1

 

Macroeconomic determinants of stock market behaviour in South Africa:

Evidence from both aggregate and sectoral level data

K. Junkin∗ and Z. Chinzara♣

                                                            ∗ Department of Economics, Rhodes University, P.O. Box 94 Grahamstown, South Africa. ♣ School of Economics and Finance, Queensland University of Technology, Garden Point Campus, GPO Box 2434 QLD 4001, Brisbane, Australia. All correspondences must be addressed to Z. Chinzara: Phone: +61 46 699 5060. Email: zchinzara@yahoo.com .

2

 

Abstract

This study analysed the long run and short run influences of macroeconomic fundamentals on the South African (SA) stock market using both aggregate and sectoral data for the period 1995 to 2008. The Johansen and Juselius (1990) multivariate cointegration, together with the Vector Error Correction and impulse response analyses were utilised. It was found that significant long run relationship exists between macroeconomic fundamentals and the stock market. Nevertheless, the speed of adjustment following disequilibria, the signs and elasticity of response tend to vary across sectors. These findings highlight the weakness of relying only on aggregate data, which is the current state of relevant research on the SA stock market.

Keywords: Macroeconomic variables, Cointegration, Price indices, JSE

3

 

1. INTRODUCTION

The link between the macro-economy and stock prices is well documented in financial and economic literature. Models such as the Dividend Growth, Capital Asset Pricing Model (CAPM) and Arbitrage Pricing Model (APT) provide the theoretical framework which explains how changes in the macroeconomy are transmitted into stock prices. These models predict that any anticipated or unanticipated arrival of new information about macroeconomic variables (e.g. GDP, industrial production, inflation, interest rate, and exchange rate, etc) will indirectly affect share prices through its impact on expected dividend, discount rate or both.

The understanding of the macroeconomic determinants is invaluable for investors and policy markers. Both individual and institutional investors would find such an understanding priceless because it would enable them to proactively act to either profit or mitigate risk in the face of macroeconomic or policy changes. A counterargument to this would be that because stock markets are generally informationally efficient, investors would not benefit from such knowledge1. However, given that the EMH hypothesis has been found to be inapplicable due to some of its lax assumptions many investors still rely on fundamental analysis. To policy makers an understanding of the link between the macroeconomy and the stock market would be useful in formulating and effectively implementing regulatory policies. Furthermore, given the increasing importance of the stock market in promoting economic growth in emerging marking markets (c.f. Levine and Zervos, 1998; Kose, et al, 2006; Deb and Mukherjee, 2008), policy makers would need this understanding if they hope to formulate policies that foster stock market development.

Theoretical literature suggests a number of macroeconomic factors that could influence the stock market. Some of these include GDP, inflation, exchange rate, interest rate, money supply and foreign GDP. Domestic GDP is expected to have a positive impact on stock prices. This is because as an economy experiences increased current or predicted growth, aggregate consumption would increase and this would increase corporate profitability. This would increase anticipated dividend and investors would be induced to purchase more shares. It is then that this increase in demand for shares will put upward pressure on stock prices.

                                                            1 Coined by Fama (1970), the market efficiency hypothesis in its semi-strong sense posit that investors cannot earn excess returns by exploiting using currently available information.

4

 

The tax-effect and the proxy-effect hypotheses outline the channels through which inflation can negatively impact on the stock market. Coined by Feldstein (1980), the tax-effect hypothesis posits that inflation imposes higher effective tax rates on corporate income and by doing so reduces real net yield that investors receive per unit of capital. This effect emanates from the accounting treatment of historical-cost depreciate and capital gains. The latter hypothesis emphasises the effects of inflation on the stock market through its impact on the real economy (Fama, 1981). Because real economic activity is positively related to stock market performance and negatively related to inflation, it is logical that inflation is negatively related to stock market prices.

Interest rates have the potential to impact on the stock market through two main channels. Firstly interest rate imposes a cost of borrowing to both corporate and stock market investors. Thus both the profitability of investments and demand for shares are likely to be negatively affected and consequently stock prices. Secondly an increase in interest rates increases the opportunity costs of investing in the stock market. This is because other financial assets like bonds become cheaper2. Thus, investors would reallocate funds from the stock market to the bond market resulting in reduction in stock prices.

Changes in money supply may have a double-sided effect on stock prices. This is due to the theoretical linkage between money supply, inflation and interest rates. On the one side, an increase in money supply would affect stock prices through its effects on current interest rates. Lower interest would reduce financing costs and thus increase the profitability of companies. The effect on stock prices in this case would depend on the interest rate sensitivity of different sectors/companies as well as their source of finance. Furthermore, a decrease in interest rate would inveigle investors to sell interest-dominated assets like bonds (as their prices would have increased) and invest the resultant earnings into the stock market, thus putting an upward pressure in the stock prices. On the other side, an increase in money supply would increase the likelihood of increased inflation and thus a decrease in future demand for goods (Mishkin, 2004:11). Since investors are forward-looking they would interpret this anticipated decrease in aggregate demand as a decrease in the future profitability of companies. Lower profitability of a firm lowers expected dividends and thus stock prices.

In an open economy like that of South Africa, the exchange rate movements affect the stock markets through its impact on imports and exports as well as on

                                                            2 The Liquidity preference theory explains how interest rates are negatively related to bond prices through expectations.

5

 

capital flows. A depreciation of a currency promotes exports and discourages imports. In the short run this would increase the earnings of exporting companies and hence their stock prices. In the long run this could also promote economic growth and thus overall stock market development. Depreciation may also increase the flow of portfolio equity capital as foreign investors try to take advantage of cheaper domestic shares. This increase in demand for domestic shares will increase share prices in the short run. However, in the long run, both increase exports and capital inflows could result in positive pressure being put on domestic interest rates thereby dampening the initial increase in stock prices. Similar effects can also result if depreciation is too much so that it fuels imported inflation. This of course would be the case if a country relies heavily on important goods for instance oil and physical capital.

The influence of foreign GDP on domestic stock prices is uncertain. In the first instance according to Jefferis and Okeahalam (2000), foreign GDP may be positively related to domestic stock prices. A possible reason for this is that if the domestic country were a strong export trading partner of the foreign country, an economic upswing of the foreign country’s economy would increase the relative attractiveness of the domestic country’s exports. The result being an increase in profitability of the domestic country’s exporting firms, thus the dividend yields of these firms and furthermore the increase in the price of their stocks. On the contrary, foreign GDP may have a negative impact on domestic stock prices. This is because an economic upswing of a foreign economy may result in the increased profitability and stock prices of foreign firms, which in turn may trigger a capital outflow from the domestic country into the more profitable foreign firms. A net capital outflow would theoretically lead to a decrease domestic stock prices and an increase in foreign stock prices. The price movement of Brent crude oil may have an influence on the prices of shares. Since the price of Brent crude oil is directly linked to the price of petroleum and thus transport costs, in South Africa, this has a knock-on effect on inflation. This would lead to increased inflationary pressures and as discussed prior, results in lower profitability and thus lower share prices.

Empirical studies have shown that there is indeed a relationship between macroeconomic variables and stock returns. Fama (1981) established a strong negative relationship between inflation and stock prices, and a positive relationship between real economic variables and the stock market. Taking an international perspective, Bodurtha et al. (1989) tested the relationship between stock returns and international macroeconomic variables using inter-battery factor analyses. They found that anticipated inflation as well as industrial

6

 

production has a significant influence on stock market returns in the developed economies studied. Using Vector Autoregression (VAR), Sadorsky (1999) investigates if oil price movements are important in determining stock return movements in the US. Using impulse response functions it was found that oil price movements are important in determining stock returns, which in turn has an impact on the real economy. From a different perspective Nikkinen et al. (2006) investigated how global stock markets are integrated with respect to U.S. macroeconomic news announcements. Using ten scheduled macroeconomic news announcements and thirty seven stock markets from around the world, they found that Latin America was least influenced by such decisions. European as well as Asian countries were most influenced by such announcements depending on their market size, foreign ownership and international trade.

Empirical studies on emerging and developing stock market have produced somewhat mixed results. For instance, Maysami et al. (2004) analysed the relationship between macroeconomic variables and the Singapore stock market using Johansen’s Vector Error Correction Model. Variables that were included in the test were the interest rate, inflation, the exchange rate and industrial production. From the results only industrial production was found to be an insignificant determinant. Inflation was surprisingly found to have a positive impact on stock prices. Using the same methodology, Adam and Tweneboah (2008) also found a positive relationship between inflation and stock returns for Ghana. They interpreted this finding as an indication that the stock market provides a partial hedge against inflation. Vuyyuri (2005) investigated the long run relationship and causality between the financial and real sectors of the Indian economy. Again, inflation, interest rates and exchange rates were found to be influential macroeconomic factors on stock returns. Their signs for all three factors conformed to prior expectations and causality was found to be unidirectional from the macroeconomic variables to the stock market. Utilising VECM and variance decomposition, Gunasekarage et al. (2004) found that consumer price index, money supply and the Treasury bill rate have significant lagged, but temporary influences on the Sri Lankan stock market. Their variance decomposition results showed that shocks to these macroeconomic factors explained only a minority of the variance error of the stock index. For South Africa (SA), three studies of this nature exist to the best of the researcher’s knowledge. These include Coetzee (2002), Moolman and Du Toit (2005), and Durodola (2006). A study by Coetzee (2002) found statistically significant evidence of a negative relationship between quarterly monetary variables such as inflation, short-term interest rate, rand-dollar exchange rate

7

 

and stock prices in both the short run and the long run for the period 1991-2001. Nevertheless, Moolman and Du Toit (2005) showed that discounted future dividend determines the long-run behaviour of the stock market behaviour while factors like short-term interest rate, the rand-dollar exchange rate and the S&P 500 index determine the short run behaviour for the period 1993-2003. Finally, using the Johansen cointegration approach, Durodola (2006) established that not only domestic macroeconomic factors but also foreign GDP influences the long-run behaviour of both the SA stock market index and stock market capitalisation. He further found that the stock market index adjusts back to equilibrium faster than market capitalisation.

In the same spirit as most of the previous studies, the current study paper analyses the long run and short run influence of macroeconomic variables and the SA stock market. The contribution of the current study is twofold. Firstly, more recent and higher frequency data is used. The use of monthly data in this study is more likely to better capture the dynamics in the stock market, given that stock markets react promptly to new information. This is in contrast to relevant studies on SA that have used quarterly data (c.f. Coetzee, 2002; Moolman and Du Toit, 2005; Durodola, 2006). Secondly, the current study departs from the existing studies as it uses both aggregate market and sector level data. It is the researchers’ considered view that different sectors that make up the South African stock market might react differently to similar macroeconomic variables. For instance while interest rate is a source of income for financial companies like banks, it is a cost of funds for industrial companies. Thus, while stocks of banks might react positively to an increase in interest rate, stocks of industrial companies might react negatively. Similarly, it is expected that exchange rate movements are likely to have a direct and higher impact on mining stocks compared to their impact on general retail stocks. In this regard it is hoped that the current study will offer better information to investors as it gives them possible information on how they could diversify in the different sectors. Likewise policy makers and regulators would benefit from the current study as the findings may offer information on how specific-industrial oriented policies can be formulated and how regulatory policies that counter the possibility of policy-arbitrage by investors can be formulated.

The rest of the paper is organized as follows: Section 2 focuses on data issues and methodology, section 3 presents and analyses the findings, and lastly section 4 concludes the study.

8

 

2. DATA AND METHODOLOGY 2.1. Selection of Variables, data and a priori expectation

Based on the discussion above and on the consideration that the SA stock market is becoming increasingly integrated into the global economy (see Chinzara and Aziakpono, 2009), the researchers propose the following macroeconomic variables: industrial production3, money supply, US GDP (a proxy for foreign GDP) and consumer price inflation, Treasury bill rate, nominal exchange rate, and oil prices. The data used for this study is for the period 1995 to 2008 obtained from Thompson DataStream 2007. All data was available in monthly frequency except for United States GDP, which was interpolated from quarterly to monthly. For the stock market, the JSE FTSE Index and four sectorial indices were used which were selected based on their relative size as well as their importance to the South African economy. Based on this data, five models were estimated with each of the stock market indices being a dependent variable. Except for interest rate, all the variables are logarithmically transformed. Table 1 summarises the variables and their measurement. Also given in the table are the expected sign of each of the macroeconomic variables. The a priori expectation is in line with the discussion in Section 1.

Table 1: Variable and model description summary

Variable Model Description Unit

LALSI 1 LOG of FTSE/ALL SHARE INDEX PRICE INDEX LFPI 2 LOG of FTSE/FINANCIAL PRICE INDEX PRICE INDEX LIPI 3 LOG of FTSE/INDUSTRIAL PRICE INDEX PRICE INDEX LGRPI 4 LOG of GENERAL RETAIL PRICE INDEX PRICE INDEX LMPI 5 LOG of FTSE/MINING PRICE INDEX PRICE INDEX LCPI LOG of CONSUMER PRICE INFLATION- INDEX (2005=100) LMS LOG of M3 MONEY SUPPLY- R MILLIONS LPCO LOG of PRICE OF BRENT CRUDE OIL- R PER BARREL LIPSA LOG of INDUSTRIAL PRODUCTION OF SOUTH AFRICA+ INDEX (2005=100) LUSGDP LOG of GDP OF UNITED STATES OF AMERICA+ $ BILLIONS LNER LOG OF NOMINAL EXCHANGE RATES+ RANDS/DOLLAR NIR NOMINAL INTEREST RATES+- TREASURY BILL RATE Notes: +,- represent the expected relationships between the dependant and explanatory variables

                                                            3 Note that industrial production is used as a proxy for GDP because GDP figures are only available on a quarterly basis.

9

 

Table 2: Pair-wise correlation the macroeconomic variables.

Model 1 LCPI LIPSA LMS LPCO LRER NIR LUSGDP LCPI 1.00 LIPSA 0.03 1.00 LMS 0.97 0.10 1.00 LPCO 0.93 0.21 0.94 1.00 LNER 0.65 -0.18 0.58 0.65 1.00 NIR -0.67 0.12 -0.63 -0.62 0.40 1.00 LUSGDP 0.56 0.01 0.65 0.93 0.62 0.68 1.00

Source: Thompson DataStream, 2009 and authors’ own estimates

Given the theoretical links among some of the macroeconomic variables, the possibility of multicollinearity among them cannot be ruled out. To this end pairwise correlation matrices for the variables were estimated. The results are presented in Table 2 and they suggest possible multicollinearity of most of the macroeconomic variables with money supply and oil prices. Specifically, money is highly correlated to CPI and oil prices while oil prices are also correlated to CPI and US GDP. These results are theoretically sound. With the wisdom that multicollinearity negatively affects efficiency of estimates, both money supply and oil prices were dropped from the empirical analysis in order to maintain efficiency in further estimation.

2.2. Methodology

The standard Ordinary Least Squares (OLS) requires that each series is integrated of order zero [i.e. I(0)]. If this is not the case then the possibility of spurious regression arises (Guajarati, 2003). However, it is possible a combination of individual I(1) series is I(0). If such a case happens then the series are said to possess a long run relationship and possibility of spurious regressions from such series is invalidated. One necessary condition of cointegration is that the series must be integrated of an order of more than zero (normally order 1) or at least have a deterministic trend. In this regard, it is logical that studies that test for the existence of a long run relationship among series first perform stationarity tests. In this study, the stationarity of the series were tested using the Augmented Dickey Fuller test. While the ADF test performs well when the serial correlation in the error terms are well approximated by a low order AR(p) process without any large negative roots, the tests is biased towards rejection of the null hypothesis in cases where the error terms follow an MA or ARMA process. Furthermore, the ADF test requires the choice of an optimal lag order, one in which guarantees that serial correlation is dealt with (Davidson & MacKinnon, 2004: 622). In this study, the

10

 

first issue was addressed by performing a robustness check using Kwiatkowski, Phillips, Schmidt, and Shin (KPSS) tests, which directly test for stationarity, unlike the unit root test which tests for stationarity in an indirect manner4. To deal with the second issue, the Schwarz information criterion was used to select the optimal lag length, the maximum lag order was set at 12 months, as the data being used was monthly data.

Two most commonly used methods for testing for cointegration are the Engle and Granger, and the Johansen cointegration approach. The former has lost popularity in recent years because among other weakness, it fails to identify multiple cointegration vectors and it suffers from the inability to accommodate the possibility of simultaneity in the causal relationship among variables. Such problems are solved by the latter approach which initially assumes that all variables are endogenous in the system. It is then possible to run a weak exogeneity test to distinguish between truly endogenous variables on which the identified cointegrating vectors will be normalised on. Given its relative superiority, the current study uses the Johansen approach. This approach involves applying the maximum likelihood to a VAR model assuming that the errors are white noise (Maddala and Kim, 1998). Following the practice in standard econometric literature, a typical )(kVAR model can be represented as:

tkit

k

ititt XXX ε+ΔΓ+Π=Δ −

=− ∑1 [1]

where tX ...),( 21 tt XX= denotes an nx1 vector of ( )1I stock market

indices and the identified macroeconomic variables5, tXΔ are all ( )0I , iΓ are n x n coefficient matrices; ktε are normally and independently distributed error terms; and Π is a long-run coefficient matrix, whose rank give the number of cointegrating vectors. Given that 1... +−ΔΔ ktt XX are all ( )0I , but tX is ( )1I , it is logical that, for equation (1) to be consistent iΠ should not be of full rank.

Otherwise, a full rank be would imply that all tX are ( )0I , thus invalidating the necessity of testing for cointegration6. A rank of iΠ = 0 is possible but it will imply that there are no long run relationships among the variables (Harris

                                                            4 Since the two techniques for testing for stationarity are widely explored in several empirical studies, their theoretical underpinning behind them will not be discussed here. For a discussion of the method see Gujarati (2003) and Brooks (2003) 5 Note that all variables are endogenous. 6 This will mean that all the variables are stationary at level terms thus OLS can be used provided the other Classical Linear Regression assumptions are satisfied.

11

 

1995). Usually iΠ has a reduced rank, that is )1( −≤ nr , in which case it can be decomposed as:

iΠ 'αβ= [2] where α is a rn× matrix and 'β is a nr × matrix. Then 1

'−tXβ are the r

cointegrated variables, 'β is the matrix of the long-run coefficients, and α has the interpretation of the matrix of error correction terms7. The rank of the matrix iΠ and the number of cointegrating relation(s) are determined using the two likelihood ratio (LR) test statistics, proposed by Johansen (1988). These are the trace statistic ( traceλ ) and the maximum eigenvalues ( maxλ ) with their test statistics given respectively as follows:

traceλ = ∑+=

∧

−−n

riiT

1

)1log( λ [3]

)1log( 1max +

∧

−−= rT λλ [4] where iλ is the thi − largest eigenvalue of the iΠ matrix in equation (2).

The trace statistic consecutively tests the null hypothesis that the number of cointegrating relations is r against the alternative of k cointegrating relations, where k is the number of endogenous variables. The maximum eigenvalue tests the null hypothesis that there are r cointegrating vectors against an alternative of r+1 (Brooks, 2002).

Once cointegration vectors are identified, it is now possible to estimate the Vector Error Correction Models (VECMs). This is done by first identifying the variables that are truly endogenous and exogenous using the weak-exogeneity test. The VECMs are now obtained by specifying the number of cointegrating vectors, trend assumptions used in identifying the vectors and normalising on the truly endogenous variables. The VECM framework restricts the long-run behaviour of the endogenous variables to converge to their cointegrating relationships, while accommodating short run adjustment dynamics. Finally, it is necessary to perform diagnostic checks on the residuals from the estimated VECM to ensure that they are white noise. Normally, serial correlation, heteroscedasticity and normality are tested. However, since we are dealing with financial data, there is no guarantee that the last two latter problems will be dealt with8. For this reason our concern in this study will be serial correlation.

                                                            7 This is Granger’s representation theorem. 8 For problems with financial data which cannot be captures by time series model, see Brooks

12

 

If cointegration is found in this study, then it would imply that although the stock market indices and macroeconomic variables are individually non-stationary, they move together in the long run. Moreover, since cointegration will imply that a valid VECM model exist, it would imply that, in contrast to weak form efficiency, lagged macroeconomic variables can be used to predict the behaviour of the stock market.

In this study, the short run relationships between the stock market and macroeconomic variables are further analysed using impulse response functions. Impulse response functions trace out the responsiveness of stock market returns to one standard deviation shocks in each of the macroeconomic variables in the VAR framework. Of importance here is the sign, magnitude and persistence of responses of stock market returns. Thus impulse responses can also be interpreted as measuring the weak form efficiency of the aggregate market and sectorial returns.

Given a VAR model such as that in equation [1] and assuming that the error terms follow a white noise process, then impulse responses are the coefficients of a moving average process that is obtained from the VAR equation (Lutkepohl and Saikkonen, 1997). The moving average takes the following form:

∑=

−+=k

sstst CX

0

εβ (5)

Where Xt denotes a linear combination of current and past one step ahead forecast error or innovations. The coefficients, βs can be interpreted as the response of one stock market returns to a one standard error shock of any of the macroeconomic variables. As in equation (1), the εt’s are also serially uncorrelated although they may be contemporaneously correlated.

The impulse responses are commonly estimated using the generalised impulse response proposed by Koop, Pesaran and Potter (1996) and Pesaran and Shin (1998), and the Cholesky decomposition proposed by Sims (1980). The former has the advantage over the latter in that it does not require orthogonolization of innovations as it does not vary with the ordering of variables in the VAR9 (Pesaran and Shin, 1998:17 and Aziakpono, 2006:8). For this reason, this study uses the generalized estimation criterion. The impulse response functions are only reliable when series are stationary.

                                                                                                                                                             (2003:380-382) 9 However, results from the two methods coincide if the shocks are uncorrelated.

13

 

3. EMPIRICAL RESULTS

3.1. Unit root tests As a preliminary analysis the graphical plots of the variables are shown in

Figure 1. Visually none of the series seem to be stationary as they all look time-varying. All the series except industrial production are trending, justifying the use of the trend assumption in testing for unit root in all cases except in the case of industrial production. Stationarity was tested and the results are reported in Table 3. As evident from Table 3, none of the series are stationary at level terms but they all become stationary at first difference. Thus all the series are integrated to the order of I(1).

Table 3: Stationarity Tests results

Variable ADF Level KPSS Level ADF 1ST Difference KPSS 1st Difference Order of Integration LALSI -1.92(0.637) 1.40(0.000) -12.11(0.000) 0.09(0.848) I(1) LFPI -2.17(0.502) 0.25(0.000) -12.34(0.000) 0.10(0.257) I(1) LIPI -1.63(0.777) 0.21(0.000) -10.67(0.000) 0.11(0.457) I(1) LMPI -1.86(0.670) 0.12(0.000) -12.09(0.000) 0.09(0.736) I(1) LGRPI -1.96(0.617) 0.27(0.000) -9.27(0.000) 0.91(0.878) I(1) LCPI -2.80(0.201) 0.25(0.000) -3.85(0.017) 0.10(0.257) I(1) LIPSA -2.18(0.213) 0.21(0.000) -3.71(0.005) 0.08(0.213) I(1) LUSGDP 1.02(0.919) 1.49(0.000) -1.90(0.056) 0.15(0.187) I(1) LNER -1.57(0.799) 0.28(0.000) -11.16(0.000) 0.14(0.796) I(1) NIR -1.48(0.834) 0.14(0.000) -14.50(0.000) 0.09(0.601) I(1)

Notes: p-values are in parenthesis. Source: Thompson Datastream (2009) and author’s own estimates using E-views 6

3.2. Johansen cointegration and VECM

Five VARs were specified for each of the stock market indices with all the macroeconomic variables, and cointegration was tested. The Johansen cointegration approach requires that an appropriate lag length and deterministic assumption is specified. While an extremely low lag length may lead to serial correlation, a lag length that is too high negatively impacts on the asymptotic properties of estimates especially if the sample size is small (Hall, 1991). Empirical studies have also shown that Johansen’s test statistics are sensitive to the chosen lag and one suggestion has been that information criteria should be used to determine the optimal lag order. Unfortunately, different information criteria tend to select conflicting VAR orders. Thus basing analysis on one information criteria could be misleading. To this end this study will initially employ information criteria, and then see the range of lags that have been selected by the information criteria. Cointegration is then sequentially tested beginning from the smallest lag length until meaningful cointegration results are

14

 

obtained10. Should no meaningful cointegration be found until the highest lag selected, this will be interpreted as an indication that there is no long run relation among the variables. Since monthly data is being used, twelve month was used as the maximum lag order as it is our considered belief that due to arbitrage, the stock market would adjust back to equilibrium should there be short run disequilibrium. The lag length results for the five models are presented in Table 4 and it is evident that different information criteria select different lags.

Table4: Lag length selection

Model 1 Model 2 Model 3 Model 4 Model 5 Lag

Criteria All Share Index Financial Index Industrial Index General Retailing Index Mining Index

LR 8 12 11 12 8 FPE 5 5 5 5 5 AIC 5 12 12 5 5 SC 3 3 3 3 3 HQ 4 3 3 3 4

Notes LR: sequential modified LR test statistic (each test at 5% level) FPE: Final prediction error AIC: Akaike information criterion SC: Schwarz information criterion HQ: Hannan-Quinn information criterion Source: Thompson DataStream (2009) and author’s own estimates using E-views 6

Table 5: Cointegration results Trace Max

Model Obs k A r < 0 r < 1 r<2 r < 0 r < 1

1 157 4 4 138.27[0.002] 93.37[ 0.022] 58.58[0.129] 44.89[0.0452] 34.79[0.121]

2 157 4 4 143.16[0.001] 92.09[ 0.028] 56.34[0.183] 51.07[0.0084] 35.75[0.101]

3 157 4 4 138.33[0.001] 88.01[0.060] 55.73[0.21] 50.32[0.0104] 26.39[0.213]

4 157 4 4 129.22[0.008] 92.29[0.027] 62.82[0.11] 36.92[0.262] NA

5 157 4 4 129.29[0.009] 93.88[0.021] 61.23[0.10] 35.42[0.341] NA

Notes: Obs is the number of observations in the model, k represents the chosen lag length, and A is the cointegration assumption used. P-values are in parenthesis. Source: Thompson DataStream (2009) and author’s own estimates using E-views 6

Cointegration was tested in the 5 models and the results were subsequently

tested for serial correlation. The results are reported in Table 5. For all the 5 models, lag order 4 under deterministic assumption 4 gave the most meaningful

                                                            10 By meaningful results here we mean cointegration and meaningful VECM should exist in that lag order and the residual obtained should be unserially correlated.

15

 

results. In models 1, 2 4 and 5, the trace statistic identified two cointegrating vectors, while the maximum eigenvalue identified one vector only in model 1 and 2 and zero cointegrating vectors for models 4 and 5. In the case of model 3, both the trace and the maximum eigenvalue identified one cointegrating vector.

Having identified cointegrating vectors, we now moved to estimating the VECM. Since estimating VECMs requires that we normalize on truly endogenous variable, weak exogeneity tests were performed for each of the models and the results are reported in Table 6. Given that our objective is to see whether macroeconomic variables explain the stock market in the long run, the main concern was whether the stock market indices were endogenous. Fortunately in all the models, the indices were endogenous and most of the macroeconomic variables were weakly exogenous, except for exchange rate and US GDP in model 3 and 4 respectively. However, exchange rate is only endogenous at 10% level. In the case of US GDP, we consider this result meaningless since macroeconomic variables of a small emerging economy like South Africa would rarely affect economic activity in the world largest economy. Furthermore, for experimentation, VECMs normalized on US GDP and the error correction coefficients were meaningless11. Thus, we only reported results for VECMs that were estimated by normalizing on the stock indices in Table 7.

Table 6: Exogeneity test results

Model LASI LCPI LIPSA LNER NIR LUSGDP

1 15.42 [0.00] 0.27 [0.60] 0.18 [0.67] 0.01 [0.90] 0.13[0.72] 0.00 [0.99]

2 6.81 [ 0.01] 0.52 [0.47] 2.59 [0.11] 0.14 [0.70] 0.38 [0.54] 2.86 [0.09]

3 18.89 [0.00] 2.70 [0.10] 0.04 [0.84] 3.12 [0.08] 0.57 [0.45] 5.05 [0.02]

4 9.46 [0.00] 0.00 [0.94] 1.60 [0.21] 0.35 [0.56] 0.08 [0.78] 8.79 [0.00]

5 4.44 [0.02] 0.47 [0.49] 0.91 [0.34] 1.37 [0.24] 0.16 [0.69] 1.63 [0.20]

Source: Thompson DataStream (2009) and Authors’ own estimates Eviews 6. P-values are in parenthesis.

                                                            11 ECM coefficient was not negative and significant. The results are not reported here but are available on request.

16

 

Table 7: VECM results Model

Intercept LCPI LIPSA LNER NIR LUSGDP R2 ECM S.Cor

1 0.45 -4.70 [-6.66]a

4.98 [7.84]a

0.39 [4.11]a

-0.01 [-0.12]

1.77 [2.02]b

0.4 -0.38 [-6.16]a

38.30 [0.37]

2 0.43 -8.82 [-6.49]a

2.67 [2.13]b

-0.62 [-3.35]a

0.19 [2.95]b

4.81 [6.38]a

0.4 -0.19 [-4.85]a

40.72 [0.27]

3 5.72 -4.42 [-7.17]a

7.67 [13.66]a

0.06 [1.78]c

-0.14 [-1.61]b

0.68 [2.01]b

0.4 -0.46 [-6.38]a

43.92 [0.17]

4 11.57 -4.03 [-2.14]b

3.58 [2.06]b

0.59 [2.31]b

-0.54 [-2.04]b

5.61 [5.24]a

0.3 -0.09 [-2.51]a

26.02 [0.89]

5 14.76 -1.54 [-0.18]c

12.23 [4.57]a

2.94 [2.75]a

-1.09 [-2.68]a

2.28 [1.44]c

0.3 -0.04 [-1.75]6

37.16 [0.42]

Notes: t-values in parenthesis. a, b, c denote the rejection of the null hypothesis of the variable being significant at 1%, 5% and 10% critical level respectively. The ECM is the short run adjustment coefficient of the VECM, and S.Cor is the serial correlation of the model. T- values in parenthesis. Source: Thompson DataStream (2009) and author’s own estimates using E-views 6

The error correction coefficients in all models are negative and significant

implying that all the stock market indices adjust back to their long run equilibriums relationship with macroeconomic variables if there is short run disequilibrium. The industrial index adjusts back to equilibrium fastest taking just over two months. Surprising, the speed of adjustment for the aggregate market index is slightly lower than that of the industrial index. The mining index adjusts slowest taking more than 2 years to adjust, with the general retail index adjusting second least taking just below 1 year to fully adjust to equilibrium. These results imply that industrial sector is the most weakly efficient sector while the mining sector is the least. The results signify that unexploited arbitrage opportunities exist mostly in the mining sectors followed by the general retail sector and least beneficial in the industrial sectors. The coefficients of determinations are quite high justifying the importance of macroeconomic variables in explaining the behavior of the stock market. All the models are robust for the serial correlation test.

The sign of coefficient for all the macroeconomic variables conform to a priori for all the models except the model for the financial index. For instance, interest rates have a positive and significant influence on the financial sector. This could be due to the fact that interest rates are more an earning than it is a cost to financial companies like banks. Thus an increase in interest rate would increase expected interest earnings. Furthermore, exchange rate negatively affects the financial sector. All the indices are elastic to changes in CPI and industrial production. Only the mining sector is elastic to changes in exchange rate. This is expected since the sector is heavily dependent on export earnings. The depreciation of the rand would therefore benefit them12. The mining sector

                                                            12 This serves as justification for the calls by miners that the South African rand is overvalued.

17

 

is also sensitive to changes in interest rates. As would be expected all the indices are very responsive to changes in world GDP. 3.3. Impulse Response Functions

To examine the signs and persistence of the short run response of the stock market to one standard error shocks in each of the macroeconomic variables, ten month ahead impulse response functions were estimated using the generalised response approach. Since impulse response functions are only reliable when series are stationary, the data on macroeconomic variables and stock market indices were differenced once to make them stationary. The results are presented in Figure 2 in the appendix.

The aggregate market and all the sectors respond negatively to changes in inflation although the financial and mining sectors seem to show insignificant positive responses in some rare instances. Responses of the financial and the general retail sectors seem to die off after eight months while the responses of the others seem to be persistent. While the mining and financial sectors seem to be non-responsive to short run shocks in industrial production, all the remaining sectors and the aggregate market respond positively but largely non-persistently. Generally, with the exception of some instances for the mining sector, all the sectors seem to negatively respond to short run shocks in exchange rate. Likewise short run innovations from the nominal interest rate tend to have mixed effects on different sectors. While the effects seem to be largely positive in the mining sector and mixed in the financial sector, it is negative in the other sectors. This finding is in line with the VECM results where a positive influence of interest rate on the financial sector was found. The finding however contradicts the VECM finding in the mining case where a negative influence was found. It is our considered view that the VECM result is more meaningful as interest rate is largely a cost rather than an earning for mining companies. Furthermore increases in interest rate are likely to strength the currency thus indirectly decreases mining earnings. However, the responses of the financial, industrial and general retail to interest rate shocks tend to be non-persistent. Finally, as expected, the aggregate market and all the sectors respond positively to short run innovations from the US GDP. Our impulse response findings are generally in line with the VECM results and with a priori expectation, and in overall the findings of this study is in line with studies (cf. Akin and Basti, 2008; Ibrahim, 2003; Pilinkus and Boguslauskas, 2009).

18

 

4. CONCLUSION This paper sets out to investigate whether there are significant long run and

short run relationships between macroeconomic variables and stock market behaviour in SA. Both aggregate market and sectoral data were used. Using Johansen’s multivariate cointegration tests, it was found that there is long run relationship between the identified macro-variables and stock market behaviour, both at an aggregate and sectoral level.

VECM models were estimated to analyse the short run dynamics of the relationship between the SA stock market behavior and macroeconomic variables. It was found that following a short run disequilibria, the industrial sector adjusts back to long run equilibrium fastest followed by the aggregate market index, with mining showing the worst sluggishness in adjustment. Generally the influence of macroeconomic variables on the SA stock market is in line with a priori expectation except in the case of the influence of nominal interest rate on the financial sector which is positive. This is reasonable given that the interest is rather more of an earning than a cost for most financial firms. The aggregate market and all sectors seem to be elastic to changes in inflation, industrial production and foreign GDP while only the mining index is elastic to changes in exchange rate. Nevertheless, the sizes of response of different across sectors to different macroeconomic variables differ. To further analyse the short run response of the stock market to short run innovations from macroeconomy, 10 month impulse functions were estimated. The signs of short run responses of the stock market (both at aggregate and sectoral levels) were largely in line with those from the VECM models and the responses were generally non-persistent.

These results have investment and policy ramifications. For instance the sluggish response of the mining and retail indices presents active investors with possible arbitrage opportunities in these sectors. On the other hand, since stock market development positively enhances economic growth, our findings are important for policy makers. Monetary and fiscal policies can be formulated in such a manner that they help foster stock market development.

One key finding of the study was that the magnitude, and also the direction of influence of macroeconomic variables on the stock market tend to vary across sectors. A question posed is what could be the reason for this? Could this be due to some micro-structural reasons or other issues that are peculiar to each of the sectors? While the current study tried to give possible explanations for these results, it should be noted that the focus of the study was merely on analyzing the influences of macroeconomic variables on the stock market.

19

 

Therefore, future research could empirically explore the reason(s) for differences in the magnitude of responses across sectors.

20

 

5. APPENDIX

Figure1: Graphical plot of each series

3.6

3.8

4.0

4.2

4.4

4.6

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LASI

1.7

1.8

1.9

2.0

2.1

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LCPI

3.6

3.8

4.0

4.2

4.4

4.6

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LFPI

3.6

3.8

4.0

4.2

4.4

4.6

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LGRPI

3.4

3.6

3.8

4.0

4.2

4.4

4.6

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LIPI

1.97

1.98

1.99

2.00

2.01

2.02

2.03

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LIPSA

3.2

3.6

4.0

4.4

4.8

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LMPI

5.4

5.6

5.8

6.0

6.2

6.4

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LMS

0.5

0.6

0.7

0.8

0.9

1.0

1.1

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LNER

1.6

2.0

2.4

2.8

3.2

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LPCO

2.88

2.92

2.96

3.00

3.04

3.08

3.12

95 96 97 98 99 00 01 02 03 04 05 06 07 08

LUSGDP

1.0

1.1

1.2

1.3

1.4

1.5

95 96 97 98 99 00 01 02 03 04 05 06 07 08

NIR

21

 

Figure 2: Impulse response functions

Impulse Response Function of LASI

Impulse Response Function of LFPI

Impulse Response Function of LIPI

Impulse Response Function of LGRPI

Impulse Response Function of LMPI

- .01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LASI) to D(LASI)

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LASI) to D(LCPI)

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LASI) to D(LIPSA)

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LASI) to D(LRER)

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LASI) to D(NIR)

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LASI) to D(LUSGDP)

Response to Generalized One S.D. Innovations

-.01

.00

.01

.02

.03

.04

2 4 6 8 10

Response of D(LFPI) to D(LFPI)

-.01

.00

.01

.02

.03

.04

2 4 6 8 10

Response of D(LFPI) to D(LCPI)

-.01

.00

.01

.02

.03

.04

2 4 6 8 10

Response of D(LFPI) to D(LIPSA)

-.01

.00

.01

.02

.03

.04

2 4 6 8 10

Response of D(LFPI) to D(LRER)

-.01

.00

.01

.02

.03

.04

2 4 6 8 10

Response of D(LFPI) to D(NIR)

-.01

.00

.01

.02

.03

.04

2 4 6 8 10

Response of D(LFPI) to D(LUSGDP)

Response to Generalized One S.D. Innovations

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LIPI) to D(LIPI)

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LIPI) to D(LCPI)

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LIPI) to D(LIPSA)

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LIPI) to D(LRER)

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LIPI) to D(NIR)

-.01

.00

.01

.02

.03

2 4 6 8 10

Response of D(LIPI) to D(LUSGDP)

Response to Generalized One S.D. Innovations

-.02

.00

.02

.04

2 4 6 8 10

Response of D(LGRPI) to D(LG RPI)

-.02

.00

.02

.04

2 4 6 8 10

Response of D(LGRPI) to D(LCPI)

-.02

.00

.02

.04

2 4 6 8 10

Response of D(LG RPI) to D(LIPSA)

-.02

.00

.02

.04

2 4 6 8 10

Response of D(LG RPI) to D(LRER)

-.02

.00

.02

.04

2 4 6 8 10

Response of D(LGRPI) to D(NIR)

-.02

.00

.02

.04

2 4 6 8 10

Response of D(LGRPI) to D(LUSG DP)

Response to Generalized One S.D. Innovations

-.01

.00

.01

.02

.03

.04

1 2 3 4 5 6 7 8 9 10

Response of D(LMPI) to D(LMPI)

-.01

.00

.01

.02

.03

.04

1 2 3 4 5 6 7 8 9 10

Response of D(LMPI) to D(LCPI)

-.01

.00

.01

.02

.03

.04

1 2 3 4 5 6 7 8 9 10

Response of D(LMPI) to D(LIPSA)

-.01

.00

.01

.02

.03

.04

1 2 3 4 5 6 7 8 9 10

Response of D(LMPI) to D(LNER)

-.01

.00

.01

.02

.03

.04

1 2 3 4 5 6 7 8 9 10

Response of D(LMPI) to D(NIR)

-.01

.00

.01

.02

.03

.04

1 2 3 4 5 6 7 8 9 10

Response of D(LMPI) to D(LUSGDP)

Response to Generalized One S.D. Innovations

22

 

6. LIST OF REFERENCES

Adam, M. and Tweneboah, G., 2008. “Macroeconomic Factors and Stock Market Movement: Evidence from Ghana.” Social Science Research Network. Akin, A. and Basti, E., 2008. “An assessment of trading behaviour and performance of foreign investors in Istanbul stock exchange.” Transformations in Business & Economics. 7, 3: 146-15. Allen, D. E. And McDonald, G., 1995. “The long-run gains from international equity diversification: Australian evidence from cointegration tests.” Applied Financial Economics. 5: 33-42. Aziakpono, M.J., 2007. “Financial and monetary autonomy and interdependence between South Africa and the other SACU countries.” Paper presented at Annual meetings of the Allied Social Science Associations, January 5-7, 2007, Chicago, IL. Bodurtha, J.N., Cho, D.C., and Subnet, L.W., 1989. “Economic Forces and the Stock Market: An International Perspectives.” The Global Finance Journal 1, 1: 21-46. Brooks, C. 2002. Introductory econometrics for finance. Cambridge: Cambridge University Press. Chinzara, Z. and Aziakpono, M., 2009. “Integration of the South African equity market into the major world stock markets: Implications for portfolio diversification.” African Finance Journal 1: 95-119. Deb, G. S. and Mukherjee, J., 2008. “Does stock market development cause economic growth? A time series analysis for Indian economy.” International Research Journal of Finance and Economics, 21.

Demirguc-Kunt, D. and Levine, R., 1996a. “Stock Markets, Corporate Finance and Economic Growth: An Overview.” The World Bank Economic Review 10, 2: 223-239. Dickey, D.A., and Fuller, W.A., 1979. “Distribution of Estimators for Time Series Regressions with Unit Root.” Journal of the American Statistical Association. 74: 427-431. Engle, R. F., and Granger, C. W. J., 1987. “Cointegration and error correction representation, estimation and testing.” Econometrica. 55: 251-276.

23

 

Fama, E. F., 1965. “Random walks in stock market prices.” Financial Analysts Journal. 21: 55-90 Fama, E., 1970. “Efficient Capital Markets: A review of theory and empirical work.” The Journal of Finance. 25, 2: 383-417. Fama, E. F., 1981. “Stock Returns, Real Activity, Inflation and Money.” The American Economic Review, 71, 4: 545-565. Feldstein, M., 1980. “Inflation and the stock market.” American Economic Review. 70, 5: 839-847. Gordon, M.J., 1959. “Dividends, Earnings and Stock Prices.” Review of Economics and Statistics. 41:99-105. Granger, C. W. J., 1986. “Developments in the study of cointegrated economic variables.” Oxford Bulletin of Economics and Statistics. 48: 213-228. Gujarati, D.N., 2003. Basic Econometrics (4e). New York: McGraw-Hill/Irwin. Gunasekarage, A., Pisedtasalasai, A., Power, D.M., 2004. “Macroeconomic influence on the Stock Market: Evidence from an Emerging Market in South Asia.” Journal of Emerging Market Finance. 3, 3: 285-304. Hall, S. G., 1991. The effects of varying length of VAR models on the maximum likelihood estimates of cointegrating vectors. Scottish Journal of Political Economy. 38, 4: 317-23. Harris, R., 1995. Using cointegration analysis in econometric modelling. London: Prentice Hall. Ibrahim, M. H., and Aziz, H., 2003. “Macroeconomic variables and the Malaysian equity market: a view through rolling subsamples.” Journal of Economic Studies, 30, 1: 6-27. Jefferis, K.R. and Okeahalam, C.C., 2000. “The Impact of Economic Fundamentals on stock markets in Southern Africa.” Development South Africa. 17, 1: 24-51. Johansen, S., 1988. “Statistical analysis of Cointegration vectors.” Journal of Economic Dynamics and Control. 12, 231-254.

24

 

Johansen, S., and Juselius, K., 1990. “Maximum Likelihood Estimation and Inference on Co integration with Application to the Demand for Money.” Oxford Bulletin of Economic and Statistics. 52: 169-210. Kose, M. A., Parasad, E. S., Rogoff, K., and Wei, S., 2006. “Financial Globalization. A reappraisal.” NBER Working paper No.12484. Cambridge, MA: NBER.

Levine, R., and S. Zervos., 1998. “Stock Markets, Banks, and Economic Growth.” American Economic Review, Vol. 88:537-58. Maysami, R.C., HOWE, L.C., and HAMZAH, M.A., 2004. “Relationship between Macroeconomic Variables and Stock Market Indices: Cointegration Evidence from Stock Exchange of Singapore’s All-S Sector Indices.” Journal Pengurusan. 24: 47-77. Maysami, R.C., and Koh, T.S., 1998. “A vector error correction model of the Singapore stock market.” International Review of Economics and Finance. 9: 79-96. Mishkin, F.S., 2004. The Economics of Money, Banking and Financial Markets (7e). Cape Town: Pearson Addison Wesley. Nikkinen, J., Omran, M., Sahlstrom, P., and Aijo, J., 2006. “Global stock market reactions to scheduled U.S. macroeconomic news announcements.” Global Finance Journal. 17: 92-104. Oludamola, D., 2006. “Macroeconomic determinants of Stock Market Behaviour.” Unpublished Masters Thesis. Rhodes University. South Africa. Pesaran, H. H. and Shin, Y., 1998. “Generalized impulse response analysis in linear multivariate models.” Economics Letter. Elsevier. 58(1), 17-29. Pilinkus, D., and Boguslauskas, V., 2009. “The short-run relationship between stock market prices and macroeconomic variables in Lithuania: An application of the impulse response functions.” Inzinerine Ekonomika-Engineering Economics, 5. Sadorsky, P., 1999. “Oil price shocks and stock market activity.” Energy Economic Journal. 21: 449-469.

Smal, M.M., and DE Jager, S., 2001. “The Monetary Transmission Mechanism in South Africa.” South African Reserve Bank Paper, 16.

25

 

Vuyyuri, S., 2005. “Relationship between Real and Financial Variables in India: A Cointegration Analyses.” Social Science Research Network.