oxcarre research paper 133 dutch disease and the oil and ... · 1 1 introduction commodity prices...

DEPARTMENT OF ECONOMICS OxCarre Oxford Centre for the Analysis of Resource Rich Economies

Manor Road Building, Manor Road, Oxford OX1 3UQ Tel: +44(0)1865 281281 Fax: +44(0)1865 271094

[email protected] www.oxcarre.ox.ac.uk

Direct tel: +44(0) 1865 281281 E-mail: [email protected]

_

OxCarre Research Paper 133

Dutch Disease and the Oil and Boom and Bust

Brock Smith

OxCarre

mailto:[email protected]

http://www.oxcarre.ox.ac.uk/

Dutch Disease and the Oil and Boom and Bust

Brock Smith

University of California, Davis

[email protected]

ABSTRACT

This paper examines the impact of the oil price boom in the 1970s and the subsequent bust on

non-oil economic activity in oil-dependent countries. During the boom, manufacturing value

added and exports increased significantly relative to non-oil dependent countries, along with

wages, employment and investment. These measures decreased, though to a lesser extent,

during the bust, displaying a positive relationship with oil prices. In contrast with the

Dutch Disease model, exportable manufacturing sectors grew faster than non-exportable

ones. However, exports of non-hydrocarbon natural resources and agricultural products

displayed a strongly negative relationship to prices. The results suggest a push towards

industrialization induced by the oil revenue windfall.

1

1 Introduction

Commodity prices have long been notable for their volatility, creating special economic

circumstances in major producing countries. One commonly studied relationship in such

countries is the effect of natural resource booms on non-resource economic activity. The

literature on this topic has almost exclusively centered on the concept of “Dutch Disease”,

a term coined by The Economist magazine in 1977 referring to the Netherlands gas boom.

The Dutch Disease hypothesis posits that a boom in the natural resource sector shrinks

the manufacturing sector through crowding out and an appreciation of the real exchange

rate. Krugman (1987), Matsuyama (1992) and Torvik (2001) have argued that this could

have harmful long-term economic consequences, arguing that growth is largely driven by

learning-by-doing in the manufacturing and agricultural sectors. This mechanism has been

cited as a possible culprit for the “resource curse” found by Sachs & Warner (1995) and

several subsequent papers, although the existence of the curse has been called into question

by a few recent papers including Brunnschweiler & Bulte (2008), Alexeev & Conrad (2009),

and Smith (2013).

The core Dutch Disease theoretical model, provided by Corden & Neary (1982) and

Corden (1984), posits a booming traded sector (assumed to be natural resources), a non-

booming traded sector (manufacturing), and a non-traded sector (services). A boom in the

resource sector can take three forms: a technology-induced rise in productivity, a windfall

discovery, or a rise in the world price. For any of these cases, the model distinguishes two

separate effects on manufacturing. The resource movement effect refers to mobile factors of

production migrating to the booming sector away from other sectors. The spending effect

occurs as the extra income from the booming sector is spent on the non-traded sector, thus

raising the price of the non-traded sector relative to the non-booming traded sector, or the

real exchange rate, and further drawing resources away from the non-booming traded sector.

2

In the simplest form of the model with only labor as a mobile factor, a resource boom

unambiguously predicts a decrease in output of the non-booming traded sector, or “de-

industrialization”. However, making more than one factor mobile between sectors can pro-

duce ambiguous implications for manufacturing. Further, Corden & Neary (1982) and Cor-

den (1984) point out real-world factors beyond the core model that could lead to a resource

boom being a boon for manufacturing. Since the increase in income will presumably be

taxed, the government may invest the extra revenue in other sectors or in infrastructure that

raises productively generally. If there is a non-booming traded sector besides manufacturing,

say agriculture, then that sector could suffer the brunt of the negative effects. Perhaps most

pertinently, since manufacturing will contain both traded and non-traded sub-sectors, the

spending effect may be a net positive for manufacturing as a whole.

Thus, even accepting the core model, the question of how a resource boom affects

manufacturing is ultimately an empirical one. In this paper, I evaluate the effect of the

1970s oil price boom and subsequent bust on manufacturing levels of oil and gas-dependent

countries. The methodology is adapted from Black et al (2005), which evaluates various

economic outcomes for coal producing counties in the United States during the similarly

timed coal boom and bust 1. The 1973 and 1979 energy crises caused world real oil prices

to more than quintuple between 1973 and 1980, creating an enormous surge in revenues

for oil-exporting countries. However, over the following eight years prices dropped back

to nearly their pre-1973 levels. Since individual countries generally do not considerably

influence the world oil price, these price shocks constitute a plausibly exogenous source of

resource revenues and thus a suitable test of the Dutch Disease hypothesis. The roughly

symmetric nature of the boom-and-bust allows us to test if price increases and decreases

have symmetric effects on manufacturing, and also reduces the possibility that the observed

effects are due to pre-existing trends. Further, the period of roughly flat prices that followed

1This paper had inconclusive results for the effects on the manufacturing sector

3

the boom and bust can arguably be interpreted as a placebo test.

I find that manufacturing value added and exports are strongly pro-cyclical with respect

to oil prices during the boom and bust period. I find, with varying levels of significance,

that wages, employment, capital formation, and productivity are also pro-cyclical. The

finding that wages and output jointly increased during the boom runs counter to the core

model. However, this is reconciled by the finding that capital formation and productivity

likewise increased, suggesting investment spillovers into manufacturing. I further analyze the

performance of different manufacturing subsectors, focusing on differences between oil-linked

and non-oil linked sectors and tradable vs. non-tradable sectors. I find that subsectors of

all types experienced pro-cyclical growth. Unsurprisingly, downstream sectors consisting of

refined products experienced especially strong growth during the boom. However, a counter-

intuitive finding is that more tradable sectors (with tradability defined as exports divided by

value added) experienced higher growth during the boom than other sectors. These sectors

likewise saw vastly higher capital formation growth, suggesting that the investment spillovers

specifically targeted tradable sectors.

Other results are indicative of the presence of Dutch Disease mechanisms. Oil-rich

countries experienced excess exchange rate appreciation during the boom period. Imports of

all types of goods were strongly and positively associated with oil prices. Most pertinently,

exports of agricultural products and non-hydrocarbon natural resources are negatively asso-

ciated with oil prices, indicating a shift towards industrialization.

Overall, income gains from the oil boom brought about a substantial rise in general

manufacturing activity in oil-producing nations. The results suggest two complementary

explanations: first, the strongly positive effects of prices on imports across all types of

sectors imply a positive shock to local demand for all types of goods, and this demand was met

through a combination of imports and local production. This mechanism does not contradict

the core theoretical model, but is simply a consequence of the fact that manufacturing and the

4

“non-booming traded sector” of the model are not one in the same, as a substantial portion

of manufactured goods are indeed locally consumed, even in poor countries. Second, part

of the oil revenue windfall was reinvested in manufacturing, possibly through taxation and

state-directed investment, and with an emphasis on exportable sectors that may be seen as

higher-growth opportunities. This mechanism is not present in the theoretical model, but as

mentioned above is discussed in Corden (1982) as a possible countering factor. Additionally,

the export results suggest that Dutch Disease mechanisms do appear to be present, but

are borne primarily by the agriculture and non-oil resource sectors. The results altogether

are indicative of the resource windfall spurring a push towards industrialization, whether

state-directed or otherwise.

The empirical literature on Dutch Disease has traditionally been far less developed than

the theoretical literature. Gelb (1988) and Sala-i-Martin (2003) use a case study approach,

but do not find evidence of oil revenues harming the manufacturing sector. Stijns (2003) finds

negative effects of oil revenues on manufacturing exports using a bilateral gravity model.

However, a number of empirical Dutch Disease papers have been produced more recently.

Caselli & Michaels (2009) exploit differences in oil endowment across Brazilian municipali-

ties, finding little effect on non-oil GDP. Allcott & Keniston (2013) use an approach similar

to this paper and Black et al (2005) and finds that oil booms increase manufacturing out-

put in oil-rich United States counties. In the cross-country literature, Rajan & Subramanian

(2011) considers aid rather than resources as the windfall, and finds that aid-dependent coun-

tries experience slower growth in tradable manufacturing sectors relative to non-tradeable

sectors during the 1980s and 1990s. Ismail (2011) and Harding & Venables (2013) both

find a negative relationship between price movements and manufacturing value added and

exports among oil-exporting countries. The results of the latter two papers contrast with

this paper, but use a different source of resource windfall variation. Both rely on within-

country variation in resource wealth among resource-rich countries, whereas this paper takes

5

a quasi-experimental treatment-control approach, comparing oil-rich countries to non-oil-rich

countries in the presence of oil price shocks. One problem with the within-country variation

approach is that variation stems from both price and production, but production levels are

internal to countries and may be endogenously determined. This paper relies on external

price shocks that affect countries with significant production prior to the shock.

The rest of this paper proceeds as follows: Section Two briefly describes the oil boom

and bust and its causes. Section Three outlines my empirical strategy and describes the

data. Section Four presents and discusses all empirical results. Section Five concludes.

2 The Oil Boom and Bust

Prior to the 1970s oil had long been abundant and relatively cheap. The era of low, stable oil

prices came to an end in 1973, when Egypt and Syria launched a surprise attack on Israel.

When the United States provided Israel with substantial military aid, the Organization

of Arab Petroleum Exporting Countries (OAPEC) declared an oil embargo to punish the

U.S. and other rich nations supporting Israel. Oil prices quadrupled between 1973 and

1974, and although the actual embargo only lasted for six months, prices remained high

as fears of unreliable supply gripped the industrialized world. Prices soared again in 1979

as a result of the political fallout from the Iranian Revolution. Starting in 1981, oil prices

crashed for several years due to a combination of oversupply, diminished economic activity in

industrialized nations and a shift to alternative energy sources. After prices finally bottomed

out in 1986, they once again remained relatively low and stable until the mid-2000s.

Figure 1 shows a graph of oil prices from 1962 to 1995, with vertical lines representing

the boom and bust periods. The boom period is defined as 1974-1980, The bust period is

defined as 1981-1988, and the years 1989-1995 are classified as “valley” years. Although

prices are not going uniformly up or down during these periods, the idea is to examine non-

6

oil outcomes during long periods of upheaval in the global oil market. This helps to mitigate

the issue of uncertain time frames of Dutch Disease mechanisms-if effects on manufacturing

and other non-oil sectors take longer than one year to manifest, this is not be captured in

studies using year-to-year changes in prices.

Figure 2 shows the average value of oil production among treatment countries (defined

in the following section). Oil earnings increase and decline in a very similar way to prices

themselves, reinforcing that these countries were subjected to dramatic earnings shocks due

to world price movements during the period studied, rather than any adjustments in output.

Figure 1: Oil price and boom/bust periods

020

4060

80re

aloi

lpric

e

1960 1970 1980 1990 2000year

Figure 2: 1972 Oil Dependence Histogram

0.2

.4.6

.81

Den

sity

5 10 15 20+ 1972 Hydrocarbon Dependence

This paper’s strategy is to exploit the boom and bust as a test of a price-driven resource

shock’s effect on the non-resource sectors. If the effects were positive during the boom

7

years and negative during the bust years, it would be evidence of positive spillovers of the

resource sector, and conversely for negative spillovers. If the effect were positive or negative

only during one of the boom or bust periods but not the other, it would be evidence of

asymmetric price effects. If effects are negative or positive during both boom and bust, it

could be evidence of spurious or no effect of price. The “valley” period can arguably be

interpreted as a kind of placebo test; since prices are not significantly changing one way

or the other, we should not expect significant differences between treatment and control

countries unless some other mechanism is present.

3 Data and Methodology

Following Black et al (2005), I use the following equation to assess the impact of the boom

and bust on several outcomes:

Δln(Ycrt) = β1(Tc ∗Boomt) + β2(Tc ∗Bustt) + β3(Tc ∗ V alleyt) + γrt + εct (1)

Where Ycrt is the outcome of interest for country c in region r on year t. The dependent

variable is thus the year-on-year growth rate of the outcome Ycrt. Tc is an indicator for

being in the treatment group (defined below), Boomt is an indicator for being between the

years 1974 and 1980, Bustt is an indicator for being between the years 1981 and 1988, and

V alleyt is an indicator for being between the years 1989 and 1995. γrt represents region-

by-year fixed effects, which controls for any common shocks experienced across a region2.

Thus β1 is the average difference in outcome growth rates between treatment and control

countries during the boom period, conditional on region-year fixed effects, with β2 and β3

2Regions assignments are adapted from World Bank country groups. A full list of included countriesin each region is given in the Appendix. One difficult region assignment is Australia and New Zealand.Because these are the only regions with data in the Oceania region, I assign them to Northern Europe.While obviously not a geographic match, it is a reasonable economic and institutional match. Droppingthese two countries altogether does not affect the results.

8

having similar interpretations for the bust and valley periods. This specification does not

include country fixed effects on the right hand side because the dependent variable is a

growth rate, and hence differences in levels across countries are controlled for. I also do not

include non-interacted effects for the boom, bust and valley periods-including these would

show the overall growth effect of these period across all countries, but would not change

β1,2,3, which are the conditional differences in growth rates between treatments and controls.

Robust standard errors clustered at the country level are used for all regressions.

Treatment assignment is based on oil and gas dependence in 1972, the year prior to

the oil embargo. For the main specifications, a country is assigned to the treatment group

if oil and gas production exceeded three percent of GDP in 1972. This rule yields a list of

19 treatment countries, listed in Table 1 along with their 1972 hydrocarbon dependence3.

However, the UNIDOmanufacturing data does not cover all countries, and coverage varies for

the countries that are covered. Five of the 19 treatment countries do not have manufacturing

output data available during the boom and bust periods, so the treatment group for this

part of the analysis is only 14 countries. Additionally, for some of these 14 countries not

every year is available during the boom and bust periods. Fortunately, the import/export

data is much more complete and does not suffer from the same balance problems.

While necessarily arbitrary, the 3% threshold effectively separates the large mass of mi-

nor producers from more exceptional cases. This is shown in Figure 1, which is a histogram

of each country’s 1972 hydrocarbon dependence, among countries with non-zero oil produc-

tion. The vertical line is the three percent treatment group threshold. Further justifying the

threshold, the treatment group includes every country with non-trivial 1972 net exports of

oil (with the possible exception of the Republic of Congo).

Countries that are only slightly below the 3 percent threshold may still be affected by

3Exact oil 1972 dependence is unknown for Libya and the United Arab Emirates because 1972 GDP datais not available for those countries. But oil production was very significant for these countries in 1972 andthere is no question of their inclusion in the treatment group.

9

Table 1: Treatment Countries

Country 1972 Dependence UNIDO data?Algeria 14.0% yBahrain 15.1% nBolivia 2.5% yEcuador 4.1% yGabon 27.1% yIndonesia 9.8% yIran 3.3% yIraq 10.9% yKuwait 7.5% yLibya unknown yNigeria 11.9% yOman 72.4% nQatar 9.4% nSaudi Arabia 65.1% nSyria 4.8% yTrinidad and Tobago 15.0% yTunisia 3.6% yUnited Arab Emirates unknown nVenezuela 19.2% y

the price shock and are therefore not desirable control countries. Therefore I only include

countries in which oil and gas production accounted for less than 1 percent of 1972 GDP

in the control group. Countries that are between 1 and 3 percent are dropped from the

main specifications. However, in alternative specifications I include these countries as “low

dependence” countries and separately test the effect of the boom and bust for low and

high-dependence countries4.

I also drop nine countries that would be in the control group by the above definition but

made significant oil discoveries during the period studied, as this makes them inappropriate

as controls. Further, I drop seven countries that experienced civil wars during the boom

or bust periods (including one country, Angola, that would have qualified as a treatment

country). These changes do not appreciably change the results.

Data on manufacturing value added, employment, wages and capital formation are

taken from the United Nations Industrial Development Organization (UNIDO) INDSTAT2

4Low dependence countries are Afghanistan, Albania, Canada, Colombia, Egypt, Malaysia, Mexico,United States.

10

series covering the years 1963-1995 (the data set actually extends to 2004 but I cut it off

at 1995 as the data becomes somewhat sparse thereafter). The UNIDO data set provides

values in current US dollars. I deflate these values using the US Producer Price Index from

International Financial Statistics (IFS).

Resource production data comes from UN Industrial Commodities Statistics, which

provides production quantities of commodities for all countries and years from 1950-2001.

Import and Export data comes from the NBER-UN World Trade Flows 1962-2000 data

set. This data set gives complete bilateral and total export and import flows organized by

the four-digit Standard International Trade Classification (SITC), Revision 2. To make the

subsector export/import analysis comparable to the manufacturing analysis, I use the many-

to-one SITC to ISIC crosswalk provided by Muendler (2009). This converts 784 4-digit SITC

revision 2 categories into 40 3-digit ISIC revision 2 categories, including 29 manufacturing

categories.

4 Results

4.1 Manufacturing Output

I begin by examining the effect of the boom and bust on total manufacturing. Col-

umn 1 of Table 2 shows the results of estimating Equation 1 with year-on-year difference

in total log manufacturing value added as the dependent variable. There is a large and

significant positive effect of the oil price boom on manufacturing output, and a significant

negative effect during the bust. The coefficients indicate that during the boom, year-on-year

total manufacturing value added growth was on average 7.8 percentage points higher in the

treatment countries relative to control countries, while growth during the bust years was 5.1

percentage points lower. There is no effect during the valley period. The smaller magnitude

11

during the valley period may be indicative of a ratchet effect. A revenue windfall may bring

about a surge in investment in manufacturing capital, but a subsequent decrease would not

cause capital destruction, but possibly restrained growth in the sector.

Table 2: All Manufacturing Results

(1) (2) (3) (4) (5)Value Added Employment Wage Cap. formation Productivity

Treat*Boom 7.216∗∗ 4.357∗∗ 2.943∗ 8.907 2.237(2.582) (1.345) (1.368) (6.975) (1.857)

Treat*Bust -4.332∗ -2.693 -1.462 0.353 -1.760(1.755) (1.933) (1.993) (3.782) (2.216)

Treat*Valley 0.629 2.142 -4.876+ 4.399 -1.600(2.842) (2.307) (2.890) (6.079) (3.115)

N 2111 2267 2066 1451 2009Countries 106 112 104 82 103R-sq 0.250 0.201 0.225 0.222 0.228

Notes: All regressions include region-year fixed effects. Robust standard errors clusteredat the country level are reported in parenthesis. +, *, ** indicates significance at a 10%, 5%,and 1% level respectively.

Columns 2-4 of Table 2 report the effects on employment and wage. Basic Dutch

Disease theory predicts that an oil boom should push up wages, which in turn reduces

employment in the non-booming sector. We do indeed see a significant positive effect on

wages during the boom, with a smaller, insignificant negative effect during the bust. However,

there is also a significant positive effect on manufacturing employment during the boom,

and a negative but insignificant effect during the bust. This could reflect increased local

demand for manufactured goods coupled with low labor intensity of the resource sector. It

could also indicate that investment in the manufacturing sector increased during the boom.

The economically large positive effect on capital formation reported in column 4 would

support this notion, although due to limited data on this particular outcome the standard

errors are large and the estimates are not significant. Column 5 reports the effect on labor

productivity5. Although again not statistically significant, productivity during the boom

period did increase by a similar amount to the wage.

5Labor productivity is not given directly in the UNIDO data, so I construct it by dividing total valueadded by total employment.

12

I next consider countries that had too low hydrocarbon dependence in 1972 to be in-

cluded in the treatment group, but high enough to be excluded from the control group in

the main specifications (these countries had 1972 hydrocarbon dependence between 1 and 3

percent). I label these as “low-dependence” countries and include them in the sample, but

add three more interaction terms to the specification: a low-dependence dummy interacted

with each of the boom, bust, and valley periods. Table 3 reports the results of this speci-

fication for the same set of outcomes as in Table 2. For the low-dependence countries, the

value-added estimates are close to zero and insignificant throughout the period studied. In

general, there are little to no significant effects in low-dependence countries, although due

to the small number of such countries the standard errors are large. This result further

strengthens the case that the 3% threshold is appropriate in identifying countries affected

by the boom and bust.

Table 3: High and Low Dependence Manufacturing Results


High-dep*Boom 7.319∗∗ 4.579∗∗∗ 2.823∗ 8.611 2.211(2.533) (1.322) (1.318) (6.775) (1.832)

High-dep*Bust -4.975∗∗ -2.621 -1.974 -0.264 -2.412(1.901) (1.879) (2.002) (3.989) (2.350)

High-dep*Valley 1.248 1.975 -3.700 4.930 -1.481(2.853) (2.251) (2.979) (6.083) (3.097)

Low-dep*Boom 0.449 0.970 -2.767+ 1.658 -0.870(2.217) (1.109) (1.596) (5.661) (2.002)

Low-dep*Bust 1.366 -1.516∗ 4.214 -0.008 3.019(3.325) (0.675) (3.518) (4.384) (3.765)

Low-dep*Valley -0.571 0.707 0.404 8.472 -6.792(5.008) (4.738) (5.906) (8.748) (4.194)

N 2281 2451 2249 1593 2176Countries 113 120 112 88 110R-sq 0.254 0.203 0.223 0.211 0.230


I next turn from total manufacturing to subsector analysis. The Dutch Disease model

predicts that the production of more tradable goods should decline relative to local goods. I

13

test this using value added data for three-digit manufacturing subsectors given in the UNIDO

data set. I use the binary exportability classifications found in Rajan & Subramanian (2011),

which are based on historical export to value-added ratios6. I sum together all exportable

and non-exportable sectors and run separate regressions for total value added for each (so

the unit of observation remains at the country-year level). The results are shown in columns

1 and 2 of Table 4. Both exportable and non-exportable sectors experience rapid growth

during the boom period and decline during the bust period. Contrary to predictions of the

core model, exportable sectors grew slightly faster than non-exportables during the boom,

although the coefficients are not statistically distinguishable from each other.

Table 4: Subsector Manufacturing Results

(1) (2) (3) (4) (5)Exp Index=1 Exp Index=0 Upstream Downstream Non-oil

Treat*Boom 9.343∗∗ 6.693∗∗ 10.532∗∗∗ 19.869∗∗ 6.448∗∗

(3.035) (2.465) (2.570) (6.835) (2.354)Treat*Bust -4.025+ -3.213+ -6.742∗∗∗ -5.405 -3.178+

(2.139) (1.876) (2.008) (4.223) (1.842)Treat*Valley 0.690 0.060 1.649 4.895 -0.256

(3.488) (2.913) (3.626) (6.316) (2.876)N 2101 2123 2083 1860 2125Countries 106 108 106 91 108R-sq 0.208 0.234 0.187 0.182 0.238

Notes: The dependent variable is the year-on-year difference in log value added. All regressionsinclude region-year fixed effects. Robust standard errors clustered at the country level arereported in parenthesis. +, *, ** indicates significance at a 10%, 5%,and 1% level respectively

I next examine the effects on upstream and downstream sectors. Upstream sectors in-

clude chemicals, metals, machinery, and transport equipment7. Downstream sectors include

refined petroleum products, rubbers and plastics8. I created an upstream index that takes

a value of one for upstream sectors and zero otherwise, and a similarly defined downstream

index. Columns 3 and 4 present results for the specification of Equation 1 when only up-

6A subsector takes an exportability index value of one if it historically had an exports to value addedratio above the median according to World Integrated Trade Solution (WITS) data. This index correspondswell to industries with high export to value-added ratios as of 1972 using the data used in this paper.

7Upstream sectors are ISIC revision 2 codes 351,352,381,382,383,384.8Downstream sectors are ISIC revision 2 codes 353,354,355,356.

14

stream and downstream sectors are included in the sample, respectively. Both upstream

and downstream industries are strongly pro-cyclical with respect to oil prices. The effect

during the boom is larger than that for non-oil sectors, particularly for downstream sectors.

This is unsurprising, as increased investment in exploration and drilling activity boosts the

upstream sectors, and the shocks to oil prices filter into the downstream sectors. However,

during the bust the negative effect is slightly larger for non-oil sectors.

Returning to exportable vs. non-exportable sectors, a weakness of the exportability

index used above is that it is rather blunt and does not necessarily identify the most ex-

portable industries in each individual country. In the following specification I attempt to

assess the differential impact of the boom and bust on industries that were most exportable

for each individual treatment country just before the boom relative to less exportable ones.

Within each country, I rank each subsector by the export to value added ratio as of 1972. I

then create a dummy that is equal to one for observations that represent a country-sector

pair such that the sector ranked as one of the five most exportable sectors in that country

in 1972. I then run the following regression at the country-sector-year level:

Δln(Ycsrt) =β1(Tc ∗Boomt) + β2(Tc ∗Bustt) + β3(Tc ∗ V alleyt) + β4(HEIcs)+

β5(Tc ∗HEIcs ∗Boomt) + β6(Tc ∗HEIcs ∗Bustt) + β7(Tc ∗HEIcs ∗ V alleyt)+

γrt + sectors + εct (2)

Where HEIcs is the high-exportability index indicator and sectors represents sector

fixed effects. The coefficients β5, β6, β7 represent the additional effect of being a high-

exportability sector in each respective period. The results of this specification are shown

in Table 5. During the boom, exportable sectors on average grew an additional 3.6 per-

centage points relative to non-exportable sectors. There are no significant additional effects

15

during the bust or valley periods. While somewhat counter-intuitive, this result does not

necessarily contradict the Dutch Disease model. As mentioned in Corden & Neary (1982),

additional investment into tradable sectors, which is not included in their model, could offset

whatever harmful exchange rate effects there might be. And indeed, as shown in Column 4

of Table 5 capital formation grew at a much higher rate for high-exportability sectors relative

to low-exportability ones (although caution is again warranted due to the lower data avail-

ability for capital formation). The results suggest the possibility that more of the windfall oil

revenues were reinvested into exportable sectors relative to local sectors. One possible reason

is the common belief that oil revenues are not widely distributed among the population, so

that there were greater non-oil growth opportunities in non-local sectors.

Table 5: High-exportability and Manufacturing results


HEI 1.297∗ 1.143∗∗ -0.209 1.154 0.060(0.554) (0.403) (0.313) (0.963) (0.365)

Treat*Boom 9.750∗∗∗ 5.802∗∗ 3.790∗∗ 11.119∗ 4.106∗∗

(2.528) (1.805) (1.353) (4.430) (1.495)Treat*Bust -4.566∗∗ -1.616 -2.837 -2.111 -3.456∗

(1.714) (1.659) (1.910) (3.742) (1.716)Treat*Valley -0.489 1.597 -5.473∗ 6.364 -2.449

(3.201) (2.326) (2.581) (5.674) (2.689)T*HEI*boom 3.567∗ 1.375 -0.167 8.784∗ 1.591

(1.726) (1.359) (0.545) (4.115) (1.219)T*HEI*bust -0.889 -1.381 0.972+ 0.185 -1.066

(2.127) (1.257) (0.556) (4.080) (1.227)T*HEI*valley 4.393 1.197 -1.200 5.346 3.492

(3.938) (1.513) (1.250) (3.662) (3.531)N 49614 54573 48991 30810 47403Country-sectors 2467 2640 2484 1865 2433R-sq 0.0681 0.0396 0.116 0.0527 0.0718


4.2 Exports/Imports

I next analyze the impact of the boom and bust on treatment country trading activity.

The effect on manufacturing exports and imports can help to shed light on the extent that

16

the increase in manufacturing output during the boom was driven by increased local demand.

If there is little or no effect on manufacturing exports and/or a positive effect on imports,

it would suggest that the increased activity is due primarily to local demand. If there is a

large effect on exports, it suggests that the boom was driving investment in traded goods as

well.

Column 1 of Table 6 shows the results of estimating equation 1 with year-on-year dif-

ference in log total manufacturing exports as the dependent variable9. During the boom,

manufacturing exports grew by an average of seven percentage points per year faster than

in control countries, a result similar to that for total value added. The effect is negative but

insignificant during the bust, and non-existent during the valley period. Column 2 shows the

results for manufacturing imports. Import activity likewise grows substantially relative to

control countries during the boom period, and decreases during the bust period, with both

effects being statistically significant.

Table 6: Total Exports and Imports

(1) (2) (3) (4) (5) (6)Manuf. exp Manuf. imp non-oil res. exp non-oil res. imp Ag. exp Ag. imp

Treat*Boom 6.989∗ 6.986∗∗∗ -11.289∗ 2.923 -9.009∗∗ 9.008∗∗∗

(3.548) (1.582) (4.390) (3.125) (3.448) (2.506)Treat*Bust -3.605 -6.114∗∗ 7.956+ 2.639 -0.522 2.506

(3.598) (2.102) (4.153) (4.180) (3.041) (3.393)Treat*Valley -0.076 -3.059 -0.498 -5.418 6.765∗ -5.934

(3.755) (2.210) (4.167) (4.797) (3.269) (4.640)N 3532 3534 3233 3263 3521 3491Countries 110 110 110 110 110 110R-sq 0.150 0.139 0.108 0.145 0.155 0.165


According to columns 1 and 2, The estimated effects of the boom on manufacturing

exports and imports are nearly identical. This yields an important interpretation. In the

year before the boom, 17 of the 19 treatment countries were net importers of manufactured

9Recall that the import/export data includes all 19 treatment countries and is balanced.

17

goods. This implies that although value-added and exports increased during the boom,

total net imports increased as well, as would be expected given a rising exchange rate and

an increase in local demand. This result would be consistent with a version of the Corden

& Neary (1982) model that specifies manufacturing as an intra-industry sector (in which

products from the same sector are both imported and exported). Intra-industry trade can

arise from product differentiation within a sector. Of course, manufacturing is an extremely

broad classification and is self-evidently an intra-traded sector. Hence the positive result for

manufacturing exports does not contradict the Corden & Neary (1982) model, but rather

serves as a caution against conflating manufacturing with the stylized “traded sector” of the

model.

I next examine the effect on production of non-oil mineral exports. For each country-

year observation, I sum together the export values for the ISIC categories “Coal Mining”,

“Metal Ore Mining” and “Other Mining” (these categories are not available in the UNIDO

data, which only includes manufacturing, so I only examine trading activity rather than

overall production). Columns 3 and 4 of Table 6 report the export and import results for

non-oil minerals. These results are the first indication thus far of a negative Dutch Disease

effect. During the boom there is a large and significant negative effect on non-oil resource

exports, and a positive significant effect during the bust. There is no significant effect on

imports in either period, suggesting that local demand is low or unaffected by the oil price

swings.

Columns 5 and 6 report the results for total agricultural exports 10. Again, during the

boom there is a large and significant negative effect on exports, with no effect during the

bust. However, there is also a significant positive effect during the valley period, suggesting

there may be structural differences between treatment and control countries for this outcome.

Interestingly, agricultural imports increased significantly during the boom. Since the demand

10All ISIC subcategories under “Agriculture, Hunting, Forestry and Fishing”.

18

for agricultural goods is presumably inelastic, these results suggest that during the boom

resources used for food production shifted to other sectors of the economy, and local food

production was displaced by imports.

Table 7 presents the effects on exports for different groups of subsectors, using the same

categories as in Table 4. Columns 1 and 2 report the export results for exportable and

non-exportable sectors, using the general index used in Table 4. Exports of both groups rose

significantly during the boom, with magnitudes similar to the value added results in Table

2. Exports likewise decreased for both groups during the bust period, but the effect is only

statistically significant for the non-exportable sectors, and only at a 10% level. Exports of

upstream and downstream sectors both increased substantially during the boom, but these

measures were particularly volatile, yielding large standard errors and lack of statistical

significance. The non-oil related sectors did see a significant positive effect during the boom

and a negative effect during the bust, further indicating spillovers beyond the oil sector.

Table 7: Subsector Exports


Treat*Boom 7.237∗ 5.128+ 6.200 10.575 5.496+

(3.635) (2.999) (7.415) (13.143) (2.839)Treat*Bust -4.889 -8.458∗∗ 2.651 -4.275 -7.864∗∗

(3.682) (2.578) (4.653) (10.076) (2.634)Treat*Valley -0.453 -1.992 3.684 4.641 -1.465

(3.945) (3.845) (3.832) (7.672) (3.842)N 3528 3533 3464 2782 3536Countries 110 110 110 110 110R-sq 0.151 0.146 0.114 0.122 0.141


Table 8 shows the results for subsector imports. Exportable and non-exportable sector

imports were likewise pro-cyclical with respect to oil prices, with estimate magnitudes similar

to those for exports. Non-oil related sector imports also behaved similarly to that of exports.

Unsurprisingly, imports of upstream sectors were strongly procyclical with respect to oil

19

prices, while there is no significant effect for downstream imports.

Table 8: Subsector Imports


Treat*Boom 6.958∗∗∗ 5.099∗∗ 7.428∗∗∗ 3.340 5.513∗∗∗

(1.529) (1.613) (2.052) (3.945) (1.571)Treat*Bust -5.757∗ -6.044∗∗ -7.943∗∗∗ 5.328 -6.660∗∗

(2.272) (2.147) (1.994) (4.894) (2.202)Treat*Valley -2.911 -2.297 -1.756 -2.000 -3.404

(2.262) (2.338) (2.453) (3.924) (2.226)N 3533 3531 3524 3510 3539Countries 110 110 110 110 110R-sq 0.136 0.185 0.153 0.161 0.173


4.3 Real Exchange Rates

Although plausibly explained by factors outside the core model, the finding that high-

exportability manufacturing sectors grew faster during the boom relative to low-exportability

sectors is somewhat surprising. One of the primary mechanisms by which traded sectors are

predicted to relatively decline is an appreciation of the real exchange rate. Therefore it is

worthwhile to check if this mechanism is detectable in the empirical setting of this paper.

I evaluate equation (1) with two different measures of exchange rate appreciation as the

dependent variable. First, I use the year-on-year difference in price level of GDP given in

Penn World Table 7.0, which is the real exchange rate relative to the United States. However,

real exchange rate movements could be a long term trend associated with income levels, due

to the Balassa-Samuelson effect. Therefore I follow Rajan & Subramanian (2011) and Rodrik

(2007) in constructing a measure of “excess appreciation”, which estimates the deviation of

the real exchange rate from what would be predicted by the Balass-Samuelson model11.

11This is done through the following procedure: for each year, I run a regression at the country level oflog price level on log PPP-adjusted GDP per capita. Excess appreciation is then the estimated residualfrom this regression. The procedure rests on the assumption of a long-run equilibrium relationship betweenincome and exchange rates based on PPP.

20

Column 1 of Table 9 shows the results of estimating equation (1) with year-on-year

difference in log price level as the dependent variable. Real exchange rates in treatment

countries increased an average of three percentage points per year during the boom relative

to control countries. The excess appreciation measure increased by nearly the exact same

amount, suggesting that all of the appreciation was so-called “excess” appreciation caused

by the spike in export revenues. During the bust there is no effect on either measure.

During the valley period, there is an economically large negative effect, though only the

excess appreciation effect is significant at a 10% level. However, these negative estimates are

largely driven by two outlier observations, one each in Iraq and Iran, with enormous drops in

price levels in the early 1990s. If just these two observations are removed, the effect during

the valley period is still negative but much smaller and insignificant.

Table 9: Exchange Rates

(1) (2)Price level Excess apprec.

Treat*Boom 3.002∗ 3.013∗

(1.326) (1.319)Treat*Bust 0.450 -0.132

(1.382) (1.168)Treat*Valley -2.350 -3.182

(2.217) (2.351)N 2934 2844Countries 124 124R-sq 0.160 0.0990


It is possible that these results represent a lag between the incidence of oil prices and

their effect on exchange rates. According to Rogoff (1996), exchange rate adjustments in

response to PPP deviations have a half-life of 3-5 years. However, Cashin (2004) challenges

this consensus and estimates that commodity currency adjustments have a half life of just

10 months. Whatever the actual lags, the bust period could contain a period of positive

currency adjustment left over from the boom period, followed by a negative adjustment,

21

averaging out at close to zero. Similarly, the valley period could contain a period of negative

adjustment left over from the bust period. In any case, the production and export results

suggest that either these currency adjustments were too small to have a discernible impact

on exportable manufacturing sectors, or were overcome by other positive factors, such as

increased investment.

5 Conclusion

This paper has analyzed the effect of the oil boom and bust on non-oil economic activ-

ity in oil-dependent countries, adding to the thin but emerging empirical evidence on the

presence of Dutch Disease or lack thereof. During the oil price spike of the 1970s, manufac-

turing output increased substantially in oil-rich countries relative to non-oil-rich countries.

Manufacturing wages also increased, satisfying a necessary condition for the Dutch Disease

phenomenon, but capital formation and productivity also increased, which is likely to at

least partially explain the rise in output. These effects hold for sectors connected or un-

connected to the oil industry, and for both high and low-exportability sectors, although

high-exportability sectors grew even faster during the boom and received a larger bump in

investment. Manufacturing exports likewise increased significantly, as did imports by a sim-

ilar amount, implying an increase in net manufacturing imports and reflecting an increase in

local demand. However, agricultural exports and non-hydrocarbon natural resource exports

fell sharply, suggesting that these sectors were subject to Dutch Disease mechanisms and

did not receive the investment spillovers that the manufacturing sector did. The converse

of all of these effects was observed during the oil price bust of the 1980s, though for many

outcomes to a lesser extent.

The results taken together suggest that manufacturing benefited from the oil boom in

two ways. First, increased local demand for manufactured goods resulting from the revenue

22

windfall was met partially by increased imports and partially by increased local produc-

tion. Second, the windfall resulted in investment spillovers into manufacturing. That these

spillovers do not appear to be present for non-hydrocarbon commodities and agricultural

products (though I do not actually observe capital formation for these sectors) may be a re-

sult of higher perceived returns in manufacturing or a directed push towards industrialization

by the government.

These results do not contradict the core Dutch Disease model, which ultimately leaves

the effect of an oil boom on other sectors ambiguous. Nor do they imply that Dutch Disease

mechanisms, namely the resource movement and spending effects, are not present. The

spending effect is positive for the share of manufactured goods that are locally consumed.

And whatever negative mechanisms exist appear to have been overwhelmed by the positive

investment spillovers, which are not considered in the core model.

As this is a study covering a certain time period, it is fair to question external validity.

Perhaps there were circumstances during the time frame and set of countries studied in this

paper that do not necessarily hold for other commodity booms. Due to data limitations, as

of this writing I am unable to study the 2000s oil boom in a similar way, but as more data

becomes available this will be a useful line of research.

23

References

Alexeev, M. and Conrad, R. (2009). The elusive curse of oil. The Review of Economics andStatistics, 91(3):586–598.

Allcott, H. and Keniston, D. (2013). Dutch disease or agglomeration? the local economiceffects of natural resource booms in modern america. Technical report, Working Paper.

Black, D., McKinnish, T., and Sanders, S. (2005). The economic impact of the coal boomand bust*. The Economic Journal, 115(503):449–476.

Brunnschweiler, C. N. and Bulte, E. H. (2008). The resource curse revisited and revised: Atale of paradoxes and red herrings. Journal of Environmental Economics and Management,55(3):248–264.

Caselli, F. and Michaels, G. (2009). Do oil windfalls improve living standards? evidencefrom brazil. Technical report, National Bureau of Economic Research.

Cashin, P., Cespedes, L. F., and Sahay, R. (2004). Commodity currencies and the realexchange rate. Journal of Development Economics, 75(1):239–268.

Corden, W. M. (1984). Booming sector and dutch disease economics: survey and consolida-tion. oxford economic Papers, 36(3):359–380.

Corden, W. M. and Neary, J. P. (1982). Booming sector and de-industrialisation in a smallopen economy. The economic journal, 92(368):825–848.

Gelb, A. H. (1988). Oil windfalls: Blessing or curse? Oxford University Press.

Harding, T. and Venables, A. (2013). The implications of natural resource exports for non-resource trade.

Ismail, K. (2010). The structural manifestation of the’dutch disease: The case of oil exportingcountries. IMF Working Papers, pages 1–36.

Krugman, P. (1987). The narrow moving band, the dutch disease, and the competitiveconsequences of mrs. thatcher: Notes on trade in the presence of dynamic scale economies.Journal of development Economics, 27(1):41–55.

Matsuyama, K. (1992). Agricultural productivity, comparative advantage, and economicgrowth. Journal of economic theory, 58(2):317–334.

Muendler, M.-A. (2009). Converter from sitc to isic. University of California-San Diego,unpublished mimeo.

Rajan, R. G. and Subramanian, A. (2011). Aid, dutch disease, and manufacturing growth.Journal of Development Economics, 94(1):106–118.

Rodrik, D. (2008). The real exchange rate and economic growth. Brookings papers oneconomic activity, 2008(2):365–412.

Rogoff, K. (1996). The purchasing power parity puzzle. Journal of Economic literature,34(2):647–668.

24

Sala-i Martin, X. and Subramanian, A. (2003). Addressing the natural resource curse: Anillustration from nigeria. Technical report, National Bureau of Economic Research.

Smith, B. (2013). The resource curse exorcised: Evidence from a panel of countries. Technicalreport, Working Papers, University of California, Department of Economics.

Stijns, J.-P. (2003). An empirical test of the dutch disease hypothesis using a gravity modelof trade. Available at SSRN 403041.

Torvik, R. (2001). Learning by doing and the dutch disease. European economic review,45(2):285–306.

25

Appendix: List of Countries in Exports/Imports Regression Samples

The following table lists the countries included in the main export/import regressions of

Table 6, separated by region. The export/import regressions are chosen because the because

the coverage is superior and includes more countries. The second column indicates if the

country is in the treatment group, and the third column indicates if the country is included

in the value-added regressions using UNIDO manufacturing data.

East Asia and Pacific Treatment Group? UNIDO data?CambodiaChina xFiji xHong Kong xIndonesia x xJapan xKorea, Republic of xLaosMacao xMongolia xPhilippines xSingapore xTaiwanThailand xVietnam

East Asia and Pacific Treatment Group? UNIDO data?Bosnia and Herzegovina xBulgariaCzech RepublicHungary xPoland xRomania x

Latin America and the Caribbean Treatment Group? UNIDO data?Argentina xBahamas xBarbados xBelize xBolivia x x

26

Brazil xChile xCosta Rica xCuba xDominican Republic xEcuador x xGuatemala xGuyanaHaiti xHonduras xJamaica xPanama xParaguay xPeru xTrinidad and Tobago x xUruguay xVenezuela x x

Middle East and North Africa Treatment Group? UNIDO data?Algeria x xBahrain xDjiboutiIran x xIraq x xJordan xKuwait x xLibya x xMorocco xOman x xQatar x xSaudi Arabia xSyria xTunisia x xTurkey xUnited Arab Emirates x x

Northern Europe Treatment Group? UNIDO data?Australia xAustria xDenmark xFinland xIceland xIreland x

27

Netherlands xNew Zealand xSweden x

South Asia Treatment Group? UNIDO data?Bangladesh xIndia xNepal xPakistan x

Southern Europe Treatment Group? UNIDO data?Cyprus xGreece xItaly xMalta xPortugal xSpain x

Sub-Saharan Africa Treatment Group? UNIDO data?Sub-Saharan AfricaBenin xBurkina Faso xBurundi xCentral African Republic xChadCongo, Dem. Rep.Ethiopia xGabon x xGambia, The xGhana xGuineaGuinea-BissauKenya xLiberiaMadagascar xMalawi xMaliMauritania xMauritius xNiger xNigeria x xRwanda xSenegal x

28

Seychelles xSierra Leone xSomalia xSouth Africa xTanzania xTogo xUganda xZambia xZimbabwe x

oxcarre research paper 133 dutch disease and the oil and ... · 1 1 introduction commodity prices...

Documents