and the mechanisms through which this effect may
operate. This essay identifies three possible mechanisms through which
diversification and regime choice may be related—namely, the shock absorption,
financial development and rent-seeking mechanisms. A direct effect of
diversification on regime choice is also hypothesised. By using a cross-country
dataset covering 125 countries from 1971 to 2000, and by adopting instrumental
variables estimation together with identification-through-heteroskedasticity
methodologies, this essay runs a ‘horse race’ among the hypothesised effects. The
relative importance of all these factors is also gauged empirically. The strongest
effects are found to be those of the direct effect and rent-seeking mechanisms. That
is, diversification makes countries less fearful of adopting flexible regimes, and, in
countries where corruption is high, concentration leads to fixed regimes, as it may
create more scope for rent-seeking by powerful elites. Diversification is also more
likely to lead to flexible exchange rate regimes, but only in developing countries
with lower levels of financial development. There is also some evidence that the
shock absorption channel moderates diversification, in which case diversification
facilitates the adoption of flexible regimes in countries experiencing greater shocks.
In consideration of the high and persistent fiscal deficit and public debt currently
marring several economies, the third essay addresses the role of exchange rate
regime choice in disciplining fiscal policy. There is a consensus at the theoretical
level that exchange rate regime choice has important bearing on the conduct of fiscal
policy; however, considerable debate continues as to which regime provides more
fiscal discipline. Empirical studies focusing only on the direct effect of regime
choice on fiscal discipline cannot provide a definitive answer. This study stresses
that, apart from its direct effects, regime choice can also affect fiscal outcomes
viii
through its interactions with trade openness and central bank independence. This
hypothesis is tested using different fiscal measures from two panel datasets covering
the periods from 1971 to 2000 and 1971 to 2007, for a large number of developed
and developing countries. This study finds strong evidence that the effect of regime
choice on fiscal discipline critically depends on the level of trade openness. More
specifically, estimated marginal effects show that fixed exchange rate regimes are
more disciplinary at a low level of trade openness, while flexible regimes exert
disciplinary effect at a high level of trade openness. However, the essay does not
find any evidence that central bank independence has an interaction effect with
regime choice to affect fiscal outcomes. There is also little evidence that the central
bank independence has a direct effect on fiscal outcomes.
1
Chapter 1: Introduction
1.1 Introduction
Exchange rate regime choice is one of the central macroeconomic policy choices for
an economy. This is because the choice of exchange rate regime has significant
influence on the path of nominal exchange rate, which is considered one of the most
important prices in any country. This price acts as a vital link between the domestic
economy and the rest of the world (Williamson 2009). Further, in the event of
domestic price stickiness, nominal exchange rate, in the short and medium term, is
the main determinant of real exchange rate, with the latter being a key determinant
of macroeconomic stability and motivation to trade (Williamson 2009; Frenkel and
Rapetti 2010). On the other hand, it is argued that a poor exchange rate regime
choice can lead to a number of detrimental outcomes for an economy. For instance,
probability, extent, and cost of the financial crises are tightly linked with exchange
rate regime choice (Domac and Martinez-Peria 2003). Consequently, the type of
exchange rate regime has been a core consideration within the International
Monetary System (IMS).
Regarding the critical roles played by the exchange rate regime in an economy,
Friedman (1953) argues that flexible exchange rate can insulate an economy from
real external shocks.1 Conversely, Williamson (2000, p 55) observes that flexible
exchange rate generates ‘short-run volatility’ as well as ‘long-run misalignments’ in
exchange rate which are not conducive for good macroeconomic performance.
Further, regime choice can influence the effectiveness of stabilisation policies,
1 Theoretically, exchange rate regimes are divided into two categories: fixed regimes and floating (or flexible) regimes. In this thesis, floating and flexible regimes are used interchangeably, which is common in exchange rate literature.
2
including monetary policy and fiscal policy. For example, according to the Mundell-
Fleming model (Mundell 1963; Fleming 1962), under perfect capital mobility,
adoption of a fixed exchange rate system renders fiscal policy fully effective in
changing output, while monetary policy becomes completely ineffective.
Conversely, a floating exchange rate makes monetary policy extremely useful and
fiscal policy fully neutral in affecting the output level. Thus, one of the implications
of this model is that the adoption of a floating regime enables countries to pursue
independent monetary policy for stabilisation, while the adoption of a fixed regime
makes this important stabilisation tool powerless. There is also argument that a fixed
exchange rate, itself, can act as a nominal anchor for monetary policy (Corden
2002). All of this highlights the essential role played by exchange rate policy in any
economy, and the importance of choosing an appropriate exchange rate regime.
These considerations together with a number of macroeconomic developments in the
global context make the exchange rate regime choice during the post–Bretton
Woods (post-BW) era a particularly important topic. Three features characterise the
exchange rate regime behaviour of countries in the post-BW period. First, the
International Monetary Fund (IMF)—the watchdog of the International Monetary
System—has granted liberty to its member countries to choose their exchange rate
regime in this period.2 This characteristic makes the post-BW era distinct from
previous international monetary arrangements in which the choice was very limited
(Klein and Shambaugh 2010). In fact, because of this liberty, countries have adopted
many variants of fixed and floating exchange rates, known as intermediate or in-
between regimes. For example, IMF currently distinguishes as many as 10 different
2 This has been possible due to a 1978 amendment of the IMF’s Articles of Agreement (see Article IV of IMF Articles of Agreement).
3
exchange rate regimes exercised by its members (see Table 1).3 ‘Hard peg’ regimes,
such as exchange rate arrangements with no separate legal tender and currency
board arrangement, and ‘independent floating’ regime appear, respectively, in the
first and last place on the list.
A second and related characteristic is that regime choice, which was a developed
country issue until the 1980s, has become an important policy choice for other
groups of countries, such as low-income, middle-income and emerging economies.
More importantly, a great degree of variation in regime choice is observed both
across and within countries and country groups over time. Thus, many different
regimes can be observed nowadays, including dollarisation (no separate legal
tender) in El Salvador, Ecuador and Kosovo; currency board arrangements in Hong
Kong, Bulgaria and Djibouti; currency union in EURO area countries; fixed regimes
in oil-exporting Arab countries; managed floating in Albania, Afghanistan, Thailand
and Turkey; and constant free-floating in the United States (US), Japan and
Australia. Latin American countries are particularly noted for their experiments with
virtually all types of exchange rate arrangements at different points in time (see
Frenkel and Rapetti 2010).
Third, a number of currency and financial crises, such as the crises in the European
Monetary System (1992), Mexico (1994), East Asia (1997) and Argentina (2001),
have brought the issue of appropriate choice of exchange rate regime to the forefront
of academic discussions and policy formulations. This debate has been stimulated
further by the influential study of Calvo and Reinhart (2002), who observed, in the
aftermath of the East-Asian financial crisis of 1997, that countries’ announced
regime choices differ considerably from their actual regimes. They argue that IMF’s
3 Table 1 also reports the number of countries affiliated to different regimes in 2009 to 2011.
4
de jure exchange rate regime classification, which is based on countries’
declarations of their regime choices, is ‘misleading’ (Reinhart and Rogoff 2004, p
4).4 This gives rise to ‘de facto’ or actual exchange rate regime classifications which
are based on observed behaviour of some variables such as nominal exchange rates
and reserve. Accordingly, three de facto classifications are proposed by Reinhart and
Rogoff (2004), Levy-Yeyati and Sturzenegger (2005), and Shambaugh (2004).5
Thus, exchange rate choice during the post-BW period is a highly debatable issue,
both at the theoretical and empirical levels.
To provide a perspective for the above-mentioned facets of exchange rate regime
choice, an overview of the Reinhart and Rogoff (2004) (RR) classification is helpful.
The RR classification distinguishes as many as 14 exchange rate categories for the
period 1946 to 2001 (see Table 2 for a list of these categories).6 Ilzetzki, Reinhart
and Rogoff (2008) extend this classification for the period 1940 to 2007 for a large
number of countries. Table 3, presenting some trends in de facto exchange rate
regime choice using Ilzetzki, Reinhart and Rogoff (2008) classification, shows that
only 8% of countries were floaters in 1975. This figure almost doubled in 1985. The
percentage of floaters increased slightly in 1990 to reach 20%, then declined slightly
in 1995, and then dropped significantly later. At the end of 2007, a little more than
4% of countries were floaters, which included constant floaters, such as the US,
Japan and Australia. In contrast, the vast majority of countries followed fixed and
4 IMF has been publishing the exchange rate arrangements of its members since 1945. 5 Since 1998, IMF also published a de facto classification. However, until mid-2000, the classification was mainly based on countries’ announced behaviour. Since 2009, IMF has announced a ‘revised classification system’, as shown in Table 1 (see also Habermeier et al.2009). However, panel data for this classification are not yet publicly available.6 RR classification is based on parallel exchange rate market data, and covers the longest periods. Two other de facto classifications—namely, Levy-Yeyati and Sturzenegger (2005) and Shambaugh (2004) have three and two categories, respectively. The main criticism of these classifications is that they rely on IMF official exchange rate data. As Rose (2012) states, these classifications ‘distrust’ IMF official classification, but rely on IMF official exchange rates.
5
intermediate regimes throughout the sample period. The percentage of fixed regime
countries dropped slightly from 48% in 1975 to 45% in 2007. Conversely, the
percentage of countries following a form of intermediate regimes increased from
43% in 1975 to 51% in 2007.
To offer an alternative picture based on country groups by income, 62% of low-
income countries adopted fixed regimes in the early period of the post-BW era. This
percentage reduced to 31% in 2007. Low-income floaters also almost halved during
the sample period, decreasing from 10.3% to 5.7%. In sharp contrast, the percentage
of low-income countries following intermediate regimes more than tripled during
the sample period to stand at 63% in 2007. The choice of fixed and intermediate
regimes of middle-income countries followed a similar pattern to low-income
countries, although changes for middle-income countries were less dramatic.
However, very few middle-income countries (only 2.4% in 2007) allowed their
currencies to float. The percentage of high-income countries following floating
regimes remained relatively stable during the sample period. The percentages were
7.7% and 6.1% in 1975 and 2007, respectively. However, in contrast to the two
other groups of countries, the share of high-income countries following fixed
regimes increased significantly, while the share for low-income countries following
intermediate regimes dropped sharply.
This overview suggests that countries’ actual regime choices exhibit a large
spectrum of fixed and intermediate regimes. Thus, contrary to Obstfeld and Rogoff
(1995), fixed exchange rate regimes are not ‘mirages’. Floating regimes, on the
other hand, appear to be a marginal category that reflects the ‘fear of floating’
behaviour first observed by Calvo and Reinhart (2002). Figures 1 and 2 present the
average values of the RR index over the years for all countries and different country
6
groups, respectively. The data in these figures also corroborate the fear of floating.
Figure 1 shows that, on average, countries moved to more flexible regimes from the
early 1970s. However, this trend seems to wane from the early 1990s. Similar trends
can be observed for the country groups of low-income, middle-income and high-
income, separately (see Figure 2).
1.2 A brief historical overview of exchange rate regime choice
To shed a further light on the choice of exchange rate regimes and their evolution
over the course of the last 150 years, a brief account of contemporary history of the
International Monetary System (IMS) is useful. This overview is intended to show
the reader the turning points in the understanding of exchange rate regimes. The said
period can be broadly classified into four distinct phases: classical gold standard
(1870 to 1914), inter-war arrangements (1914 to 1945), Bretton Woods fixed-but-
adjustable exchange rate system (1946 to 1973) and Post–Bretton Woods floating
era (1973 onwards). However, exchange rate regime during the early periods of IMS
was relatively simple because there was only one dominant system from which to
choose: fixed rate (Moosa 2005).7 Moreover, the choice was imperative mainly for
‘core’ countries; ‘peripheral’ countries sought to follow the core in most cases
(Bordo 2003).8
1.2.1 Classical Gold Standard: The Gold Standard, in essence, was a fixed
exchange rate system under which each country expressed the value of its currency
7 Those countries that failed to follow the dominant system (such as Austria-Hungary and Spain during gold standard periods) were viewed with disfavour (Bordo 2003). The failure to follow the dominant systems reflected these countries’ inability to do so, rather than a deliberate choice (Nurkse 1944).8 The core countries were Britain, France, Germany, the Netherlands, the US and some western European countries, while countries such as India, China and Austria-Hungary were considered peripheral (Bordo 2003).
7
in terms of a fixed amount of gold, and the currency was fully convertible in gold.9
However, this convertibility could be suspended temporarily in the event of shocks,
such as wars or financial disturbances. Once the disturbances were over, the
convertibility was restored at the pre-disturbance parity. This characteristic of the
Gold Standard was termed ‘escape clause’, and contributed to the smooth operation
of this regime (Eichengreen 1994, p 42).
The Gold Standard era began when Britain, the economic and political superpower
of that time, officially adopted that regime in 1870. Gradually, all other core
countries and many peripheral countries followed (Bordo 2003). Eventually,
countries subscribing to the Gold Standard accounted for approximately 67% of the
world Gross Domestic Product (GDP) and 70% of world trade (Chernyshoff, Jacks
and Taylor 2009, p. 196).
It is believed that capital flow during the Gold Standard period was free and high,
which resulted in the automatic adjustment of balance of payments.10 Despite high
capital mobility, exchange rate stability could be successfully maintained during the
Gold Standard period. This period was also relatively successful in absorbing
shocks. These were both possible due to a number of social and political factors (see
Eichengreen and Sussman 2000, p. 22). First, output stabilisation and unemployment
reduction were not issues, given that there was no generally established theory
relating monetary policy to the economic conditions. As such, these issues were
subordinate to maintaining exchange rate stability. Second, because of low labour
unionisation, nominal variables, such as wages and prices, were flexible downward.
In the event of internal and external shocks, this could adjust economies and
9 When each country announced the value of its currency in terms of gold, the (fixed) exchange rate between two currencies could be obtained by cross-referencing.10 This is known in the literature as ‘price-specie flow mechanism’.
8
maintain stable exchange rates. The ‘escape clause’, as mentioned above, was also
instrumental. At the beginning of World War I, gold convertibility was suspended.
This suspension was ultimately not temporary because the war lasted longer than
anticipated.
1.2.2 Inter-war arrangements: During World War I, countries following the Gold
Standard system were forced to postpone the commitments of backing their
currencies by gold (Eichengreen 2008). Instead, the countries issued ‘fiat-money’ to
meet war expenses, which caused extensive fluctuations in their exchange rates
(Eichengreen 2008, p. 45). Nurkse (1944) observes that, after the war, countries
attempted to return to the Gold Standard at pre-war gold parity, and in most cases at
new parity. However, he maintains, concern among policy makers about the short
supply of gold resulted in the adoption of a ‘gold exchange standard’, under which
convertible foreign currency could be used alongside gold as reserves.
The gold exchange standard broke down at the end of the 1920s. This occurred
because France, who accumulated huge surplus from the repatriation of capital and
from the current balance of payments, decided to accept only gold in settlement of
the surplus in 1928 (Nurkse 1944). This led Britain and other gold exchange
standard countries to abandon gold parity, and allow their currencies to fluctuate
(Nurkse 1944). He maintains that countries tried to control these fluctuations from
time to time, which reflected their reservations about free-floating exchange rates,
and their preference for exchange rate stability.
Though classical Gold Standard was relatively successful in maintaining a stable
exchange rate and absorbing shocks, the inter-war monetary system was noted for its
instability. This instability resulted from the fact that the economic and political
circumstances that were conducive for pre-war stability—such as macroeconomic
9
flexibility and less concern for output stabilisation or unemployment reduction—
were not present during inter-war periods (Eichengreen 2008, p. 43).
1.2.3 The Bretton Woods system: The Bretton Woods system emerged from the
negotiations between Britain and the US at the end of the World War II. The
objective of the negotiation was to set a monetary system that would be free from
the instability of the inter-war period and would ensure stable exchange rates,
national full employment policies and cooperation (Bordo 1993). To this end, the
case for a ‘fixed-but-adjustable’ exchange rate was made, which would combine the
favourable characteristics of the Gold Standard and flexible exchange rate
systems—namely, exchange rate stability and monetary and fiscal independence,
respectively (Bordo 1993).
Under this fixed-but-adjustable system, each country had to declare a fixed parity to
the US dollar, which was, in turn, fixed to a certain amount of gold. However, the
fixed parity could fluctuate within plus or minus 1% of the declared parity. The
fixed parity of a country could be changed in the face of ‘fundamental
disequilibrium’11 in a country’s balance of payments, and with IMF’s approval.
The BW fixed exchange rate system was different from the Gold Standard fixed
system in a number of ways (see Eichengreen 2008, p. 91). First, the BW-period
fixed exchange rates were adjustable in the event of ‘fundamental disequilibrium’ in
the balance of payments. Second, controls on international capital flows were
permitted in order to avoid destabilising speculation. Third, a new institution, IMF,
was formed to supervise national economic policies and help countries facing
problems in their balance of payments.
11 This term was not defined in the IMF Articles of Agreement. Williamson (2009) defines this as an imbalance in payments that makes the cost of maintaining the existing parity very high.
10
Bordo (1993), through reviewing historical data, states that overall macroeconomic
performance during the BW era was superior to its predecessors of the Gold
Standard and inter-war periods. However, the system had flaws that caused its
demise in the early 1970s. Bordo (1993) gives the following three reasons for the
collapse of the BW system. First, the gold parity of the dollar put the US under a
convertibility crisis, which led the US to adopt policies that made convertibility even
harder. Second, the US monetary expansion policy caused worldwide inflation,
which was not appropriate for the system. Third, adjustable peg was un-workable
under increasing capital mobility.
1.2.4 Post–Bretton Woods era: After the collapse of the Bretton Woods system in
the early 1970s, a new IMS emerged, in which countries could freely choose their
exchange rate arrangements. The result of this freedom of choice was that ‘the
United States, Japan, Europe, and developing and emerging market economies
embarked on different courses’ (Ghosh, Gulde and Wolf 2002, p 19). This ‘modern
era’ continues to be the longest era in the history of IMS (Klein and Shambaugh
2010).
This historical account of the IMS provides ample evidence of the role of capital
mobility and financial development in the adoption of prevailing dominant exchange
rate regimes and in the breakdown of regimes. To quote Bordo (2003, p. 32), ‘the
dynamics of the international monetary system and the evolution of exchange rate
regime is driven by financial development and international financial integration’.
Further, Bordo and Flandreau (2003) observe that, during the nineteenth century,
core countries (more developed countries) preferred to fixed their exchange rates,
whereas the same developed countries have opted for a floating regime since the
early 1970s. The case was more or less opposite for the peripheral (less developed)
11
countries during these two periods, Bordo and Flandreau maintain. They argue that
financial maturity was critical for a country to adhere to the Gold Standard, and the
same is important to maintain a floating regime in the post-BW period. These
observations point to the critical role played by financial development in exchange
rate regime choice.
1.3 A brief account of the determinants and consequences of exchange rate regime choice
A rich history of the exchange rate regime system spanning the last 150 years, as
well as a wide spectrum of exchange rate regime choices in the post-BW era as
shown above, have not gone unnoticed by academic researchers. The key point of
investigation has been the factors affecting countries’ regime choices across fixed,
floating and intermediate regimes. Considering that the exchange rate regime choice
and its macroeconomic consequences are interrelated—‘they are flip sides of the
same coin: a rational choice of the exchange rate regime presumably reflects the
properties that the regime promises’ (Ghosh, Gulde and Wolf 2002, p 23) - two
strands of empirical literature have developed. One strand has examined the factors
that affect regime choice, while the other has investigated the macroeconomic
consequences of that choice. This section provides a concise review of the literature
that investigates exchange rate regime choice and its macroeconomic consequences.
This section also highlights some important gaps in the literature.
Since the late 1970s, empirical studies have identified a large number of economic,
political, institutional, and historical factors as possible determinants of exchange
rate regime choice. For example, Juhn and Mauro (2002) identify 30 such
determinants. Motivated by optimum currency area (OCA) literature proposed by
Mundell (1961), and extended by McKinnon (1963) and Kenen (1969), early
12
empirical studies (e.g. Heller 1978) mainly examined the role of economic factors
such as trade openness, country size, economic development, capital mobility,
inflation and export concentration. Subsequent empirical studies examined the other
economic factors that include, but are not limited to, shocks, international reserves,
GDP growth, current account balance, external debt, and remittance flows. The
political, and institutional factors examined by empirical studies include: political
instability, political freedom, interest groups, democracy, institutional quality,
central bank independence, and so on (for a list of these factors, see Juhn and Mauro
2002; Rizzo 1998; von Hagen and Zhou 2007; Carmignani, Colombo and Tirelli
2008; Levy-Yeyati, Sturzenegger, and Reggio 2010).12 Some common and robust
determinants across the empirical studies are: country size, trade openness, shocks,
economic development, political instability, and institutional quality.
Some authors (e.g. Husain, Mody and Rogoff 2005; Klein and Shambaugh 2008;
Rose 2012) argue that exchange rate regime choice can be persistent over time.
However, empirical studies invariably fail to document this persistence with formal
econometric techniques. There is also an important theoretical linkage between the
health of the financial sector, the degree of capital account openness, and the
exchange rate regimes (see Calvo and Mishkin 2003). However, a systematic and
thorough treatment is yet to be undertaken for these factors given their demonstrated
role in recent financial crises in emerging market economies, particularly that in
East Asia. Analogously, predicated on the idea that the real sector can impact on
some central macroeconomic policy options, the role of product diversification has
12 Here it should be mentioned that Exchange rate classifications categorising countries into different regimes over time are crucial for the analysis of the regime choice. Given the limitations of IMF de jure exchange rate classification, recent empirical studies have focused on ‘de facto’ or actual classifications.
13
been long acknowledged as a possible determinant of regime choice (see Kenen
1969). However, the effects of this factor have not yet been investigated either
directly or in conjunction with several other national indicators.
Regarding the macroeconomic consequences of exchange rate regime choice, Baxter
and Stockman (1989) were the first to empirically examine the behaviour of
macroeconomic outcomes, including industrial production, consumption, export and
import, under different exchange rate regimes. However, Baxter and Stockman find
little systematic relationship between exchange rate regimes and those variables.
Subsequent empirical studies use better exchange rate classification data and
advanced econometric techniques, and examine the consequences of exchange rate
regime choice on such variables as growth, growth volatility, shock absorption
capacity, bilateral trade, and inflation. For instance, Levy-Yeyati and Sturzenegger
(2003) investigate the growth effect of exchange rate regime choice and find that
floating regime is associated with higher growth. However, this effect seems to be
more relevant for non-industrial countries. Output volatility is also found to be
higher under fixed regimes (Levy-Yeyati and Sturzenegger 2003; Ghosh, Gulde and
Wolf 2002). Empirical studies further establish that flexible regime can act as a
shock absorber in the event of terms-of-trade shocks (Edwards and Levy-Yeyati
2005; Broda 2004) and world output and world real interest rate shocks (Hoffmann
2007). There is also evidence that fixed exchange rates are conducive to higher
bilateral trade (Klein and Shambaugh 2006; Frankel and Rose 2002) and lower
inflation (Klein and Shambaugh 2010).
Along this line, the issue of an exchange rate regime’s effect on fiscal discipline is
important and timely, given the high and persistent budget deficits currently
affecting many countries. There is a theoretical consensus, which dates back to
14
Keynes (1923), that exchange rate regimes can discipline fiscal policy. However,
empirical studies that only examine the direct effect of regime choice on fiscal
discipline cannot provide a definitive answer as to which exchange rate regime
provides more fiscal discipline. The reason for this may be that these studies fail to
account for the fact that exchange rate regime choice may interact with other
theoretically important variables to affect fiscal discipline. This conjecture is
justifiable, given that regime choice is a highly complicated issue. These gaps in
empirical research must be properly addressed.
1.4 Thesis overview
This thesis consists of three essays that investigate the role of capital account
openness, financial sector health and product diversification on de facto exchange
rate regime choice, and the effect of the de facto choice on fiscal discipline during
post-BW periods. In so doing, this thesis adopts RR de facto exchange rate
classification with 14 categories and treats it as a continuous variable. Exchange rate
classification data for the period from 1971 to 2007 are used. RR classification
contains regime choice information for 178 countries, out of which 41 are low-
income countries, 88 are middle-income countries and 49 are high-income countries.
Data for other relevant variables are obtained accordingly, based on availability.
This thesis exploits a panel dataset that covers almost the entire post-BW period.
The first essay, entitled The Role of Capital Account Openness and Financial
Development in Exchange Rate Regime Choice, examines how countries’ capital
account openness, financial sector development and financial fragility affect their de
facto exchange rate regime choice. Particular attention is given to the regime choice
of different country groups—namely, low-income, middle-income and high-income
countries. The major contribution of this essay is that it investigates the persistence
15
of regime choice, which previous studies have invariably ignored. In so doing, this
essay uses system GMM technique. Fixed effect/random effect estimation technique
has also been adopted for comparison purposes.
The important and robust finding of this essay is that regime choice is generally
persistent over time. This suggests that a country’s regime choice for the previous
period is a good predictor of regime choice for the current period. However, this
persistence is particularly pronounced in low-income and high-income countries.
Middle-income countries show very little persistence in their regime choice. This
implies that middle-income countries change their regimes relatively often. This
essay also finds robust evidence that higher capital account openness leads countries
towards more fixed regimes—particularly low-income and middle-income countries,
suggesting that countries prefer fixed regime induced exchange rate stability when
their capital account openness becomes higher. There is also some evidence that
financial sector development influences exchange rate regime choice. This leads
countries towards flexible regimes in general; however, this effect is not always
robust across different estimation techniques and country groups. There is little
evidence that financial fragility, the other measure of financial sector health, affects
regime choice.
The second essay An Empirical Inquiry into the Role of Product Diversification in
Exchange Rate Regime Choice, focuses on one theoretically important determinant
of regime choice: product diversification. The essay hypothesises that product
diversification affects exchange rate regime choice both directly as well as through
the real shock, financial sector development, and rent-seeking channels.
Considering the long-term nature of product diversification, this study is conducted
in a cross sectional setting. The endogeneity of product diversification is tackled by
16
instrumental variable (IV) and identification-through-heteroskedasticity technique
proposed by Lewbel (2012). Neighbours’ size-weighted diversification is used as an
IV for diversification. It is found that product diversification has a direct effect on
exchange rate regime choice. The effect is such that, as countries become more
diversified, they tend to adopt flexible regimes. This finding is contrary to Kenen’s
(1969) presumption that diversified economies are better candidates for fixed
regimes. However, it supports the idea that more diversified countries experience
lower output volatility, and thus, exhibit less fear of floating.
This essay also finds strong evidence that, in countries where the level of corruption
is high, concentration leads to fixed regimes, as it may create more scope for rent-
seeking on the part of powerful elites. Diversification is also more likely to lead to
flexible exchange rate regimes in developing countries with lower levels of financial
development. There is also some evidence that diversification facilitates the
adoption of flexible regimes in countries that are experiencing greater shocks.
The third essay Exchange Rate Regime and Fiscal Discipline: The Joint Role of
Trade Openness and Central Bank Independence, investigates the consequences of
exchange rate regime choice on fiscal discipline, given that previous studies
examining the direct effect of regime choice offer inconclusive results as to which
exchange rate regime provides more fiscal discipline. This essay states that an
exchange rate regime can affect fiscal discipline directly and indirectly through its
interaction with other variables. Specifically, this essay hypothesises that exchange
rate regimes interact with trade openness and central bank independence (CBI) to
jointly affect fiscal discipline.
This essay offers strong evidence that exchange rate regimes have a direct effect and
an interaction effect with trade openness to influence fiscal discipline. Therefore, the
17
effect of regime choice on fiscal discipline critically depends on the levels of a
country’s trade openness. Estimated marginal effects show that, at low levels of
trade openness, fixed regimes provide more fiscal discipline, while flexible regimes
have more disciplinary effects at high levels of trade openness. This essay does not
find any evidence that regime choice interacts with CBI to affect fiscal discipline.
There is also little evidence that CBI has any direct effect on fiscal discipline. These
findings are robust across different measures of fiscal outcomes, and are supported
by several robustness checks.
18
References
Baxter, M and Stockman, A 1989, ‘Business cycles and the exchange rate regime’,
Journal of Monetary Economics, vol. 23, pp. 377-400.
Bordo, M 2003, ‘Exchange rate regime choice in historical perspective’, NBER
Working Paper no. 9654, National Bureau of Economic Research, Cambridge.
Bordo, M 1993, ‘The Gold Standard, Bretton Woods and other monetary regimes: a
historical appraisal’, Federal Reserve Bank of St. Louise Review, March/April 1993.
Bordo, M and Flandreau, M. 2003, ‘Core, periphery, exchange rate regimes, and
globalization’, in Michael D. Bordo, Alan M. Taylor and Jeffry G. Williams (eds),
Globalization in Historical Perspective, the University of Chicago Press, Chicago.
Broda, C 2004, ‘Terms of trade and exchange rate regimes in developing countries’,
Journal of International Economics, vol. 63, pp. 31-58.
Calvo, G and Mishkin, F 2003, ‘The mirage of exchange rate regimes for emerging
market countries’, Journal of Economic Perspectives, vol. 17, no. 4, pp. 99-118.
Calvo, G and Reinhart, C 2002, ‘Fear of floating’, Quarterly Journal of Economics,
vol. 117, no. 2, pp. 379–408.
Carmignani, F, Colombo, E. and Tirelli, P. 2008, ‘Exploring different views of
exchange rate regime choice’, Journal of International Money and Finance, vol. 27,
pp. 1177-1197.
Chernyshoff, N, Jacks D and Taylor, A 2009 ‘Stuck on gold: real exchange rate
volatility and the rise and fall of the gold standard, 1875-1939’, Journal of
International Economics, vol. 77, pp. 195-205.
Corden, W 2002, Too sensational: on the choice of exchange rate regimes, The
MIT Press, Cambridge, Massachusetts.
Domac, I and Martinez-Paria, M 2003, ‘Banking crises and exchange rate regimes:
is there a link?’, Journal of International Economics, vol. 61, pp. 41–72.
Edwards, S and Levy-Yeyati, E 2005, ‘Flexible exchange rates as shock absorbers’,
European Economic Review, vol. 49, pp. 2079-2105.
19
Eichengreen, B 2008, Globalizing capital - a history of the international monetary
system, 2nd Edition, Princeton University Press, Princeton and Oxford.
Eichengreen, B 1994, International monetary arrangements for the 21st century,
Brookings Institution, Washington.
Eichengreen, B and Sussman, N 2000, ‘The international monetary system in the
(very) long run’, IMF Working Paper no. 00/43, International Monetary Fund,
Washington.
Fleming, J 1962, ‘Domestic financial policies under fixed and under floating
exchange rates’, IMF Staff Papers, vol.9, no.3, pp. 369-380.
Frankel, J and Rose, A 2002, ‘An estimation of the effect of common currencies on
trade and income’, Quarterly Journal of Economics, vol. 117, no. 2, pp. 437-466.
Frenkel, R and Rapetti, M 2010, ‘A concise history of exchange rate regime in Latin
America’, Working Paper no. 2010-01, Economics Department Working Paper
Series, University of Massachusetts, Amherst.
Friedman, M, 1953, ‘The case for flexible exchange rates’, Essays in Positive
Economics, The University of Chicago Press, Chicago.
Ghosh, A, Gulde, A and Wolf, H 2002, Exchange rate regimes – choices and
consequences, The MIT Press, Cambridge, MA.
Habermeier, K, Kokenyne, A, Veyrune, R and Anderson, H 2009, ‘Revised system
for the classification of exchange rate arrangements’, Working Paper no. 9/211,
Monetary and Capital Markets Department, International Monetary Fund,
Washington.
Heller, R 1978, ‘Determinants of exchange rate practices’, Journal of Money, Credit
and Banking, vol. 10, no. 3, pp. 308–321.
Hoffmann, M 2007, ‘Fixed versus flexible exchange rate – Evidence from
developing countries’, Economica, vol. 74, no. 295, pp. 425-449.
20
Husain, A, Mody, A and Rogoff, K 2005, ‘Exchange rate regime durability and
performance in developing versus advanced economies,’ Journal of Monetary
Economics, vol. 52, pp. 35-64.
Ilzetzki, E, Reinhart, C and Rogoff, K 2008, ‘Exchange rate arrangements in the
21st century: which anchor will hold?’, viewed 20 August 2009,
http://terpconnect.umd.edu/~creinhar/Papers.html.
IMF Annual Review – various years, International Monetary Fund, Washington.
IMF ‘Articles of agreement’, International Monetary Fund, Washington.
Juhn, G and Mauro, P 2002, ‘Long run determinants of exchange rate regimes: a
simple sensitivity analysis’, Working Paper no. 2/104, International Monetary Fund,
Washington.
Kenen, P 1969, ‘The theory of optimum currency areas: an eclectic view’, in R
Mundell and A Swoboda (eds.), Monetary problems of the international economy,
University of Chicago Press, Chicago.
Keynes, J 1923, A tract on monetary reform, Macmillan and Co., London.
Klein, M and Shambaugh, J 2006, ‘Fixed exchange rates and trade’, Journal of
International Economics, vol. 70, pp. 359–383.
Klein, M and Shambaugh, J 2008, ‘The dynamics of exchange rate regimes: fixes,
floats and flips’, Journal of International Economics, vol. 75, pp. 70–92.
Klein, M and Shambaugh, J 2010, Exchange rate regimes in the modern era, The
MIT Press, Cambridge, MA.
Levy-Yeyati, E and Sturzenegger, F 2003, ‘To float or to fix: evidence on the impact
of exchange regimes on growth’, American Economic Review, vol. 93, no. 4, pp.
1173–1193.
Levy-Yeyati, E and Sturzenegger, F 2005, ‘Classifying exchange rate regimes:
deeds vs. words’ European Economic Review, vol. 49, pp. 1603-1635.
Levy-Yeyati, E, Sturzenegger, F and Reggio, I, 2010, ‘On the endogeneity of
exchange rate regimes’, European Economic Review, vol. 54, pp. 659-677.
21
Lewbel, A 2012, ‘Using heteroskedasticity to identify and estimate mismeasured
and endogenous regressor models’, Journal of Business and Economic Statistics,
vol. 30, pp. 67-80.
McKinnon R 1963, ‘Optimum currency area’, American Economic Review, vol. 53,
no. 4, pp. 717–725.
Moosa, I 2005, Exchange rate regimes: fixed, flexible or something in between?,
Palgrave Macmillan, Hampshire and New York.
Mundell, R 1961, ‘A theory of optimum currency area’, American Economic
Review, vol. 41, no. 4, pp. 657–665.
Mundell, R 1963, ‘Capital mobility and stabilization policy under fixed and flexible
exchange rates’, Canadian Journal of Economics and Political Science, vol.29, no.4,
pp. 475-485.
Nurkse, R 1944 International currency experience: lessons of the inter-war period,
Economic, Financial and Transit Department, League of Nations, Geneva.
Obstfeld, M and Rogoff, K. 1995, ‘The mirage of fixed exchange rates’, Journal of
Economic Perspectives, vol. 9, no. 4, pp.73-96.
Reinhart, C and Rogoff, K 2004, ‘The modern history of exchange rate
arrangements: a reinterpretation’, Quarterly Journal of Economics, vol. 119, no. 1,
pp. 1-48.
Rizzo, J 1998, ‘The economic determinants of the choice of an exchange rate
regime: a probit analysis, Economics Letters, vol. 59, pp. 283–287.
Rose, A 2012, ‘Exchange rate regime in modern era: fixed, floating, and flaky’,
Journal of Economic Literature, vol. 49, no. 3, pp. 652–672.
Shambaugh, J. 2004, ‘The Effects of fixed exchange rates on monetary policy’,
Quarterly Journal of Economics, vol. 119, no. 1, pp. 301–352.
Von Hagen, J and Zhou, J 2007, ‘The choice of exchange rate regimes in developing
countries: a multinomial panel analysis’, Journal of International Money and
Finance, vol. 26, pp. 1071–1094.
22
Williamson, J 2009. ‘Exchange rate economics’, Open Economics Review, no. 20,
pp. 123 – 146.
Williamson, J 2000, Exchange rate regimes for emerging markets – reviving the
intermediate options, Institute for International Economics, Washington.
23
Tables and Figures
Table 1: Number of countries under different exchange rate regimes (IMF taxonomy)
Exchange rate regimes April 2009 April 2010 April 2011
No separate legal tender 10 12 13
Currency board arrangements 13 13 12
Conventional peg arrangements 42 44 43
Stabilised arrangements 13 24 23
Crawling pegs 5 3 3
Crawl-like arrangements 1 2 12
Pegged exchange rates within crawling bands 4 2 1
Other managed arrangements 21 21 17
(Managed) floating 46 38 36
Free-floating 33 30 30
Total countries 188 189 190
Source: IMF Annual Reports, various years.Note: IMF has been following this classification taxonomy since 2009. For previous taxonomies, see Habermeier et al. (2009).
24
Table 2: List of Reinhart-Rogoff (2004) de facto exchange rate classifications
Category Exchange Rate Arrangements/Regimes
1 No separate legal tender
2 Pre announced peg or currency board arrangement
3 Pre announced horizontal band that is narrower than or equal to +/-2%
4 De facto peg
5 Pre announced crawling peg
6 Pre announced crawling band that is narrower than or equal to +/-2%
7 De facto crawling peg
8 De facto crawling band that is narrower than or equal to +/-2%
9 Pre announced crawling band that is wider than or equal to +/-2%
10 De facto crawling band that is narrower than or equal to +/-5%
11 Moving band that is narrower than or equal to +/-2% (i.e., allows for both appreciation and depreciation over time)
12 Managed floating
13
14
Freely floating
Freely Falling (Hyper float)
Source: Reinhart and Rogoff (2004)Note: This classification is not directly comparable with IMF taxonomy in Table 1.
25
Tab
le 3
: Per
cent
age
of c
ount
ries
in d
iffer
ent e
xcha
nge
rate
reg
imes
ove
r tim
e (R
R c
lass
ifica
tion)
All
coun
tries
Low
-inco
me
Mid
dle-
inco
me
Hig
h-in
com
e
Yea
rFi
xed
Inte
r-M
edia
teFl
oatin
gFi
xed
Inte
r-m
edia
teFl
oatin
gFi
xed
Inte
r-m
edia
teFl
oatin
gFi
xed
Inte
r-m
edia
teFl
oatin
g
1975
48.3
43.3
8.3
62.1
27.6
10.3
53.8
38.5
7.7
30.8
61.5
7.7
1980
42.4
48.0
9.6
56.7
36.7
6.7
48.2
42.9
8.9
23.1
64.1
12.8
1985
34.1
48.8
17.1
39.3
42.9
17.9
32.1
48.2
19.6
33.3
53.8
12.8
1990
36.0
43.9
20.1
31.3
46.9
21.9
35.9
37.5
26.6
39.5
51.2
9.3
1995
38.8
41.9
19.4
35.3
41.2
23.5
35.4
40.5
24.1
46.8
44.7
8.5
2000
43.7
48.1
8.2
35.3
50.0
14.7
40.3
53.2
6.5
55.3
38.3
6.4
2005
44.4
50.9
4.7
27.0
64.9
8.1
41.0
56.6
2.4
63.3
30.6
6.1
2007
44.9
50.9
4.2
31.4
62.9
5.7
39.8
57.8
2.4
63.3
30.6
6.1
Sour
ce: C
alcu
late
d fr
om Il
zetz
ki, R
einh
art a
nd R
ogof
f (20
08).
Not
e: F
ixed
regi
me
incl
udes
cat
egor
ies o
neto
four
;int
erm
edia
te in
clud
es c
ateg
orie
s fiv
eto
12;
and
float
ing
regi
me
incl
udes
cat
egor
ies
13 a
nd 1
4 of
RR
cla
ssifi
catio
n.C
ount
ries a
re g
roup
ed b
ased
on
the
Wor
ld B
ank
clas
sific
atio
n.
26
Figure 1: Average values of RR index over time (all Countries)
Figure 2: Average values of RR index over time (Country Groups)
45
67
8R
egim
e C
hoic
e (R
R In
dex)
1970 1980 1990 2000 2010Year
24
68
Reg
ime
Cho
ice
(RR
Inde
x)
1970 1980 1990 2000 2010Year
RR for high-income RR for middle-incomeRR for low-income
27
Chapter 2: The Role of Capital Account Openness and Financial Development in Exchange Rate Regime Choice
2.1 Introduction
Understanding the choice of the exchange rate regime is crucial given its important
effects on key macroeconomic variables such as growth rate (Ghosh, Gulde and
Wolf 2002; Levy-Yeyati and Sturzenegger 2003), output volatility (Levy-Yeyati and
Sturzenegger 2003), and inflation (Klein and Shambaugh 2010; Ghosh, Qureshi and
Tsangarides 2011). However, the determination of the exchange rate regime itself
remains one of the least understood areas in international macroeconomics, despite a
recent surge in research. One of the reasons for this lack of clear understanding is
the presence of diverse de facto exchange rate arrangements across countries, as
opposed to the theoretical classification of fixed versus floating regimes. Further,
data on this detailed arrangement have been made available only recently. The
econometric difficulty in incorporating many different choices into analysis in a
tractable manner has led to a corpus of research to collapse them, which results in
loss of information, or even failure to explain the true choice made by a country.
This study investigates two important determinants of exchange rate regime choice -
capital account openness and financial sector performance: both are firmly rooted in
theoretical grounds. The linkages among the degree of capital account openness,
health of the financial sector and exchange rate regime choice have come to
forefront after the financial crises in emerging market economies, including those in
the East Asian countries (Domac and Martinez Paria 2003). Further, the
international community, including the International Monetary Fund (IMF),
pressures countries to move towards a greater exchange rate flexibility and liberalise
international capital movements (Cooper 1999). The Independent Evaluation Office
28
of the IMF (2006, p 3) also observes that ‘since the Argentine crisis the IMF has
been seen by many as strongly favouring a flexible exchange rate underpinned by
inflation targeting as the only viable regime under most circumstances’. However, in
many instances, the IMF prescriptions are considered suspicious and
counterproductive in the developing world.
Although there are some recent studies that investigate the role of capital account
openness and financial development, this study departs from the extant literature in
several distinct ways. First, it takes into account the persistence of exchange rate
choice that previous studies have categorically ignored. It also exploits the most
detailed 14 de facto exchange range arrangements compiled to date. A country does
not frequently switch from fixed to floating exchange rate (or vice versa); rather, its
actual regime lies in between. It also stays with a particular arrangement for quite
some time before switching to another that may not be considerably different from
the previous arrangement. Only a rich classification of exchange rate arrangements
can account for this behaviour. Second, the study investigates the role of financial
sector fragility, in addition to financial development, to capture the health of the
financial sector that previous studies have ignored. It has now been recognised that
financial development alone cannot account for the health of the financial sector
(Loayza and Ranciere 2006). In addition, most studies investigating the determinants
of exchange rate regime use the ratio of M2 to gross domestic product (GDP) as a
proxy for financial development, which captures only the degree of monetisation in
the financial system and ignores the degree of bank intermediation. We use credit
disbursed to the private sector by banks and other financial institutions relative to
GDP; an advantage of this proxy is that it excludes credit extended to the public
sector and the funds provided from central or development banks. Finally, the role
of capital account openness has not so far been investigated in a systematic way;
29
previous studies have included this variable in regressions as a control, ignoring its
theoretical background.
There is a discrepancy between the de jure and de facto exchange rate regime
choices, referring, respectively, to the exchange rates announced by a country and
the actual one it practices. The IMF classifies exchange rate regimes based on the
announcement, but a country often deviates from its declared regime choice (Calvo
and Reinhart 2002). In this study, we adopt the de facto exchange rate regime
classification compiled by Reinhart and Rogoff (2004), henceforth RR, who
characterise countries in 14 ordered categories in terms of the actual exchange rate
choices ranging from fully fixed to fully flexible. However, the presence of too
many categories also poses econometric difficulty in estimating an ordered probit
model; this becomes a near impossible task in the context of dynamic panel data.
Nevertheless, 14 categories themselves enable treating the regime choice as
continuous,13 facilitating estimation of a dynamic panel data model by the system
GMM developed by Arellano and Bover (1995) and Blundell and Bond (1998). Data
are averaged over five-year periods because the system GMM is appropriate for a
large cross-section and a shorter time period; in our case this is also intended to test
the exchange rate persistence in the medium-run. To address the heterogeneity
among countries, we also estimate the model separately for countries, which, to a
great extent, are homogenous in terms of income levels. We do this because
developing, middle-income or developed countries encounter different
13 This kind of treatment is quite common in empirical literature. For example, Polity IV indicators or Freedom House political rights and civil liberties measures are often used as both the dependent variable, as well as the lagged dependent variable or explanatory variables on the right-hand side (see, Acemoglu et al. 2008; Bruckner and Ciccone 2011). Similarly, Bhattacharyya and Hodler (2010) treat corruption index with seven categories as continuous dependent variable, and apply system GMM. On the other hand, Aghion et al. (2009) consider Reinhart-Rogoff (2004) classification as continuous and use it as an explanatory variable.
30
macroeconomic problems that also influence their exchange rate choice. It is also
highly likely that the persistence of exchange rate choice may differ by the level of
development, especially because the middle-income or emerging countries have
gone through more frequent changes than others in their exchange rate
arrangements, thus exhibiting little persistence. We therefore estimate the model
ignoring the persistence, which in turn allows us to compare our results with the
existing literature.
Our results document that capital account openness is a significant determinant of
exchange rate regime choice, leading countries towards more fixed regimes,
especially for the low-income and middle-income countries. This result suggests that
higher capital account openness and free-floating regime are not compatible for
these groups of countries. The results also show that financial development leads
towards floating regimes, but the result is not robust across specifications. By
contrast, financial development leads to a fixed exchange regime for low-income
countries, suggesting that a floating regime is not a viable option for such countries
where the financial sector is underdeveloped. Financial sector fragility has no effect
when all countries are considered together; however, it leads to a fixed exchange
rate regime for low-income countries and a flexible regime for middle-income and
high-income countries (although the latter results are not quite robust across
specifications). Most importantly, we document a significant persistence of
exchange rate regime choice for low-income and high-income countries, but almost
no persistence for the middle-income countries, suggesting frequent changes in their
exchange rate regime.
The paper is organised as follows: section 2.2 reviews the relevant theoretical and
empirical literature and develops the motivation of the paper. Section 2.3 discusses
31
the data. Section 2.4 discusses the estimation strategy that includes system GMM,
fixed and random effect models. Section 2.5 presents the results. Finally, section 2.6
provides a conclusion. All tables and figures are reported at the end of the paper.
2.2 Relevant literature and motivation
The literature on the optimum currency area (OCA), first discussed by Mundell
(1961) and later extended by McKinnon (1963) and Kenen (1969), links labour
mobility, trade openness, country size and export diversification to exchange rate
regime choice. Subsequent literature suggests other determinants, which include,
among others, nominal and real shocks (Melvin 1985), political stability (Edwards
1996) and institutional quality (Alesina and Wagner 2006).
Theoretical arguments also suggest that development of the financial sector and its
stability and capital account openness are important determinants of regime choice.
Adoption of a floating exchange rate requires a reasonably developed financial
system, because exchange rate volatility is higher in countries with underdeveloped
financial markets, which in turn imposes economic costs and makes floating costlier
(Eichengreen 1994). This argument can be understood in the following way
(Eichengreen 1994, p 85):
a temporary disturbance that leads some investors to sell domestic-currency-
denominated assets will cause the exchange rate to plummet if liquidity constraints
or other financial market imperfections prevent other investors from purchasing
those assets in anticipation of a subsequent recovery in their value. The absence of
forward markets similarly renders it difficult for firms and households to hedge
exchange risk.
There is another strand of theoretical literature that stresses the role of financial
sector fragility on regime choice in the light of financial crises in emerging market
32
economies. When bank deposits are denominated in local currency and a central
bank serves as a lender of last resort, a flexible exchange rate can prevent a bank run
or crisis (Chang and Velasco 2000). The implication of this argument is that
countries with a fragile financial sector would be better off choosing a flexible
exchange rate. Conversely, Eichengreen and Hausmann (1999, p 6) purport that the
‘moral hazard’ problem is commonplace to varying degrees in all financial systems,
where ‘banks are leveraged, banks have limited liabilities, markets have asymmetric
information about the risk banks take, and banks are rescued with some probability
when they get into trouble’. This moral hazard problem leads banks to take
excessive risks, which is the main cause of financial fragility, Eichengreen and
Hausmann maintains. They consider a pegged exchange rate as a source of moral
hazard, in the sense that pegging is an implicit form of guarantee, which encourages
unhedged short-term foreign-currency-denominated borrowing. The choice of
exchange rate in this situation should be more flexible ones, which can limit short-
term capital inflows (Eichengreen and Hausmann 1999).
The ‘bipolar’ view of exchange rate arrangements asserts that with increasing capital
mobility countries should move towards either ends (hard peg or independent
floating) of the exchange rate arrangements; intermediate regimes will disappear and
floating will dominate between the two corner solutions (Eichengreen 1994; Fischer
2001). This view has its origin in the ‘impossible trinity’ theorem of international
economics, which states that a country cannot have exchange rate stability (a fixed
exchange rate), perfect capital movement and independent monetary policy at the
same time. Accordingly, given that capital is highly mobile, a country has to adopt a
floating regime if it wants to exercise monetary policy for stabilisation purposes.
Frankel (1999, p 7), however, purports that countries can choose intermediate
regimes even under perfect capital mobility, and achieve ‘half-stability and half-
33
independence’. Conversely, Williamson (2000) argues that as capital is highly
mobile only for industrial and emerging market countries, intermediate regimes can
still be viable options for a large number of low income countries, which are
effectively isolated from the international capital market.
Earlier empirical studies, such as Heller (1978), Dreyer (1978) and Holden, Holden
and Suss (1979), test the role of the OCA factors that include trade openness,
country size and export concentration, but these studies estimate cross-section
specifications. Other cross-section studies go beyond the OCA criteria to examine
the role of other determinants.14 Von Hagen and Zhou (2007) examine the role of a
large number of determinants for about 100 developing countries using a panel
dataset for the periods 1980 to 2000. They allow only three exchange rate choices—
fixed, intermediate and flexible—and estimate static and dynamic random effect
multinomial panel models using a simulation-based technique. Among the large
number of determinants, they find that OCA fundamentals (trade openness,
geographical concentration of trade, country size and level of economic
development), macroeconomic stabilisation considerations (inflation, relative price
shock and domestic monetary shock) and currency crisis factors (international
reserve adequacy, public finance performance and current account positions) are
important determinants of regime choice. They also find that financial development,
measured by the ratio of M2 to GDP, leads to a fixed regime, which is contrary to
theory.
One common limitation of these studies is that they use the IMF de jure exchange
rate classification, based on the announced regime choice by countries. However,
the accuracy of de jure classification is now questioned (see Calvo and Reinhart
14 For details, see Melvin (1985), Rizzo (1998), Juhn and Mauro (2002) and Poirson (2001).
34
2002) because of the significant difference between a country’s announced and
actual regime choices. Considering the limitations of de jure exchange rate
classification, subsequent studies use de facto classification. For example, Alesina
and Wagner (2006) use RR de facto exchange rate classification; however, they
choose a coarse classification of only five categories to investigate the role of
institutional quality on exchange rate regime choices and estimate an ordered logit
model. Lin and Ye (2011) examine the role of financial development on de facto
regime choice using logit and survival analyses. They find that financial
development increases the likelihood of shifting towards flexible regimes, although
at the low level of financial development a country adopts a fixed regime. However,
they estimate pooled regressions from the panel data without accounting for country
specific heterogeneity and other associated time-invariant unobservables.
The above review demonstrates that previous studies employ either cross-section or
pooled regressions and estimate a probit or logit model. Most importantly, these
studies do not take into account the dynamics of the exchange rate choice. It is
recognised that accounting the dynamics in discrete choice models is
computationally challenging. This study overcomes this difficulty by using the RR
14 category de facto exchange rate regime classification, which, as argued, can be
treated as a continuous measure given the large spectrum over which it classifies the
exchange rate regimes.
2.3 Data
We use a panel dataset for 145 countries for the period 1971 to 2007. The RR de
facto exchange rate regime classification that we adopt ranges from one to 14,
representing the most fixed and the most flexible regimes, respectively.15. Although
15 A complete list of the classification is provided in Table 2 in Chapter 1.
35
data for this classification are available from 1940 to 2007, this study uses data for
the period 1971 to 2007 (Ilzetzki, Reinhart and Rogoff 2008), because data
availability for most of the control variables starts from 1971 or later.16
Financial development is proxied by credit disbursed to the private sector by banks
and other financial institutions relative to GDP. Following Loayza and Ranciere
(2006), we use the standard deviation of growth rate of the ratio of private credit to
GDP as the proxy for financial sector fragility. We use the measure of financial
openness proposed by Chinn and Ito (2008) to proxy capital account openness. This
is a de jure measure of capital account openness, which reflects the policy intentions
of the countries. The index is normalised ranging from zero to one, where higher
values indicate that countries are more open to cross-border transactions. A full list
of the variables included in the analysis and their data sources are presented in
Appendix A.1.
2.4 Estimation strategy
2.4.1 Model specification
Given the persistence in exchange rate choice, the dynamic panel model is the most
appropriate estimation strategy. The persistence has also been interpreted as state
dependence in exchange rate regime choice (Von Hagen and Zhou 2007). Hence, we
estimate the following dynamic panel model:
= + + + , + + + + + … … . . (1)16 Also, before the 1970s (during the Bretton Woods period), regime choice was very limited; countries were following some kinds of fixed regimes. After the breakdown of the Bretton Woods system in the early 1970s, countries began to choose different types ofexchange rate regimes.
36
where is the exchange rate regime choice for country i in period (interval) t. represents country fixed effects, denotes time fixed effects, which are captured by
time dummies, and the error term is assumed not be correlated across countries.
CAO, PCR and FSF are capital account openness, private credit and financial sector
fragility, respectively.
The variables in the vector have been chosen based on economic theory and
empirical literature discussed in section 2.2. The vector includes: i) openness
measured as the share of export-import in GDP, ii) real shocks, iii) nominal shocks,
iv) product concentration, v) inflation, vi) current account balance as a percentage of
GDP, vii) quality of institution proxied by polity score (polity2) and viii) country
size measured by (log) GDP.
The real shock is measured by the standard deviation of the terms of trade (TOT),
while the nominal shock is measured by the average absolute deviation of the
transformed growth rate of broad money (M2) from the three-year backward moving
average (see, Von Hagen and Zhou 2007). Standard deviation measures the
volatility of a variable. We use a five-year period to calculate the volatility; annual
data for all variables are averaged over five years accordingly. Five-year averaging
gives seven time intervals: 1971 to 1975, 1976 to 1980, 1981 to 1985, 1986 to 1990,
1991 to 1995, 1996 to 2000 and 2001 to 2007. However, data for the entire period
are not available for many countries, so we estimate an unbalanced panel.
2.4.2 Estimation method
In estimating equation 1 we consider RR classification as a continuous variable
which can be justified as follows. Countries choose a set of exchange rate
37
arrangements among a continuum of policy choices to comprise their exchange rate
regimes. The de facto classification of Reinhart-Rogoff categorises this spectrum
into 14 groups. If two countries differ only slightly in their exchange rate
arrangements, they are likely to be allocated into the same group. To be allocated
into different groups, they must have made sufficiently different policy choices. The
question that arises here is whether these 14 groups should be treated as ordinal
variables, or continuous variables rounded to integers. Even if one views these
categories as ordinal, it seems that 14 different regimes make the underlying
categorisation almost indistinguishable from a continuous measure.17 It is also
important to mention that the five-year averaging of regime choice, which is a
discrete variable, transforms it into a literally continuous variable. 18
We then estimate equation (1) by the Arellano and Bover (1995) and Blundell and
Bond (1998) system GMM method.19 This method assumes that there is no
autocorrelation in the errors, ,i t and requires the initial condition that the panel-
level effects, i be uncorrelated with the first difference of the first observation of
the dependent variable ( ,i ty ). We report the Arellano-Bond (1991) test statistic for
second-order serial correlation in the first-differenced errors. The estimators are
consistent only if the moment conditions are valid. We test the validity of the
overidentifying moment conditions by the Sargan statistic. However, the asymptotic
distribution of the Sargan statistic is unknown when the standard errors are corrected
17 Monte Carlo simulations of Johnson and Creech (1983) demonstrate that grouping a continuous measure into finite number of categories is unlikely to create measurement error problem if the number of categories is greater than five. Ferrer-i-Carbonell and Frijters (2004) find that cardinality vs ordinality assumptions regarding happiness scores make little difference to the results.18 As shown in footnote 13, treating categorical variables as continuous dependent variables is quite common in empirical literature.19 For a comprehensive discussion of the estimation of dynamic panel data, see Roodman (2009).
38
for heteroskedasticity. We therefore report the Sargan statistic by estimating the
model without such correction. All variables except TOT volatility and polity are
treated as endogenous. The exogeneity of TOT volatility is also supported by
empirical evidence (Mendoza 1995). It is also unlikely that a country’s exchange
rate choice will affect its institutional development; other similar studies, such as
Alesina and Wagner (2006), have also treated institutions as exogenous to exchange
rates.
We also estimate the model by three country groups separately—low-income,
middle-income and high-income—to understand how regime choice differs by the
level of development. Countries also vary considerably in terms of capital account
openness and financial development. Lastly, we also estimate our model ignoring
the persistence of the exchange rate choice (excluding , 1i ty in equation [1]). The
reason is that persistence, although relevant for countries in general, may not be
large enough for the middle-income countries. This conjecture is also supported by
our empirical results in the next section. Further, this estimation will be helpful to
compare our results with the previous literature. We estimate a fixed and random
effect model depending on the Hausman test. To account for endogeneity, the first
lag of all variables, except those of TOT volatility and polity, has been included in
the regression.
2.5 Results
2.5.1 Descriptive statistics
Before turning to the estimation results, we present some descriptive statistics of
main variables for the five-year averaged panel. First, the statistics for explanatory
39
variables are presented by three exchange rate regime groups20 (fixed, intermediate
and floating) (see, Table 1). It can be seen from the Table 1 that countries that
follow fixed exchange rates have more open capital accounts on average. Financial
development does not vary much by exchange rate regimes, but the financial sector
is more fragile in countries with a floating regime (see Table 1). Then, the
descriptive statistics of dependent variable (RR index) and explanatory variables are
presented by three country groups such as low-income, middle-income and high-
income countries in Table 2. Countries of different income groups are not much
different in terms of their exchange rate arrangements (see Table 2). Capital account
is more open in high-income countries, followed by middle-income countries.
Again, high-income countries are more financially developed than other groups.
Financial sector fragility is the highest in low-income countries and the lowest in
high income countries. The descriptive statistics of the other explanatory variables
also varies widely across different income groups.
2.5.2 Graphical analysis
In this section, we show relationships between exchange rate regime choices and
three main explanatory variables—namely, capital account openness, financial
development and financial fragility—with the aid of scatter graphs. In each case, we
present the relationships for full samples and for three country groups categorised by
income level.
Figure 1 shows the relationship between exchange rate regimes and capital account
openness. Overall, the relationship is negative, which implies that higher capital
account openness should lead countries towards fixed regimes (see Figure 1a). The
20 The reason for aggregation is that descriptive statistics by 14 exchange rate categories are difficult to follow. The following is the aggregation from the 14 categories discussed in Table 2 in Chapter 1: fixed—1 to 4, intermediate—5 to 12 and floating—13 and 14.
40
same relationship can be observed for three country groups as well (see Figure 1b to
Figure 1d). However, the relationship seems to be somewhat weaker for high-
income countries.
Figure 2 documents the association between exchange rate regime choices and
financial sector development, as measured by private sector credit to GDP. The
overall positive relationship between these two variables reflects that countries adopt
flexible regimes as they become more financially developed (see Figure 1a).
However, this relationship varies widely across country groups. For instance, as can
be seen from figures 1b to 1d, the positive relationship is pronounced for high-
income countries only. Interestingly, for low and middle-income countries the
relationship is negative; that is, financial development in these countries is
associated with fixed regimes.
The relationship between exchange rate regime choices and financial fragility also
varies across country groups, as can be seen from Figure 3. Higher financial fragility
leads to flexible regimes in low- and medium-income countries, but to fixed regimes
in high-income countries.
2.5.3 Results for the dynamic panel estimation
In the following, we first present the results for the dynamic panel estimation. Our
first specification includes only the three main variables—capital account openness,
private credit and financial sector fragility. The second specification includes two
OCA variables, such as trade openness, and country size. The final specification
adds product concentration, institutional quality and factors related to
macroeconomic and currency crisis, such as institution, both real and nominal
shocks, inflation and current account balance. The sample size reduces in the full
41
specification because of the lack of data for product concentration, institution and
TOT volatility for several countries, so we also estimate another specification
excluding these three variables to take advantage of a larger sample.
The results for all countries are reported in Table 3. In our first two specifications,
the AR(2) test does not pass the suggested criteria. In the following, we report only
the results in which both AR(2) and Sargan tests are valid. The results for the
specification with all controls (see column 3) show that capital account openness has
a significant negative coefficient, suggesting that more open capital account leads to
fixed exchange rate regimes. Conversely, financial development leads to more
flexible exchange rate, as indicated by its positive and significant coefficient.
Financial fragility has been found to have no effect. Among other controls, only
nominal shock has a significant effect, and it leads a country to choose more flexible
exchange rates. In our final specification with fewer controls, but with a larger
number of countries, only capital account openness remains significant (see column
4). It is important to note that in all specifications the coefficient of the lagged
dependent variable is large, ranging between 0.5 and 0.8, and significant at any
conventional level.
We now estimate the model separately for three country groups—low-income,
medium-income and high-income. The results are presented in Table 4. For low-
income countries, the results for the specification with all controls show that capital
account openness leads to fixed exchange rates, while greater financial development,
unlike in the full sample, leads to fixed exchange rates (see column 1). A country
with financial fragility opts for fixed exchange rates. Among other determinants,
product concentration is related to fixed exchange rates and nominal shocks lead to
flexible exchange rates. Country size also leads low-income countries to flexible
42
regimes. However, only financial fragility and country size continue to be
significant in the larger sample with higher number of countries (see column 2). In
both specifications, coefficients of lag dependent variables are quite high (0.6 and
0.7, respectively), and highly significant, which suggests the persistence of exchange
rate regime choices for low-income countries. The results for the middle-income
countries are reported in column 3. However, the coefficient of the lagged dependent
variable is insignificant and its magnitude is very low at less than 0.2. This suggests
that a dynamic panel model is not appropriate for this group of countries. This
implies that persistence in exchange rate regimes is insignificant, indicating middle-
income countries change their exchange rate arrangements more frequently than
low-income and high-income countries. The results for the high-income countries
are presented in columns 4 to 5. In the specification with all controls, none of the
determinants is significant. However, there are only 13 countries in this
specification. When the above three variables are excluded, the number of countries
increases to 26. Contrary to the low-income countries, both financial development
and its fragility lead to greater exchange rate flexibility, and capital account
openness has no significant effect. However, regime choice for this group of
countries is also highly persistent, as can be seen from the large and significant
coefficient of lag dependent variables.
Although we treat the exchange rate arrangements as continuous, it is difficult to
give a quantitative interpretation of the results, because marginal change from one
arrangement to another qualitatively differs across different arrangements. However,
we can interpret the results by calculating beta coefficient which is -0.13 for capital
account openness in column 4 of Table 3. To interpret it, we take the case of Brazil.
During 2001 to 2007 periods, Brazil’s regime choice was at the flexible end of RR
classification index, and its average capital account openness was roughly equal to
43
the mean value of the capital account openness index (0.41). Then, according to our
model, one standard deviation increase in the Brazil’s capital account openness will
bring the country’s regime choice next to Chile or Columbia. Similarly, South
Africa was a free-floater and had a relatively closed capital account (at 25%
percentile value) on an average during 2001 to 2007. Our model predicts that one
standard deviation increase in the capital account openness should bring South
Africa’s regime choice at Thailand or Mexico’s level.
2.5.4 Results for the fixed or random effect estimation
Although we have found significant persistence in the choice of exchange rate
regimes, we now estimate the model by ignoring the dynamics for the following
reasons. First, persistence of the exchange rate regime varies across country groups;
as documented above, it is indeed not persistent for the middle-income countries.
Second, this investigation helps understand whether an inclusion of dynamics in the
exchange rate choice model influences the significance of other determinants.
Finally, this allows us to compare our results with the literature. We estimate the
model by fixed or random effects depending on the Hausman test. Our estimation
still departs from others in the literature in terms of using both panel data and
disaggregated classification of exchange rate regimes.
The results for all countries are reported in Table 5. We estimate the same
specifications as in previous section. The results indicate that capital account
openness leads to fixed exchange rate choices, as in the dynamic model. The
magnitude of the coefficient is smaller in both full and reduced samples (columns 1
and 2, respectively) compared to those in the dynamic panel estimations. Financial
development leads to more flexible exchange rate regimes, but financial fragility has
no effect; these results are in line with the previous results. However, in contrast to a
44
dynamic panel model, we find that TOT volatility and better institution leads a
country to more flexible exchange rates. Large countries also opt for more flexible
exchange rates.
However, as in the dynamic panel model, results vary across countries of different
income levels, evident in Table 6. For example, capital account openness contributes
to fixed exchange rates only for middle-income countries; financial development
contributes to flexible exchange rates only for high-income countries. Nominal
shocks lead a country to choose fixed exchange rates for the high-income countries.
A low-income country experiencing high inflation would choose fixed exchange
rates, which establishes the role of a fixed exchange rate regime as a nominal anchor
to inflation.
We also find some contrasting results in the two different estimation methods, which
are due to the failure of incorporating the dynamics in exchange rate choices.
2.6 Conclusion
Determination of exchange rate regimes remains to be one of the least understood
areas in international macroeconomics, despite a recent surge in research in the area.
A number of studies consider different factors, but overlook capital account
openness and financial sector development and fragility. This paper investigates the
role of current account openness and overall performance of the financial sector in
exchange change regime determination in a dynamic panel framework, accounting
for the dynamic persistence of the exchange rate choice. Adopting a detailed
classification of 14 de facto exchange rate arrangements, estimations are carried out
across low-income, middle-income and high-income country groups. These features
distinguish this paper from the extant literature.
45
One important contribution of this paper is to uncover significant persistence in
exchange rate choice over time. However, the persistence varies considerably across
different country groups. Exchange rate regime choice appears to be quite persistent
for low- and high-income countries. Middle-income countries do not exhibit any
persistence at all, pointing to frequent changes from one exchange rate arrangement
to another. Once this persistence is controlled for, capital account openness is
documented to be leading countries towards more fixed regimes. This effect is
especially pronounced for the low-income and middle-income countries. Financial
development leads towards floating regimes in general, but to fixed exchange
regimes for low-income countries. The latter result suggests that floating is not a
viable option for low-income countries with an underdeveloped financial sector.
Financial sector fragility is found to have no effect when all countries are considered
together; however, there are some evidences that it leads to fixed exchange rate
regimes for low-income countries and flexible regimes for middle- and high-income
countries. The latter finding is not robust across different estimation methods. We
also document that failure to account for the persistence in exchange rate choices
may lead to misleading results, especially in relation to the effect of TOT volatility
and institutional quality.
Future research in this area would benefit from exploring how exchange rate regime
choice can be determined in a non-linear fashion, given that the effects of some
determinants on the choice may depend critically on the levels of other theoretically
important variable(s).
46
References
Acemoglu, D, Johnson, S, Robinson J, and Yared, P 2008, ‘Income and democracy’,
American Economic Review, vol. 98, no. 3, pp. 808-842.
Aghion, P, Bacchetta, P, Ranciere, R and Rogoff, K 2009, ‘Exchange rate volatility
and productivity growth: the role of financial development’, Journal of Monetary
Economics, vol. 56, no. 4, pp. 494–513.
Alesina, A and Wagner, A 2006, ‘Choosing (and reneging on) exchange rate
regimes’, Journal of European Economic Association, vol. 4, no. 4, pp. 770–799.
Arellano, M and Bond S 1991, ‘Some tests of misspecification for panel data: Monte
Carlo evidence and an application to employment equations’, Review of Economic
Studies, vol. 58, no. 2, pp. 277–297.
Arellano, M and Bover, O 1995, ‘Another look at the instrumental variable
estimation of error-components models’, Journal of Econometrics, vol. 68, pp. 9–51.
Beck, T and Demirgüç-Kunt, A May 2009, ‘Financial institutions and markets
across countries and over time: data and analysis’, Working Paper 4943, World
Bank Policy Research, Washington.
Bhattacharyya, S. and Hodler, R. 2010, ‘Natural resources, democracy and
corruption’, European Economic Review, vol. 54, pp. 608-621.
Blundell, R and Bond, S 1998, ‘Initial conditions and moment restrictions in
dynamic panel data models’, Journal of Econometrics, vol. 87, pp. 115–143.
Bruckner, M. and Ciccone, A 2011, ‘Rain and the democratic window of
opportunity’, Econometrica, vol. 79, no. 3, pp. 923–947.
Calvo, G and Reinhart, C 2002, ‘Fear of floating’, Quarterly Journal of Economics,
vol. 117, no. 2, pp. 379–408.
Chang, R and Velasco, A 2000, ‘Financial fragility and the exchange rate regime’,
Journal of Economic Theory, vol. 92, pp. 1–34.
47
Chinn, M and Ito, H 2008, ‘A new measure of financial openness’, Journal of
Comparative Policy Analysis, vol. 10, no. 3, pp. 309–322.
Cooper, R 1999, ‘Exchange rate choices’, Discussion Paper No. 1877, Harvard
Institute of Economic Research, Harvard University, Cambridge.
Domac, I and Martinez-Paria, M 2003, ‘Banking crises and exchange rate regimes:
is there a link?’, Journal of International Economics, vol. 61, pp. 41–72.
Dreyer, J 1978, ‘Determinants of exchange rate regimes for currencies of developing
countries: some preliminary results’, World Development, vol. 6, pp. 437–445.
Edwards, S 1996, ‘The determinants of the choice between fixed and flexible
exchange rate regimes’, NBER Working Paper no. 5756, National Bureau of
Economic Research, Cambridge.
Eichengreen, B and Hausmann, R 1999, ‘Exchange rate and financial fragility’,
NBER Working Paper no. 7418, National Bureau of Economic Research,
Cambridge.
Eichengreen, B 1994, International monetary arrangements for the 21st century,
Brookings Institution, Washington.
Ferrer-i-Carbonell, A and Frijters, P 2004, ‘How important is methodology for the
estimates of the determinants of happiness?, Economic Journal, vol. 114, pp. 641–
659.
Fischer, S 2001, ‘Exchange rate regimes: is the bipolar view correct?, Journal of
Economic Perspectives, vol. 15, no. 2, pp. 3–24.
Frankel, J 1999, ‘No single currency regime is right for all countries or at all the
times’, NBER Working Paper no. 7338, National Bureau of Economic Research,
Cambridge.
Ghosh, A, Gulde, A and Wolf, H 2002, Exchange rate regimes: choices and
consequences, MIT Press, Cambridge.
Ghosh, A, Qureshi, M and Tsangarides, C 2011, ‘Words vs. deeds: what really
matters?’, IMF Working Paper 11/112, International Monetary Fund, Washington.
48
Heller, R 1978, ‘Determinants of exchange rate practices’, Journal of Money, Credit
and Banking, vol. 10, no. 3, pp. 308–321.
Holden, P, Holden, M and Suss, E 1979, ‘The determinants of exchange rate
flexibility: an empirical investigation’, Review of Economics and Statistics, vol. 61,
no. 3, pp. 327–333.
Ilzetzki, E, Reinhart, C and Rogoff, K 2008, ‘Exchange rate arrangements in the
21st century: which anchor will hold?’, Background Materials (available at
http://terpconnect.umd.edu/~creinhar/Papers.html, accessed on 20/08/09).
IMF 2006, ‘The IMF’s advice on exchange rate policy’, issues paper for an
evaluation by the Independent Evaluation Office (IEO), International Monetary
Fund, Washington, viewed http://www.ieo-imf.org/eval/ongoing/041906.pdf.
Johnson, D and Creech, J 1983, ‘Ordinal measures in multiple indicator models: a
simulation study of categorization error’, American Sociological Review, vol. 48, pp.
398–407.
Juhn, G and Mauro, P 2002, ‘Long run determinants of exchange rate regimes: a
simple sensitivity analysis’, IMF Working Paper no. 2/104, International Monetary
Fund, Washington.
Kenen, P 1969, ‘The theory of optimum currency areas: an eclectic view’, in R
Mundell and A Swoboda (eds), Monetary problems of the international economy,
University of Chicago Press, Chicago.
Klein, M and Shambaugh, J 2010, Exchange rate regimes in the modern era, MIT
Press, Cambridge.
Levy-Yeyati, E and Sturzenegger, F 2003, ‘To float or to fix: evidence on the impact
of exchange regimes on growth’, American Economic Review, vol. 93, no. 4, pp.
1173–1193.
Lin, S and Ye, H 2011, ‘The role of financial development in exchange rate regime
choice’, Journal of International Money and Finance, vol. 30, no. 4, pp. 641–659.
Loayza N and Ranciere, R 2006, ‘Financial development, financial fragility, and
growth’, Journal of Money, Credit and Banking, vol. 38, no. 4, pp. 1051–1076.
49
McKinnon R 1963, ‘Optimum currency area’, American Economic Review, vol. 53,
no. 4, pp. 717–725.
Melvin, M 1985, ‘The choice of an exchange rate regime and macroeconomic
stability’, Journal of Money, Credit and Banking, vol. 17, no. 4, pp. 467–478.
Mendoza, E 1995, ‘The terms of trade, the real exchange rate, and economic
fluctuations,’ International Economic Review, vol. 36, no. 1, pp. 101–37.
Mundell, R 1961, ‘A theory of optimum currency area’, American Economic
Review, vol. 41, no. 4, pp. 657–665.
Poirson, H 2001, ‘How do countries choose their exchange rate regime’, IMF
Working Paper no.1/46, International Monetary Fund, Washington.
Reinhart, C and Rogoff, K 2004, ‘The modern history of exchange rate
arrangements: a reinterpretation’, Quarterly Journal of Economics, vol. 119, no. 1,
pp. 1-48.
Rizzo, J 1998, ‘The economic determinants of the choice of an exchange rate
regime: a probit analysis, Economics Letters, vol. 59, pp. 283–287.
Roodman, D 2009, ‘How to do xtabond2: an introduction to difference and system
GMM in stata,’ Stata Journal, vol. 9, no. 1, pp. 86–136.
Von Hagen, J and Zhou, J 2007, ‘The choice of exchange rate regimes in developing
countries: a multinomial panel analysis’, Journal of International Money and
Finance, vol. 26, pp. 1071–1094.
Williamson, J 2000, Exchange rate regimes for emerging markets – reviving the
intermediate options, Institute for International Economics, Washington.
.
50
Tables and Figures
Table 1: Descriptive statistics of explanatory variables under different exchange rate regimes
Variables Fixed Regimes IntermediateRegimes
Floating Regimes
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.Capital account openness 0.49 0.35 0.42 0.33 0.34 0.35Financial development 0.44 0.38 0.42 0.35 0.46 0.51Financial fragility 10.60 10.82 9.42 9.20 19.04 56.51openness 94.52 52.95 78.35 54.28 52.87 36.17Level of development 8.58 1.17 8.67 1.037 8.58 1.16GDP (log) 22.17 2.50 23.51 1.96 24.10 2.62Product concentration 0.16 0.11 0.13 0.10 0.11 0.07Real shock 8.31 10.63 7.22 10.63 11.13 13.41Nominal shock 0.42 2.27 0.10 2.28 0.45 2.66Polity2 0.17 7.50 2.56 6.91 2.44 6.52Inflation 0.07 0.08 0.10 0.08 0.30 0.24Current account balance -4.90 11.79 -2.81 9.56 -3.90 5.59
Note: The following is the aggregation of exchange rate regimes from the 14 categories listed in Table 2 in chapter 1: fixed – 1 to 4, intermediate – 5 to 12, and floating – 13 to 14.
Table 2: Descriptive statistics of dependent variable and explanatory variablesby different country Groups
Variables Low-incomecountries
Middle-incomecountries
High-incomecountries
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.Regime choice 6.84 4.31 6.87 4.21 6.27 4.01Capital account openness 0.25 0.23 0.38 0.32 0.68 0.34Financial development 0.14 0.09 0.32 0.24 0.75 0.42Financial fragility 13.08 8.11 12.98 27.99 6.88 8.92openness 61.49 34.48 85.15 43.60 89.48 67.17Level of development 7.20 0.526 8.47 0.665 9.89 .621GDP (log) 21.63 1.39 22.32 2.39 24.42 2.49Product concentration 0.21 0.14 0.15 0.10 0.10 0.05Real shock 13.14 15.88 7.98 8.05 4.06 5.80Nominal shock 0458 2921 0.14 0.44 0.13 0.33Polity2 -2.56 5.18 0.583 7.01 5.12 7.69Inflation 0.12 0.12 0.14 0.16 0.06 0.06Current account balance -6.30 7.97 -4.46 9.05 -.359 12.39
Note: Countries are grouped based on World Bank classification.
51
Table 3: Determinants of exchange rate choice: dynamic panel estimation
Variables (1) (2) (3) (4)Lag dependent variable 0.818*** 0.797*** 0.511*** 0.755***
(0.087) (0.076) (0.132) (0.079)Capital account openness -1.269* -1.681** -3.438** -1.580*
(0.694) (0.772) (1.398) (0.908)Financial development 0.730 0.680 2.711*** 0.669
(0.722) (0.738) (0.924) (0.671)Financial fragility -0.038 -0.019 -0.017 -0.036
(0.037) (0.025) (0.036) (0.029)Openness -0.004 -0.005 -0.002
(0.007) (0.009) (0.007)GDP (log) 0.048 0.230 0.267
(0.223) (0.311) (0.191)Product concentration 3.005
(4.364)Inflation 4.147 0.370
(3.360) (1.951)Nominal shock 0.191** 0.052
(0.078) (0.049)Current account balance 0.030 0.004
(0.043) (0.017)Polity2 0.080
(0.059)Real shock 2.498
(3.917)Constant 2.557*** 1.900 -2.489 -2.937
(0.875) (5.480) (8.135) (4.537)Number of observations 630 617 206 528Number of countries 136 133 68 119Number of instruments 46 66 76 96p-value of the AR(2) coefficient
0.04 0.04 0.16 0.15
p-value of Sargan statistics 0.33 0.14 0.30 0.55Note: Figures in the parentheses are Arellano-Bond (1991) robust standard errors. ***, ** and * are significant at 1%, 5% and 10% level, respectively. All equations contain time dummies, but they are not reported.
52
Table 4: Determinants of exchange rate choice: dynamic panel estimation by country income group
Low-income Middle-income High-incomeVariables (1) (2) (3) (4) (5)Lag dependent variable 0.617*** 0.710*** 0.180 0.608*** 0.792***
(0.116) (0.091) (0.114) (0.172) (0.089)Capital account openness -3.756** 0.253 -3.163*** 1.412 0.453
(1.790) (1.363) (1.211) (1.424) (1.002)Financial development -7.655* -2.725 3.585*** 0.750 1.701*
(4.176) (3.031) (1.073) (1.158) (0.916)Financial fragility -0.146*** -0.065* -0.011 0.096 0.025**
(0.045) (0.036) (0.026) (0.059) (0.013)Openness 0.001 -0.013 -0.021* 0.002 -0.000
(0.019) (0.014) (0.012) (0.004) (0.005)GDP (log) 0.381* 0.506** 0.037 0.356 0.333
(0.207) (0.228) (0.448) (0.430) (0.241)Product concentration -10.552*** 8.050** -8.341
(3.535) (3.403) (6.411)Inflation 1.741 8.697 10.018*** 2.629 0.898
(8.309) (7.911) (2.903) (2.350) (1.882)Nominal shock 0.084** 0.014 -0.101 -1.379 0.318
(0.038) (0.058) (0.720) (1.063) (0.611)Current account balance -0.034 0.001 0.024 -0.048 -0.017**
(0.076) (0.042) (0.041) (0.038) (0.009)Polity2 0.037 0.163*** -0.050
(0.138) (0.055) (0.044)Real shock 3.995 3.980 3.572
(2.812) (3.792) (4.625)Constant -2.263 -8.250 3.774 -6.157 -7.411
(5.714) (5.815) (12.055) (10.905) (5.632)Number of observations 44 126 118 44 118Number of countries 18 30 37 13 26Number of instruments 66 96 76 72 96p-value of the AR(2) coefficient
0.15 0.61 0.17 0.55 0.82
p-value of Sargan statistics 0.98 0.90 0.28 0.97 0.44Note: Figures in the parentheses are Arellano-Bond (1991) robust standard errors. ***, ** and * are significant at 1%, 5% and 10%, respectively. All equations contain time dummies, but they are not reported. For middle-income countries regression results for full model are not reported because AR(2) and Sargan tests are not valid for the full model.
53
Table 5: Determinants of exchange rate choice: random effect estimation
Variables (1) (2) (3) (4)Capital account openness -2.869*** -2.868*** -1.987** -1.617**
(0.716) (0.700) (0.931) (0.717)Financial development 1.525* 1.229 2.800*** 1.894**
(0.924) (0.890) (0.909) (0.965)Financial fragility -0.000 0.003 0.028 0.012
(0.008) (0.008) (0.018) (0.010)Openness -0.012 -0.009 -0.012
(0.008) (0.010) (0.008)GDP (log) 0.581*** 0.581** 0.626***
(0.140) (0.261) (0.162)Real shock 5.734***
(1.992)Nominal shock 0.125*** 0.006
(0.027) (0.063)Product concentration -0.584
(3.090)Inflation -3.067 -0.259
(2.365) (1.585)Current account balance -0.035 0.019
(0.026) (0.014)Polity2 0.068*
(0.038)Constant 7.205*** -5.847 -6.831 -7.186*
(0.494) (3.694) (6.723) (4.161)
Number of observations 592 578 258 488Number of countries 145 142 74 127p-value of Hausman statistic 0.33 0.78 0.11 0.57Note: Figures in the parentheses are White (1980) robust standard errors. ***, ** and * are significant at 1%, 5% and 10% level, respectively. All equations contain time dummies, but they are not reported.
54
Table 6: Determinants of exchange rate choice: random (fixed) effect estimation by country income group
Low-income Middle-income High-incomeVariables (1) (2) (3) (4) (3) (6)
RE RE RE RE FE RECapital account openness -0.698 1.375* -2.786** -2.813*** -0.264 0.800
(1.869) (0.819) (1.175) (0.929) (1.182) (1.324)Financial development -0.340 -4.203 2.659* 2.226 2.368* 2.330*
(8.938) (3.436) (1.550) (1.454) (1.255) (1.315)Financial fragility -0.013 -0.021 0.042*** 0.018 0.154** 0.028
(0.055) (0.042) (0.014) (0.011) (0.059) (0.031)Openness 0.009 -0.030 -0.019 -0.019** 0.051* 0.003
(0.035) (0.023) (0.014) (0.009) (0.028) (0.008)GDP (log) 0.660 1.092** 0.688* 0.571*** 0.445 0.673*
(0.806) (0.531) (0.408) (0.212) (1.630) (0.357)Real shock -3.270 4.731* 13.406*
(5.763) (2.606) (6.659)Nominal shock 0.171*** 0.033 0.291 0.362 -3.019*** -1.354**
(0.049) (0.065) (0.443) (0.410) (0.494) (0.544)Product concentration -6.268 3.322 0.975
(9.862) (2.271) (5.035)Inflation 12.669*** 7.369** -5.464* -1.798 -2.608 -2.377
(4.751) (2.988) (3.014) (2.091) (1.873) (2.415)Current account balance 0.036 0.086 -0.029 0.025 -0.021 0.015
(0.072) (0.064) (0.031) (0.035) (0.044) (0.019)Polity2 0.032 0.119** 0.091
(0.112) (0.047) (0.083)Constant -7.715 -16.779 -9.396 -4.773 -8.064 -11.849
(19.207) (13.181) (10.575) (5.358) (42.156) (8.661)56 118 147 260 55 110
Number of observations 20 33 40 68 14 26Number of countries 0.31 0.13 0.30 0.39 0.01 0.32
Note: Figures in the parentheses are White (1980) robust standard errors. ***, ** and * are significant at 1%, 5% and 10% level, respectively. RE denotes random effect while FE denotes fixed effect estimation. All equations contain time dummies, but they are not reported.
55
Figures
Figure 1: Exchange rate regime choice and capital account openness
1a: All countries
1b: Low-income countries
05
10
15
Re
gim
e C
hoic
e
0 .2 .4 .6 .8 1Capital A / C Openness
05
10
15
Re
gim
e C
hoic
e
0 .2 .4 .6 .8 1Capital A / C Openness
1c: Middle-income countries
1d: High-income countries0
510
15
Re
gim
e C
hoic
e
0 .2 .4 .6 .8 1Capital A / C Openness
05
10
15
Re
gim
e C
hoic
e
0 .2 .4 .6 .8 1Capital A / C Openness
56
Figure 2: Exchange rate regime choice and financial development
1a: All countries
1b: Low-income countries
05
1015
Reg
ime
Cho
ice
0 .5 1 1.5Financial Sector Development
05
1015
Reg
ime
Cho
ice
0 .1 .2 .3Financial Sector Development
1c: Middle-income countries
1d: High-income countries
05
10
15
Re
gim
e C
hoic
e
0 .2 .4 .6 .8 1Financial Sector Development
05
10
15
Re
gim
e C
hoic
e
0 .5 1 1.5 2Financial Sector Development
57
Figure 3: Exchange rate regime choice and financial fragility
1a: All countries
1b: Low-income countries
05
1015
Re
gim
e C
hoi
ce
0 10 20 30 40Financial Fragility
05
1015
Re
gim
e C
hoi
ce
0 10 20 30 40Financial Fragility
1c: Middle-income countries
1d: High-income countries
05
10
15
Re
gim
e C
hoic
e
0 10 20 30 40Financial Fragility
05
10
15
Re
gim
e C
hoic
e
0 10 20 30 40Financial Fragility
58
Appendix A.1: Variable descriptions and data sourcesVariables Description Data source Data period
Regime choice (dependent variable)
Exchange rate regime choice, Reinhart-Rogoff ‘fine’ classification
Reinhart and Rogoff (2004) and Ilzetzki, Reinhart and Rogoff (2008)
1971–2007
Financial sector development
Private sector credit to GDP in decimal
Beck and Demirgüç-Kunt, 2009
1971–2007
Financial fragility Standard deviation of the growth rate of private credit
Beck and Demirgüç-Kunt, 2009
1971–2007
Capital account openness (CAO)
A CAO index, which takes values between 0 to 1
Chinn and Ito (2008) 1971–2007
Openness The sum of export and import of goods and services as percentage of GDP
Pen World Table version 6.3
1971–2007
Size of the economy
Log of GDP United Nations Statistics Division
1971–2007
Level of development
Log of per capita GDP in USD Pen World Tableversion 6.3
1971–2007
Product concentration
Herfindahl index, which takes values between 0 to 1
UNCTAD 1971–2000
Real shock Standard deviation of the terms of trade
WDI 1980–2007
Nominal Shock Average absolute deviation of the transformed growth rate of broad money (M2) from the three-year backward moving average
WDI 1971–2007
Polity2 A Polity index (-10 to 10), proxy for institutional quality
Polity IV Database 1971–2007
Inflation Transformed CPI inflation. Transformation formula: ((inflation/(1+inflation))
WDI 1971–2007
Current account balance
Current account balance as percentage of GDP (proxy forcurrency crisis factors)
WDI 1971–2007
59
Chapter 3: An Empirical Inquiry into the Role of Product Diversification in Exchange Rate Regime Choice
3.1 Introduction
There is a considerable debate over what determines exchange rate regime choices. A
corpus of literature has come up with no single theory or empirical model that
encapsulates all of the possible determinants of exchange rate policies. Given that the
factor(s) which play the most significant role in the choice of a regime are not known a
priori, studies have utilised a myriad of variables in their quest for the determinants
(see, inter alia, Juhn and Mauro 2002, Von Hagen and Zhou 2007, and Carmignani,
Colombo and Tirelli 2008, for a list).
This study focuses on product diversification as one of the key determinants of
exchange rate regimes. The core idea behind this paper is that production patterns in the
real sector are likely to shape some central macroeconomic policy choices, such as
exchange rate arrangements. Despite this intuitive point, the role of product
diversification in exchange rate regime determination has, surprisingly, not been
subjected to any empirical scrutiny. Our principal objective is to fill this gap.21 Also, as
is foreshadowed in the theoretical discussions below, product diversification may affect
the exchange rate regime not only directly, but also through a combination of several
factors. This line of reasoning leads us to explore the interactions between
diversification and shock absorption, financial market development, and rent-seeking.
Exploring these factors, then, enables us to investigate the “mechanisms” through which
they affect the exchange rate regimes. This approach is in contrast to the extant
21 Trade diversification has been considered in the literature for exploring different questions (see Michaely 1958, Holden, Holden, and Suss 1979, Savvides 1990).
60
empirical work, which assumes that the exchange rate regime is determined in a linear
fashion, i.e., only through a direct effect, by the hypothesised variable(s).
Going back to our core idea, patterns in the real economy, and the policy impacts
thereof, can be investigated through product (or sectoral) diversification.22 This is
helped by the definition of diversification, which refers to the economy’s engagement in
production activities in various sectors, instead of concentrating on only a few. Often
measured using metrics which capture the relative disparities among sectoral shares,
diversification has been shown, in an influential paper by Imbs and Wacziarg (2003), to
be non-monotonically related to the level of development. That is, economies initially
experience rising levels of diversification, then tend to re-concentrate after reaching a
certain level of income. This curious finding suggests that concentration follows a U-
shaped pattern with respect to per capita income (see also Koren and Tenreyro 2007).
Another important feature of the concept of diversification is that it is at the centre of
one of the longest-standing debates in the economics literature. On the one hand,
several economic theories, such as the Ricardian and Heckscher-Ohlin theories of
international trade, promote specialization, in order to reap the benefits of comparative
advantage and productivity gains. On the other hand, a number of studies, motivated by
a “portfolio” approach, emphasize the importance of diversification highlighting the
risks associated with specialization, and favouring diversification over specialization.
For instance, Burns (1960) considers the sectoral composition of the economy as one of
predictors of output volatility, because some activities, such as agriculture, tend to be
riskier than others, like services. Thus, product diversification is seen as a means of
22 On a terminology-related note, we refer to both sectoral diversification and concentration in this paper, with the latter being the inverse of diversification, i.e., a lack of diversification.
61
spreading production risks over a number of activities, and as a recipe for output
volatility and employment fluctuations (see Imbs and Wacziarg 2003, and Kenen
1969).23
As the discussion above also hints, the importance of product diversification as one of
the key factors in exchange rate regime determination has been acknowledged in the
context of its shock absorbing role in the economy. For instance, the output cost of
exchange rate volatility, as discussed by Lahiri and Végh (2001), can be mitigated
through diversification because a diversified economy is characterized by less volatile
terms of trade and real exchange rates. On the other hand, Kenen (1969), one of the
early studies of optimum currency areas (OCA), argues that product diversification
enables countries to adopt fixed regimes because it ensures a stable external sector and
terms of trade, and thus obviates the need for frequent changes in nominal exchange
rates. Despite having been observed in principle, this shock absorption effect has not yet
been put to an explicit empirical test. Furthermore, other possible mechanisms through
which product diversification may affect regime choice have been overlooked
categorically, both theoretically and empirically. This paper identifies the financial
development and rent-seeking channels as possible mechanisms behind exchange rate
regime determination, in addition to the shock absorption channel. The idea here is that
shock absorption, financial market development, and rent-seeking may, clearly, have
their own independent effects on exchange rate regime choice (see, for instance, Von
23 Cameron (1978), on the other hand, shows some ways in which industrial concentration can be risk-reducing. He argues that a high industrial concentration (i.e., a larger share of production and employment in a few firms) facilitates higher unionization and provides a wider scope for collective bargaining. This creates stronger labour confederations, ensuring larger income supplements in the form of social security schemes, health insurance, unemployment benefits, job training, and employment subsidies from the government. These income supplements help to mitigate the external risks.
62
Hagen and Zhou 2007; Alesina and Wagner 2006; and Lin and Ye 2011); however,
depending on their levels, they may also facilitate or inhibit the impact of
diversification, and/or may differ in their degree of influence at different levels of
diversification. The direct effect of product diversification, on the other hand, does not
depend on any of the three interaction variables above, but can still exert a significant
direct impact on the choice of exchange rate regime. We provide theoretical
underpinnings for the three mechanisms identified and the direct effect by drawing on
several studies. Thus, taken together, this study fills two major gaps in the literature.
First, it provides the first empirical assessment of the impact of product diversification
on the de facto exchange rate regime choice; and, second, it explores whether the
exchange rate regimes are shaped by a direct effect and/or (an) interaction effect(s) with
real shock absorption, financial development, and rent-seeking, associated with
diversification. The paper runs a ‘horse race’ among the three identified factors and the
direct effect in order to evaluate their relative levels of importance. In summary, the
main contribution of this study is to shed a unique light on the way in which changing
production patterns in the real sector shape a core factor in macroeconomic policy,
namely exchange rate regimes.
Our empirical analysis uses a cross sectional dataset covering 125 countries over the
period 1971–2000. We adopt Reinhart and Rogoff’s (2004) de facto24 exchange rate
regime classification, which ranges from 1 to 14, representing the most fixed and the
most flexible regimes, respectively. The Herfindahl index, a commonly used metric of
24 The IMF classification of exchange rate regimes, which is based on member countries’ announcements, is termed the de jure classification. However, countries often deviate from their announced behaviour (see Calvo and Reinhart 2002), which gives rise to the de factoclassification, reflecting the actual behaviour of exchange rates.
63
concentration, is utilised for measuring product diversification. An investigation based
on a five-year averaged pooled panel has also been conducted. Instrumental variable
(IV) estimation, together with ‘identification through heteroskedasticity’, as proposed
by Lewbel (2012), is used to tackle possible endogeneity due to reverse causation from
regime choice to product diversification. In drawing empirical inferences, we rely
primarily on marginal effects of the interaction terms rather than on point estimates;
these are illustrated graphically, portraying the portion of the constituent variable over
which the effect is statistically significant.
Our findings are very insightful. Product diversification has a direct, robust and
significant effect on regime choice, leading to flexible regimes. The rent-seeking
channel is robustly significant, with the result that, in countries with higher corruption
levels, sectoral concentration (or lower diversification) leads to fixed regimes, which
may provide more scope for rent-seeking among the powerful elites. The financial
development channel is also critical, as diversified economies will choose flexible
regimes, conditional on the level of financial development. However, this effect is
significant only for developing countries with relatively less developed financial
markets. Finally, diversification facilitates the adoption of flexible regimes in countries
which are experiencing greater shocks, a finding which is in contrast to Kenen’s (1969)
prediction. All of these results hold even after several variables and country-fixed
factors have been controlled for. Our main conclusions in relation to the marginal effect
of diversification on regime choice are generally consistent across the different
identification methods and the cross-sectional and panel datasets. In summary, this
study finds that product diversification is a critical component of exchange rate
determination, leading to flexible regimes in developing countries which are facing
64
higher levels of real shock, and where financial development is low and corruption is
relatively high.
The rest of the paper is arranged as follows. Section 3.2 discusses the possible links
between the regime choice and diversification. Section 3.3 describes the data, and
Section 3.4 the methodology. Section 3.5 presents the estimation results. Some
conclusions are drawn in Section 3.6. All tables and figures are reported at the end of
the paper
3. 2 Theoretical underpinnings
The direct effect of diversification on exchange rate regime choice is not well-
documented in the literature. Lahiri and Végh (2001) argue that the output costs of
exchange rate fluctuations explain the stylised facts about the exchange rate behaviour
well. This sheds light on the reluctance of most countries to float their exchange rates,
which is termed the “fear of floating” by Calvo and Reinhart (2002). Along these lines,
Levy-Yeyati and Sturzenegger (2007) argue that this “fear of floating” is actually a
“fear of appreciation”, which can be attributed to the ‘Dutch disease’ phenomenon, i.e.,
loss of competitiveness and serious setbacks to export diversification (Reinhart 2000).
The output costs due to excessive swings in the exchange rate under floating regimes
can be stabilised by diversified domestic and the external sectors, given that the terms
of trade and real exchange rate are less volatile in such economies. Thus, one can
hypothesise that product diversification would reduce the fear of floating, and therefore
lead to more flexible exchange rate regimes.
65
3.2.1 Real external shocks, diversification, and exchange rate regime choice
The direct effect of external shocks on the exchange rate regime should lead to flexible
exchange rate regimes, given that such regimes would insulate the economy against
negative consequences. This follows from the automatic stabilizer role of flexible
exchange rates in the event of a real shock (Friedman 1953). When domestic prices are
sticky, a negative real shock (say, a deterioration in the terms of trade) leads to a
depreciation of the nominal exchange rate, which in turn reduces the relative price of
tradable goods. Therefore, the negative effect of the shock is partially offset by the
resulting fall in the price of tradable goods (Broda, 2001).25
The interaction effect between diversification and shocks follows from the shock
absorbing role of diversification, which protects an economy from excessive
volatility.26 An example of the shock absorbing role of diversification is given by
Kenen (1969), who argues that a well-diversified economy will not have to undergo
frequent changes in its terms of trade, because product diversification is likely to be
reflected in export diversification, which keeps a country’s aggregate exports relatively
stable. This is unlikely in a concentrated economy, because sector-specific external
shocks will not be averaged out. Therefore, a well-diversified economy is more capable
of adopting a fixed exchange rate regime. Also, Corden (2002) argues that the costs of
25 The insulation property of flexible exchange rate regimes has been established empirically by Broda (2001), Broda (2004), Hoffmann (2007) and Ramcharan (2007). Broda (2004), using a panel VAR, finds that, in the event of terms-of-trade shocks, the adjustment to the economy is better under a flexible exchange rate regime. Hoffmann (2007) extends this analysis to an examination of whether this better adjustment capacity of flexible regimes can be established for other kinds of external shocks, such as world output and world interest rate shocks. 26 Ramcharan (2005) empirically examines the effectiveness of diversification for mitigating the costs of shocks. He measures the cost of shocks as the impact of earthquake shocks on a country’s consumption. Using a panel of 39 countries over the period 1971–2001, he finds that sectoral specialization greatly magnifies the cost of shocks.
66
abandoning independent monetary and exchange rate policies are likely to be lower in a
diversified economy because of the relative stability of the terms of trade and real
exchange rates.27
All of these studies suggest that the direct effect of real shocks which leads to the
adoption of flexible regimes is weakened by the presence of a concentrated economic
structure, resulting in a lack of insulation for the real economy. Similarly, the ‘fear of
floating’ effect associated with concentration is magnified in economies which are
facing higher real external shocks. Therefore, the sign of the interaction term between
shock and product concentration (diversification) is expected to be negative (positive).
3.2.2 Financial development, diversification, and exchange rate regime choice
It is now established that financial development has a direct impact on the exchange
rate regime choice. Financial development mitigates exchange rate volatility, and is
therefore conducive to reaping the benefits of flexible exchange rate (Eichengreen,
1994). This has also been established empirically by Lin and Ye (2011), among others.
The interaction effect between financial development and diversification in the
exchange rate regime determination is not clear-cut, because the argument could go
either way. For example, Saint-Paul (1992) argues that developed financial markets lead
to greater specialization, because risks arising from specialization can be insured in the
financial market. In the absence of financial markets, diversification can be an avenue
27 It must be noted, however, that both Kenen and Corden provide these arguments in support of the formation of a currency union (which is itself a fixed exchange regime from the viewpoint of the member country). Although a currency union may have other benefits in a specific environment, such as increased trade with the union members (Frankel and Rose, 2002), the above arguments do not necessarily imply that a currency union is the best choice for an economy facing real external shocks.
67
by which to mitigate the risks of production. Kalemli-Ozcan, Sorensen and Yosha
(2003) empirically support the argument above that countries with developed financial
markets are better able to insure against idiosyncratic shocks, and thus can exploit the
benefits of specialisation based on their comparative advantage. This suggests that
financial development itself leads to more specialization (or less diversification). It also
points to a negative (positive) sign for the interaction effect between concentration
(diversification) and financial development, given that the economy can tolerate more
specialisation, leading to the adoption of fixed regimes. In other words, the ‘fear of
floating’ effect of diversification, which normally leads to flexible regimes, is mitigated
in developed financial markets.28 On the other hand, Acemoglu and Zilibotti (1997)
observe that economies in the early stages of their development (as well as less
developed countries which are also financially underdeveloped) have fewer
opportunities to diversify, because they have less capital to invest. They argue that this
lack of capital allows them to undertake only a limited number of imperfectly correlated
projects. Moreover, these countries will seek insurance by investing in safe but less
productive assets. All of these factors reduce the room for diversification. However, as
financial markets develop, the concentration is reduced, due to a variety of investments,
and more flexible regimes can be adopted.29
Given the role of product diversification in exchange rate determination, as discussed
earlier, it can therefore be argued that the interaction between diversification and
financial market development is a valid mechanism for exchange rate choice, but the
direction of the effect is an empirical question.
28 Admittedly, this effect is more applicable to developed countries with developed financial markets.29 This argument is more applicable to developing countries.
68
3.2.3 Rent-seeking, diversification, and exchange rate regime choice
The direct effect of rent-seeking on the regime choice is likely to act through the
political economy channel. Concentration may be associated with only a handful of
interest groups in an economy, who may lobby the government for protection through
the exchange rate regime. Protection may be granted to such groups by fixing the
currency at a devalued rate in order to cover their lack of competitive ability in
international markets. Thus, in a political equilibrium, a fixed regime may emerge as an
outcome of what is now very well-known to be special interest politics (see the seminal
work of Grossman and Helpman 1994). The mechanism here implies that, analogous to
the determination of trade policy with tariff rates as an instrument of protection, a
lobbying process may be behind the exchange rate regime choice. Conversely,
diversification implies that there is a greater number of interest groups, greater political
competition, and a limited scope for rent-seeking (or limited rewards from a given
lobbying activity), as is reflected in the choice of a more flexible regime.
In terms of the interaction effect, the diversification levels of an economy depend
crucially on rent-seeking. Ades and Di Tella (1999) document the fact that corruption,
the most common proxy for rent-seeking in the literature, is higher in countries which
are dominated by a few firms. More specialised production structures facilitate the
capture of the resources by the powerful elite, thus increasing the scope for rent-
seeking. De Waldemar (2010) establishes the negative effect of corruption on
diversification through the channel of innovation, as corruption retards innovation.
Alesina and Wagner (2006) hypothesise and show empirically that institutional quality,
which is closely associated with corruption, has a U-shaped relationship with exchange
69
rate regime flexibility. The above arguments suggest that product diversification is
likely to be associated with low levels of corruption, which facilitates the choice of a
flexible exchange rate policy.
3.3 Data
This study exploits a cross-sectional dataset covering the period 1971 to 2000 for 125
developed and developing countries. As diversification is a long-run phenomenon,
cross-country analysis is well-suited to the examination of its relationship with regime
choice. We also adopt a five-year averaged pooled panel, with the objective of
examining the temporal dimension in the relationship and checking for the robustness of
cross-sectional findings. Reinhart and Rogoff’s (2004) de facto exchange rate regime
classification is commonly used in the literature for classifying exchange rate systems,
and hence will be adopted in this study too. The categorisation of regimes is based on
the actual behaviour of exchange rates observed from parallel or dual exchange rate
markets.30 The product diversification measure is the commonly used metric, the
‘Herfindahl index (HI)’, which is expressed as the summation of the squared shares of
all sectors in the total employment. The index takes values between 0 and 1, where
lower values indicate a higher level of diversification. In an economy with a low value
of the Herfindahl measure, production activities are dispersed across a wide range of
sectors. We have constructed HI from UNIDO (2003), which provides manufacturing
data at 3-digit level of disaggregation, in accordance with International Standard
Industrial Classification (ISIC). The data are available as employment shares across
different manufacturing sectors, and cover a wide range of countries over a long period
of time. Table 1 describes all of the other variables, as well as their data sources.
30 A list of the exchange rate regimes is provided in Table 2 in Chapter 1.
70
Table 2 presents the descriptive statistics of the key variables. Because HI is the inverse
of diversification, we use ‘concentration’ as the summary indicator of the real economy
from now on. A scatter plot of regime choice against concentration is displayed using
the locally weighted smoothing technique. Figure 1 shows that the relationship between
regime choice and concentration is linear and negative. The parametric quadratic fitted
line confirms the fact that the relationship between regime choice and diversification is
indeed negative (Figure 2). Table 3 shows the correlation matrix of the variables used in
the empirical analysis. Among the four main explanatory variables, the proxy for
financial development (findev) is somewhat correlated with corruption, with a
correlation coefficient of 0.55.
3.4 Empirical analysis
3.4.1 Model specification
We aim to examine the direct effect of product diversification on regime choice, as well
as the interaction effects between diversification and shock, financial development, and
corruption, based on the theoretical discussions above. In the language of interaction
models, concentration is the ‘focal independent’ variable, and shock, financial
development, and corruption are the ‘moderator variables’. For an interaction effect to
exist, the effect of the focal independent variable on the outcome variable (regime
choice) must vary based on the level of moderator variable. Therefore, we formulate
the following empirical model:
= + + + + + + + + + … … … (1)
71
where i is a country index; concen is a measure of sectoral concentration (the opposite
of sectoral diversification); shock is the terms of trade shock, measured as the standard
deviation of terms of trade within the sample period; findev is the share of private credit
in GDP, proxying the development of the financial sector; and corrup is a measure of
corruption, proxying the rent-seeking mechanism. Finally, X is a vector which includes
other controls and some fixed factors which can affect exchange rate regime choice
directly.31
As per the control variables, some common determinants of regime choice in the
literature include the level of development, country size, trade openness (openness),
capital account openness (kaopen), and current account balance. To minimize the
multicollinearity among the explanatory variables, we go back to the correlation matrix
(Table 3). Per capita income (lgdpc), representing the level of development is highly
correlated with diversification à la Imbs and Wacziarg (2003), but it is also highly
correlated with financial development (findev) and corruption (corrup). Therefore, we
do not include it among the base controls. Log GDP, a measure of country size, is
highly correlated with findev, but another measure, namely log population, does not;
and therefore, we use the log population as a proxy for country size. Trade openness is
an important determinant of regime choice, in that highly open economies may find
fixed regimes more attractive, because fluctuating exchange rates under flexible regimes
may be costly for highly open economies, where trade and investment are crucial
(Corden 2002). However, it is closely correlated with the log population in our dataset,
and therefore we do not include it among the base controls. Countries with highly open
31 The list of possible determinants for regime choice is quite long. For example, Juhn and Mauro (2002) identify as many as 30 different determinants from empirical studies.
72
capital account systems should choose either hard peg or fully flexible regime,
according to the ‘Bi-polar view’ of exchange rate regime choice (see Eichengreen
1994). Meanwhile, the current account balance acts as a proxy for the currency crisis
factor (see Von Hagen and Zhou 2007); a high and persistent current account deficit
may not be conducive to fixed regimes. Based on these correlations, we select only log
population, capital account openness, and current account balance as our base controls.
Variables which are commonly used in the macro-comparative development literature
to represent country-fixed factors, and which can be associated with omitted variables,
are also included in our models. These include two geographic variables, namely,
landlockedness and latitude (see Ramcharan 2010). Landlockedness can act as a proxy
for natural barriers to openness (see Poirson 2001), while latitude controls for climate
and its impact on work conditions. Likewise, we also include dummies for countries’
colonial past, to capture historical institutional factors and other ties that countries may
have with their former rulers. A dummy for major oil exporting countries is also
incorporated throughout the models, as its omission may cause endogeneity, due to its
possible simultaneous effect on both regime choice and diversification.
As was discussed in Section 3.2, the sign of 1, reflecting the direct effect of
concentration, should be negative i.e., a lower concentration (higher diversification)
should lead to flexible regimes. Given the theoretical discussion, the anticipated sign for
the direct effect of financial development is such that it should lead to floating regimes.
The predicted signs of the direct effects of real shock and corruption are ambiguous.
The signs of the coefficients of the interaction terms between concentration and real
shock, and between concentration and financial development, are negative and
73
ambiguous, respectively. On the other hand, the sign of the coefficient of the interaction
term between concentration and corruption is likely to be negative.
In a model with interaction term(s), the main focus should be on the marginal effect of
the focal explanatory variable on the dependent variable; singular attention to the
magnitude and significance of the individual coefficients may lead to the omission of
important information. For instance, it is possible that the marginal effect of
concentration on regime choice is significant only at some values of the shock, in which
case, insignificance of the point estimate of 2 in equation (1) would not necessarily
mean that the effect is insignificant for all values of the shock: it just means that the
average direct effect is insignificant. Therefore, for a careful examination, the focus
should be on the joint impact of the relevant estimates, e.g., the joint significance of 1
and 3 in the case above (see Wooldridge 2002; Hadad, Lim and Saborowsky 2010).
Consequently, we are interested in the marginal effect of concentration on regime
choice, which is given by the following derivative term:
= + + + … … . (2) In order to find the values of shocks at which concentration may have a significant
impact on regime choice, this derivative is evaluated at the mean of findev and corrup.
The corresponding standard errors for marginal effects can be calculated using the
‘delta method’. The results can also be presented by a confidence interval (CI) graph,
illustrating the range of shocks over which concentration has an impact on regime
choice. The marginal effect of concentration with respect to findev and corrup is
investigated in a similar vein.
74
3.4.2 Estimation issues
In terms of panel treatment, a fixed effects model delivers completely insignificant
results, which is not surprising, given that diversification is likely to change slowly over
time (at least for most countries), and/or its relationship with regime choice is
dependent on the between-country variation. Thus, we choose the cross-sectional
dataset as our baseline data. Nevertheless, we also adopt a pooled panel analysis for a
robustness check, controlling for common time effects, a common time trend, and
clustering of the standard errors at the country level. Our results are qualitatively similar
across cross-sectional and (pooled) panel datasets.
Another estimation issue is related to the type of the dependent variable. Countries
choose a set of exchange rate arrangements from a continuum of policy choices to make
up their exchange rate regimes. The de facto classification of Reinhart-Rogoff
categorizes this spectrum into 14 groups. If two countries differ only slightly in their
exchange rate arrangements, they are likely to be placed in the same group. For
countries to be allocated into different groups, they must have made sufficiently
different policy choices. The question that arises here is whether these 14 groups should
be treated as ordinal variables, or as continuous variables rounded to integers. Even if
one views these categories as ordinal, it seems that having 14 different regimes makes
the underlying categorization almost indistinguishable from a continuous measure.32
32 The Monte Carlo simulations of Johnson and Creech (1983) demonstrate that grouping a continuous measure into a finite number of categories is unlikely to create a measurement error problem if the number of categories is greater than five. Ferrer-i-Carbonell and Frijters (2004) find that cardinality vs. ordinality assumptions regarding happiness scores make little difference to the results.
75
Averaging these integers over 30 years also makes them practically continuous. Thus,
without much loss of generality, we treat the dependent variable as continuous.33
Equation (1) is initially estimated via OLS. However, endogeneity may result because
of reverse causation, in that fixed exchange rate regimes may keep economies
concentrated in a few sectors, due to the ‘protection’ they provide. Thus, we address
this problem through instrumental variables and identification-through-
heteroskedasticity methods.
3.4.2.1 Instrumental variables estimation
Ideally, an instrumental variable should approximate a randomized experiment, but it is
difficult to find such an experiment in the case of diversification. Following the
common practice, we therefore resort to variables that satisfy the desirable properties
for an instrument conditional on some covariates. Countries’ neighbours are likely to
have these properties, predicated on the identifying assumption that the diversification
level of a typical country is positively related to those of its neighbours. Bordering
countries may share joint histories, similar geography and initial conditions, and parallel
development trajectories. Our aim is to formulate a specification whereby a country’s
neighbours constitute a ‘reference group’, the development trajectory of which has not
33 Aghion et al. (2009) also consider the RR classification as continuous, and use it as explanatory variable by taking its five-year average. Further, several other measures in the literature which are constructed in a similar way are treated as continuous. For instance, Polity IV indicators or Freedom House political rights and civil liberties measures are often used as both the dependent variable, as well as the lagged dependent variable or explanatory variables on the right-hand side.
76
been affected by the country’s own exchange rate regime.34 Thus, we take advantage of
neighbours’ size-(GDP)weighted diversification as an instrument.
Our identification strategy relies on the following two assumptions. First, for most
countries, their neighbours are likely to constitute a control group which possesses a
parallel development trajectory. Second, conditional on covariates, neighbours’
trajectories are not affected by the country’s exchange rate regime. If these two
assumptions are true, then neighbours could provide the diversification path that a
country would have followed if its own diversification structure had not been affected
by its exchange rate regime.35 For this identification strategy to work, we need to satisfy
two conditions. First, any factors that might prevent the neighbours from being an
exogenous counterfactual should be controlled for. Second, conditional on the
covariates, neighbours should affect a country only through diversification, and the
factors that shaped the neighbours’ diversification structures in the past should not
affect a country’s exchange rate regime choice today. These conditions are commonly
known as the exclusion restriction and exogeneity properties, respectively, for an
instrument. With respect to the first condition above, we already control for possible
observable factors such as landlockedness, colonial past, and country size. In addition to
these, we also control for the neighbours’ shares in the overall trade, and include
dummy indicators as to whether the countries are the members of regional trade
agreements, including the European Union (EU), Commonwealth of Independent States
34 This measure is like WD, where W is a spatial weighting matrix (obtained using a contiguity matrix, which takes a value of 1 if a country i borders country j and 0 otherwise, which is weighted by the neighbours’ sizes, i.e., GDP) and D is the neighbours’ diversification levels. Using neighbours’ surface areas or populations as measures of size provides similar results.35 This trajectory does not have to be the same trajectory as the country’s; it suffices that it be parallel, or have a reasonably similar slope.
77
(CIS), North American Free Trade Agreement (NAFTA), Association of South East
Asian Nations (ASEAN), and Gulf Cooperation Council (GCC). Controlling for these
variables aims to ensure that the development paths of the neighbours are not distorted
by the country in question due to interactions that might take place among the
neighbours in the sample period. To check for the second condition above, we include
the neighbours’ GDP-weighted exchange rate regime in the first- and second-stage IV
regressions, and find that this variable is always estimated to be insignificant. This
finding is not entirely surprising, given that diversification has a broader scope than a
macroeconomic policy that may change more rapidly.36 Overall, our chosen instrument
seems to be doing a good job of identifying the effect of diversification; however, we
will check its robustness below using a different methodology.37
First-Stage and Reduced Form Results: Table 4 presents a battery of test results
relating to the reliability of neighbours’ GDP-weighted concentration. For cross-
sectional data, the first-stage results show that a country’s own concentration is robustly
36 Motivated by both the trade and transport engineering literature, as well as by the new economic geography which stresses the influence of transport costs on production patterns, Ramcharan (2010) uses countries’ topographical characteristics as the instruments of product diversification. The idea is that topographical characteristics, such as terrain variability and soil drainage, which affect the transport cost of production across space, can affect product diversification. Ramcharan uses two topography-related variables, namely the distribution of land area by elevation, and bioclimatic classes, which are available for about 70 countries. Another topography-related variable is the terrain ruggedness index (‘ruggedness’), constructed by Nunn and Puga (2012), which quantifies topographic heterogeneity. Data for this later variable are available for a larger number of countries. However, none of these topography-related variables make strong instruments according to the Stock and Yogo (2005) rule, and therefore they cannot be used in our estimation. 37 Of the three moderator variables in equation (1), real shock can be considered as exogenous, but the exchange rate literature suggests that financial sector development and corruption may be endogenous. We instrument these variables with their initial values in the cross-sectional estimation. The interactions between these variables and product diversification are also endogenous and are instrumented with their respective initial values, interacted with the neighbours’ size-weighted diversification. We use the initial values of two baseline controls, such as capital account openness and current account balance, in the cross-sectional analysis. In five-year averaged panel, we adopt the pre-determined values of the explanatory variables (except for shock, in which case we use the contemporaneous value).
78
and positively related to its neighbours’. In column 1, the instrument is significant at the
1% level when heteroskedasticity is not corrected. The F-statistic of the excluded
instrument for this estimation is 17.25. The F-statistic becomes somewhat smaller (i.e.,
12.70) when base controls and fixed factors are included in the regressions (column 2).
With heteroskedasticity-robust standard errors, the estimate remains significant at the
5% level (column 3), with the F-statistic being 6.37. When three baseline controls and
fixed factors are included in the first stage, the F-statistic on the excluded instrument is
5.85 (column 4). However, the results confirm that the explanatory power of our
instrument is not reduced very much in the second stage when controls are included.
The scatter plots in Figures 3 and 4 (both non-parametric and parametric) also confirm
the positive relationship between concentration and our IV. The Shea partial R2 value of
0.13 in the first stage regressions is not very high, suggesting that the instrument may be
somewhat weak. However, the instrument is not under-identified, as can be seen from
the under-identification tests in the last row of Table 4. The reduced form specifications
which examine the link between regime choice and the instruments directly show that
our instrument is significant both with and without controls (columns 5 and 6). As we
get mixed messages about the strength of our instrument based on the first-stage F-
statistics and the Shea partial R2, we estimate our models via Limited Information
Maximum Likelihood (LIML), which is robust to the weak instruments problem.38
38 In the full model, the F-statistics of excluded instruments for six endogenous variables, namely concentration, financial development, corruption, and interactions between concentration and shock, and financial development, and corruption, are 7.88, 29.54, 56.94, 5.02, 7.91, and 13.28, respectively.
79
3.4.2.2 Identification through heteroskedasticity
Although we have argued that our external instrument (neighbours’ size-weighted
diversification) is plausible, it would be useful to cross-check the identification obtained
this way. For instance, one could argue that unobservable omitted variables, such as
historical trade links, entrepreneurship, and various other determinants of economic and
political equilibria, may also play a role in the determination of the right-hand-side
variable and exchange rate regime choice. We employ an alternative identification
which was proposed recently by Lewbel (2012), and which does not rely on exclusion
restrictions but exploits heteroskedastic disturbances for identification. In this approach,
identification is obtained through a set of exogenous variable(s) (Z), which can be a
subset of exogenous regressors (X) in the regression equation (Z can also include
traditional instruments), whereby, in the first stage, endogenous variables are regressed
on the Z vector, followed by the retrieval of the residuals . Using these residuals, (Z –
)* , where is the mean of Z, is constructed, and can then be used as instruments in
the second stage. For this method to work, the residuals must be heteroskedastic. The
second-stage regression can be estimated through the usual IV methods.39
Equation (1) includes six endogenous variables, i.e., concentration, financial
development, corruption, and their three interactions. In the first stage of the method,
we regress the endogenous variables on the log population, initial share of the current
account balance in GDP, initial capital account openness, and latitude, which are
exogenous in equation (1). For our main endogenous variable, i.e., concentration, the Z
vector also includes neighbours’ sizes, weighted concentration as the standard
39 For other studies which use the Lewbel method, see Emran and Hou (2011) and Sabia (2007).
80
instrument, but our results are robust to the complete exclusion of this variable from the
Lewbel approach. The Breusch-Pagan test rejects the null hypothesis of
homoskedasticity for all endogenous variables at the 1% level of significance. The
second stage instruments are obtained using the construction described above.
Once the instruments have been generated a la Lewbel (2012), we estimate our models
by GMM (which is more efficient than 2SLS). As we obtain more than one instrument
for each endogenous variable, our model is over-identified. The major advantage of the
IVs obtained via this method is that they are free from the exclusion restrictions
concerns, and thus, over-identification tests, such as Hansen’s J (or Sargan) test,
produce statistically reliable results. The point to emphasize here is that our results in
relation to the marginal effects of concentration on exchange rate regime choice are
stable across different identification methodologies.
3.5 Estimation results
3.5.1 OLS, instrumental variables, and identification through heteroskedasticity
We now discuss the estimation results for equation (1) using OLS, LIML, and the
Lewbel method. We first consider the direct effect of diversification on regime choice,
as well as each channel in isolation. In LIML regressions, we include the neighbours’
trade share in the overall trade and dummies for regional trade agreement membership
in the second stage regressions, along with base controls and fixed factors. Including
neighbours’ share in overall trade reduces the sample size substantially. We check
thoroughly as to whether this variable affects the inference regarding the main
explanatory variables, and find that it does not drive the results. Thus, we exclude this
variable from all second stage regressions presented.
81
Table 5 presents the results for equation (1), examining only the direct effect, in the
absence of the interaction terms. The table reports the findings with the OLS, LIML and
the Lewbel estimator. The OLS regression, which also includes the initial values of base
controls and fixed factors, indicates a highly significant negative relationship between
regime choice and concentration. The corresponding IV estimates using LIML, reported
in columns 3 and 4, also support this finding. Table 5 also reports the estimation results
obtained using the Lewbel technique, estimated with GMM (column 5 and 6). These
GMM estimates are more efficient than the OLS estimates, as is evidenced by the lower
standard errors of the former estimates. The over-identification tests affirm the
statistical reliability of our standard IV, given that instruments generated by the Lewbel
method provide a sound basis for these tests.
The negative coefficient of concentration implies that a higher level of sectoral
diversification (concentration) leads to flexible (fixed) regimes. This finding shows that
product diversification alters countries’ ‘fear of floating’ behaviour, which was first
observed by Calvo and Reinhart (2002). In other words, diversified economies are less
fearful of floating.
Table 6 displays the results for each interaction effect in isolation. The first three
columns show the results using OLS, LIML and the Lewbel method for the real shock
absorption channel, including base controls and fixed factors. It can be seen that all
coefficients have the same signs across different estimation methods. The signs of both
the concentration coefficient and the interaction term are negative, implying that the
marginal effect of concentration (diversification) on regime, conditional on the impact
of the shock, is negative, i.e., leading to fixed (flexible) regimes. The marginal effect
82
will be more negative as the impact of real shocks increases. The coefficients of the
concentration index and the shock are significant in the OLS regressions in column 1
and the Lewbel regressions in column 3, but the interaction term is not. The instruments
used for Lewbel method estimation are also valid, as can be seen from the over-
identification test results.
The results for the financial development mechanism in isolation are presented in
columns 4 to 6 in Table 6. The direct effect of concentration is always significant. The
coefficient of the interaction is significant only with the Lewbel regression in column 6.
The coefficients of concentration and financial development are also significant in the
Lewbel estimates, and over-identifying restrictions are valid for the Lewbel method
regression as well. The joint significance of the coefficient of concentration and the
interaction is also robust across different models and different estimation techniques,
suggesting that the financial development has a strong influence on concentration and
regime choice.
The last three columns of Table 6 present the results on the rent-seeking or corruption
channel. Both the interaction and the direct effect are robustly significant across
different estimation techniques, yielding strongly significant coefficients at the 1%
level. The over-identification restrictions are also valid. The coefficients of
concentration and the interaction term are jointly significant in all specifications. The
effect of concentration conditional on the corruption level is negative, suggesting that
concentration leads to fixed regimes in more corrupt economies. The effect will be less
negative and will tend to zero at higher values of the corruption index; that is, the effect
becomes smaller as the corruption level decreases.
83
The full model includes concentration, all three of the moderator variables, and the three
interaction terms, running a ‘horse race’ among the channels identified. We have
estimated OLS and LIML regressions, as well as the Lewbel method using GMM, and
report two regressions for each method: one including the base controls, with the six
main variables mentioned above, and the other including both the base controls and the
fixed factors. The results are presented in Table 7 in columns 1 to 6. The point
estimates of the concentration and corruption channels are significant in OLS
regressions at the 5% level, but the shock and financial development channels are not
(column 1). The concentration and corruption channels remain significant at the 5%
level after controlling for the fixed factors (column 2). However, the coefficients of
concentration and all three mechanisms are jointly significant in both regressions. The
LIML regressions find no point estimate to be individually significant in column 3 and
4; however, the corresponding Lewbel method estimates in column 5 and 6 show that
coefficients of the concentration and corruption channels are significant. The direct
effect of shock is also significant in columns 5 and 6. These regressions pass over-
identification tests as well. The coefficients of concentration and the three interaction
terms are also jointly significant in both regressions in columns 5 and 6.
It is noticeable that the shock channel is never significant in any of the regressions in
the full model. It was not also significant when we estimated each channel in isolation
in Table 6. This could be ascribed to multicollinearity, or the shock channel’s effect
may be able to be explained by other factors which are included in the model. However,
it may also mean that a short-run phenomenon like shock is completely insignificant in
a cross-sectional dataset which captures relatively more long-run effects. Columns 1 to
6 are then re-estimated, omitting the insignificant real shock channel. This exercise also
84
helps us to gain about 20 additional data points. The results are presented in Table 7 in
columns 7 to 12. The corresponding OLS results improve somewhat, as can be seen
from columns 7 and 8, where the concentration becomes significant at higher levels.
The LIML results improve significantly, as the concentration coefficient is now
significant in columns 9 and 10 at the 5% and 10% levels, respectively. Moreover, the
corruption channel also becomes significant in column 10 at the 10% level. Another
positive aspect of the exclusion of the shock channel is that the coefficients of
concentration and two remaining channels are now jointly significant in both
regressions. The corresponding Lewbel estimates in columns 11 and 12 remain similar
to those in columns 5 and 6 with respect to the coefficient significance, over-
identification restrictions validity, and the joint significance of the concentration and
interaction coefficients.
We also re-estimated all of the OLS and LIML regressions with neighbours’-size
weighted concentration as IV by using the five-year averaged panel data and including
time dummies and a common time trend in each regression.40 The standard errors are
also clustered at the country level. We do not replicate the Lewbel method estimates,
because this method has not been proven to be suited to panel data. Table 8 reports the
results for the direct effect and the full model. Conclusions similar to those reached for
the cross-sectional dataset largely hold. That is, the direct effect of concentration on
regime choice is negative, leading to fixed regimes (column 1 and 2). For the full model
with three interactions (columns 3 and 4), the corruption channel is significant in both
the OLS and LIML regressions and the shock channel is significant in the LIML
40 The inclusion of current account balance to relative GDP reduces the sample size drastically in pooled panel estimation, so we use only capital account openness and log population as the base controls.
85
regressions, but the financial development channel is largely insignificant. In order to be
consistent with the cross-sectional regressions, we run the full model without the shock
channel (regressions 5 and 6). The corruption channel is significant only in regression 6
when estimated by LIML. The coefficients of the concentration and interaction terms
are also jointly significant for this regression (column 6).
3.5.2 The marginal effect of concentration on regime choice, conditional on the
moderator variables
The regression results presented in the section above only provide the point estimates.
They do not tell us much about the marginal effect of concentration, conditional on
different values of the moderator variables. In this section, we present the effects of
concentration on regime choice, conditional on the values of shocks, financial
development, and corruption, graphically, with the aid of confidence interval (CI)
graphs. We construct the CI figures based on the full model in Table 7 using the Lewbel
method, which includes all relevant controls. The CI figures based on corresponding IV
estimations largely tell the same story.
Figure 5 presents the CI graph which represents the marginal effect of concentration at
different levels of shocks. The figure is drawn using the estimates from column 6 in
Table 7. In the figure, the solid horizontal line is the zero line, and the solid downward
sloping line represents the different values of the marginal effect of concentration on
regime choice when the incidence of shocks varies. For the marginal effect of
concentration conditional on shock to be significant, the upper and lower bounds of the
CI (the dashed lines) should be simultaneously either above or below the zero line.
Figure 5 shows that the marginal effect of concentration on regime choice is negative
86
throughout the whole range of shocks, meaning that a higher concentration leads to
fixed regimes. However, the effect is significant only at the higher levels of shocks.
More specifically, the effect is significant when the impact of the real shock, as
measured by the standard deviation of terms of trade, is above 14. A list of the countries
which fall in this range is provided below the figure. With the exception of Japan and
Spain, these countries are mainly the emerging and less developed economies of the
world. It can also be seen that these countries are relatively concentrated: the mean
value of their concentration is 0.17, which is higher than the mean value of
concentration for other countries outside the range (0.13), and slightly above the sample
mean of 0.16. This may imply that the impact of shocks is likely to be amplified if
countries are concentrated, which will be reflected in their choice of a fixed exchange
rate regime.
Figure 6 shows the marginal effect of concentration on regime choice, conditional on
financial sector development, from the full model excluding shocks (column 12 in
Table 7). As the marginal effect in this regression depends on the level of financial
development and corruption, we draw the graph by fixing the value of corruption at its
mean, and let the financial development level vary. It can be seen from Figure 6 that the
negative marginal effect is significant for those countries where the level of financial
development, as measured by private sector credit as percentage of GDP, is 40% or
lower. The negative marginal effect implies that concentration will lead to fixed
regimes. A list of countries for which the effect is significant is provided after Figure 6.
Again, these are all developing countries, with the exception of Denmark and Iceland.
The marginal effect is positive when private sector credit is 73% of GDP, i.e., at a high
level of financial development. However, the positive marginal effect is not significant.
87
These results are consistent with Acemoglu and Zilibotti’s (1997) argument, given in
Section 2, that countries with weaker financial markets are unable to diversify, and
hence, tend to adopt fixed regimes.
The marginal effect of concentration on regime choice, conditional on the corruption
level, and calculated using the estimates from column 12 in Table 7, is presented in
Figure 7. To draw the graph, we keep the value of financial development fixed at its
mean value, and let the corruption level vary. The graph shows that the marginal effect
is significant for those countries where the corruption level is relatively high (i.e., where
the values of corruption are relatively lower). The exact threshold value in Figure 7 is
3.7. Thus, the marginal effect of concentration is negative and significant, leading to a
fixed regime when the value of the corruption index is lower than 3.7 (higher values
indicate a lower level of corruption). The magnitudes of the negative marginal effect
become smaller and smaller as the corruption level decreases (higher values in the
graph). The implication of this finding is that, in countries with higher levels of
corruption, concentration, by leading to more fixed regimes, creates more rent-seeking
opportunities, presumably for a few elites. The list of 73 the countries for which the
marginal effect is significant is provided below Figure 7. With one exception (i.e.,
Italy), all these countries are developing countries, including low, medium and high
income oil exporting countries, such as Saudi Arabia, Kuwait, and Bahrain.
Figure 7 also shows that the marginal effect of concentration becomes positive, and thus
leads to a flexible regime, for very low corruption countries, i.e., countries with a
corruption index higher than about 5.7. In our sample, this means a few developed
countries such as Canada, Denmark, Finland, Iceland, Luxembourg, the Netherlands,
88
New Zealand, Norway, Sweden, and Switzerland. However, the marginal effect of
concentration is not significant for these countries.
Overall, the CI figures reveal that the marginal effect of concentration on regime choice
is significant for those developing countries which experience larger shocks, and have a
low level of financial development and a relatively high corruption level. In all cases,
the marginal effect of concentration is negative, leading to fixed regimes; or,
alternatively, higher diversification leads to flexible regimes.
3.6 Conclusion
This study examines the effect of product diversification on exchange rate regime
choice using a cross-sectional dataset covering 125 countries over the period 1971–
2000. It provides very useful evidence as to the ways in which changing production
patterns in the real sector can shape one of the central macroeconomic policies of a
country. The paper identifies three mechanisms through which diversification and
regime choice may be related, namely the real shock absorption, financial sector
development, and rent-seeking channels. Along with a hypothesised direct effect, the
paper then runs a ‘horse race’ among the identified mechanisms. A dual identification
strategy involving instrumental variables estimation, with the neighbours’ GDP-
weighted diversification as the instrument, together with Lewbel’s (2012) ‘identification
through heteroskedasticity’, have been used to tackle the possible endogeneity. Our
results in relation to the marginal effect of product diversification are robust and stable
across both different identification methods, and cross-sectional and panel datasets.
It is found that product diversification has a direct and significant effect on regime
choice, leading countries towards flexible regimes. This finding is robust to the
89
inclusion of several controls in the models, and implies that a diversification of
production activities makes countries less likely to feel the fear of adopting a floating
regime which Calvo and Reinhart (2002) argued. The results also reveal that
diversification affects regime choice through a rent-seeking channel, which is proxied
by the level of corruption in the country. Robustly significant across different
specifications, this finding suggests that, in countries with higher levels of corruption,
sectoral concentration leads to fixed regimes, as this type of regime may provide more
scope for rent-seeking for the powerful elite. There is also some evidence for the shock
absorption channel across various model specifications, suggesting that the real shock
channel moderates the diversification opportunity for developing countries which
experience higher incidences of real shocks. The financial development channel is
found to be significant as well, especially for developing countries with low levels of
financial development. In general, it can be concluded that the three channels identified
in this paper do affect diversification, and thus exchange rate regime choice is more
pronounced for developing countries.
Finally, our study contributes to a deeper understanding of the concept of diversification
and how it interacts with other national indicators in shaping a key macroeconomic
policy choice. In view of recent surge in interest for diversification and its links with
economic development, structural change, volatility, and productivity growth, we
believe that the mechanisms put forward in this paper can shed light on the appropriate
policy options available for countries that are at different stages of productive capacity
and institutional development.
90
References
Acemoglu, D and Zilibotti, F 1997, ‘Was Prometheus unbound by chance? risk,
diversification, and growth’, Journal of Political Economy, vol. 105, no. 4, pp. 709-751.
Ades, A and Di Tella, R 1999, ‘Rents, competition, and corruption’, American
Economic Review, vol. 89, no. 4, pp. 982-993.
Aghion, P, Bacchetta, P, Ranciere, R and Rogoff, K 2009, ‘Exchange rate volatility and
productivity growth: the role of financial development’, Journal of Monetary
Economics, vol. 56, no. 4, pp. 494–513.
Alesina, A and Wagner, A 2006, ‘Choosing (and reneging on) exchange rate regimes’,
Journal of the European Economic Association, vol. 4, no. 4, pp. 770-799.
Broda, C 2001, ‘Coping with terms-of-trade shocks: pegs versus floats’, American
Economic Review, vol. 91, no. 2 (Papers and Proceedings of the Hundred Thirteenth
Annual Meeting of the American Economic Association), pp. 376-380.
Broda, C 2004, ‘Terms of trade and exchange rate regimes in developing countries’,
Journal of International Economics, vol. 63, pp. 31–58.
Burns, A 1960, ‘Progress towards economic stability’, American Economic Review, vol.
50, no. 1, pp. 1-19.
Calvo, G and Reinhart, C 2002, ‘Fear of floating’, Quarterly Journal of Economics, vol.
117, no. 2, pp. 379–408.
Cameron, D 1978, ‘The expansion of the public economy: a comparative analysis’,
American Political Science Review, vol. 72, no. 4, pp. 1243-1261.
Carmignani, F, Colombo, E and Tirelli, P 2008, ‘Exploring different views of exchange
rate regime choice’, Journal of International Money and Finance, vol. 27, pp. 1177-
1197.
Chinn, M and Ito, H 2008, ‘A new measure of financial openness’, Journal of
Comparative Policy Analysis, vol. 10, no. 3, pp. 309-322.
91
Corden, W 2002, Too sensational: on the choice of exchange rate regimes, The MIT
Press, Cambridge, Massachusetts.
De Waldemar, S 2010, ‘How costly is rent-seeking to diversification – an empirical
approach’, paper presented in CSAE Conference 2010, 21st – 23rd March, 2010, St.
Catherine’s College, Oxford.
Eichengreen, B 1994, International monetary arrangements for the 21st century, The
Brookings Institution, Washington.
Emran, S.and Hou, Z 2011, ‘Access to markets and household consumption: evidence
from rural China’, Review of Economics and Statistics (forthcoming).
Ferrer-i-Carbonell, A and Frijters, P 2004, ‘How important is methodology for the
estimates of the determinants of happiness?’, Economic Journal, vol. 114, pp. 641-659.
Frankel, J and Rose, A 2002, ‘An estimation of the effect of common currencies on
trade and income’, Quarterly Journal of Economics, vol. 117, no. 2, pp. 437-466.
Friedman, M 1953, ‘The Case for flexible exchange rates’, Essays in Positive
Economics, The University of Chicago Press, Chicago.
Grossman, G and Helpman, E 1994, ‘Protection for sales’, American Economic Review,
vol. 84, no. 4, pp. 833-50.
Hadad, M, Lim, J and Saborowsky, C 2010, ‘Trade openness reduces growth volatility
when countries are well diversified’, Policy Research Working Paper No. 5222, The
World Bank, Washington.
Hoffmann, M 2007, ‘Fixed versus flexible exchange rate – evidence from developing
countries’, Economica, vol. 74, no. 295, pp. 425-449.
Holden, P, Holden, M and Suss, E 1979, ‘The determinants of exchange rate flexibility:
an empirical investigation’, Review of Economics and Statistics, vol. 61, no. 3, pp. 327-
333.
92
Ilzetzki, E, Reinhart, C and Rogoff, K 2008, ‘Exchange rate arrangements in the 21st
century: which anchor will hold?’, viewed 20 August 2009,
http://terpconnect.umd.edu/~creinhar/Papers.html.
Imbs, J and Wacziarg, R 2003, ‘Stages of diversification’, American Economic Review,
vol. 93, no. 1, pp. 63-86.
Johnson, D and Creech, J 1983, ‘Ordinal measures in multiple indicator models: a
simulation study of categorization error’, American Sociological Review, vol. 48, pp.
398-407.
Juhn, G and Mauro, P 2002, ‘Long run determinants of exchange rate regimes: a simple
sensitivity analysis’, IMF Working Paper no. 2/104, International Monetary Fund,
Washington.
Kalemli-Ozcan, S, Sorensen, B and Yosha, O 2003, ‘Risk sharing and industrial
specialization: regional and international evidence’, American Economic Review, vol.
93, no. 3, pp. 903-918.
Kenen, Peter B 1969, ‘The theory of optimum currency areas – an eclectic view’, in
Monetary Problems of the International Economy by Mundell, Robert A. and
Alexander K. Swoboda (eds.), The University of Chicago Press, Chicago.
Koren, M and Tenreyro, S 2007, ‘Volatility and development’, Quarterly Journal of
Economics, vol. 122, no. 1, pp. 243-287.
Lahiri, A and Végh, C 2001, ‘Living with the fear of floating: an optimal policy
perspective’, in Preventing Currency Crises in Emerging Markets, S. Edwards and J.
Frankel (eds.), Chicago: University of Chicago Press for the National Bureau of
Economic Research.
La Porta, R, Lopez-de- Silanes, F, Shleifer, A. and Vishny, R. 1999, ‘The quality of
government’, Journal of Law, Economics and Organization, vol. 15, pp. 222–279.
Levy-Yeyati, E and Sturzenegger, F 2007, ‘Fear of appreciation’, Policy Research
Working Paper No 4387, The World Bank, Latin America and Caribbean Region.
93
Lewbel, A 2012, ‘Using heteroskedasticity to identify and estimate mismeasured and
endogenous regressor models’, Journal of Business and Economic Statistics, vol. 30,
pp. 67-80.
Lin, S and Ye, H 2011, ‘The role of financial development in exchange rate regime
choice’, Journal of International Money and Finance, vol. 30, no. 4, pp. 641-659.
Michaely, M 1958, ‘Concentration of exports and imports: an international
comparison’, Economic Journal, vol. 68, pp. 722-736.
Nunn, N and Puga, D. 2012, ‘Ruggedness: the blessings of bad geography in Africa’,
Review of Economics and Statistics, vol. 94, no. 1, pp. 20-36.
Poirson, H 2001, ‘How do countries choose their exchange rate regime?’, IMF Working
Paper no. 1/46, International Monetary Fund, Washington.
Ramcharan, R 2005, ‘how big are the benefits from diversification? evidence from
earthquakes’, IMF Working Paper no 5/48, International Monetary Fund, Washington.
Ramcharan, R 2007, ‘Does the exchange rate regime matter for real shocks? evidence
from windstorms and earthquakes’, Journal of International Economics, vol. 73, pp.
31–47.
Ramcharan, R 2010, ‘The Link between the economic structure and financial
development’, B.E. Journal of Macroeconomics, vol. 10, no. 1, pp. 1-35.
Reinhart, C 2000, ‘The mirage of floating exchange rates’, American Economic Review,
vol. 90, no. 2, pp. 65-70.
Reinhart, C and Rogoff, K 2004, ‘The modern history of exchange rate arrangements: a
reinterpretation’, Quarterly Journal of Economics, vol. 119, no. 1, pp. 1-48.
Saint-Paul, G 1992, ‘Technological choice, financial market and economic
development’, European Economic Review, vol. 34, pp.763-781.
Sabia, J 2007, ‘The effect of body weight on adolescent academic performance,’
Southern Economic Journal, vol. 73, pp. 871-900.
94
Savvides, A 1990, ‘Real exchange rate variability and the choice of exchange rate
regime by developing countries’, Journal of International Money and Finance, vol. 9,
pp. 440-454.
Stock, J and Yogo, M 2005, ‘Testing for weak instruments in linear IV regression’, in
Identification and Inference for Econometric Models: Essays in Honor of Thomas
Rothenberg, D.W.K. Andrews and J.H. Stock (eds.), Cambridge: Cambridge University
Press.
UNIDO 2003, Industrial Statistics Database, United Nation’s Industrial Development
Organisation.
Von Hagen, J and Zhou, J 2007, ‘The choice of exchange rate regimes in developing
countries: a multinomial panel analysis’, Journal of International Money and Finance,
vol. 26, pp. 1071-1094.
Wooldridge, J 2002, Econometric analysis of cross section and panel data, MIT Press,
Cambridge, MA.
95
Tables and Figures
Table 1: Variables and data sourcesVariables Description Data SourceRegime Exchange rate regime choice, Reinhart-Rogoff fine
classification (1 to 14 scale), higher values indicate more flexible regimes.
Ilzetzki, Reinhart and Rogoff.(2008)
concen Herfindahl index, a measure of sectoral concentration, 0 to 1scale, higher values indicate higher concentration (lower diversification).
UNIDO (2003)
concen_neigh Neighbours’ GDP-weighted concentration index. Instrument for concentration index.
Calculated by theauthors
findev Private sector credit to GDP, a proxy for financial sector development.
WB Financial Structure Database
shock The standard deviation of terms of trade, a proxy for real shock.
WDI
corrup Corruption, proxy for rent-seeking, 0 to 6 scale, higher value indicates lower corruption.
ICRG, the PRS Group
kaopen Capital account openness index (0 to 1), higher valuesindicate higher levels of capital account openness.
Chinn and Ito (2008)
open Trade openness, the share of export and import of goods and services in GDP.
Penn World Tables 6.3
lgdp Log of GDP in current dollar, measuring size of the economy.
WDI
lgdpc Log of per capita GDP in USD. Penn World Tables
lpop Log population. WDIcabal Current account balance as percentage of GDP, a proxy for
currency crisis factor. WDI
landlocked A landlocked country dummy. CIA World FactBook
Colonial past Dummy for former colonial past. La Porta et al.(1999)
rugged Terrain ruggedness. Nunn and Puga (2012)
lat The absolute value of the latitude of each country,normalised to 0–1.
Ramcharan(2010)
trade_neigh Neighbours’ share in overall trade, calculated by the authors using data from the Direction of Trade Statistics (DOTS).
DOTS (IMF)
96
Table 2: Descriptive statistics of main variables (cross-sectional data)
Variable Obs. Mean Std. Dev.
Min Max
Exchange Rate Regime Choice (Regime) 163 7.10 3.71 1 14Concentration Index (concen) 155 0.16 0.11 0.06 0.75Real Shock (shock) 106 19.98 17.43 1.31 85.17Financial Development (findev) 155 0.33 0.28 0.01 1.46Corruption (corrup) 137 3.29 1.22 0.28 6
Table 3: Correlation matrix (cross-sectional data)
1 2 3 4 5 6 7 8 9 10 11
1. concen 1.002. shock 0.17 1.003. findev –0.37 –0.41 1.004. corrup –0.32 –0.47 0.55 1.005. kaopen –0.28 –0.22 0.42 0.46 1.006. open 0.09 –0.05 0.14 –0.02 0.16 1.007. lgdp –0.52 –0.31 0.59 0.50 0.44 –0.33 1.008. lpop –0.25 0.01 0.07 –0.13 –0.04 –0.51 0.64 1.009. lgdpc –0.45 –0.37 0.67 0.72 0.57 0.08 0.66 –0.13 1.0010. cabal –0.29 –0.25 0.49 0.25 0.41 0.11 0.52 0.11 0.58 –0.18 1.00
Note: See Table 1 above for a description of the variables.
97
Table 4: First stage (FS) and reduced form (RF) results: neighbours’ GDP-weighted concentration as IV (cross-sectional data)
First Stage Reduced form(1) (2) (3) (4) (5) (6)
VARIABLES Non-robust Non-robust Robust RobustConcen_neigh 0.327*** 0.260*** 0.327** 0.260* –9.631*** –7.637*
(0.079) (0.076) (0.141) (0.143) (3.541) (4.289)Observations 125 125 117 117 125 117R2 0.12 0.12 0.34 0.34 0.04 0.184Base controls No Yes No Yes No YesFixed Factors No Yes No Yes No YesF-statistic of excluded instrument
17.25 12.70 6.37 5.85 - -
Shea partial R2 0.13 0.11 0.13 0.11 - -Under-ID test (p-val.) 15.38
(0.00)11.63(0.00)
4.75(0.02)
3.83(0.04)
- -
Note: Robust standard errors are given in parentheses for regressions 3, 4, 5 and 6. ***, **, and * indicate p < 0.01, p < 0.05, and p < 0.1, respectively. The dependent variable for the FS regressions is the Herfindahl concentration index (concen), and that for RF is regime choice. Concen_neigh is neighbours’GDP weighted concentration index (IV). Base controls are initial values of capital account openness (kaopen), current account balance as a percentage of GDP (cabal), and log population (lpop). Fixed factors include a landlocked country dummy, latitude, dummies pertaining to a colonial past, and a dummy for major oil exporting countries. The under-ID test with robust standard errors is Kleibergen-Paap’s (2006) rk statistic, the under-ID test without robust standard errors is Anderson’s (1951) canonical correlations test.
Table 5: The simple relationship between concentration and regime choice (cross-sectional data)
(1) (2) (3) (4) (5) (6)VARIABLES OLS OLS LIML LIML Lewbel GMM Lewbel GMMconcen –19.64*** –19.488*** –29.31*** –32.115*** –26.385*** –20.722**
(5.319) (5.423) (6.456) (9.594) (5.995) (8.604)Observation 122 121 117 116 116 116R2 0.246 0.391 - - - -Base controls Yes Yes Yes Yes Yes YesFixed factors No Yes No Yes No YesOver ID Test: p-val. - - - 0.40 0.36
Note: The dependent variable is regime choice. Each model has a constant. Robust standard errors are given in parentheses; and ***, **, and * indicate p < 0.01, p < 0.05, and p < 0.1, respectively. The baseline controls include the initial values of capital account openness, the ratio of current account balance to GDP,and log population. The fixed factors include a landlocked country dummy, latitude, dummies pertaining to a colonial past, and a dummy for major oil exporting countries.
98
Tab
le 6
:The
inte
ract
ion
effe
ctso
f sho
ck, f
inan
cial
dev
elop
men
t and
ren
t-se
ekin
g on
reg
ime
choi
ce(c
ross
-sec
tiona
l dat
a)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
VA
RIA
BLE
SO
LSLI
ML
Lew
bel G
MM
OLS
LIM
LLe
wbe
l GM
MO
LSLI
ML
Lew
bel G
MM
conc
en–1
7.08
1**
–14.
71–2
0.33
7***
–21.
542*
**–2
2.72
8–3
8.06
8***
–53.
018*
**–8
3.01
6**
–67.
740*
**(7
.263
)(1
0.54
)(5
.676
)(7
.840
)(1
4.07
7)(9
.076
)(1
9.66
7)(3
5.56
8)(1
3.39
7)sh
ock
0.06
7*0.
042
0.04
8*(0
.037
)(0
.068
)(0
.029
)co
ncen
*sho
ck–0
.410
–0.2
75–0
.255
(0.2
48)
(0.4
3)(0
.186
)fin
dev
–2.5
580.
770
–7.0
93*
(3.7
66)
(4.3
18)
(3.9
36)
conc
en*f
inde
v3.
941
–16.
590
69.5
78*
(37.
701)
(44.
635)
(36.
688)
corr
up–1
.155
*–2
.107
–1.6
62**
*(0
.587
)(1
.399
)(0
.451
)co
ncen
*cor
rup
11.2
31**
16.9
97*
14.9
01**
*–1
.155
*(9
.997
)(3
.237
)O
bser
vatio
ns89
8888
116
109
109
106
103
103
R20.
388
--
0.25
8-
-0.
221
--
Bas
e co
ntro
lsY
esY
esY
esY
esY
esY
esY
esY
esY
esFi
xed
fact
ors
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Ove
r ID
Tes
t: p-
val.
--
0.29
--
0.67
--
0.28
Join
t Sig
. (F
-sta
t, p-
val.)
11.6
5(0
.00)
7.91
(0.0
2)26
.67
(0.0
0)6.
54(0
.00)
3.83
(0.0
3)12
.24
(0.0
0)4.
28(0
.02)
6.39
(0.0
0)11
.46
(0.0
0)
The
depe
nden
t var
iabl
e is
regi
me
choi
ce.E
ach
mod
elha
s a c
onst
ant.
Rob
ust s
tand
ard
erro
rs a
re g
iven
in p
aren
thes
es.*
**,*
*,an
d*
indi
cate
p<
0.01
, p<
0.05
, and
p<
0.1,
resp
ectiv
ely.
The
base
line
cont
rols
incl
ude
the
initi
al v
alue
sof c
apita
l acc
ount
ope
nnes
s,th
e ra
tio o
fcur
rent
acc
ount
bal
ance
to G
DP
and
log
popu
latio
n.Th
e fix
ed fa
ctor
s inc
lude
a la
ndlo
cked
cou
ntry
dum
my,
latit
ude,
dum
mie
s per
tain
ing
to a
col
onia
l pas
t, an
d a
dum
my
for m
ajor
oil
expo
rting
co
untri
es. T
he jo
int s
ig. t
est i
s the
test
that
coe
ffic
ient
s of c
once
n an
d th
e in
tera
ctio
n te
rm a
re jo
intly
zer
o.
99
Tab
le 7
:The
inte
ract
ion
effe
cts o
f sho
ck, f
inan
cial
dev
elop
men
t and
ren
t-se
ekin
g on
reg
ime
choi
ce —
full
mod
el (
cros
s-se
ctio
nal d
ata)
Full
Mod
el w
ith th
ree
inte
ract
ions
Full
Mod
el w
ithou
t sho
ck(1
)(2
)(3
)(4
)(5
)(6
)(7
)(8
)(9
)(1
0)(1
1)(1
2)V
AR
IAB
LES
OLS
OLS
LIM
LLI
ML
Lew
bel
GM
MLe
wbe
l G
MM
OLS
OLS
LIM
LLI
ML
Lew
bel
GM
MLe
wbe
l G
MM
Con
cen
–50.
730*
*–5
4.61
5**
–43.
897
–17.
462
–48.
825*
*–5
7.03
0***
–50.
364*
**–4
8.10
4**
–85.
737*
–103
.424
**–5
9.91
7***
–69.
518*
**(2
1.54
1)(2
3.07
9)(5
2.18
7)(6
9.29
5)(2
0.21
5)(1
6.96
8)(1
8.19
2)(1
9.32
7)(4
7.02
9)(4
5.26
0)(1
4.05
0)(1
6.06
9)co
ncen
*sho
ck–0
.359
–0.2
29–0
.150
–0.3
38–0
.524
–0.3
87(0
.463
)(0
.468
)(0
.758
)(0
.723
)(0
.383
)(0
.259
)co
ncen
*fin
dev
–16.
238
–10.
019
–6.3
722.
628
–22.
617
10.0
10–3
0.79
2–3
9.60
813
.453
13.2
5525
.490
49.6
37(4
5.26
6)(4
3.54
8)(6
2.74
7)(8
1.71
2)(3
7.12
0)(2
9.23
7)(4
9.53
6)(5
0.50
9)(7
0.42
1)(3
6.83
9)(5
4.78
4)(4
6.48
2)co
ncen
*cor
rup
11.3
81**
12.3
48**
9.65
81.
313
13.2
52**
*13
.287
***
11.2
01**
11.7
64**
16.4
4121
.078
*10
.996
**10
.209
**(4
.837
)(5
.089
)(9
.931
)(1
1.85
6)(3
.979
)(3
.704
)(5
.445
)(5
.354
)(1
7.91
1)(1
0.86
2)(4
.438
)(4
.452
)fin
dev
1.58
51.
613
3.02
21.
803
2.82
90.
143
0.71
81.
859
–0.7
56–0
.408
–2.9
29–5
.874
(4.7
52)
(4.7
94)
(5.9
08)
(7.2
55)
(3.9
72)
(2.8
95)
(5.3
15)
(5.4
25)
(7.7
91)
(3.5
33)
(5.7
76)
(4.4
75)
corr
up–0
.924
–0.8
22–1
.019
0.38
7–0
.741
–0.7
90–0
.982
–1.0
02–2
.158
–2.4
87–0
.650
–0.4
60(0
.675
)(0
.753
)(1
.224
)(1
.670
)(0
.706
)(0
.650
)(0
.711
)(0
.700
)(2
.239
)13
.255
(0.7
57)
(0.6
93)
shoc
k0.
078
0.05
60.
050
0.06
60.
117*
*0.
084*
*(0
.062
)(0
.062
)(0
.097
)(0
.102
)(0
.060
)(0
.039
)O
bser
vatio
ns78
7877
7778
7810
210
197
9799
99R2
0.32
90.
401
--
--
0.24
20.
240
--
--
Bas
e co
ntro
lY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esFi
xed
fact
ors
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
No
Yes
Ove
r ID
Tes
t: p-
val.
--
--
0.30
0.16
--
--
0.76
0.33
Join
t Sig
.(F
-sta
t, p-
val.)
6.67
(0.0
0)4.
26(0
.00)
1.24
(0.3
0)1.
54(0
.20)
14.8
5(0
.00)
14.0
1(0
.00)
4.53
(0.0
1)3.
73(0
.01)
5.61
(0.0
0)11
.13
(0.0
0)7.
26(0
.00)
6.73
(0.0
0)Th
e de
pend
ent v
aria
ble
is re
gim
e ch
oice
.Eac
h m
odel
has a
con
stan
t.R
obus
t sta
ndar
d er
rors
are
giv
en in
par
enth
eses
.***
,**,
and
*in
dica
te p
<0.
01, p
<0.
05, a
nd p
<0.
1,re
spec
tivel
y.Th
e ba
selin
e co
ntro
ls in
clud
e th
e in
itial
val
ues o
f cap
ital a
ccou
nt o
penn
ess,
the
ratio
of c
urre
nt a
ccou
nt b
alan
ce to
GD
P an
d lo
g po
pula
tion.
The
fixed
fact
ors i
nclu
de a
land
lock
ed c
ount
ry d
umm
y, la
titud
e, d
umm
ies p
erta
inin
g to
a c
olon
ial p
ast,
and
a du
mm
y fo
r maj
or o
il ex
porti
ng
coun
tries
. The
join
t sig
. tes
t is t
he te
st th
at c
oeff
icie
nts o
f con
cent
ratio
n an
d th
e in
tera
ctio
n te
rms a
re jo
intly
zer
o.
100
Table 8: The interaction effects of shock, financial development and rent-seekingon regime choice — five-year averaged panel
Direct effect Full Model with three interactions
Full Model withoutShock
(1) (2) (3) (4) (5) (6)VARIABLES OLS LIML OLS LIML OLS LIMLconcen –17.390*** –36.549** –41.083*** –11.940 –23.260* –48.231*
(4.047) (18.000) (14.543) (29.408) (11.737) (25.688)shock 0.108 0.313**
(0.073) (0.153)findev –0.726 2.528 0.767 1.750
(3.278) (2.204) (3.078) (2.382)corrup –0.160 –0.122 –0.081 –0.813
(0.485) (0.489) (0.388) (0.516)concen*shock –0.348 –2.075*
(0.512) (1.127)concen*findev 14.568 –9.874 –6.455 –18.465
(27.289) (26.820) (26.354) (25.113)concen*corrup 7.450** 8.898** 4.881 11.508**
(3.305) (3.958) (3.053) (4.544)Observations 514 399 243 218 358 257R2 0.227 - 0.249 - 0.224 -Base Controls Yes Yes Yes Yes Yes YesFixed factors No Yes No Yes No YesJoint Sig. test(F-stat, p-val)
- - 3.88(0.00)
9.99(0.04)
1.57(0.20)
2.24(0.09)
The dependent variable is regime choice. Robust standard errors, clustered to country level, are given in parentheses. ***, **, and * indicate p < 0.01, p < 0.05, and p < 0.1, respectively. Each model has a constant, time dummies, and a common time trend. The baseline controls are the one period lag of capital account openness and log population. The fixed factors include a landlocked country dummy, latitude, dummies pertaining to a colonial past, and a dummy for major oil exporting countries. The joint sig. test is the test that the coefficients of concen and the interaction terms are jointly zero.
101
Figure 1: Regime choice and product concentration — non-parametric fit
Figure 2: regime choice and product concentration — quadratic fit
05
1015
Reg
ime
Cho
ice
0 .1 .2 .3 .4Concentration
bandwidth = .8
Lowess smoother
05
1015
Reg
ime
Cho
ice
(rrf)
0 .1 .2 .3 .4Concentration Index
rrf Fitted values
102
Figure 3: Product concentration and Neighbours’ GDP-weighted concentration
Figure 4: Product concentration and neighbours’ GDP-weighted concentration — quadratic fit
0.1
.2.3
.4C
once
ntra
tion
.05 .1 .15 .2Neighbour's GDP weighted Concentration
bandwidth = .8
Lowess smoother
AFG
ALB DZA AGO
ARG
ARM
AUS
AUT
AZEBRB
BEL
BLZ
BEN
BTN
BOL
BIH
BWA
BRABGR
BFA
BDI
KHM
CMR
CAN
CAF
CHL CHNCOL
ZAR
COG
CRICIV
HRV
CYPCZE DNK
ECUEGY
SLV
GNQ
ETH
FJI
FINFRA
GAB
DEUGHA
GRC
GTM HND
HUN
ISL
INDIDN IRNIRQ
IRL ISRITA
JAM
JPN JOR
KEN
KORKWT
KGZ
LAOLVA
LBN
LSO
LBY
LTU
LUX
MWI
MYSMLT
MUS
MEX
MDAMNGMAR
MOZ NPL
NLDNZL
NICNGA
NOR
OMN
PAK
PAN PNG
PRY
PERPHL
POLPRT
QAT
ROM RUS
RWA
SAU
SRB
SGP
SVKSVN
SOM
ZAFESP
LKA
SDN
SUR
SWZ
SWECHE
SYR
TZA
THA
TGO
TTO TUNTUR
UGA
UKR
ARE
GBRUSA
URY
VEN
YEM
ZMB
ZWE
0.1
.2.3
.4C
once
ntra
tion
Inde
x
.05 .1 .15 .2Neighbour's GDP weighted Concentartion
u323 employment share HI Fitted values
103
Figure 5: The effect of concentration on regime choice, conditional on the levelof shocks (cross-sectional data)
Countries where diversification leads to flexible regimes at a higher level of shocks (from Figure 5)
Algeria El Salvador Malaysia SpainBolivia Ethiopia Mexico TanzaniaBrazil Gabon Mozambique TogoBurkina Faso Gambia, The Myanmar Trinidad and TobagoChile Ghana Nicaragua UgandaCongo, Rep. Guatemala Nigeria Venezuela, RBCote d'Ivoire Haiti Paraguay ZambiaDominican Republic Indonesia PeruEcuador Japan PolandEgypt, Arab Rep. Malawi Senegal
-76-7
2-68-6
4-60-5
6-52-4
8-44-4
0-36-3
2-28-2
4-20-1
6-12-
8-4
04
812
1620
24
Mar
gina
l Effe
ct o
f Con
cent
ratio
n
0 10 20 30 40 50 60 70 80
Shocks
Marginal Effect of Concentration95% Confidence Interval
104
Figure 6: The effect of concentration on regime choice, conditional on the level of financial sector development (cross-sectional data)
Countries where diversification leads to flexible regimes at a lower level of financial development (from Figure 6)Albania Croatia Iran PeruAlgeria Denmark Iraq PhilippinesArgentina Dominican Republic Indonesia PolandArmenia Ecuador Iran, Islamic Rep. Saudi ArabiaBahrain Egypt, Arab Rep. Jamaica SenegalBangladesh El Salvador Kenya Slovak RepublicBelgium Ethiopia Latvia SloveniaBolivia Gabon Libya Sri LankaBotswana Gambia, The Madagascar SudanBrazil Ghana Malawi SurinameBulgaria Greece Mexico TogoBurkina Faso Guatemala Mongolia Trinidad and TobagoCameroon Guyana Morocco TurkeyColombia Haiti Mozambique UruguayCongo, Dem. Honduras Myanmar UruguayCongo, Rep. Hungary Nigeria Venezuela, RBCosta Rica Iceland Pakistan ZambiaCote d'Ivoire India Papua New Guinea
-120
-105
-90
-75
-60-
45-3
0-15
015
3045
6075
9010
5120
Mar
gina
l Effe
ct o
f Con
cent
ratio
n
0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 1.1 1.2 1.3 1.4 1.5
Financial Sector Development
Marginal Effect of Concentration95% Confidence Interval
105
Figure 7: The effect of concentration on regime choice, conditional on the level of corruption (cross-sectional data)
Countries where concentration leads to fixed regimes at a higher level of corruption (from Figure 7)
Albania Dominican Republic Korea, Rep. SenegalAlgeria Ecuador Kuwait Sri LankaArgentina Egypt, Arab Rep. Latvia SudanArmenia El Salvador Libya SurinameAzerbaijan Ethiopia Lithuania Syrian Arab RepublicBahamas, The Gabon Malawi TanzaniaBahrain Gambia, The Malta ThailandBangladesh Ghana Mexico TogoBolivia Guatemala Moldova Trinidad and TobagoBotswana Guyana Morocco TunisiaBrazil Haiti Nigeria TurkeyBurkina Faso Honduras Pakistan UgandaCameroon India Panama UkraineChile Indonesia Papua New Guinea UruguayChina Iran, Islamic Rep. Paraguay Venezuela, RBColombia Italy Peru ZambiaCongo, Rep. Jamaica Philippines ZimbabweCote d'Ivoire Jordan Russian FederationCroatia Kenya Saudi Arabia
-110
-100
-90
-80
-70
-60
-50
-40
-30
-20
-10
010
2030
4050
Mar
gina
l Effe
ct o
f Con
cent
ratio
n
0 .5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Level of Corruption
Marginal Effect of Concentration95% Confidence Interval
106
Chapter 4: Exchange Rate Regime and Fiscal Discipline: The Joint Role of Trade Openness and Central Bank Independence
4.1 Introduction
There is a theoretical consensus regarding the effect of exchange rate regime choice
on the conduct of fiscal policy (Rose 2012). However, considerable debate remains
as to which exchange rate regime provides more fiscal discipline in reality.41 One
strand of literature, termed as the ‘traditional view’, argues that fixed exchange rate
regime can discipline fiscal policy. This follows Keynes (1923, p 188), who
considers the ‘Gold Standard’, a form of fixed exchange rate, as a way to ‘strap
down the Ministers of Finance’ who are always tempted to create budget deficits.
Along this line, fixed exchange rates mean more fiscal discipline since failure to
maintain fiscal discipline under these regimes may be economically and politically
costly for any government (see, among others, Aghevli, Khan and Montiel 1991;
Frenkel and Goldstein 1988). The other strand of literature posits that a flexible
regime is associated with more fiscal discipline (for example, Tornell and Velasco
1995; Tornell and Velasco 2000). This happens because fiscal indiscipline has an
almost instantaneous adverse effect under the flexible regime.
Though high and persistent budget deficits and the resulting public debt are a long-
standing macroeconomic policy concern, a refreshed analysis of their determinants
has become a pressing need in the light of recent serious concerns for many
developed and developing countries. For example, overall fiscal deficit stood at
7.6% of gross domestic product (GDP) in advanced countries in 2010 (IMF 2011).
41 The fiscal discipline refers to the ability of exchange rate regimes to contain budget deficit, government expenditure or public debt.
107
Similarly, on average, public debt as a percentage of GDP has escalated to about
100% in advanced countries, and in countries like Greece, public debt has become
as high as 143% of GDP (IMF 2011). Budget deficits in emerging market economies
and low-income countries were 3.7% and 3% of GDP respectively in 2010 (IMF
2011). Though average public debt in emerging countries is not very high, it ranges
from 70% to 80% of GDP in countries like India, Brazil and Hungary. It is argued
(see Fatas 2010; Eichengreen et al. 2011) that this surge in deficit and public debt is
not solely the result of the recent crisis of 2008, but started much earlier and reflects
the failure of governments to accumulate enough surpluses in good times.
The renewed interest on macroeconomic imbalance and fiscal consolidation is also
rooted in their impact on the other parts of the economy. For instance, ‘aggressive
use of discretionary fiscal policy’ induces macroeconomic instability such as output
volatility, which in turn reduces economic growth (Fatas and Mihov 2003, p 1440).
Further, budget deficit can lead towards current account deficit, as suggested by
‘twin deficit’ hypothesis. Bluedorn and Leigh (2011) provide fresh evidence on this
hypothesis that if budget deficit to GDP decreases by 1% current account balance to
GDP also decreases by 0.6%. As Fatas (2010, p 2) states, the ‘surge in government
debt among many OECD countries in recent years is a wake-up call for their
government to increase fiscal discipline’.
Within the above context, the present paper carries out a systematic and thorough
empirical analysis on how countries’ de facto or actual exchange rate regime choices
affect fiscal policy outcomes. In doing so, this study, for the first time in the related
literature, disentangles both the direct and some of the interaction effects of the
regime choice on fiscal discipline. Our focus on the interaction effects of regime
choice is motivated by inconclusive picture offered by prior empirical studies
108
regarding the direct effect of exchange rate regimes on fiscal discipline (see El-
Shagi 2011; Fatas and Rose 2001; Hamann 2001; Tornell and Velasco 2000). One
reason for this inconclusiveness may be that the studies generally ignore the fact that
an exchange rate regime’s effect on fiscal outcomes may depend critically on the
levels of other theoretically important variable(s).
Consequently, this study aims to investigate the critical joint role of two important
determinants of fiscal discipline in moderating the effect of exchange rate regime
choice: trade openness and central bank independence (CBI). The findings of this
study can inform countries of their appropriate choice of exchange rate regime that
is commensurate with their current levels of trade openness and CBI for maintaining
fiscal discipline.
There are good reasons to believe that central macroeconomic policy choices and
fiscal outcomes can be shaped by trade openness. Focusing on the role of higher
trade openness on budget balance, Combes and Saadi-Sedik (2006) argue that higher
openness may reduce rent-seeking opportunity in a country and thus can improve
budget balance. They further argue that higher trade openness, by increasing
inequality, may deteriorate government budget balance. Rodrik (1998) and Cameron
(1978) discuss the direct effect of trade openness on government size. They suggest
the important role of government expenditure as a mitigating factor in higher
external risk resulting from higher trade openness. There are also compelling
arguments that higher trade openness can determine directly the level of exchange
rate regime choice. For example, exchange rate fluctuations may be costly for highly
open economies where trade and investment becomes crucial (Corden 2002). As a
result, when countries become more and more open, they should move towards fixed
regimes. Alternatively, higher openness can make countries more vulnerable to real
109
shocks and thus can make flexible regimes more attractive (Magud 2010). The net
effect is an empirical question.
As to CBI, which has exhibited a significant improvement across the world since the
1980s (Cukierman 2008; Crowe and Meade 2008), it is often argued that an
independent central bank keen on maintaining price stability should resist the
pressure of monetising government budget deficit. As a result, a truly independent
central bank can bring more fiscal discipline by resisting the government’s
temptation to generate persistent budget deficit (see Sikken and de Haan 1998).
Conversely, international political economy literature considers fixed exchange rate
regimes and independent central banks as two institutional arrangements to achieve
price stability (see Bodea 2010; Bearce 2008; Bernhard, Broz and Clark 2002). The
literature also suggests that these two institutions can be complementary or
substitute to each other. In the former case, higher CBI will result in more flexible
regimes, while in the latter case, higher CBI will mean fixed regimes.
From the above arguments, it is clear that apart from their direct effect, both trade
openness and CBI can moderate the effect of exchange rate regimes on fiscal
outcomes. In fact, it can be readily seen from figures 1a to 1c and 2a to 2c that the
relationship between budget deficit and exchange rate regimes varies significantly,
depending on the levels of trade openness and CBI, and suggest the possibility of an
interaction effect between the two variables and the exchange rate regime.
Therefore, one testable hypothesis would be how the interaction effect of exchange
rate regime and trade openness affects fiscal discipline. The second hypothesis
would be if there is any effect coming from the interaction between exchange rate
regime and CBI. This study tests the hypotheses by applying formal econometric
techniques.
110
More specific contributions of this study are as follows. First, unlike many other
studies which confine exchange rate regime choice to fixed and flexible regime
dichotomy, this study uses the Reinhart-Rogoff (2004) ‘fine’ exchange rate regime
classification, which distinguishes as many as 14 categories of exchange rate
regimes. The categories are ordered on a scale of 1 to 14, where 1 refers to the least
flexible and 14 refers to the most flexible regime. Using this index, this study
examines whether an exchange rate regime’s effect on fiscal policy depends on the
level of trade openness and CBI. To account for the effect of these channels’
interaction, exchange rate regime and CBI and trade openness enter non-linearly in
our empirical model. Second, in contrast to several other studies, we consider six
fiscal policy related variables, which allows us to comprehensively check the
robustness of our findings.42 These variables are overall budget balance, primary
budget balance, cash surplus, total government expenditure, primary government
expenditure and government consumption expenditure. Third, with some exceptions
(such as Fatas and Rose 2001), previous empirical studies confine themselves to a
small group of countries (for example, Sub-Saharan African or Caribbean countries),
which may result in sample selection bias. This study, in contrast, explores a large
number of countries’ most recent panel dataset. Finally, unlike many other studies
where endogeneity of exchange rate regimes has not been directly tackled, this study
addresses the possible endogeneity problem by instrumental variables (IV)
estimation and system GMM techniques.
To preview our findings, the results show significant and robust evidence to the
hypothesis that the effect of the exchange rate choice on fiscal variables depends
critically on the level of a country’s trade openness. There is a threshold level of
42 Except for Fatas and Rose (2001), all empirical studies examine the effect of regime choice on one or two fiscal variables.
111
trade openness, below which a fixed regime provides fiscal discipline and above the
threshold, a flexible regime has a disciplinary effect. This finding has important
policy implications. For instance, based on our result, Brazil, a relatively low open
economy, would likely to improve its overall budget balance to Columbia or Sri
Lanka’s level if it moves from its current flexible regime to a more fixed one.
Conversely, budget balance of a highly open economy like Singapore may
deteriorate to Australia’s level if the country abandons the current flexible regime
and embraces a fixed regime. We do not find evidence for our second hypothesis
that there is any interaction effect between central bank independence and regime
choice to affect fiscal outcomes. There is also little evidence that the central bank
independence directly influences any fiscal outcome variables. The above findings
remain robust across different measures of fiscal outcomes, and in presence of a
number of controls, which include country specific recession dummy, the level of
development, institutional quality and capital account openness.
The paper is organised in the following way. Section 4.2 outlines the existing
theoretical and empirical literature. The empirical methodology is described in
section 4.3, which includes data sources and descriptions, model specifications and
estimation methods. Section 4.4 contains detailed discussion of results. Section 4.5
concludes. All tables and figures are reported at the end of the paper.
4.2 Literature review
4.2.1 Theoretical literature
Theoretically, the direct effect of exchange rate regimes on fiscal policy is at best
ambiguous. The conventional idea is that a fixed exchange rate regime provides
more fiscal discipline; that is, it lowers budget deficit (see, inter alia, Aghevli, Khan
112
and Montiel 1991; Frenkel and Goldstein 1988; Chari and Kehoe 2008 and
Duttagupta and Tolosa 2006). This notion can be traced back to Keynes (1923, p
188), who perceives the ‘Gold Standard’, a form of fixed exchange rate, as a
mechanism to ‘strap down Ministers of Finance’. The reason is that finance
ministers are generally prone to generating budget deficits. However, the Gold
Standard (or any metallic standard) limits the ability of central banks to monetise
government budget deficits (see Corden 2002), and thus can maintain fiscal
discipline.
The extant studies point to various channels through which exchange rate can
influence fiscal outcome. For example, fixed exchange rate can be treated as a
credible promise from the government. The fiscal policy should be commensurate to
fulfil this commitment. A chronic fiscal deficit resulting from imprudent fiscal
policy will lead to high domestic inflation and reserve loss if deficit is monetised.
Budget deficit can be inflationary, even if a government finances it by issuing
nominal bonds, which are not indexed to price level (Leeper and Walker 2012).
In the event of high domestic inflation, the real exchange rate will appreciate if a
fixed exchange rate is maintained. This will result in a loss of competitiveness and a
higher trade deficit. To maintain competitiveness, the devaluation of a currency is an
option. This means breaking the fixed exchange rate promise, which can be
economically costly (see Frankel 2005). Further, the public can easily monitor
exchange rate change and detect a ‘broken promise’ (Canavan and Tommasi 1997).
The breakdown of peg may be politically costly for the governments as well.43 It is
argued that the economic and political costs of deficit-induced domestic inflation
should induce the government or fiscal authority to refrain from pursuing lax fiscal
43 See Edwards and Santaella (1993) for evidence on these costs.
113
policy. Hefeker (2010) develops a model to show that credibly fixed exchange rates
in a low inflation country can reduce corruption and improve the fiscal system,
because low inflation will result in lower seigniorage revenue, which will induce a
country to fight against corruption and the looting of the budget.
Conversely, Tornell and Velasco (1995; 2000) and Sun (2003) refute the
conventional view by arguing that lax fiscal behaviour involves costs both under
fixed and flexible exchange rate regimes, but inter-temporal distribution of these
costs is different under alternative exchange rate regimes. They argue that under
fixed regimes, lax fiscal policies result in declining reserves or soaring debts which
will punish the policy makers when the situation becomes unsustainable in the
future. Conversely, flexible regimes let the consequences of imprudent fiscal
policies to manifest themselves instantly through fluctuations in the exchange rate
and the price level. The conclusion of their arguments is that ‘if inflation is costly
for the fiscal authorities, then flexible rates, by forcing the costs to be paid up-front,
can provide more fiscal discipline’ (Tornell and Velasco 1995, p 401).
Ghosh, Gulde and Wolf (2002) and Fatas and Rose (2001) point to another channel
through which fiscal policy and exchange rate regimes can be related. Their
argument is that under fixed exchange rate regimes and high capital mobility,
monetary policy becomes ineffective as a stabilisation tool. Consequently, fiscal
policy has to shoulder the entire burden of macroeconomic stabilisation, the
implication of which is that fiscal policy has to be larger and more responsive to
business cycles under a fixed regime. From the above discussion, it is clear that at
the theoretical level there is no consensus as to which exchange rate provides more
fiscal discipline and in which way.
114
4.2.1.1 Trade openness, exchange rate regime and fiscal policy
This section focuses on the direct effect of trade openness on fiscal policy outcomes.
Combes and Saadi-Sedik (2006) show several channels through which trade
openness may affect budget balance, but conclude that the direction of trade
openness’s effect on budget balance is ambiguous.44 First, higher openness by
ensuring higher competition in the product markets may result in lower rent-seeking
or corruption, which in turn may result in higher budget balance. Second, higher
openness increases inequality, which increases the demand for public goods, but
reduces a government’s ability to collect taxes and ultimately results in a higher
budget deficit. They also point to the exposure of highly open country to external
shocks in affecting budget balance. Rodrik (1998) and Cameron (1978) posit that
higher trade openness results in larger government size, which can lead to larger
budget deficit. Rodrik (1998) opines that more open countries are vulnerable to
higher external shocks or risks, and higher government expenditure can mitigate the
risk by providing more social insurance.
The interaction effect between trade openness and exchange rate regime choice can
be justified from the theoretical argument that higher trade openness can
significantly influence exchange rate regime choice.45 For example, Corden (2002)
argues that, in a highly open economy, exchange rate fluctuations, by affecting trade
and capital movement, will exert a larger negative effect. To avoid this negative
effect, countries would prefer to adopt fixed exchange rate regimes. On the other
hand, Magud (2010) hypothesises that flexible exchange rate regimes can act as a
44 However, in the empirical section of the paper, Combes and Saadi-Sedik (2006) find that higher openness increases budget deficit.45There appears to be a possible feedback from the exchange rate regime choice on the level of trade openness. For example, a fixed exchange regime made by keeping the exchange rate stable and thus reducing the uncertainty and transaction costs induces higher trade (see Frankel and Rose 2002; Qureshi and Tsangarides 2012).
115
shock absorber in a sufficiently open economy facing real shocks and flexible
regimes should be preferable for this economy. In these ways, higher trade openness
could most definitely affect the exchange rate regime choice, though the direction of
the effect is ambiguous. The researcher can argue that higher trade openness, by
affecting exchange rate regime choice, can moderate the latter’s effect on fiscal
outcomes. If trade openness restrains countries’ exchange rate regime choices, then
higher openness would weaken the disciplinary effect of an exchange rate regime.
4.2.1.2 Central bank independence, exchange rate regime and fiscal policy
The direct effect of the central bank independence (CBI) on fiscal discipline is likely
to work in the following way. A highly independent central bank would resist the
propensity to monetise extravagant government expenditure, thus bringing in more
fiscal discipline. An independent central bank, by affecting the rate of growth of
money and credit, can influence other macroeconomic variables, such as inflation
and budget deficit (Cukierman, Webb and Neyapti 1992). Sargent and Wallace
(1981) argue that if fiscal policy dominates monetary policy, a monetary authority
may be forced to monetise a budget deficit that cannot be fully financed by issuing
bonds. It follows from this argument that an independent central bank, to achieve its
price level stability objective, may be reluctant to monetise government deficit. This
is likely to make governments more cautious in pursuing lax fiscal policy (Sikken
and de Haan 1998). This argument is especially true for developing countries, where
the possibility of financing a government deficit by selling bonds is limited (Sikken
and de Haan 1998).
The interaction effect between CBI and exchange rate regime choice follows from
international political economy literature, which stresses the association between
116
CBI and fixed exchange rate regimes. This literature considers fixed exchange rate
regimes and independent central banks as two institutional arrangements to achieve
price stability (see Bodea 2010; Bearce 2008; Bernhard, Broz and Clark 2002).
These studies also suggest that the above two institutions can be complementary or
substitute to each other. In the complementarity case, one institution reinforces the
other, while in the latter one institution obviates the need for the other. Berger,
Sturm and de Haan (2000) argue that an independent central bank will result in
lower inflationary bias, which will lower the credibility gain from a pegged
exchange rate. Thus, higher CBI may lead towards flexible regimes. Frieden, Ghezzi
and Stein (2000) argue that an independent central bank may want to tie its hands by
adopting a peg regime so that it can pursue its price stability objective. On the
contrary, Bodea (2010) claims that if there are imperfections in these institutions, in
the sense that fixed exchange rates are adjustable or CBI lack transparency, then
both fixed regimes and independent central banks can be chosen. The above
arguments suggest the possibility that higher CBI can determine the level of an
exchange rate regime, which in turn will affect fiscal outcomes. However, the
direction of the effect is ambiguous.
4.2.2 Empirical literature
In this section, we discuss some of the empirical studies that examine the correlation
between fiscal variables and exchange rate regime. Tornell and Velasco (2000)
analyse data for 25 Sub-Saharan African (SSA) countries to test their theoretical
idea that flexible but not fixed regimes provide more fiscal discipline. The positive
signs of the coefficients of the dummies representing exchange rate regime choice
suggest that a flexible regime provides more fiscal discipline. However, the
coefficients remain largely insignificant across different time periods.
117
Fatas and Rose (2001) examines the effect of exchange rate regime choice on a
number of fiscal policy related variables, such as total government expenditure,
current revenue, overall budget surplus, general government consumption and tax
revenue. It also considers disaggregated measures of revenue and expenditure. Fatas
and Rose (2001) use a panel dataset covering the period 1960 to 1998 for a large
number of countries. The estimation method used is ordinary least squares (OLS),
including time dummies. However, this study considers dummies for only hard peg,
ignoring other fixed regimes, and finds that while a currency board arrangement
provides some fiscal discipline, currency union and dollarisation do not.
Endogeneity of exchange rate regime has not been accounted for directly.
Duttagupta and Tolosa (2006) examine free-riding behaviour under fixed and
flexible regimes in 15 Caribbean countries. They find that fiscal balance is worse in
countries with a fixed regime. They use lag values of controls to avoid endogeneity,
but consider regime choice to be exogenous. Kim (2003) finds little evidence that
exchange rate regime has any significant effect on budget deficit. Hamann (2001)
does not find evidence that a fixed regime provides more fiscal discipline.
El-Shagi (2011) concentrates on the disciplinary effect of exchange rate regime
choices for more than 100 countries from 1975 to 2004. The study is confined only
to government indebtedness measured by the change in the ratio of government debt
to GDP. Unlike other studies, El-Shagi (2011) includes current as well as lagged
regime choices; the latter cover the possibility that the final effects of changing
regimes on debt do not manifest immediately. By using a least-square dummy
variable (LSDV) method and dynamic panel with a two-step system GMM, this
study finds that new debt initially decreases when a pegged is introduced, but returns
to the original level over time. Aside from being confined to only new indebtedness,
this study only examines the direct effect of regime choice on new indebtedness. In
118
so doing, it considers regime choice in arbitrarily chosen fixed versus flexible
settings.
The existing empirical studies generally suffer from a number of shortcomings.
First, they focus exclusively on the direct effect of regime choice on fiscal
outcomes. Second, the findings of these studies are not conclusive and robust. Third,
no previous work considers the possibility that trade openness and central bank
independence can moderate the effect of exchange rate regime choices in affecting
fiscal outcomes.
4.3 Data and methodology
4.3.1 Data
We employ two datasets for fiscal policy related variables. The first dataset is from
the government finance statistics (GFS) of the International Monetary Fund (IMF),
and covers the period 1971 to 2000 for 97 developed and developing countries. This
dataset provides observations on overall budget balance, primary budget balance, as
well as total and primary government expenditure.46 Government cash surplus,
which is similar to the overall budget balance but is not directly comparable, is also
used as another indicator of fiscal policy variable. The cash surplus data are from
1971 to 2007.47 The second data source is the world development indicator (WDI).
WDI provides the indicator for government size, which is proxied by the
government consumption expenditure. We have data for 126 countries for the period
1971 to 2007. All variables are expressed as a percentage of GDP.
46 These fiscal policy related variables are constructed according to the GFS 1986 manual. We use this dataset because it provides data for a wide range of fiscal policy related variables for a large number of countries. 47 Current GFS database reports data on fiscal variables according to the GFS 2001 manual. However, the database provides fiscal balance data for a few countries for a few years.
119
Exchange rate regime choice is the main explanatory variable, which is proxied by
the Reinhart and Rogoff (2004) de facto exchange rate classification index. The
index has 14 categories arranged from the least flexible to the most flexible regimes.
Higher values of this index indicate more flexible regimes. We use two measures for
Central Bank Independence. First is the central bank governor turnover rate (TOR),
which reflects de facto CBI (see Cukierman, Webb and Neyapti 1992) and is
commonly used in empirical studies regarding CBI. The higher TOR indicates a
lower CBI. Yearly data for the actual number of governor turnover are obtained
from Dreher, Sturm and de Haan (2008). TOR for the five-year period has been
calculated as the total number of governor change in a five-year period divided by
five. Second, we also use another measure, the legal CBI index. This measure was
first constructed by Cukierman, Webb and Neyapti (CWN) in 1992 on a decade
basis, and has values between 0 and 1—higher values indicate higher CBI. Polillo
and Guillen (2005) have extended the CWN index up to 2000 on an annual basis for
more than 90 countries. Crowe and Meade (2008) further extend the CWN index up
to 2003. Trade openness is measured by export plus import as a percentage of GDP.
Data for this variable are obtained from the Penn World Table version 6.3.
Now we turn to the controls which are assumed to have a direct effect on fiscal
discipline, and are commonly used in the literature. We use two proxies separately
to account for the effect of cyclical behaviour of the economy on fiscal policy. First,
a country specific recession dummy is generated using the method proposed by
Bruckner and Ciccone (2011). Second, the output gap is calculated based on log real
GDP using the Hodrick-Prescott (HP) filter. The calculated output gap is a
continuous variable with higher values indicating better economic performance. It is
120
expected that fiscal policy is counter-cyclical,48 that is, during a recession,
government expenditure will tend to increase and revenue decrease resulting in a
higher budget deficit. We use a log of per capita GDP as a proxy for a country’s
level of development. One would expect that more developed countries with better
fiscal and other institutions, in general, would be capable of limiting their budget
deficit.
Good governance or institutional factors are also important determinants of budget
deficit and can maintain fiscal discipline (Woo 2003; Abiad and Baig 2005). The
Polity2 variable from the Polity IV database has been used as a proxy for
institutional quality. A country’s level of development and CBI may also capture the
institutional factors. Capital account openness may impose a discipline on
government in pursuing fiscal policy because ‘the prospect of rising interest rate and
capital flight may discourage larger public sector deficit’ (Obstfeld and Taylor 2004,
p 9). This variable is measured by a capital account openness index proposed by
Chinn and Ito (2008). This index takes the values 0 to 1, where higher values
indicate more open capital account. A complete list of all variables, definitions of
the variables and data sources is in Table 1.
4.3.2 Model specification
Based on the theoretical discussions, we specify the following empirical model:
= + + + + + + + … … … … … … … … … … … … … … … … . . (1)
48 Some studies observe that fiscal policy can be pro-cyclical in developing countries (seeAlesina, Tabellini and Campante 2008).
121
is fiscal policy related variables, Regime is exchange rate regime choice, as
measured by the Reinhart-Rogoff flexibility index, CBI is an indicator of CBI and
Open refers to trade openness. We are particularly interested in the interaction
effects, i.e., the interaction between regime choice and CBI and interaction between
regime choice and trade openness. Regime choice is the ‘focal independent variable’
and trade openness and CBI are ‘moderator variables’ in the model. The interaction
effects will capture how a regime choice’s effect on fiscal variables depends on the
level of CBI and level of trade openness. is a vector of other controls, which
have a direct effect on fiscal outcomes (see, section 4.3.1). is the error term,
which follows the standard Gaussian assumptions.
The expected signs of the coefficients, as discussed in the theoretical section, are
ambiguous in most of the cases. Moreover, the signs will be different depending on
the type of fiscal variables considered (budget balance or government expenditure).
If a government budget balance is the dependent variable and if conventional view
holds (that is, that fixed regimes provide more fiscal discipline), one would expect
< 0 , indicating an improvement of the budget balance with fixed regimes.
Alternatively, >0 if the non-conventional view holds, in which case flexible
regimes are associated with improved budget balance. If the government
expenditure related variables are the dependent variables, then > 0 for a
conventional view and < 0 for a non-conventional view. The sign of is
expected to be positive as higher CBI is expected to provide more fiscal discipline.
However, if expenditure is the dependent variable, is likely to be negative.
Similarly, the sign of , which depicts the direct effect of openness on budget
balance is theoretically ambiguous. The expected signs of the coefficients of the two
122
interaction terms ( and ) are also ambiguous in the theory. Their expected
signs ultimately become an empirical question.
It must be noted that in a model with interaction term(s), the significance of the
point estimates of the coefficient(s) are not directly relevant. Rather, the focus
should be on the marginal effect of the focal explanatory variable (Greene 2003).
Therefore, we are particularly interested in the marginal effect of exchange rate
regime choice on fiscal outcomes, which will depend on the level of CBI and trade
openness and can be written as follows:
Marginal effect of regime choice = + CBI + OpenThe marginal effects of regime choice can be calculated at different values of CBI,
while keeping openness at its mean value. The standard error of each marginal effect
can be calculated and the significance of each marginal effect can be examined by
constructing upper and lower bounds for a confidence interval. Finally, all marginal
effects, along with their confidence interval, can be presented in a graph. The
marginal effects of regime choice at different values of trade openness can also be
evaluated in a similar manner.
4.3.3 Estimation method
In estimating equation 1, we treat regime choice with 14 distinct categories as a
continuous variable. This can be justified in the following way. Very often,
countries choose a set of exchange rate arrangements among a continuum of policy
choices to make up their exchange rate regimes. The de facto classification of
Reinhart-Rogoff categorises this spectrum into 14 groups. If two countries differ
only slightly in their exchange rate arrangements, they are likely to be allocated into
the same group. To be allocated into different groups, they must have made
123
sufficiently different policy choices. The question in this case is whether these 14
groups are ordinal variables, or continuous variables rounded to integers. Even if
one views these categories as ordinal, 14 different regimes make the underlying
categorisation almost indistinguishable from continuous measures.49
4.3.3.1 Pooled OLS
The fixed effect estimation of equation 1 will provide insignificant estimates,
because one of the important explanatory variables in the equation—the CBI—
changes very little overtime for all countries. Therefore, the equation is estimated by
pooled OLS using year dummies (common time effects), which capture common
shocks for all countries. Country-specific time trends are also included to control for
the time-trending behaviour of both dependent and explanatory variables. In this
way, the estimated coefficients capture the effects of explanatory variables on the
dependent variable outside their long-term trending relationship. In other words, our
model captures the deviations from long-run trends of both explanatory and
dependent variables, which, being a restrictive specification, is likely to take us
away from pure correlations towards more reliable relationships. The standard errors
are clustered at the country level.
Though few empirical studies on exchange rate regime choice have considered fiscal
outcomes as possible determinants of regime choice, endogeneity resulting from
reverse causation from fiscal outcome to exchange rate regime choice may be a
concern. The reverse causality may arise, for example, if a country experiencing a
higher budget deficit adopts a particular exchange rate regime to ensure the fiscal
49 Aghion et al. (2009) adopt the RR ‘coarse’ classification of exchange rate regimes with five categories and treat it as a continuous variable. Further, several other measures in the literature which are constructed in a similar way are treated as continuous. For instance, Polity IV indicators, Freedom House political rights and civil liberties measures, and corruption index are often used as explanatory variables on the right-hand side.
124
discipline that the regime will provide. Trade openness may be endogenous as well
(see Rodrik 1998). Endogeneity of CBI and institutional quality is less of a concern
because these variables change very little over time. Controls such as capital account
openness may also be endogenous (see Kim 2003). In order to tackle the
endogeneity problem, one period lag of all explanatory and control variables (except
that of the recession dummy) are used in estimating equation 1. Pooled OLS is thus
our benchmark specification.
It can be argued, however, that taking one period lag of endogenous variables is not
a proper way of correcting endogeneity. We, therefore, check whether our
benchmark results change if endogeneity is corrected with more formal techniques
described below.
4.3.3.2 Identification and IV estimation
The following three methods are used for further endogeneity correction. First,
equation 1 is estimated by two-stage least squares (2SLS), where either first or
second lags of the main endogenous explanatory variables (regime choice and
openness) are used as instruments. Other variables enter directly with a lag. Second,
we resort to an instrumental variable estimation technique using external instruments
for exchange rate regime choice. Finding suitable variables, which will fulfil all the
necessary conditions to be valid instruments for exchange rate regime, is a difficult
task. This is because the determinants of an exchange rate regime choice may also
directly affect fiscal outcomes and thus will violate the exclusion principle of good
instruments. For example, Aghion et al. (2009) use an institutional variable as an
instrument of exchange rate regime while examining the effect of regime choice on
labour productivity. Levy-Yeyati and Sturzenegger (2003), while examining the
effect of regime choice on growth, propose a number of exogenous variables, such
125
as the geographical area of the country, the proportion of a country’s GDP to US
GDP, an island dummy and the average exchange rate indicator of a country’s
neighbours (reflecting regional exchange rate indicators) as instruments for regime
choice. Similarly, Harms and Kretschmann (2009) use land area, a major oil
exporter dummy, central bank TOR and a variable related to political and
institutional instability. As mentioned earlier, most of these instruments are unlikely
to comply with the exclusion restriction principle in the present paper’s context. One
exogenous variable proposed by Levy-Yeyati and Sturzenegger (2003)—the average
exchange rate indicator of a country’s neighbours or the regional exchange rate
indicator—can fulfil the exclusion restriction principle. ‘The regional exchange rate
may indicate explicit or implicit exchange rate coordination with countries that
typically share strong trade links, as the trade literature has profusely illustrated
through the use of gravity models’ (Levy-Yeyati and Sturzenegger 2003, p 1185).
This single instrument, though relevant and exogenous, might explain very little of
the complex issue of exchange rate regime choice. Consequently, this variable alone
is likely to make a weak instrument in our estimation.50 Nonetheless, the model is
estimated with this single IV using the limited information maximum likelihood
(LIML) method, which is robust to weak instrument. To fulfil exclusion restriction,
we control for other variables such as trade with neighbours and a neighbours’ size-
weighted institutional quality in the second stage. Finally, we estimate the dynamic
version of equation 1 by the system GMM proposed by Arellano and Bover (1995)
and Blundell and Bond (1998). This method allows us to use internal instruments to
account for endogeneity (see El-Shagi 2011). Sargan/Hansen test is used to check
the validity of the internal instruments.
50 In general, the coefficient of this instrument in the first stage is positive and significant at 1% level. With this instrument, our model is not under-identified as suggested by the Kleibergen-Paap rk LM test. First stage F-statistics of the excluded instrument is about eight, giving evidence that the instrument is rather weak.
126
4.3.3.3 Robustness analysis
In order to check robustness of the benchmark results, equation 1 is also estimated
after omitting observations by regions, which accounts for the unobserved
characteristics of the omitted regions and mitigates endogeneity due to omitted
factors. In this connection, we have categorised countries into eight regions,
including East Asia and Pacific, Eastern Europe and Central Asia, Latin America,
Middle East and North Africa (MENA), South Asia, Sub-Saharan Africa (SSA),
Western Europe and North America and the Caribbean. As further robustness check
the equation is estimated by OLS including region dummies to capture regional
fixed effects and region specific time trends. Regional dummies are likely to capture
the structural characteristics related to geographical location (Woo 2003). Further,
exchange rate choice varies widely across regions.51 For example, oil exporting
Middle Eastern countries follows fixed regimes, some African countries pegged
their currency to their former colonial ruler, some Caribbean countries form
currency board, and Latin American countries have experimented with different
exchange rate regimes.
Common time effects are also included in all regressions to account for common
shocks.
4.4 Results and discussion
In this section we discuss the estimation results of equation 1. The signs and
magnitude of the coefficients of the main explanatory variables vary regarding the
choice of dependent variables. First, we discuss the results when dependent
variables are related to budget balance (that is, overall budget balance, primary
51 We have elaborated this in Table 3.
127
budget balance and cash surplus) in tables 4 to 8. Then we discuss findings when the
dependent variables are related to government expenditure, such as total
expenditure, primary expenditure and consumption expenditure (see Tables 9 and
10). In each case, we discuss only the coefficients of the main explanatory variables
and interaction terms.
4.4.1 Descriptive statistics
In this section we explore the potential direct and indirect links between fiscal
variables and main explanatory variables (exchange rate regime choice, trade
openness and CBI) and interaction effects by examining descriptive statistics and
scatter plots.
First, we present the average budget balance and other fiscal policy related variables
across different exchange rate regimes, levels of trade openness and CBI. Table 2
shows that the average overall budget balance is worst for flexible regimes (-4.89%
of GDP) followed by intermediate regimes. Budget balance is relatively better (-
3.38% of GDP) under fixed regimes. The same trend is observed for primary budget
balance and cash surplus. Conversely, government expenditure (total government
expenditure, primary expenditure and consumption expenditure) is higher under
fixed regimes than flexible regimes. This opposite behaviour of budget balance and
government expenditure under an alternative exchange rate is consistent with the
ambiguous direct effect of regime choice on fiscal outcomes.
Columns 4 to 6 in Table 2 show that average budget balances, as well as a cash
surplus position, is better for highly open countries as compared to low open
countries. However, average expenditure is consistently higher for highly open
countries. Part three of Table 2 shows that, consistent with the theory, low central
128
bank governor TOR (which corresponds to a high CBI) is associated with better
budget balance. Again, expenditure behaviour is not consistent with the theory.
Table 3 outlines considerable differences regarding budget balance across different
regions of the world. Overall budget balance, primary balance and cash surplus are
worse in the South Asia region, followed by Sub-Saharan African countries.52 The
table also shows the wide variation in exchange rate regime choices across regions.
It can also be noted that two regions with the worst budget balance are also the
regions following relatively fixed regimes.
In order to explore the possible interactions between regime choice and trade
openness on overall budget balance, we plot overall budget balance against regime
choice for observations pertaining to lower, two middles and upper quartiles of trade
openness. These plots are shown in figures 1a to 1c. It is observed that, depending
on the level of trade openness, the relationship between budget balance and regime
choice varies significantly, suggesting the possibility of an interaction effect. At a
low level of openness the relationship is negative, while at high openness the
relationship appears to be positive. At a medium level of trade openness the
relationship is rather weak. A similar relationship between primary budget balance
and regime choice prevails at different levels of trade openness. Similarly, figures 2a
to 2c plot budget deficit and regime choice at lower, middle and upper quartiles of
central bank governor TOR. The figures again reveal possible interaction effects
between budget balance and regime choice. We wish to establish this link more
formally by using the econometric techniques outlined previously.
52 A list of all regions is in the previous section.
129
4.4.2 Overall budget balance
Table 4 presents the benchmark OLS regression results of equation 1 with the
overall budget balance as the dependent variable. Column 1 shows the regression
results including common time effect (year dummies) but excluding controls. It can
be seen that the direct effect of a regime choice enters with a negative coefficient
and is significant at 5% level. The interaction between regime choice and trade
openness enters with a positive coefficient and is highly significant at 1% level,
suggesting that a regime choice’s effect on budget balance depends on the level of
trade openness. The direct effect of trade openness is insignificant with a negative
coefficient. CBI, as measured by the central bank governor TOR, enters with a
positive coefficient, contrary to our expectation. However, the coefficient is not
significant. Interaction between regime choice and TOR is not significant either,
reflecting the possibility that there may not be an interaction effect between TOR
and regime choice to affect budget balance.53 The inclusion of country specific time
trends in the column 1 regression makes direct effect of regime choice insignificant,
though the coefficient maintains its negative sign (see column 2). The interaction
between regime choice and trade openness remains significant at 5% level. This
gives strong evidence that trade openness affects budget balance through regime
choice. Other explanatory variables remain insignificant as before.
Columns 3 to 7 in Table 4 present regression results by including one control at a
time alongside the main explanatory variables and interaction terms. The controls
are country specific recession dummy (see column 3), output gap (see column 4),
institutional quality related variable (see column 5), level of development (see
column 6) and capital account openness (see column 7). In each regression,
53 If this interaction effect is not really significant, it may distort the direct effects. We explore this issue later.
130
interaction terms between regime choice and trade openness always enter
significantly at 5% or better levels. Conversely, interaction between regime choice
and TOR is largely insignificant, except for columns 3 and 4 when it remains
significant only marginally (at 10%). Finally, column 8 reports the full model, which
includes three controls—recession dummy, level of development and capital account
openness. The output gap is excluded because it is a similar measure of cyclical
behaviour of economy as a recession dummy. Institutional quality (polity) is also
excluded because its inclusion reduces the sample size and it is likely to be highly
correlated with the level of development. This model shows that the coefficient of
interaction between regime choice and trade openness is still positive and remains
significant at 1% level. Regime choice enters with a negative coefficient and is
significant at 5% level. Trade openness, which was insignificant before, becomes
marginally significant as well and its coefficient is negative. None of the coefficients
of TOR and the interaction between regime choice and TOR remain significant in
this full model.
The negative coefficients of regime choice in columns 1 to 8 in Table 4 signal that
as countries move towards fixed (flexible) regimes, budget balance improves
(deteriorates). This finding provides some support to the traditional view that fixed
regimes are associated with better fiscal discipline. However, the significant and
positive coefficient of interaction term between regime choice and trade openness
suggests that this disciplinary effect of fixed regime depends critically on the levels
of trade openness. To elaborate, at a higher level of openness, the disciplinary effect
of fixed regimes will be weakened and at a very high level of openness, flexible
regimes may be disciplinary. This can be justified by Magud (2010) argument that at
very high level of openness flexible regimes have more shock absorption capacity.
Flexible exchange rates thus will obviate the need for higher government
131
expenditure required for mitigating the effects shocks resulting from higher
openness.
Significant point estimates in columns 1 to 8 show strong evidence of an interaction
effect between regime choice and trade openness, but little evidence of the existence
of an interaction effect between regime choice and TOR. In a model with interaction
term(s), the direct effect of explanatory variable(s) on a dependent variable is less
relevant because the marginal effect of the focal explanatory variable depends on the
level of moderator variable(s). Therefore, it may be possible that marginal effects of
regime choice on budget balance may be significant for some values of TOR or
trade openness. In order to investigate this possibility, we calculate the marginal
effects of regime choice for the whole range of trade openness, while keeping TOR
at its mean value from the full model (column 8). Similarly, we calculate the
marginal effects of regime choice at different levels of TOR by keeping trade
openness at its mean.
The marginal effects of regime choice conditional on the levels of trade openness
are presented in Figure 3. In this figure, the solid horizontal line is the zero line. The
solid upward line shows the marginal effect of regime choice at different values of
trade openness. Two dotted lines show the upper and lower bounds of confidence
interval (CI) for each marginal effect. Marginal effect is significant if both CIs are
simultaneously above or below the zero line. The figure clearly shows that the
marginal effects are significant for a large number of values of trade openness.
Similarly, we present the marginal effects of regime choice conditional on the levels
of TOR in Figure 4. Unlike the previous graph, it is seen that marginal effect is not
significant for any values of TOR, which is consistent with our findings from the
point estimates.
132
To confirm the above findings, we re-estimate regressions 1 to 8 in Table 4 with
another measure of CBI: the legal independence of the central bank. When we use
this measure of CBI, the sample size, both in terms of the number of observations
and number of countries, drops significantly. Neither legal CBI nor its interaction
with regime choice is significant in any of the regressions. Nonetheless, in each case
point estimates of interaction between regime choice and trade openness are
significant at a 5% or better level. We report only the results from the full model in
column 9 in Table 4, which corresponds to the results in column 8 in the same table.
It should be noted that signs of the coefficients of CBI and interaction between CBI
and regime choice in column 8 and 9 are opposite to each other. These are expected,
given that CBI measured by TOR and legal CBI index are inverse to each other.
That is, higher values of TOR indicate lower CBI, while higher values of legal CBI
index indicate higher CBI. We have also calculated the marginal effects of regime
choice from column 9 and drawn CI graphs corresponding to Figure 3 and Figure 4.
The graphs also reveal that marginal effects of regime choice are not significant for
any level of the legal CBI index, but are significant for a large number of values of
trade openness.
Both point estimates and CI graphs of marginal effects of exchange rate regime
choice strongly suggest that there is no interaction effect between TOR (or legal
CBI) and regime choice. Therefore, we estimate column 8 of Table 4, excluding the
insignificant interaction term.54 This will allow us to estimate the effect of other
explanatory variables more precisely. Governor TOR is retained because it is an
important explanatory variable in our model. The results are presented in column 1
in Table 4a. It is seen that the coefficient of the direct effect of regime choice is now
54 We choose column 8 instead of column 9 because in the latter case, as mentioned, sample size drops significantly.
133
larger and significant at better level. The coefficient of trade openness also becomes
somewhat larger, but its significance level does not change. Magnitude and
significance of the interaction term do not change. We conduct further endogeneity
correction and robustness check with this model.
4.4.2.1 Endogeneity correction
In this sub-section we report endogeneity corrected results using 2SLS, LIML and
system GMM. The objective is to check whether our benchmark results change
because of endogeneity correction with more formal techniques. We estimate the
full model in column 1 in Table 4a by 2SLS, with the first lag values of main
explanatory variables as instruments. Controls enter with lag values. The 2SLS
results are in column 1a in Table 4a, which show that the coefficients of regime
choice and interaction terms are somewhat larger but remain significant at 1% level.
The coefficient of openness is significant at a better level (5%) and also becomes
larger.55 Second, LIML results using the neighbour’s size weighted regime choice as
an external instrument for regime choice are reported in column 1b. It is seen that
none of the coefficient is individually significant. However, the signs of the
coefficients are the same as coefficients of column 1. Third, results estimated by
system GMM are reported in column 1c in Table 4a. The country specific time trend
is not included in this regression. The interaction term is significant at a 5% level
with a positive sign. Regime choice and openness are significant at a 5% and 1%
level, respectively, and enter with the same sign as in column 1. Overall, 2SLS,
LIML and system GMM results broadly support the results of our full model in
column 1.
55 The use of second lags of explanatory variables as instruments gives similar results.
134
4.4.2.2 Robustness check
To check the robustness of the results we estimate the full model in column 1 in
Table 4a for sub-samples of countries, excluding observations from regions or
country group. The results are presented in column 1 to 7 in Table 5. Results in each
column are estimated by excluding countries from one region at a time. Column 1
and column 2 present results for developing and OECD countries, respectively.
Column 3, 4, 5, 6 and 7 exclude countries of SSA, Latin America, Caribbean,
Middle East and North Africa and Eastern Europe, respectively. Except for column
2, the results are qualitatively similar to those presented in column 1 in Table 4a.
Regime choice always enters with a negative sign and significant. The interaction
term is positive and significant at 1% level. In column 2, which presents results for
OECD countries, explanatory variables enter with the same signs as in other
regressions in the table, but none of the coefficients is significant. This may be
because this regression uses a relatively smaller number of countries and
observations. This may also result from the fact that exchange rate choice variation
is stronger in developing countries than developed countries.
The last three regressions (columns 8 to 10) in Table 5 show the results of our full
model, including region fixed effects, region time trends, and common time effects.
Column 8 reports results for all available observations. Column 9 and 10 report
results with observations from developing and OECD countries, respectively. Again,
results are similar to what were obtained before. Regime choice enters with negative
coefficient and is significant. The coefficient of interaction term is also highly
significant with consistent positive signs. The coefficient of trade openness is
negative, but significant only in columns 8 and 9.
135
Overall, we find strong evidence that regime choice has a direct effect on overall
budget balance. Moreover, it has an interaction effect on overall budget balance
through trade openness. That is, a regime choice’s effect depends on the level of
trade openness. Conversely, trade openness has a direct effect on budget deficit, but
that effect is not robust across different specifications.
4.4.2.3 Marginal effect of exchange rate regime choice on overall budget
balance
This section examines the marginal effects of regime choice on overall budget
balance conditional on the level of trade openness only. When interaction between
regime choice and trade openness is considered, the marginal effect of regime
choice is as follows:
Marginal effect of regime choice on budget balance = + .
Results in column 1 in Table 4a are chosen for discussion. In the extreme situation
when a country does not trade with rest of the world at all (openness is zero), the
marginal effect of regime choice on budget balance is -0.33. That is, if this country
moves towards fixed regimes by on unit in the Reinhart-Rogoff flexibility scale, its
budget balance will improve by 0.33%. If trade openness of this country increases
from zero to 45% (roughly the twenty-fifth percentile value) the marginal effect
decreases to -0.11. At the median value of trade openness (about 70%), the marginal
effect of regime choice becomes positive (0.02) and moving towards flexible
regimes provides more fiscal discipline.
As the signs of the coefficients of regime choice and interaction between regime
choice and openness are opposite to each other, higher level of openness will
weaken the effect of regime choice. In this case, the marginal effect of a regime may
136
not be entirely negative or positive, but a threshold level of trade openness may be
obtained, after which marginal effect may change sign. This is evident from Figure
5, which shows that at a low level of trade openness (below 61%), the marginal
effect of regime choice on budget balance is negative. When trade openness is above
61% of GDP, the marginal effect becomes positive. However, the marginal effect is
significant, as indicated by the two dotted lines lying simultaneously above or below
the zero line, when trade openness is equal or below 30% and equal or above 103%.
To make this finding more concrete, we consider the case of Brazil. On average,
Brazil is the least open country in our dataset (15%) and facing a high budget deficit
(7.5% of GDP) during 1971 to 2000 periods. This country is at the highest end of the
Reinhart-Rogoff exchange rate regime flexibility index (12), a managed floater. Our
model predicts that, given its current level of trade openness, if Brazil moves
towards fixed regimes, its fiscal balance will improve to Columbia or Sri Lanka’s
level. Conversely, Singapore is the most open country, with average openness above
300% of GDP during 1971 to 2000 period. The actual regime choice of this country
is at the floating end of the Reinhart-Rogoff flexibility index (category 11). The
average overall budget surplus (5.15%) of this country is one of the largest during
the same period. According to our model, at its current level of openness, if
Singapore’s actual regime choice becomes less flexible, its budget balance will
deteriorate to Australia’s level.
4.4.3 Primary budget balance and cash surplus
This section investigates whether the above findings for overall budget balance hold
for the other two measures of budget balance: primary budget balance and cash
surplus.
137
4.4.3.1 Primary budget balance
We have replicated the regressions in tables 4, 4a and 5 with primary budget balance
as dependent variable. The results are qualitatively similar to what we have found
regarding the overall budget balance. For brevity, we report pooled OLS, 2SLS,
LIML and system GMM for the full model only. The results are in the first part of
Table 6.
Column 1 of Table 6, which corresponds to column 8 in Table 4, reports the OLS
results when both interaction terms are included. It can be seen that regime choice
has a significant direct effect, as well as an interaction effect, on trade openness. The
interaction effect between regime choice and TOR appears non-existent. Relevant
CI graphs of marginal effects (not reported) also confirm these findings.56
Consequently, we report the OLS results in column 2, excluding the interaction
between regime choice and TOR. This column shows that signs and significance of
coefficients are in conformity with Table 4a results. The coefficients of regime
choice and interaction terms between regime choice and openness are highly
significant, but coefficients become somewhat smaller. 2SLS results with first lag as
instruments confirm the sign and significance of the above coefficients as well (see
column 2a). LIML estimates in column 2b confirm the sign of the coefficients, but
in contrast to column 1b in Table 4a, the interaction term becomes marginally
significant. System GMM results in column 2c are also in conformity with previous
results. The direct effect of trade openness, though entering with a negative
coefficient, is significant only in column 2a. This gives evidence that trade openness
affects budget balance mainly through it interaction with regime choice.
56 We have estimated column 1 with legal CBI index as well. Our conclusion on the
interaction effects does not change.
138
4.4.3.2 Robustness check: We estimate regression 2 in Table 6 with sub-samples
and regional fixed effects and present the results in Table 7. It is seen that the results
are not driven by any particular regions or group of countries, including OECD
countries (see columns 1 to 7). Results with a regional fixed effect are similar but
less robust when we conduct this analysis for developing and developed countries.
With a developing country sample, interaction term and openness enter significantly,
but regime choice does not. For OECD samples, none of the coefficients is
significant; this is not surprising because variation among the OECD countries is
likely to be smaller in exchange rate regime, trade openness, and so on.
4.4.3.3 Marginal effects of regime choice on primary budget balance: We have
calculated the marginal effects of regime choice on primary budget balance
conditional on the different values of trade openness based on column 2 in Table 6
and presented them in a graph (see Figure 6). The graph is similar to Figure 5. That
is, the marginal effect is negative (the fixed regime is disciplinary) at a relatively
low level of trade openness and positive at a higher level of openness. The threshold
value of trade openness at which marginal effect changes sign is 72%, which is
somewhat higher than what was obtained from Figure 5. The marginal effect is
significant when trade openness is lower than 40% or higher than 114%.
4.4.3.4 Cash surplus
This section outlines the results when cash surplus is the dependent variable. The
results of the full model are presented in the second part of Table 6. With both
interaction terms, OLS results in column 3 shows that the direct effect of regime
choice and interaction effect are significant with negative and positive coefficients,
respectively. While direct effect is significant at a 10% level, the interaction effect
is highly significant at 1% level. Column 4 excludes interaction between regime
139
choice and TOR. In this column, the direct effect of regime choice and its interaction
with trade openness are still significant but at lower levels. The direct effect of trade
openness is not significant, as usual. 2SLS results in column 4a are similar to OLS
results. In contrast, none of the coefficients are significant when LIML and system
GMM are considered (see columns 4b and 4c). However, the results do confirm the
signs of the coefficients; that is, a negative coefficient for regime choice and a
positive coefficient for the interaction term.
4.4.3.5 Robustness check: An analysis with sub-samples, including regional fixed
effects is shown in Table 8. Again, significant direct and interaction effects of
regime choice are confirmed in columns 1 to 7. Only in column 4 when Latin
American countries are omitted direct effect of regime choice is not significant. Of
course, the interaction effect is significant at 1% level in column 4. Results with
regional fixed effects presented in columns 8 to 10 broadly confirm our findings.
4.4.3.6 Marginal effects of regime choice on cash surplus: The marginal effects of
regime choice on cash surplus calculated from column 4 of Table 6 are presented in
Figure 7. The marginal effects are negative at lower levels of openness, suggesting
that fixed regimes are better able to provide a higher cash surplus. The effects are
significant when trade openness levels are 52% of GDP or lower. The magnitude of
marginal effects becomes positive when trade openness is very high, such as 133%
of GDP. However, as compared to marginal effect of regime choice on overall and
primary budget balance, the marginal effect on regime choice is not significant at
higher levels of trade openness.
140
4.4.4 Government expenditure
This section shows the results when government expenditure related variables are
dependent variables. We use three such variables: total government expenditure,
primary expenditure and government consumption expenditure. For each case we
present two OLS regressions—one with two interaction terms and the other with one
interaction term. We also present corresponding 2SLS, LIML and system GMM
results for the latter OLS regression. The results are presented in tables 9 and 10.
First, we consider the results when total government expenditure is the dependent
variable (see the first part of Table 9). As expected, the signs of the coefficients of
the main explanatory variables are now opposite to what was found in tables 4 to 8.
The coefficient of regime choice is now positive and significant in all cases except
for LIML results. Taken alone, the positive sign of the coefficient implies that
moving towards fixed regimes results in lower government expenditure; that is,
disciplining government expenditure. The sign of the interaction term between
regime and openness is negative and significant in all cases except LIML. This
implies that at a higher level of trade openness, the direct disciplinary effect will be
weaker. The trade openness variable also remains significant, with a positive
coefficient implying that higher openness alone will result in higher government
expenditure. Similar patterns are observed when dependent variables are primary
budget balance (see the second part of Table 9) and government consumption
expenditure (see Table 10). It is seen that the coefficient of TOR is significant with a
negative coefficient in some cases, but its sign is contrary to our expectations.57
57 We estimate the full model with one interaction term for three government expenditure related variables for sub-samples and with regional fixed effects. The results, not presented here, are highly consistent with the results in tables 9 and 10. However, the inclusion of regional fixed effects makes TOR insignificant, which was significant before in some cases.
141
4.4.4.1 Marginal effects of regime choice on government expenditure
This section discusses column 2 in Table 9. As the interaction effect is significant,
the marginal effect of regime choice on total government expenditure depends on
the level of trade openness. Moreover, as signs of the coefficients of regime choice
and interaction term are opposite to each other, a threshold level of trade openness
can be calculated, after which marginal effects will change sign. To find the
threshold, we present the marginal effects of regime choice for each level of trade
openness in a graph (see Figure 8).
It can be seen from Figure 8 that the marginal effects of regime choice on total
government expenditure are positive and significant when trade openness is below
27% of GDP; that is, at a low level of trade openness. The positive marginal effect
implies that at a low level of trade openness, fixed regimes (flexible regimes) result
in lower (higher) government expenditure. Marginal effect changes sign and
becomes negative when trade openness is 56% or higher, at which point fixed
regimes are associated with higher government expenditure. Negative marginal
effects are significant when trade openness is 87% or higher. We have also
calculated the marginal effects of regime choice on primary government expenditure
and government consumption expenditure from column 4 in Table 9 and column 2
in Table 10, respectively. The graphs of the marginal effects show similar patterns
as above.
All the above findings strongly suggest that the exchange rate regime choice has a
direct effect on fiscal outcomes, such as government budget balance and
expenditure. Moreover, regime choice interacts with trade openness to affect fiscal
outcomes. Conversely, trade openness has only an interaction effect on fiscal
outcomes. Consequently, marginal effects of regime choice on fiscal variables
142
depend critically on the level of trade openness. At a low level of openness, fixed
regimes result in a higher budget balance or lower government expenditure; that is,
fixed regimes have a more disciplinary effect. In contrast, when trade openness is
high, flexible regimes are conducive for maintaining fiscal discipline.
4.5 Conclusion
Against the backdrop of high and persistent fiscal deficit and public debt marring
several economies, this study addresses the role of exchange rate regime choice in
disciplining fiscal policy. The theoretical literature strongly justifies the role of an
exchange rate regime in disciplining fiscal policy outcomes, such as budget deficit
and government expenditure. However, considerable debate remains as to which
exchange rate provides more fiscal discipline. Empirical studies, which examine
only the direct effect of an exchange rate regime choice on fiscal outcomes, also fail
to provide a distinctive answer. In order to contribute to this debate, this study, for
the first time, stresses that an exchange rate regime has a direct effect, as well as
interaction effects through other variables, on fiscal outcomes. Particularly, we focus
on the strong theoretical arguments regarding the direct effect of trade openness and
CBI, which have increased significantly in recent decades, on fiscal discipline. Apart
from the direct effect of exchange rate regimes, we hypothesise that both CBI and
trade openness can moderate the effect of exchange rate regime choice, and thus can
jointly affect fiscal outcomes.
In order to test the empirical validity of our hypotheses, we use two annual panel
datasets covering the periods 1971 to 2000 and 1971 to 2007 for a large number of
developed and developing countries. We use six measures of fiscal outcomes,
including overall government budget balance, primary budget balance, cash surplus,
143
total government expenditure, primary expenditure and government consumption
expenditure. We estimate our empirical model by pooled OLS, including year
dummies, country specific time trends and with lags of all explanatory variables, to
begin. We account for the endogeneity of exchange rate regime choice by using
neighbours’ size-weighted exchange rate regimes as instruments and using the
limited information maximum likelihood estimation technique. The system GMM
method with internal instruments has also been applied. Sub-sample analysis and
separate pooled OLS estimation, including region fixed effects and region specific
time tends, have also been conducted.
We find strong and robust evidence that exchange rate regime choice has a direct
effect on fiscal outcomes, in that fixed regimes provide more fiscal discipline. This
finding supports the conventional view, which associates fixed exchange rate
regimes with more fiscal discipline (see Aghevli, Khan and Montiel 1991; Frenkel
and Goldstein 1988). We also find convincing evidence for the hypothesis that trade
openness interacts with regime choice to affect fiscal variables. Consequently, an
exchange rate regime’s marginal effects on fiscal discipline depend critically on the
level of a country’s trade openness. Particularly, we find that at low level of trade
openness fixed regimes provide more fiscal discipline, and there is a threshold level
of trade openness after which flexible regimes become more disciplinary. The
effects are significant at a relatively low or high level of openness, but not for a
middle level of openness. These findings are in conformity with the theoretical idea
of Magud (2010), who documents the shock-absorbing role of flexible regimes at a
high level of trade openness. Shock emanating from higher openness is
accommodated by flexible regimes, and thus governments have to bear few
consequences of the shock.
144
We do not find any evidence for our second hypothesis that CBI affects fiscal
discipline through its effect on exchange rate regime. The direct effect of CBI on
fiscal discipline is also not robust. This latter finding is consistent with the recent
trend of soaring budget deficits in many OECD countries, where central banks enjoy
reasonable degrees of independence. This may be because central bank activities are
not transparent enough and are difficult to monitor, especially in the short run
(Bodea 2010). Consequently, CBI may not always imply independence of monetary
policy over a government’s fiscal decision (Sargent 1999).
145
References
Abiad, A and Baig T 2005, ‘Underlying factors driving fiscal effort in emerging
market economies’, IMF Working Paper 106, International Monetary Fund,
Washington.
Aghevli, B, Khan, M and Montiel, P 1991, ‘Exchange rate policy in developing
countries: some analytical issues’, IMF occasional paper 78, International Monetary
Fund, Washington.
Aghion, P, Bacchetta, P, Ranciere, R and Rogoff, K 2009, ‘Exchange rate volatility
and productivity growth: the role of financial development’, Journal of Monetary
Economics, vol. 56, pp. 494–513.
Alesina, A, Tabellini, G and Campante, F 2008, ‘Why is fiscal policy often
procyclical?’, Journal of the European Economic Association, vol. 6, no. 5, pp.
1006–1036.
Arellano, M and Bover, O 1995, ‘Another look at the instrumental variable
estimation of error-components models’, Journal of Econometrics, vol. 68, pp. 9–51.
Bearce, D 2008, ‘Not complements, but substitutes: fixed exchange rate
commitments, central bank independence, and external currency stability’,
International Studies Quarterly, vol. 52, pp. 807–24.
Berger, H, Sturm, J and de Haan, J 2000, ‘An empirical investigation into exchange
rate regime choice and exchange rate volatility’, Working Paper 263, CESifo
Working Paper Series, Munich.
Bernhard, W, Broz, J and Clark, W 2002, ‘The political economy of monetary
institutions’, International Organization, vol. 56, no. 4, pp. 693–723.
Bluedorn, J and Leigh, D 2011, ‘Revisiting the twin deficits hypothesis: the effect of
fiscal consolidation on the current account’, IMF Economic Review, vol. 59, no. 4,
pp. 582–602.
Blundell, R and Bond, S 1998, ‘Initial conditions and moment restrictions in
dynamic panel data models’, Journal of Econometrics, vol. 87, pp. 115–143.
146
Bodea, C 2010, ‘Exchange rate regimes and independent central banks: a correlated
choice of imperfectly credible institutions’, International Organization, vol. 64, pp.
411–462.
Bruckner, M and Ciccone, A 2011, ‘Rain and the democratic window of
opportunity’, Econometrica, vol. 79, no. 3, pp. 923–947.
Cameron, D 1978, ‘The expansion of the public economy: a comparative analysis’,
American Political Science Review, vol. 72, pp. 1243–1261.
Canavan, C and Tommasi, M 1997, ‘On the credibility of alternative exchange rate
regimes’, Journal of Development Economics, vol. 54, pp. 101–122.
Chari, V and Kehoe, P 2008, ‘Time inconsistency and free-riding in a monetary
union’, Journal of Money, Credit and Banking, vol. 40, no. 7, pp. 1329–1355.
Chinn, M and Ito, H 2008, ‘A new measure of financial openness’, Journal of
Comparative Policy Analysis, vol. 10, no. 3, pp. 309–322.
Combes, J and Saadi-Sedik, T 2006, ‘How does trade openness influence budget
deficits in developing countries?’, Journal of Development Studies, vol. 42, no. 8,
pp. 1401–1416.
Corden, W 2002, Too sensational: on the choice of exchange rate regimes, MIT
Press, Cambridge.
Crowe, C and Meade, E 2008, ‘Central bank independence and transparency:
evolution and effectiveness’, European Journal of Political Economy, vol. 24, pp.
763–77.
Cukierman, A 2008, ‘Central bank independence and monetary policymaking
institutions: past, present and future’, European Journal of Political Economy, vol.
24, pp. 722–736.
Cukierman, A, Webb, S and Neyapti, B 1992, ‘Measuring the independence of
central banks and its effect on policy outcomes’, World Bank Economic Review, vol.
6, no. 3, pp. 353–398.
147
Dreher, A, Sturm, J and de Haan, J 2008, ‘Does high inflation cause central bankers
to lose their job? Evidence based on a new dataset’, European Journal of Political
Economy, vol. 24, pp. 778–787.
Duttagupta, R and Tolosa, G 2006, ‘Fiscal discipline and exchange rate regimes:
evidence from the Caribbean’, IMF Working Paper 6/119, International Monetary
Fund, Washington.
Edwards, S and Santaella, J 1993, ‘Devaluation controversies in the developing
countries: lessons from the Bretton Woods era’, in M Bordo and B Eichengreen
(eds), Retrospective on the Bretton Woods system, University of Chicago Press,
Chicago.
Eichengreen, B, Feldman, R, Liebman, J, von Hagen, J and Wyplose, C 2011,
‘Public debts: nuts, bolts and worries’, Geneva report on the world economy 13,
International Centre for Monetary and Banking Studies (ICMB), Geneva.
El-Shagi, M 2011, ‘The impact of fixed exchange rates on fiscal discipline’, Scottish
Journal of Political Economy, vol. 58, no. 5, pp. 685–709.
Fatas, A 2010, ‘The economics of achieving fiscal sustainability’, paper presented at
the academic consultants meeting on fiscal sustainability, Board of Governors,
Federal Reserve, April 9, 2010.
Fatas, A and Mihov, I 2003, ‘The case for restricting fiscal policy discretion’,
Quarterly Journal of Economics, vol. 118, no. 4, pp. 1419–1447.
Fatas, A and Rose, A 2001, ‘Do monetary handcuffs restrain Leviathan? Fiscal
policy in extreme exchange rate regimes’, IMF Staff Papers, vol. 47 (Special issue),
pp. 40-61.
Frankel, J and Rose, A 2002, ‘An estimate of the effect of common currencies on
trade and income’, Quarterly Journal of Economics, vol. 117, no. 2, pp. 437–466.
Frankel, J 2005, ‘Contractionary currency crashes in developing countries’, NBER
Working Paper 11508, National Bureau of Economic Research, Cambridge.
148
Frenkel, J and Goldstein, M 1988, ‘The international monetary system:
developments and prospects’, NBER Working Paper 2648, National Bureau of
Economic Research, Cambridge.
Frieden, J Ghezzi, P and Stein, E 2000, ‘Politics and exchange rates in Latin
America’, Research Network Working Paper R-421, Inter-American Development
Bank, Washington.
Ghosh, A, Gulde, A and Wolf, H 2002, Exchange rate regimes: choices and
consequences, MIT press, Cambridge.
Greene, W. 2003, Econometric Analysis, 5th Edition, Pearson Education Inc.,
Singapore.
Hamann, A 2001, ‘Exchange rate based stabilization: a critical look at the stylized
facts’, IMF Staff Papers, vol. 48, no. 1, pp. 110–138.
Harms, P and Kretschmann, M 2009, ‘Words, deeds and outcomes: a survey on the
growth effect of exchange rate regimes’, Journal of Economic Surveys, vol. 23, no.
1, pp. 139–164.
Hefeker, C 2010, ‘Taxation, corruption and the exchange rate regime’, Journal of
Macroeconomics, vol. 32, pp. 338–346.
Ilzetzki, E, Reinhart, C and Rogoff, K 2008, ‘Exchange rate arrangements in the
21st century: which anchor will hold?’, viewed 20 August 2009,
http://terpconnect.umd.edu/~creinhar/Papers.html.
IMF 2011, ‘Fiscal monitor update’, Fiscal Affairs Department, International
Monetary Fund, Washington.
Keynes, J 1923, A tract on monetary reform, Macmillan and Co., London.
Kim, W 2003, ‘Does capital account liberalization discipline budget deficit?’,
Review of International Economics, vol. 11, no. 5, pp. 830–844.
Leeper, E and Walker, T 2012, ‘Perceptions and misperceptions of fiscal inflation’,
NBER Working Paper 17903, National Bureau of Economic Research, Cambridge.
149
Levy-Yeyati, E and Sturzenegger, F 2003, ‘To float or to fix: evidence on the impact
of exchange rate regimes on growth’, American Economic Review, vol. 93, no. 4, pp.
1173–1193.
Magud, N 2010, ‘Currency mismatch, openness and exchange rate regime choice’,
Journal of Macroeconomics, vol. 32, pp. 68–89.
Obstfeld, M and Taylor, A 2004, Global capital markets: integration, crisis, and
growth, Cambridge University Press, Cambridge.
Polillo, S and Guillen, M 2005, ‘Globalisation pressure and the state: the global
spread of central bank independence’, American Journal of Sociology, vol. 110, no.
6, pp. 1764–1802.
Qureshi, M and Tsangarides, C 2012, ‘Hard or soft pegs? Choice of exchange rate
regime and trade in Africa’, World Development, vol. 40, no. 4, pp. 667–680.
Reinhart, C and Rogoff, K 2004, ‘The modern history of exchange rate
arrangements: a reinterpretation’, Quarterly Journal of Economics, vol. 119, no. 1,
pp. 1-48.
Rodrik, D 1998, ‘Why do more open economies have bigger governments?’,
Journal of Political Economy, vol. 106, no. 5, pp. 997–1032.
Rose, A 2012, ‘Exchange rate regime in modern era: fixed, floating, and flaky’,
Journal of Economic Literature, vol. 49, no. 3, pp. 652–672.
Sargent, T and Wallace, N 1981, ‘Some unpleasant monetarist arithmetic’, Federal
Reserve Bank of Minneapolis Quarterly Review, no. 5, pp. 1–17.
Sargent, T 1999, ‘A primer on monetary and fiscal policy’, Journal of Banking and
Finance, vol. 23, no. 10, pp. 1463–1482.
Sikken, B and de Haan, J 1998, ‘Budget deficit, monetization, and central bank
independence in developing countries’, Oxford Economic Papers, vol. 50, pp. 493–
511.
Sun, Y 2003, ‘Do fixed exchange rates induce more fiscal discipline?’, IMF
Working Paper 3/78, International Monetary Fund, Washington.
150
Tornell, A and Velasco, A 1995, ‘Fiscal discipline and choice of exchange rate
regime’, European Economic Review, vol. 39, pp. 759–770.
Tornell, A and Velasco, A 2000, ‘Fixed versus flexible exchange rates: which
provides more fiscal discipline?’, Journal of Monetary Economics, vol. 45, pp. 399–
436.
Woo, J 2003, ‘Economic, political, and institutional determinants of public deficits’,
Journal of Public Economics, vol. 87, pp. 387–426.
151
Tables and Figures
Table 1: Variables, data sources and availability
Variables Description Data sourceFiscal variablesOverall budget balance Revenue minus expenditure excluding
grant1971–2000, GFS, IMF
Primary budget balance Overall budget balance net of interest payment
1971–2000, GFS, IMF
Cash surplus Similar to the overall balance except that net cash outflows from lending and repayment transactions are not subtracted
1971–2007, GFS, IMF
Total govt expenditure Total expenditure including interest payment
1971–2000, GFS, IMF
Primary govt expenditure
Total expenditure excluding interest payment
1971–2000, GFS, IMF
Govt consumption Expenditure
Government final consumption expenditure
1971–2007, WDI
Other variablesRegime Exchange rate regime choice, Reinhart-
Rogoff fine classification (1 to 14 scale),higher values indicate more flexible regimes
1971–2007, Ilzetzki,Reinhart and Rogoff(2008)
Open Trade openness, the share of export and import of goods and services in GDP
1971–2007, Penn World Table version 6.3
TOR Central bank governor turnover rate,higher TOR indicates lower CBI
1971–2007
CBI Legal central bank independence, 0 to 1 index, higher value indicates higher CBI
1980–2003
Level of development Log of per capita GDP in USD 1971- 2007, Penn World Table version 6.3
Polity Proxy for institutional quality 1971–2007, Polity IV database
LGDP Log of GDP in dollar measuring size of the economy
WDI
Recession dummy Calculated by the author from log GDPOutput gap Calculated by the author from log GDP
using H-P filterCapital A/C Openness Capital account openness index (0 to 1),
higher values indicate higher levels of openness
1971–2007, Chinn and Ito (2008)
Neighbours’ Regime choice
Neighbours’ size weighted exchange rate regime, calculated by authors
Trade_neigh Neighbours’ trade share in overall trade,calculated by authors
DOTS (IMF)
152
Tab
le 2
:Ass
ocia
tion
betw
een
fisca
l var
iabl
es a
nd e
xpla
nato
ry v
aria
bles
Fixe
d re
gim
esIn
term
-ed
iate
regi
mes
Flex
ible
re
gim
esL
ow
open
ness
Med
ium
op
enne
ssH
igh
open
ness
Low
T
OR
(h
igh
CB
I)
Med
ium
T
OR
(m
ediu
m
CB
I)
Hig
h T
OR
(lo
w
CB
I)O
vera
ll bu
dget
bal
ance
-3.3
8-3
.68
-4.8
9-4
.17
-3.8
3-2
.83
-3.5
1-3
.85
-4.1
7(5
.28)
(6.3
4)(5
.51)
(4.3
7)(5
.21)
(8.7
0)(7
.21)
(5.5
2)(5
.02)
Prim
ary
budg
et b
alan
ce-0
.83
-1.0
8-1
.31
-1.7
1-1
.12
-0.1
9-0
.83
-1.1
6-1
.39
(4.4
7)(4
.79)
(5.2
1)(4
.26)
(4.4
8)(6
.06)
(5.1
8)(4
.42)
(4.8
1)C
ash
surp
lus
-1.3
7-1
.39
-3.1
3-2
.89
-1.7
4-0
.31
-1.4
1-1
.72
-1.9
0(5
.11)
(9.4
2)(4
.69)
(3.6
9)(5
.47)
(11.
27)
(11.
41)
(3.8
3)(4
.22)
Tota
l go
vern
men
t exp
endi
ture
30.3
331
.48
28.2
524
.70
32.0
537
.56
33.5
631
.44
28.7
1(1
2.27
)(1
3.36
)(1
2.88
)(9
.95)
(12.
40)
(13.
83)
(13.
87)
(12.
91)
(11.
82)
Prim
ary
gove
rnm
ente
xpen
ditu
re28
.30
28.7
824
.97
22.6
029
.10
33.8
830
.31
28.7
226
.15
(11.
35)
(10.
9)(1
1.81
)(9
.41)
(10.
88)
(11.
42)
(11.
12)
(11.
54)
(11.
21)
Gov
ernm
ent c
onsu
mpt
ion
expe
nditu
re16
.94
15.5
815
.90
13.4
316
.66
19.0
117
.20
16.2
014
.75
(6.3
9)(6
.37)
(6.7
9)(5
.08)
(6.9
0)(7
.45)
(6.4
1)(6
.27)
(6.1
1)N
ote:
For
eac
h va
riabl
e in
the
left
hand
side
col
umn,
the
figur
es in
the
firs
t row
show
s the
ave
rage
val
ue o
f tha
t var
iabl
e w
hile
figu
res i
n th
e pa
rent
hese
s ar
e st
anda
rd d
evia
tions
. Low
refe
rs to
25t
h pe
rcen
tile
valu
es o
r low
er, h
igh
mea
ns 7
5th
perc
entil
e va
lues
or h
ighe
r and
med
ium
is a
bove
25t
h pe
rcen
tile
and
belo
w 7
5th
perc
entil
e va
lues
. Thr
ee e
xcha
nge
rate
regi
mes
are
gen
erat
ed b
y co
llaps
ing
Rei
nhar
t-Rog
off 1
4 ca
tego
ry fi
ne c
lass
ifica
tion.
Fixe
d re
gim
es
incl
ude
the
hard
peg
s (ca
tego
ries 1
to 4
) of R
R c
lass
ifica
tion,
Fle
xibl
e re
gim
es in
clud
e fr
eely
floa
ting
and
free
ly fa
lling
(cat
egor
ies 1
3 an
d 14
), al
l oth
er
cate
gorie
s are
con
side
red
as in
term
edia
te re
gim
es.
153
Table 3: Average values of fiscal variables and exchange rate regime choice across regions
Regions Overall budget balance
Primary budget balance
Cash surplus
Regime choice
East Asia and Pacific -1.63(4.47)
0.30(4.87)
0.29(4.92)
7.50(4.25)
Eastern Europe and Central Asia -1.95(3.56)
0.48(3.25)
-1.34(4.79)
9.32(3.99)
Latin America -3.72(6.97)
-0.78(5.54)
-2.00(4.48)
7.91(4.59)
Middle East and North Africa -3.26(10.11)
-2.42(5.89)
-0.90(10.82)
6.91(3.98)
South Asia -6.46(4.09)
-3.89(4.35)
-2.97(3.08)
5.82(2.67)
Sub-Saharan Africa -4.82(5.36)
-2.49(5.25)
-2.61(4.84)
5.82(4.52)
Western Europe and North America -3.88(4.22)
-0.48(3.54)
-2.29(4.24)
6.51(4.10)
Caribbean -2.44(4.70)
0.07(3.95)
-1.30(2.80)
3.51(3.33)
Note: For each region figures in the first rows are average values of respective column variables, while figures in the parentheses are standard deviations.
154
Tab
le 4
:Int
erac
tion
effe
cts o
f exc
hang
e ra
te r
egim
e ch
oice
,and
CB
Iand
trad
e op
enne
ss o
n ov
eral
l bud
get b
alan
ceD
epen
dent
var
iabl
e: O
vera
ll bu
dget
bal
ance
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
OL
SO
LS
OL
SO
LS
OL
SO
LS
OL
SO
LS
OL
S
Reg
ime
-0.2
47**
-0.1
88-0
.184
-0.1
82-0
.227
-0.2
27*
-0.2
52*
-0.2
58**
-0.3
54*
(0.1
18)
(0.1
27)
(0.1
30)
(0.1
30)
(0.1
38)
(0.1
25)
(0.1
31)
(0.1
27)
(0.1
97)
CB
I1.
564
2.32
63.
383*
3.41
1*2.
036
2.86
71.
346
2.71
3-4
.780
(1.8
49)
(2.1
61)
(1.8
90)
(1.8
89)
(2.3
27)
(2.0
77)
(2.0
32)
(1.7
37)
(3.8
24)
Reg
ime*
CB
I-0
.182
-0.3
25-0
.370
*-0
.383
*-0
.281
-0.2
61-0
.105
-0.1
810.
206
(0.1
87)
(0.2
07)
(0.1
98)
(0.1
97)
(0.2
15)
(0.1
92)
(0.2
00)
(0.1
75)
(0.4
07)
Ope
n-0
.018
-0.0
16-0
.024
-0.0
23-0
.016
-0.0
14-0
.030
-0.0
33*
-0.0
14(0
.016
)(0
.021
)(0
.019
)(0
.019
)(0
.023
)(0
.020
)(0
.021
)(0
.019
)(0
.009
)R
egim
e*O
pen
0.00
4***
0.00
4**
0.00
5***
0.00
5***
0.00
4**
0.00
3**
0.00
5***
0.00
5***
0.00
4***
(0.0
01)
(0.0
02)
(0.0
02)
(0.0
02)
(0.0
02)
(0.0
02)
(0.0
02)
(0.0
02)
(0.0
01)
Com
mon
tim
e ef
fect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Cou
ntry
spec
ific
time
trend
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Oth
er c
ontro
lsN
one
Non
eR
eces
sion
du
mm
yO
utpu
tga
pPo
lity
Leve
l of
deve
lopm
ent
Cap
ital
A/C
open
ness
All
cont
rols
All
cont
rols
Obs
erva
tions
1,73
11,
731
1,71
91,
719
1,60
51,
731
1,62
11,
609
1,24
0R
-squ
ared
0.08
90.
124
0.16
90.
166
0.13
80.
177
0.18
50.
222
0.28
8N
o of
cou
ntrie
s97
9797
9789
9791
9167
Not
e: R
obus
t sta
ndar
d er
rors
clu
ster
ed a
t cou
ntry
leve
l in
pare
nthe
ses *
** p
<0.0
1, *
* p<
0.05
, * p
<0.1
. All
cont
rols
incl
ude
coun
tryre
cess
ion
dum
my,
leve
l of
deve
lopm
ent a
nd c
apita
l acc
ount
ope
nnes
s, al
l var
iabl
es e
xcep
t rec
essio
n du
mm
y in
col
umn
1 to
9 e
nter
with
a la
g. C
olum
ns 1
to 8
use
cen
tral b
ank
gove
rnor
TO
R w
hile
col
umn
9 us
es le
gal C
BI i
ndex
as m
easu
res o
f CB
I.
155
Tab
le 4
a:In
tera
ctio
n ef
fect
s of e
xcha
nge
rate
reg
ime
choi
ce a
nd tr
ade
open
ness
on
over
all b
udge
t bal
ance
Dep
ende
nt v
aria
ble:
Ove
rall
budg
etba
lanc
e(1
)(1
a)(1
b)(1
c)O
LS
2SL
SL
IML
Syst
emG
MM
Reg
ime
-0.3
27**
*-0
.392
***
-0.2
83-0
.354
**(0
.120
)(0
.145
)(0
.929
)(0
.159
)C
BI
1.23
01.
140
0.94
70.
318
(1.1
17)
(1.0
75)
(1.3
22)
(0.4
30)
Ope
n-0
.035
*-0
.044
**-0
.086
-0.0
41*
(0.0
20)
(0.0
22)
(0.0
79)
(0.0
24)
Reg
ime*
Ope
n0.
005*
**0.
006*
**0.
011
0.00
4**
(0.0
02)
(0.0
02)
(0.0
08)
(0.0
02)
Com
mon
tim
e ef
fect
Yes
Yes
Yes
Yes
Cou
ntry
spec
ific
time
trend
Yes
Yes
Yes
No
Oth
er c
ontro
lsA
llco
ntro
lsA
ll co
ntro
lsA
ll co
ntro
lsA
ll co
ntro
ls
Inst
rum
ents
Firs
tlag
of
expl
anat
ory
varia
bles
as
inst
rum
ents
Nei
ghbo
urs’
exch
ange
ra
te a
nd la
g of
end
o va
r as
inst
rum
ents
Inte
rnal
inst
rum
ents
Obs
erva
tions
1,60
91,
604
1538
1545
R-s
quar
ed0.
220
——
—N
o of
cou
ntrie
s91
9187
90In
stru
men
ts21
7H
anse
n te
st1
AR
(2) p
-val
0.46
Not
e: R
obus
t sta
ndar
d er
rors
clu
ster
ed a
t cou
ntry
leve
l in
pare
nthe
ses f
or a
ll re
gres
sion
s exc
ept s
yste
m G
MM
. Are
llano
-Bon
d (1
991)
robu
st st
anda
rd e
rror
s are
repo
rted
for s
yste
m G
MM
, ***
p<0
.01,
**
p<0.
05, *
p<0
.1. A
ll co
ntro
ls in
clud
e co
untry
rece
ssio
n du
mm
y, le
vel o
f dev
elop
men
t and
cap
ital a
ccou
nt o
penn
ess,
all v
aria
bles
ex
cept
rece
ssio
n du
mm
y in
col
umn
1 en
ter w
ith a
lag.
Coe
ffic
ient
of l
ag d
epen
dent
var
iabl
e is
not
repo
rted
for s
yste
m G
MM
resu
lts. A
ll re
gres
sion
s use
TO
R a
s a
mea
sure
of C
BI.
156
Tab
le 5
:Int
erac
tion
effe
cts o
f exc
hang
e ra
te r
egim
e ch
oice
and
trad
e op
enne
ss o
n ov
eral
l bud
get b
alan
ce (r
obus
tnes
s che
ck)
Dep
ende
nt v
aria
ble:
Ove
rall
budg
et b
alan
ce
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
OL
SO
LS
OL
SO
LS
OL
SO
LS
OL
SO
LS
OL
SO
LS
Reg
ime
-0.2
93**
-0.3
11-0
.330
**-0
.270
*-0
.344
***
-0.2
68**
-0.3
37**
*-0
.421
***
-0.4
27**
-0.6
59**
(0.1
20)
(0.1
88)
(0.1
36)
(0.1
43)
(0.1
22)
(0.1
12)
(0.1
24)
(0.1
00)
(0.1
24)
(0.2
14)
CB
I0.
880
0.00
11.
498
0.81
51.
080
0.90
81.
212
0.47
30.
317
1.59
8*(1
.242
)(1
.953
)(1
.225
)(1
.240
)(1
.114
)(1
.214
)(1
.138
)(0
.746
)(0
.805
)(0
.739
)O
pen
-0.0
44*
-0.0
23-0
.047
**-0
.013
-0.0
35*
-0.0
35*
-0.0
38*
-0.0
45**
-0.0
53**
-0.0
66(0
.025
)(0
.035
)(0
.023
)(0
.016
)(0
.021
)(0
.021
)(0
.021
)(0
.014
)(0
.020
)(0
.034
)R
egim
e*O
pen
0.00
5***
0.00
40.
007*
**0.
004*
**0.
005*
**0.
005*
**0.
006*
**0.
006*
**0.
006*
**0.
013*
*(0
.002
)(0
.005
)(0
.002
)(0
.001
)(0
.002
)(0
.002
)(0
.002
)(0
.001
)(0
.001
)(0
.005
)C
omm
on ti
me
effe
ctY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esC
ount
ry sp
ecifi
c tim
e tre
ndY
esY
esY
esY
esY
esY
esY
esN
oN
oN
oR
egio
n fix
ed e
ffec
tY
esY
esY
esR
egio
n tim
e tre
ndY
esY
esY
es
Oth
er c
ontro
lsA
llA
llA
llA
llA
llA
llA
llA
llA
llA
llO
mitt
ed o
bser
vatio
nO
ECD
om
itted
Dev
elop
ing
coun
tries
omitt
ed
SSA
omitt
edLa
tinA
mer
ica
omitt
ed
Car
ibbe
anom
itted
Mid
dle
East
omitt
ed
East
ern
Euro
peom
itted
Non
eO
ECD
omitt
edD
evel
opin
gco
untri
esom
itted
Obs
erva
tions
976
633
1,43
51,
271
1,53
51,
440
1,55
21,
609
976
633
R-s
quar
ed0.
271
0.22
50.
234
0.23
70.
220
0.23
80.
223
0.23
90.
272
0.41
4N
o of
cou
ntrie
s61
3077
7385
8379
——
—N
ote:
Rob
ust s
tand
ard
erro
rs c
lust
ered
at c
ount
ry le
vel i
n pa
rent
hese
s in
colu
mns
1 to
7, w
hile
stan
dard
err
ors a
re c
lust
ered
at r
egio
n le
vel i
n co
lum
ns 8
to 1
0, *
**
p<0.
01, *
* p<
0.05
, * p
<0.1
. All
cont
rols
incl
ude
coun
try re
cess
ion
dum
my,
leve
l of d
evel
opm
ent a
nd c
apita
l acc
ount
ope
nnes
s, al
l var
iabl
es e
xcep
t rec
essi
on d
umm
y en
ter w
ith a
lag.
All
regr
essi
ons u
se T
OR
as a
mea
sure
of C
BI.
157
Tab
le 6
:Int
erac
tion
effe
cts o
f exc
hang
e ra
te r
egim
e ch
oice
,CB
I and
trad
eop
enne
ss o
n pr
imar
y bu
dget
bal
ance
/cas
h su
rplu
sD
epen
dent
var
iabl
e: P
rim
ary
budg
et b
alan
ceD
epen
dent
var
iabl
e: C
ash
surp
lus
(1)
(2)
(2a)
(2b)
(2c)
(3)
(4)
(4a)
(4b)
(4c)
OLS
OLS
2SLS
LIM
LSy
stem
GM
MO
LSO
LS2S
LSLI
ML
Syst
em G
MM
Reg
ime
-0.2
25**
-0.2
73**
*-0
.333
***
-0.3
32-0
.402
**-0
.264
*-0
.279
**-0
.430
**-1
.884
-0.0
08(0
.093
)(0
.097
)(0
.117
)(0
.695
)(0
.166
)(0
.136
)(0
.136
)(0
.184
)(1
.410
)(0
.081
)C
BI
1.87
5*0.
855
0.76
50.
446
0.75
12.
292
-0.2
800.
026
0.04
6-0
.056
(1.0
59)
(0.7
50)
(0.7
18)
(0.9
32)
(0.5
25)
(1.4
41)
(0.8
00)
(0.7
91)
(0.0
30)
(0.3
61)
Reg
ime*
CB
I-0
.123
-0.2
33(0
.131
)(0
.158
)O
pen
-0.0
17-0
.019
-0.0
25*
-0.0
71-0
.032
-0.0
20-0
.015
-0.0
19-0
.370
0.00
9(0
.013
)(0
.013
)(0
.015
)(0
.054
)(0
.020
)(0
.013
)(0
.016
)(0
.021
)(0
.228
)(0
.007
)R
egim
e*O
pen
0.00
4***
0.00
4***
0.00
4***
0.01
0*0.
004*
*0.
005*
**0.
002*
0.00
3**
0.04
60.
001
(0.0
01)
(0.0
01)
(0.0
01)
(0.0
06)
(0.0
02)
(0.0
01)
(0.0
01)
(0.0
02)
(0.0
30)
(0.0
01)
Com
mon
tim
e ef
fect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Cou
ntry
time
trend
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Oth
er c
ontro
lsA
llco
ntro
lsA
llco
ntro
lsA
ll co
ntro
lsA
ll co
ntro
lsA
ll co
ntro
lsA
llco
ntro
lsA
llco
ntro
lsA
ll co
ntro
lsA
ll co
ntro
lsA
ll co
ntro
lsIn
stru
men
tsFi
rstl
ag o
f ex
plan
ator
yva
riabl
es a
sin
stru
men
ts
Nei
ghbo
urs’
exch
ange
ra
te a
nd
lag
of e
ndo
var a
s in
stru
men
ts
Inte
rnal
inst
rum
ents
Firs
tlag
of
expl
anat
ory
varia
bles
as
inst
rum
ents
Nei
ghbo
urs’
exch
ange
ra
te a
nd
lag
of e
ndo
var a
s in
stru
men
ts
Inte
rnal
inst
rum
ents
Obs
erva
tions
1,60
01,
497
1,49
314
3114
171,
678
1,67
81,
672
1,59
816
10R
-squ
ared
0.19
30.
331
——
—0.
255
0.49
4—
—-
No
of c
ount
ries
9690
9086
8995
9595
9091
Inst
rum
ents
217
277
Han
sen
test
11
AR
(2) p
-val
0.01
50.
419
Not
e: R
obus
t sta
ndar
d er
rors
clu
ster
ed a
t cou
ntry
leve
l in
pare
nthe
ses f
or a
ll re
gres
sion
s exc
ept s
yste
m G
MM
. Are
llano
-Bon
d (1
991)
robu
st st
anda
rd e
rror
s are
repo
rted
for s
yste
m
GM
M, *
** p
<0.0
1, *
* p<
0.05
, * p
<0.1
. All
cont
rols
incl
ude
coun
try re
cess
ion
dum
my,
leve
l of d
evel
opm
ent a
nd c
apita
l acc
ount
ope
nnes
s, al
l var
iabl
es e
xcep
t rec
essi
on d
umm
y in
col
umn
1, 2
, 3 a
nd 4
ent
er w
ith a
lag.
Coe
ffic
ient
of l
ag d
epen
dent
var
iabl
es a
re n
ot re
porte
d fo
r sys
tem
GM
M re
sults
. All
regr
essi
ons u
se T
OR
as a
mea
sure
of C
BI.
158
Tab
le 7
:Int
erac
tion
effe
cts o
f exc
hang
e ra
te r
egim
e ch
oice
and
trad
e op
enne
ss o
n pr
imar
y bu
dget
bal
ance
(rob
ustn
ess c
heck
)D
epen
dent
var
iabl
e: P
rim
ary
budg
et b
alan
ce
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
OL
SO
LS
OL
SO
LS
OL
SO
LS
OL
SO
LS
OL
SO
LS
Reg
ime
-0.2
77**
-0.3
26**
-0.3
02**
*-0
.312
***
-0.2
87**
*-0
.202
**-0
.278
***
-0.2
64*
-0.2
70-0
.275
(0.1
23)
(0.1
40)
(0.1
09)
(0.1
15)
(0.0
99)
(0.0
92)
(0.1
02)
(0.1
39)
(0.1
87)
(0.2
10)
CB
I0.
982
-0.3
690.
924
0.73
50.
750
0.51
10.
817
0.93
31.
255
-0.1
12(0
.941
)(1
.354
)(0
.813
)(0
.740
)(0
.796
)(0
.803
)(0
.773
)(0
.548
)(0
.732
)(1
.275
)O
pen
-0.0
25-0
.017
-0.0
27*
-0.0
13-0
.020
-0.0
17-0
.019
-0.0
21-0
.025
*-0
.005
(0.0
16)
(0.0
17)
(0.0
14)
(0.0
16)
(0.0
14)
(0.0
14)
(0.0
14)
(0.0
12)
(0.0
12)
(0.0
21)
Reg
ime*
Ope
n0.
004*
**0.
005*
0.00
5***
0.00
4**
0.00
4***
0.00
4***
0.00
4***
0.00
4**
0.00
4**
0.00
6(0
.001
)(0
.003
)(0
.001
)(0
.001
)(0
.001
)(0
.001
)(0
.001
)(0
.001
)(0
.001
)(0
.003
)C
omm
on ti
me
effe
ctY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esC
ount
ry sp
ecifi
c tim
e tre
ndY
esY
esY
esY
esY
esY
esY
esN
oN
oN
oR
egio
n Fi
xed
Effe
ct—
——
——
——
Yes
Yes
Yes
Reg
ion
Tim
e Tr
end
——
——
——
—Y
esY
esY
es
Oth
er C
ontro
lsA
llA
llA
llA
llA
llA
llA
llA
llA
llA
llO
mitt
ed o
bser
vatio
nO
ECD
om
itted
Dev
elop
ing
coun
tries
omitt
ed
SSA
omitt
edLa
tin
Am
eric
a om
itted
Car
ibbe
anO
mitt
edM
iddl
e ea
stom
itted
East
ern
Euro
pe
omitt
ed
Non
eO
ECD
omitt
edD
evel
opin
gco
untri
esO
mitt
edO
bser
vatio
ns87
562
21,
356
1,17
11,
433
1,35
21,
440
1,49
787
562
2R
-squ
ared
0.36
90.
295
0.31
50.
381
0.33
40.
336
0.32
20.
319
0.36
00.
384
No
of c
ount
ries
6030
7672
8483
78—
——
Not
e: R
obus
t sta
ndar
d er
rors
clu
ster
ed a
t cou
ntry
leve
l in
pare
nthe
ses i
n co
lum
ns 1
to7,
whi
le st
anda
rd e
rror
s are
clu
ster
ed a
t reg
ion
leve
l in
colu
mns
8 to
10,
***
p<0
.01,
**
p<0
.05,
* p
<0.1
. All
cont
rols
incl
ude
rece
ssio
n du
mm
y, le
vel o
f dev
elop
men
t and
cap
ital a
ccou
nt o
penn
ess,
all v
aria
bles
exc
ept r
eces
sion
dum
my
ente
r with
a la
g. A
ll re
gres
sion
s use
TO
R a
s a m
easu
re o
f CB
I.
159
Tab
le 8
:Int
erac
tion
effe
cts o
f exc
hang
e ra
te r
egim
e ch
oice
and
trad
e op
enne
ss o
n ca
sh su
rplu
s (ro
bust
ness
che
ck)
Dep
ende
nt v
aria
ble:
Cas
h su
rplu
s
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
OL
SO
LS
OL
SO
LS
OL
SO
LS
OL
SO
LS
OL
SO
LS
Reg
ime
-0.3
65**
*-0
.565
**-0
.397
***
-0.2
52-0
.345
***
-0.2
89**
-0.3
21**
-0.3
20*
-0.3
78**
*-0
.716
*(0
.128
)(0
.231
)(0
.121
)(0
.152
)(0
.122
)(0
.110
)(0
.128
)(0
.140
)(0
.103
)(0
.312
)C
BI
0.27
0-0
.297
0.62
71.
067
0.19
70.
011
0.28
10.
290
0.23
01.
350
(1.1
87)
(1.2
97)
(0.9
45)
(1.1
29)
(0.9
68)
(0.9
81)
(0.9
98)
(0.7
23)
(1.1
85)
(0.6
74)
Ope
n-0
.021
-0.0
78**
-0.0
31**
-0.0
12-0
.022
-0.0
22-0
.022
-0.0
19-0
.016
-0.0
82(0
.016
)(0
.036
)(0
.012
)(0
.017
)(0
.014
)(0
.014
)(0
.014
)(0
.012
)(0
.016
)(0
.048
)R
egim
e*O
pen
0.00
5***
0.00
7**
0.00
6***
0.00
4***
0.00
5***
0.00
5***
0.00
5***
0.00
4***
0.00
4***
0.01
4(0
.001
)(0
.003
)(0
.001
)(0
.001
)(0
.001
)(0
.001
)(0
.001
)(0
.001
)(0
.001
)(0
.007
)C
omm
on ti
me
effe
ctY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esC
ount
ry sp
ecifi
c tim
e tre
ndY
esY
esY
esY
esY
esY
esY
esN
oN
oN
oR
egio
n fix
ed e
ffec
tY
esY
esY
esR
egio
n tim
e tre
ndY
esY
esY
es
Oth
er c
ontro
lsA
llA
llA
llA
llA
llA
llA
llA
llA
llA
llO
mitt
ed o
bser
vatio
nO
ECD
om
itted
Dev
elop
ing
coun
tries
omitt
ed
SSA
omitt
edLa
tin
Am
eric
a om
itted
Car
ibbe
anom
itted
Mid
dle
East
omitt
ed
East
ern
Euro
pe
omitt
ed
Non
eO
ECD
omitt
edD
evel
opin
gco
untri
esom
itted
Obs
erva
tions
1,01
366
51,
555
1,36
21,
607
1,49
11,
494
1,67
81,
013
665
R-s
quar
ed0.
304
0.58
90.
270
0.26
50.
258
0.28
50.
264
0.26
80.
348
0.37
9N
o of
cou
ntrie
s63
3283
8090
8776
--
-N
ote:
Rob
ust s
tand
ard
erro
rs c
lust
ered
at c
ount
ry le
vel i
n pa
rent
hese
s in
colu
mns
1 to
7, w
hile
stan
dard
err
ors a
re c
lust
ered
at r
egio
n le
vel i
n co
lum
ns 8
to 1
0, *
** p
<0.0
1,
** p
<0.0
5, *
p<0
.1. A
ll co
ntro
ls in
clud
e re
cess
ion
dum
my,
leve
l of d
evel
opm
ent a
nd c
apita
l acc
ount
ope
nnes
s, al
l var
iabl
es e
xcep
t rec
essi
on d
umm
y en
ter w
ith a
lag.
All
regr
essi
ons u
se T
OR
as a
mea
sure
of C
BI.
160
Tab
le 9
:Int
erac
tion
effe
cts o
f exc
hang
e ra
te r
egim
e ch
oice
and
CB
Iand
trad
e op
enne
sson
tota
l/pri
mar
y go
vern
men
t exp
endi
ture
D
epen
dent
var
iabl
e:T
otal
gov
t exp
endi
ture
Dep
ende
nt v
aria
ble:
Pri
mar
y go
vt e
xpen
ditu
re(1
)(2
)(2
a)(2
b)(2
c)(3
)(4
)(4
a)(4
b)(4
c)O
LS
OL
S2S
LS
LIM
LSy
stem
GM
MO
LS
OL
S2S
LS
LIM
LSy
stem
GM
MR
egim
e 0.
765*
0.94
6**
1.15
4**
4.02
60.
590*
0.62
60.
790*
*0.
966*
*1.
546
0.23
3*(0
.417
)(0
.369
)(0
.454
)(7
.447
)(0
.307
)(0
.403
)(0
.362
)(0
.440
)(4
.376
)(0
.135
)C
BI
-10.
771*
**-6
.841
**-6
.914
**-5
.517
-1.0
90-9
.856
**-6
.363
**-6
.331
**-6
.691
*-0
.585
(4.0
75)
(2.9
00)
(2.7
68)
(4.8
68)
(1.0
06)
(3.8
92)
(2.6
59)
(2.5
29)
(3.8
09)
(0.4
67)
Reg
ime*
CB
I0.
477
0.42
0(0
.455
)(0
.430
)O
pen
0.18
2***
0.18
9***
0.21
9***
0.95
40.
071*
0.14
9***
0.15
5***
0.18
0***
0.52
40.
021
(0.0
56)
(0.0
56)
(0.0
63)
(1.1
20)
(0.0
40)
(0.0
53)
(0.0
54)
(0.0
59)
(0.4
85)
(0.0
19)
Reg
ime*
Ope
n-0
.016
***
-0.0
17**
*-0
.020
***
-0.1
00-0
.007
*-0
.014
***
-0.0
14**
*-0
.017
***
-0.0
53-0
.002
(0.0
05)
(0.0
05)
(0.0
06)
(0.1
18)
(0.0
04)
(0.0
05)
(0.0
05)
(0.0
06)
(0.0
50)
(0.0
02)
Com
mon
tim
e ef
fect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Cou
ntry
tim
e tre
ndY
esY
esY
esY
esN
oY
esY
esY
esY
esN
o
Oth
er c
ontro
lsA
llco
ntro
lsA
llco
ntro
lsA
ll co
ntro
lsA
ll co
ntro
lsA
ll co
ntro
lsA
llco
ntro
lsA
llco
ntro
lsA
ll co
ntro
lsA
ll co
ntro
lsA
ll co
ntro
ls
Inst
rum
ents
Firs
tlag
of
expl
anat
ory
varia
bles
as
inst
rum
ents
Nei
ghbo
urs’
exch
ange
ra
te a
nd
lag
of e
ndo
var a
s in
stru
men
ts
Inte
rnal
inst
rum
ents
Firs
tlag
of
expl
anat
ory
varia
bles
as
inst
rum
ents
Nei
ghbo
urs’
exch
ange
ra
te a
nd
lag
of e
ndo
var a
s in
stru
men
ts
Inte
rnal
inst
rum
ents
Obs
erva
tions
1,62
81,
628
1,62
315
5715
631,
499
1,49
91,
495
1433
1419
R-s
quar
ed0.
321
0.31
9—
——
0.30
60.
305
——
—N
o of
cou
ntrie
s91
9191
8690
9090
9086
89In
stru
men
ts21
721
7H
anse
n te
st1
1A
R(2
) p-v
al0.
680.
080
Not
e: R
obus
t sta
ndar
d er
rors
clu
ster
ed a
t cou
ntry
leve
l in
pare
nthe
ses f
or a
ll re
gres
sion
s exc
ept s
yste
m G
MM
. Are
llano
-Bon
d (1
991)
robu
st st
anda
rd e
rror
s are
repo
rted
for s
yste
m
GM
M, *
** p
<0.0
1, *
* p<
0.05
, * p
<0.1
. Con
trols
are
rece
ssio
n du
mm
y, le
vel o
f dev
elop
men
t and
cap
ital a
ccou
nt o
penn
ess,
all v
aria
bles
exc
ept r
eces
sion
dum
my
colu
mn
in 1
, 2, 3
an
d 4
ente
r with
a la
g. C
oeff
icie
nts o
f lag
dep
ende
nt v
aria
bles
are
not
repo
rted
for s
yste
m G
MM
resu
lts. A
ll re
gres
sion
s use
TO
R a
s a m
easu
re o
f CB
I.
161
Table 10: Interaction effects of exchange rate regime choice and CBI and trade openness on government consumption expenditure
Dependent variable: Government consumption expenditure
(1) (2) (2a) (2b) (2c)OLS OLS 2SLS LIML System
GMMRegime 0.321* 0.298* 0.358** 3.011 0.155*
(0.169) (0.153) (0.180) (3.263) (0.082)CBI -2.532* -3.120*** -3.121*** -4.108* -0.523
(1.393) (0.858) (0.852) (2.258) (0.333)Regime* CBI -0.069
(0.148)Open 0.051*** 0.050*** 0.056*** 0.354 0.017*
(0.018) (0.018) (0.020) (0.370) (0.009)Regime* Open -0.005*** -0.005** -0.006*** -0.040 -0.002*
(0.002) (0.002) (0.002) (0.043) (0.001)Common time effect Yes Yes Yes Yes YesCountry specific time trend
Yes Yes Yes Yes No
Other controls Allcontrols
Allcontrols
All controls
All controls
All controls
Instruments First lag of explanatoryvariables asinstruments
Neighbours’exchange rate and lag of endo var as instruments
Internalinstruments
Observations 3,374 3,374 3,357 3249 3396R-squared 0.242 0.242 - - -No of countries 126 126 126 120 126Instruments 284Hansen test 1AR(2) p-val 0.734Note: Robust standard errors clustered at country level in parentheses for all regressions except system GMM. Arellano-Bond (1991) robust standard errors are reported for system GMM, *** p<0.01, ** p<0.05, * p<0.1. Controls are recession dummy, level of development and capital account openness, all variables except recession dummy in column 1and 2 enter with a lag. Coefficient of lag dependent variable is not reported for system GMM results. All regressions use TOR as a measure of CBI.
162
Figure 1: Exchange rate regime choice and overall budget balance conditional
on trade openness
1a: Trade openness low
1b: Trade openness medium
1c: Trade openness high
-20
-10
010
Ove
rall
Bud
get B
alan
ce
0 5 10 15Regime Choice
-20
-10
010
20O
vera
ll B
udge
t Bal
ance
0 5 10 15Regime Choice
-20
-10
010
20O
vera
ll B
udge
t Bal
ance
0 5 10 15Regime Choice
Figure 2: Exchange rate regime choice and overall budget balance
conditional on CBI
2a: Governor TOR high (low CBI)
2b: Governor TOR medium (medium CBI)
2c: Governor TOR low (high CBI)
-20
-10
010
Ove
rall
Bud
get B
alan
ce
0 5 10 15Regime Choice
-20
-10
010
20O
vera
ll B
udge
t Bal
ance
0 5 10 15Regime Choice
-20
-10
010
20O
vera
ll B
udge
t Bal
ance
0 5 10Regime Choice
163
Figure 3: Marginal effect of exchange rate regime choice on overall budget balance conditional on trade openness
Note: This graph is based on regression eight in Table 4, when both interaction between exchange rate regime choice and CBI (TOR) and trade openness are included.
Figure 4: Marginal effect of exchange rate regime choice on overall budget balance conditional on level of CBI (Governor TOR)
Note: This graph is based on regression eight in Table 4, when both interactions between exchange rate regime choice and CBI (TOR) and trade openness are included.
-2-1
.8-1
.6-1
.4-1
.2-1
-.8-.6
-.4-.2
0.2
.4.6
.81
1.2
1.4
1.6
1.8
2
Mar
gina
l Effe
ct o
f Reg
ime
Cho
ice
0 10 20 30 40 50 60 70 80 90 100110120130140150160170180190200210220230240250260270280290300
Level of Trade Opennness
-1-.9
-.8-.7
-.6-.5
-.4-.3
-.2-.1
0.1
.2.3
.4.5
.6.7
.8.9
1
Mar
gina
l Effe
ct o
f Reg
ime
Cho
ice
0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
Level of Governor Turnover Rate
164
Figure 5: Marginal effect of exchange rate regime choice on overall budget balance conditional on trade openness
Note: This graph is based on regression one in Table 4a, when only interaction between exchange rate regime choice and trade openness is included.
Figure 6: Marginal effect of exchange rate regime choice on primary budget balance conditional on trade openness
Note: This graph is based on regression two in Table 6, when only interaction between exchange rate regime choice and trade openness is included.
-2-1
.8-1.6
-1.4-
1.2-1
-.8-.6
-.4-.2
0.2
.4.6
.81
1.21.4
1.61.8
2
Marg
inal E
ffect
of Re
gime C
hoice
0 10 20 30 40 50 60 70 80 90 100110120130140150160170180190200210220230240250260270280290300
Trade Openness
-2-1
.8-1.
6-1.4-
1.2
-1-.8
-.6-.4
-.20
.2.4
.6.8
11.
21.4
1.61
.82
Mar
gina
l Effe
ct o
f Reg
ime
Choi
ce
0 10 20 30 40 50 60 70 80 90 100110120130140150160170180190200210220230240250260270280290300
Trade Openness
165
Figure 7: Marginal effect of exchange rate regime choice on cash surplus conditional on trade openness
Note: This graph is based on regression four in Table 6, when only interaction between exchange rate regime choice and trade openness is included.
Figure 8: Marginal effect of exchange rate regime choice on total government expenditure conditional on trade openness
Note: This graph is based on regression two in Table 9, when only interaction between exchange rate regime choice and trade openness is included.
-2-1
.8-1.
6-1.4-
1.2
-1-.8
-.6-.4
-.20
.2.4
.6.8
11.
21.4
1.61
.82
Mar
gina
l Effe
ct o
f Reg
ime
Choi
ce
0 10 20 30 40 50 60 70 80 90 100110120130140150160170180190200210220230240250260270280290300
Trade Openness
-6-5
.5-5
-4.5
-4-3
.5-3
-2.5
-2-1
.5-1
-.50
.51
1.5
22.
53
3.5
44.
55
5.5
6
Mar
gina
l Effe
ct o
f Reg
ime
Cho
ice
0 10 20 30 40 50 60 70 80 90 100110120130140150160170180190200210220230240250260270280290300
Trade Openness
166
Chapter 5: Conclusion
Exchange rate regime has always been at the heart of the contemporary history of
the International Monetary System (IMS). Factors determining exchange rate regime
choice and the macroeconomic consequences of that choice have been fundamental,
yet controversial, issues in the post-Bretton Woods (BW) era of IMS. This thesis has
focused on some regime choice determinants that are theoretically important, yet
have received scant attention in empirical studies on exchange rate regime choice
during the post-BW period.
More specifically, this thesis examined the role of capital account openness and
financial sector health as crucial determinants of regime choice. In so doing,
persistence of regime choice over time was also examined. Moreover, the thesis
explored the role of product diversification and how its interactions with the three
mechanisms of shock absorption, financial development and rent-seeking affect
exchange rate regime choice. Further, this thesis also investigated the effect of
exchange rate regime choice on fiscal discipline—an important and timely issue,
given the high and persistent budget deficits affecting many countries in recent
years. To explore this issue, it is stressed that exchange rate has both direct effects
and interaction effects on fiscal discipline, through trade openness and CBI.
Given the limitations of the IMF de jure exchange rate regime classification, the
thesis adopted the Reinhart-Rogoff de facto or actual exchange rate classification in
conjunction with a post-BW panel dataset for 178 countries covering the period
from 1971 to 2007. Advanced econometric and statistical techniques were used to
explore the identified issues of regime choice and its consequences.
167
The first essay of the thesis finds strong evidence that exchange rate regime choice
is relatively persistent over time. This suggests that previous years’ exchange rate
regimes are a good predictor for the current year’s regime choice. It was also found
that persistence of regime choice is particularly pertinent for low-income and high-
income countries. Interestingly, persistence is relatively low for middle-income
countries, which suggests that this group of countries changes their exchange rate
regimes relatively more often. Capital account openness appears to be a significant
factor affecting exchange rate regime choice, such that higher capital account
openness leads countries to adopt relatively fixed regimes. This finding applies to all
country groups and is robust across different estimation techniques. The implication
of this finding is that, as countries open their capital account, they prefer stability in
their exchange rates.
There is also some evidence that financial sector development affects regime choice.
This effect is such that financial sector development leads countries to use more
flexible regimes—particularly in high-income and middle-income countries. Thus,
countries with underdeveloped financial sectors are not good candidates for floating
exchange rates. However, this finding is not always robust across estimation
techniques and country groups. The effects of another measure of financial sector
health—namely, financial fragility—on regime choice are not robust across different
country groups and estimation techniques.
The second essay establishes that product diversification is an important determinant
of exchange rate regime choice, in that more diversified economies tend to adopt
more flexible regimes. Product diversification can thus act as a factor to mitigate the
fear of floating that is common among countries. There is also strong evidence that
lower levels of diversification (i.e., concentration) leads to fixed regimes in
168
countries where the level of corruption is high. This may create more scope for rent-
seeking on the part of powerful elites in the developing countries for which this
finding is more applicable. Product diversification is also more likely to lead to
flexible exchange rate regimes in developing countries with lower levels of financial
development. There is also some evidence that diversification facilitates the
adoption of flexible regimes in countries that are experiencing greater shocks.
Overall, the findings are more applicable for those developing countries that
experience high corruption, have low levels of financial development, and face high
real shocks.
Finally, the third essay finds strong evidence that the exchange rate regime choice
has both direct and indirect effects on fiscal discipline. Particularly, this essay finds
that regime choice interacts with trade openness to affect fiscal discipline. This
suggests that a regime choice’s effects on fiscal discipline depend on the levels of a
country’s trade openness. This finding is further explored by estimating the marginal
effects of regime choice on fiscal discipline at different levels of trade openness. It is
found that fixed exchange rate regimes create fiscal discipline when a country’s
trade openness is lower. Conversely, flexible regimes become disciplinary when a
country is highly open to trade. This study does not find any evidence that regime
choice interacts with CBI to influence fiscal discipline. Nor does it find robust
evidence that CBI has any direct effect on fiscal discipline. These findings are
consistent across different measures of fiscal outcomes and CBI, and are supported
by several robustness checks.