The Kuala Lumpur stock market and economic factors: a general-to-specific error correction modeling test.

ABSTRACT

This study applies the Error-Correction Modeling technique to examine the short-run dynamic adjustment and the long-run equilibrium relationships between the four macroeconomic variables and stock returns of Kuala Lumpur stock market. The macroeconomic variables analyzed are interest rate, inflation rate, exchange rate, and the industrial productivity using monthly observations from January 1990 through June 2002. In addition, a dummy variable is included to capture the impact of the 1997 Asian financial crisis. Econometric results support the existence of short-run dynamic adjustment and the long-run, equilibrium relationships between the macroeconomic variables and the Kuala Lumpur stock market.

1. INTRODUCTION

Studies on the short and the long-run relationships between economic variables are abundant, especially with respect to developed countries. Schwert's (1981) study shows that growth of industrial production is a major determinant of long-run stock returns. Geske and Roll (1983) showed that the real interest rate affect on stock return was significant but often small in most countries of their studies. The development of cointegration technique by Granger (1986) and Engle and Granger (1987) has encouraged many researchers to examine the relationships between economic variables and stock markets. A set of time-series variables are said to be cointegrated if they are integrated of the same order and a linear combination of them is stationary. Such linear combination would then point to the existence of a long-term relationship among the variables. An advantage of cointegration analysis is that through building an error-correction model (ECM), the dynamic co-movement among variables and the adjustment process towards long-term equilibrium may be examined. Fama (1990) argued that stock price reflects expectations of earnings, dividends, interest rates, and information about future real economic activity may be reflected in the stock price before it occurs. Moreover, stock returns will affect the wealth of investors which in turn affect the level of consumption and investment. The findings of Asprem (1989), Fama (1990, Bulmash and Trivoli (19991) show that there is a negative relationship between interest rates and stock returns in Korea. A significant positive relationship is observed between industrial production and Japanese stock returns in the long-run by Gjerde and Sattem (1999), fama (1990) and Asprem (1989). Asprem (1989), for example, compared the effects of economic factors on the stock markets of 10 European countries while Bulmash and Trivoli (1991) did similarly in the US market. Peiro (1996) compared such relationships in three European countries with the U.S. Cheng (1995) and Poon and Taylor (1991) examined the UK market, and Gjerde and Settem (1999) researched on Norwegian data. The number of similar studies using data from Asian markets is considerably more limited, and with few exceptions, the research has concentrated on the Asia's largest economy--Japan (Hamao, 1988; Brown and Otsuki, 1990; Kaneko and Lee, 1995). Most recently Maysami and Hui (2001) examined the short-run and the long-run relationships between stock returns and interest rate, inflation, money supply, exchange rate, real economic activities of Japan and South Korea. Using general-to-specific approach to error correction modeling during the period Q1 1986 to Q4 1998, their results suggest the existence of equilibrium relationships between macroeconomic variables and the stock returns of the two countries. Their findings of the positive relationship between industrial production and Korean stock returns are similar to those of Kwon et al. (1997), Kaneko and Lee (1995), and Mukherjee and Naka (1995). The aim of the study of Maysami and Hui (2001) is to extend Mukharjee and Naka's (1995) analysis of the Japanese market; and Kwon at al. (1997) examination of the Korean case.

Malaysia, one of the high performing economies in Southeast Asia achieved an impressive growth over the past three decades. Her strong economic performance continued during the 1990s prior to the Asian financial crisis in 1997. GDP averaged about 8.5 per cent a year; unemployment was below 3 per cent; prices and exchange rate remained stable; and international reserves were robust. From 1970 to the mid 1990s, the country's high investment ratio facilitates the dramatic shift in the structure of the economy from agriculture and mining to a growing manufacturing. The strong performance in the manufacturing sector contributed to the rapid growth of the country's GDP. Given its strong linkages with other sectors in the economy, this had spillover effects which continued to support strong growth in several sub-sectors of the economy. In the mid 1990s exports declined and a large current account deficit developed in the context of a gradual appreciation of the effective exchange rate. While the investment-led growth strategy was successful in raising output and income, investment quality had deteriorated. This eventually led to weakness in the banking and corporate sectors, exposing the economy to the contagion of the Asian crisis. As market confidence increasingly diminished large portfolio outflows took place, and equity and property values declined substantially. The ringgit came under tremendous pressure. This heightened upward pressure on domestic interest rates, intensified outflows of ringgit funds, and exacerbated banks' liquidity problems and overall financial distress. The initial response of the authorities was to increase interest rates and curb fiscal deficit in an attempt to anchor market confidence in the financial system. Anticipation of further devaluation of the ringgit increased. By the summer of 1998, the stock market had fallen to its lowest level in recent history. However, the growth of real economy recorded strong performance in 2000 and 2001 with real GDP growth of 8.8 percent. The favorable performance of the real sector had contributed towards enhancing the health of the financial sector and facilitating efforts in restructuring the corporate sector. The overall macroeconomic fundamentals is expected to remain strong in future.

This study thus extends the literature by examining the case of Malaysian market applying Hendry's (1986) general-to-specific approach to error correction modeling (ECM) analysis. The empirical results are based on Augmented Dickey-Fuller (ADF) & Phillips-Peron (PP) unit root tests of stationarity, and the Error-Correction Modeling (ECM) tests. The rest of this paper is organized as follows. Section 2 discusses data sources and the definition of variables. The hypotheses and the methodology are discussed in section 3. Empirical results are discussed in section 4, and the last section V concludes the paper.

2. THE DATA SERIES AND THE VARIABLES

This study focuses on five macroeconomic variables, such as short term interest rates ORS), exchange rates of Malaysian ringgit (EXR) in terms of US dollar, inflation rate (INF), industrial productivity (INDP), and the Kuala Lumpur composite stock price index (KSP). The data on these variables are obtained from the Bloomberg Financial Databank, and consists of monthly series from January 1990 to June 2002. We used the change in industrial production (INDP) as a proxy for GDP growth rate since monthly series for the latter is not available. Definitions of variables with their transformations are presented in Table 1.

The stock indices are not adjusted for dividend payments. Although stock returns consist of both price change and dividends, the study considers only the price variation component since dividends in absolute terms tend to be stable over time (Asprem, 1989). It is the movements in price, which constitute the volatile component of the stock returns. Since the objective of this study is to explain the variability in stock returns, the omission of the dividend payments should not pose a problem. As pointed out by Fama (1981), a shortcoming of regressing the changes in stock returns and the macroeconomic variables is the use of actual growth rates of the variables instead of anticipated growth rates. Asprem (1989). However, argued that most investors would form their expectations based on all current available information. Therefore, it would be reasonable to assume that the actual values of the variables are unbiased estimates of the ex-ante expectations.

3. HYPOTHESIS AND METHODOLOGY

3.1 Hypothesis

I hypothesize a relationship between the Kulala Lumpur stock market and four macroeconomic variables: exchange rate, interest rate, inflation rate, and industrial productivity; and additionally a dummy variable to capture the effects of the Asian financial crisis.

Depreciation of domestic currency leads to a relative decrease in price of a country's products in foreign markets, an increase in exports and a decrease in imports; and hence increase cash flows into the country. The relationships between exchange rates and stock prices is thus become an issue for empirical studies. The intuition behind the relationship between interest rates and stock prices is straightforward. An increase in the rate of interest raises the opportunity cost of holding cash and is likely to lead to a substitution effect between stocks and other interest bearing securities. Additionally, changes in both short-term and long-term rates are expected to affect the discount rate in the same direction via their effect on the nominal risk-free rate (Mukherjee and Naka, 1995).

Since inflation is positively related to money growth rate (Fama, 1981), an increase in the money supply may lead to an increase in the discount rate, and lower stock prices. However, this negative effect may be countered by the economic stimulus provided by money growth which would likely increase cash flows and stock prices (Mukherjee and Naka, 1995). Economic theory suggests a positive relationship between the unexpected inflation and stock returns as equities are "hedges" against (unanticipated) inflation since they represent claims to real assets. Fisher (1931), citing market efficiency and noting that the nominal interest rate can be expressed as the sum of an expected real return and an expected inflation rate, concluded the same. In addition, he believed that the real and monetary sectors of the economy are largely independent and that the expected real return is determined by real factors such as the productivity of capital, investor time preferences and tastes for risk. Hence, the expected real return is determined solely by real factors and expected real returns are independent of inflationary expectations. On average, investors are compensated for changes in purchasing power. Hence, the hypothesized relationship between stock prices and inflation is positive.

Finally, stock valuation involves discounting cash flows or expected dividend streams over long periods in the future, the price of a firm's stock reflects investor's expectations of future earnings, which are likely to be influenced by measures for real activity. Kwon et al. (1997) found the Korean stock market to be significantly related to real economic activities while Kaneko and Lee (1995) and Mukherjee and Naka (1995) found positive relationships between the industrial production and stock returns. Thus I hypothesize a positive relationship between industrial growth and the stock return.

3.2. Methodology

Economic analysis suggests that there is a long-run equilibrium relationship between the economic variables involved in economic theory under investigation. Applied econometric analysis in trying to estimate these long-run relationships implicitly considers the 'constancy doctrine" of the variables involved, in terms of means and variances being constant while not dependent on time. However, empirical research has shown that in most cases the 'constancy doctrine' is not satisfied by the time-series data. Recent econometric theory and empirical studies have given mounting evidence to this problem in properties of time series regression analysis. Nelson and Posser (1982) and Perron (1988) reported that a large number of macroeconomic time series data for the U.S. are characterized by unit root non-stationary processes. Under these circumstances, the conventional t and F tests based on estimation methods which consider the 'constancy doctrine' (assumed staionarity) are no longer valid giving misleading inferences, i.e., 'spurious' regression (Granger and Newbold, 1974). Banerjee et al. (1993) even argued that neither t nor F tests have the standard distribution of stationary series, so there is a tendency to reject the null hypothesis when in fact it should not be rejected.

A possible way to avoid this problem is first-differencing all economic data before running a regression (Box and Jenkins, 1970). Although this can help solve the question of spurious regressions, Enders (1995), suggested that valuable information concerning the long-run relationships among the levels of series postulated by economic theory may be removed following this methodology. As a result researchers in the field after several revisions included the systematic verification of non-stationarity of the time series economic variables and cointegration, among others. Cointegration analysis (Granger, 1986; Engle and Granger, 1987; Johansen, 1988; Johansen and Juselius, 1990) has been regarded as perhaps the most revolutionary development in econometrics since the mid 1980s. Cointegration analysis refers to a group of variables that drift together, although individually they are non-stationary in the sense that they tend upwards and downwards over time. This common drifting of variables makes linear relationships between these variables over long period of time thus translating into equilibrium relationships of economic variables. If these linear relationships do not hold over long period of time then the corresponding variables are 'not-cointegrated'. Generally, cointegration analysis is a technique used in the estimation of the long-run or, equilibrium parameters in a relationship with non-stationary variables and is used for testing the dynamic (error-correction) models (ECM) in order to verify the validity of underlying economic theories.

Thus the solution is the use of cointegration and error-correction modeling approaches (Hendry, 1986; Engle and Granger, 1987). Augustine and Shwiff (1993) summarize the four desirable features of ECM: (i) it avoids the possibility of spurious correlation among strongly trended variables; (ii) the long-run relationships that may be lost by expressing the data in differences to achieve stationarity are captured through inclusion of lagged levels of the variables on the right-hand side; (iii) the specification attempts to distinguish between short-run (first- differences) and long-run (lagged-levels) effects; and (iv) it provides a more general lag structure, and does not impose too specific of a structure on the model. Granger (1986) showed that a necessary condition to conclude that a long-term relationship exists is that the series must be cointegrated. For example, two time series, [x.sub.t], and [y.sub.t], both of which are individually non-stationary or 1(1). if there exists a non-zero constant [beta] such that [z.sub.t] = [x.sub.t] - [beta]'[y.sub.t] is a stationary or 1(0) process, then [x.sub.t] and [y.sub.t] are said to be cointegrated with a cointegrating parameter ('. This implies that the set of 1(1) variables (xt, yt) does not diverge over time since zt has no trend in mean.

Combined with ECM, cointegration provides the tools to quantify both the long-run relationship and the short-run deviations from equilibrium. If cointegrating relationships exist among a set of 1(1) variables, then Granger Representation Theorem suggests there is a dynamic error correction representation of the data. This implies one can estimate an ECM that takes into account the short-run dynamics of all variables included in the cointegrating regression. In the simple case of two variables, xt and yt, the ECM can be expressed as

(1) (xt = -(1zt-1 + {lagged((xt, (yt)}(1t

(2) (yt = -(2zt-1 + {lagged((xt, (yt)}(2t

where zt = xt - ('yt is 1(0) and represents the error-correction term which displays the size of the preceding error derived from the long-run equation, (1 and (2 are white noises and either (1 or (2 (or both) is non-zero. In this setting, xt and [y.sub.t] are each function of distributed lags of first differences of [x.sub.t] and [y.sub.t] as well as the one period lag of the error-correction term. The error-corrections in equations (I) and (2) may be interpreted as the dis-equilibrium mechanism that forces the economy to the equilibrium [x.sub.t] = [beta][y.sub.t] in the long-run.

The error-correction stock return equation in this study consists of two parts. The first is a long-run equilibrium stock return (cointegration equation) that may be written as follows;

(3) LKS[P.sub.t] = [[beta].sub.0] + [[beta].sub.1] LIRSt + [[beta].sub.2] IN[F.sub.t] + [[beta].sub.3] LEX[R.sub.t] + [[beta].sub.4] LIND[P.sub.t] + [e.sub.t]

where KSP is the desired Kuala Lumpur stock prices, IRS is the short term interest rate, INF is inflation, EXR is exchange rate, and INDP is industrial production. L is the natural log.

Estimation of equation (3) using non-stationary data will lead to unreliable t-statistics, as the underlying time series would theoretically have infinite variances. Applied studies usually use the augmented Dickey and Fuller (ADF) and Phillips-Perron (PP) tests to check for stationarity in data series. Both ADF and PP tests on the variables in levels and first-differences to check for stationarity or unit roots. A desirable feature of the ADF test is that it allows for heteroskedasticity as well as serial correlation in the error terms, thus compensating for the misspecification of the dynamic structure of time series.

ADF requires running a regression of the first difference of the series against the series lagged once, lagged difference terms, and a constant with a time trend such as

(4) [DELTA][X.sub.t] = [[lambda].sub.0] + [[lambda].sub.1] [X.sub.t-1] + [[lambda].sub.2] + [SIGMA] [[lambda].sub.i] [DELTA][X.sub.t-i] + [[epsilon].sub.t]

i = 1 .... k

where [DELTA] is the first difference operator, [[epsilon].sub.t] is an error term, and k is the number of lagged first differenced term and is determined such that [[epsilon].sub.t] is approaching white noise. The [H.sub.0] hypothesis that [X.sub.t] is non-stationary time series translates into [H.sub.t]: [[lambda].sub.1] = 0. The output of the ADF test consists of the [iota](tau)-statistic on estimated coefficient of the lagged variable ([[lambda].sub.1]) and the critical values for the test of a zero coefficient. If the estimated ADF statistic is larger (in absolute) than its critical value then the null is rejected suggesting that the series is stationary (i.e., the [H.sub.0] is rejected if [[lambda].sub.1] is significantly negative). The choice of optimal lag-lengths used in the unit root tests is determined by applying Akaike (AIC) and Schwarz (SC) information criteria.

An alternative test for a unit root was developed by PP. Like the ADF test, the PP test is a test of the hypothesis [rho] = 1 in the following equation:

(5) [DELTA][X.sub.t] = [[alpha].sub.0] + [[alpha].sub.1]T + [rho] [X.sub.t-1] + [[??].sub.t]

Unlike the ADF test, there are no lagged difference terms in PP test. Instead, the equation is estimated by OLS and the t-statistic of coefficient [rho] is corrected for serial correlation in [[??].sub.t].

The next step is to test whether the time series concerned are cointegrated given that they are all I(1). In other words, even if each individual time series is level-non-stationary, it has to be determined if the linear combination of these series as suggested by equation (3) is level-stationary. This requires the application of OLS techniques to the estimation of the cointegration regression. The residual from such cointegrating regression (equation 3) is then examined using the ADF test to see whether it is I(0).

(6) [DELTA][e.sub.t] = -[rho][e.sub.t-1] + [SIGMA][[gamma].sub.t][DELTA][e.sub.t-1] + [e.sub.t]

The rejection of the null hypothesis [rho]=0 requires that the estimated residual series be stationary. No intercept or time trend is included in equation (6) since the et must have a zero mean and we do not expect it to have a deterministic trend. If [e.sub.t], is 1(0), then it can be concluded that the stock return equation is cointegrated and has a long-run cointegrating relationship. Intuitively, actual stock prices do not always equal what investors wish to hold on the basis of long-run factors specified in equation (3). Therefore, the second part of our stock price model is a dynamic error-correction equation (ECM) of the form

(7) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where AFC is a dummy variable for Asian financial crisis, [[epsilon].sub.t] is the short-run random disturbance term, [DELTA] refers to the first-difference operator, Ji represents the number of lags, and [e.sub.t-1] is the lagged value of the long-run random disturbance term.

Equation (7) yields the short-run determinants of stock returns, which include among others, current and past changes in the macroeconomic variables and the lagged value of the residual from long-run stock price function from equation (3), The parameter [lambda] which appears with [e.sub.t-1] in equation (7) is the error-correction coefficient. The presence of [e.sub.t-i] reflects the presumption that actual stock price does not always equal what the investor expects on the basis of long-run macroeconomic factors. In the short-run, the stock price attempts to correct any short-run disequilibrium and adjusts to its long-run equilibrium. The parameter [lambda] measures the role such disequilibrium play in explaining the short-run movements in stock price (speed of adjustment) and it is expected to be negative.

The Engle-Granger (1987) procedure has been criticized on the grounds of the small-sample bias present in the ordinary least square (OLS) estimation of the cointegrating equation. This bias carries over into the prediction of the dis-equilibrium errors and hence into the second stage estimates of the short-run parameters. Therefore, Benerjee et al. (1986) suggest that it may be preferable to carry out the estimation of long- and short-run parameters in a single step. Our study will formulate the ECM using Hendry's (-1986) general-to-specific approach which starts with a general framework and test down to a suitably final model. Both long- and short-run elasticities are estimated together. Hendry's (1986) approach allows the restrictions implied by the cointegrating regression to be relaxed. That is the [e.sub.t-1] in equation (7) is replaced by the lagged levels of the variables in equation (3), so that the short- and long-run parameters are jointly estimated. Estimates of the level variables then reveal the long-run effects of the repressors. Substitution of equation (3) into equation (7) yields a combined ECM equation.

(8) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [[delta].sub.0] = [[PSI].sub.0] - [[beta].sub.0][lambda] and [[PHI].sub.1] = [lambda]

Equation (8) can be estimated using consistent estimation procedures, and all parameters of equation (3) and (7) may be recovered from those of equation (8). For example, the error correction coefficient [lambda] is [[PHI].sub.1] the long-term interest rate elasticity [[PHI].sub.2]/[[PHI].sub.1], long term inflation elasticity is [[PHI].sub.3]/[[PHI].sub.1], the long-term exchange rate elasticity is [[PHI].sub.4]/[[PHI].sub.1] and the long-term industrial production elasticity is [[PHI].sub.5]/[[PHI].sub.1].

Although InKS[P.sub.t-1], In IR[S.sub.t-1], InIN[F.sub.t-1], In EX[R.sub.t-1] and In IND[P.sub.t-1] have unit root, nonstationary variables, Thomas (1997) postulates OLS estimates can still be applied since assuming cointegration, there is a linear combination of variables that is I(0). The existence of such a linear combination can be checked by testing the residuals from equation (8) for stationarity. There is evidence that small-sample properties of estimates obtained in this way are better than those of the two-stage Engle-Granger (1987) estimates.

The general-to-specific approach has the advantage that if the general model is rigorously tested for misspecification, the possibility of any dynamic misspecification is reduced in the final model (Thomas, 1997). Certain tests of misspecification should then be performed to assess the reliability of the final 'best' models. We remove variables or lags from equation (8) using linear restriction imposed and goodness-of-fit tests until a reliable 'best' model is achieved.

4. EMPIRICAL RESULTS

ECM and cointegration analysis involves four steps. First, the order of integration for each of the economic variables of our study must be determined. Second, cointegrating regressions should be estimated with OLS method using the variables with the same order of integration. Third is the test for unit root of the residuals of the cointegrating regressions. The fourth step is to construct the error-correction models.

4.1. Unit Root Tests

In order to test for the presence of stochastic non-stationarity (unit root) in the data, we first investigate the integration order of individual series using ADF and PP tests with drift and trend.

Table 2 provides the statistical results for both tests in levels and first difference series along with Mackinnon critical values. Statistical results do not reject the null hypothesis of unit root for level series of LINDP, LEXR, LIRS, LKSP, but reject the hypothesis in the case of INF. Thus both ADF and PP tests confirm unit root in level series in all cases except for inflation and stationary series in first differences. Thus the results show that all macroeconomic variables (except INF) under study are integrated of order one.

4.2 Error Correction Model

The objective of the general-to-specific model is to begin with a framework that builds a 'nested' model embracing a series of simpler models.

These models represent all alternative hypotheses that require consideration. Results of the ECM are presented in table 2. We first estimated the model with 4 months lags for each variable and subsequently removed lags that were insignificant. It is important that the models which are eventually selected based on the general-to-specific approach must be correctly specified in that the residuals should be white noise. We reject the hypothesis of autocorrelated error using higher order serial correlation (up to 4 lags) Lagrangian multiplier (LM) test to examine if all coefficients of AR(4) for the residuals are equal to zero. The hypothesis of heteroscedasticity examined by ARCH (autoregressive conditional heteroscedasticity) test is also not supported, Jarque and Bera (1980) tests shows that all residuals in the stock returns equations are normally distributed. Regression specification test (RESET) of Ramsey (1969) for misspecification due to incorrect functional form and omitted variables bias is used, and the test results suggest that the models do not suffer from any misspecification. The null hypothesis of non-cointegration is also tested because Granger (1986) theorem suggests that cointegration is not only a necessary ut also a sufficient condition for an error correction representation to exist. The estimated ADF statistics as presented in table 3 rejects the null-hypothesis that the residuals from non-stationary series. We therefore conclude that residuals are stationary, indicating cointegrating relationships between the stock returns and the other economic variables.

Estimated result supports the hypothesis of a negative relationship between interest rate and the stock returns. This is consistent with the findings by Asprem (1989), Fama (1990), Bulmash and Trivoli (1991), and Maysami and Hui (2001) that follows the theory of the financial valuation model. An increase in interest rate increases the cost of borrowings to the firms and the investors expect a lower profitability. There is a decrease in demand for the firm's shares and as a consequence a decline in stock prices. On the other hand, a positive relationship between inflation and the stock returns is detected. This is in support of the Fisher hypothesis that holding stocks may be an effective hedge against inflation. The inflation rate was contained within a limit in Malaysia over the period of study. With inflationary expectations and a healthy economy prior to the crisis, Kuala Lumpur stock market did well. Marshall (1992) suggested that if inflation was caused by a money shock, the relationship between stock returns and inflation would be positive as higher liquidity would lower the rate of interest, causing investors to transfer their cash holdings to stock for potential capital gains. The increase in demand for stock drives up prices. Another justification was put forward by Kaul (1987) who argued pro-cyclical monetary policy is the cause of the positive relationship between the inflation and the stock prices. Yet another explanation is given by Fama and Gibbons (1982) who suggest that increases in expected inflation may also indicate the market of potential increases in real activity and greater production, and hence higher stock returns.

The ringgit (Malaysian dollar) and stock returns are inversely related in the short-run. The depreciation of the ringgit against the U.S dollar would make Malaysian Exports more competitive compared to their close substitutes produced in the U.S. The potential benefits of exports in firms' cash flows may be reflected in higher stock returns. That implies the economic exposure is the dominant force in exchange rate-stock return relationship. In the long-run, however, the relationship is positive. A plausible justification is that the exchange rate is a reflection of the economy's health, a weaker ringgit could mean a troubled economy and future decline in real activities. The depreciation of the ringgit value during the financial crisis after which the Kuala Lumpur stock market collapsed validates this hypothesis. Moreover, foreigners might loss confidence to invest in Kuala Lumpur equity market should they expect the depreciation of the ringgit value to continue in the long-run.

A statistically significant positive relationship is found between industrial production and the Kuala Lumpur stock returns both in the short run and in the long-run. Fama (1990) argued that stock price reflects expectations of earnings, dividends, interest rates and future activity and the information about future real activity may be reflected in the stock price before it occurs. Moreover, stock returns will have wealth effect which in turn affect the level of demand for consumption and investment goods. In Malaysia, the development of the financial sector has lagged behind the real sector of the economy. The positive relationship between industrial production and the stock returns, even in the short-run, is not surprising. This is in support of Schwert's (1991) assertion that growth rate of industrial production is a major determinant of long-horizon stock returns.

The error correction term indicates the speed of adjustment towards the long-rum equilibrium, and is found to be statistically significantly negative in the stock return. The larger the value of the error-correction term, the faster the disequilibria is adjusted in the short -run so that long-run equilibrium relationship holds. The speed of adjustment is 0.533, implying that about 53 percent of the previous deviation between the actual and the desired stock prices are corrected in each month. Finally, the Asian financial crisis (AFC) is found to have statistically negative effect on the stock returns.

5. CONCLUSIONS

The short-run (dynamic) and the long-run (equilibrium) relationships between stock returns of Kuala Lumpur stock market and the short term interest rate, inflation, exchange rate, industrial productivity and a dummy variable to reflect the Asian financial crisis were investigated in this study using the monthly data from January 1990 to June 2002. Applying Hendry's (1986) general-to-specific approach to error-correction model, our statistical results suggest the existence of stable long-run (cointegrating) relationships between macroeconomic variables and the stock returns. All macroeconomic variables' affect the stock returns are consistent with the hypothesized signs and are also found to be statistically significant at the traditional level of significance. Granger (1986) argued that the existence of cointegration relationship and the fact that asset prices might be predicted would mean that the stock market is not entirely efficient. Defining market efficiency as the absence of arbitrage opportunities, Dwyer and Wallace (1992), however, stressed that cointegration does not necessarily violate the notion of information efficiency. The short-run affect of the variables with some lags are also found to be statistically significant with expected signs. The error-correction model is, however, known to be sensitive to the choice of lags and the dimensions. Due to a limited number of observations and to take into account the impact of the financial crisis, our study uses four macroeconomic variables and one dummy variable. Future studies may consider other time series data.

TABLE 1: Definitions of Variables and Time Series Transformations

Variables:   Definition of Variables:

LKSP:        Natural logarithm of the Kuala Lumpur composite Index
             (KLCI). KLCI is a broad-based capitalization weighted
             index of 100 stocks designed to measure the performance
             of the Kuala Lumpur Stock Exchange. The Index has a base
             value of 95.83 as of January 3, 1977.

LINDP:       Natural logarithm of the month-end industrial production
             index. This index is used as a proxy of the level of
             economic activity in the country and figures are
             seasonally adjusted.

LIRS:        Natural logarithm of the month-end interbank offer rate.

LCPI:        Natural logarithm of the month-end Consumer Price Index.

LEXR:        Natural logarithm of the month-end the Malaysian ringgit
             per US dollar.

                                              Definition of
First Difference Transformation:              Transformation:

[DELTA]KSP = KS[P.sub.t] - KS[P.sub.t-1]      Monthly return of the
                                              KSP index.

[DELTA]IRS = IR[S.sub.t] - IR[S.sub.t-1]      Monthly change of the
                                              interbank rate.

INF = CP[I.sub.t] - CP[I.sub.t-1]             Monthly realized
                                              inflation rate.

[DELTA]EXR = EX[R.sub.t] - EX[R.sub.t-1]      Monthly change in
                                              exchange rate.

[DELTA]INDP = IND[P.sub.t] - IND[P.sub.t-1]   Monthly growth rate of
                                              industrial production.

TABLE 2: Unit Root Tests

Variable   ADF            Mackinon   ADF            Mackinon
           Coefficients   critical   Coefficients   critical
           in levels      value      in first       value
                                     difference

           [[tau].sub.    [[tau].    [[tau].sub.    [[tau].
           [mu]           sub.t]     [mu]           sub.t]

LINDP      -2.47          -3.146     -5.39 **       -4.05
LKSP       -1.51          -3.44      -5.77 **       -4.03
LIRS       -1.63          -3.47      -4.17 **       -4.08
LEXR       -1.97          -3.45      -3.69 *        -3.46
INF        -3.98 *        -3.48      --             --

Variable   Phillips-      Mackinnon   Phillips-      Mackinnon
           Perron         critical    Perron         critical
           Coefficients   value       Coefficients   value
           in levels                  in First
                                      Difference

           [[tau].sub.    [[tau].     [[tau].sub.    [[tau].
           [mu]           sub.t]      [mu]           sub.t]

LINDP      -2.83          -3.45       -11.51 **      -4.054
LKSP       -1.81          -3.44       -10.59 **      -4.03
LIRS       -1.89          -3.47       -6.41 **       -4.08
LEXR       -1.90          -3.45       -7.78 **       -4.05
INF        -6.92 **       -4.09       --             --

Note: **(*) indicate rejection of null hypothesis of unit root
(non-stationary) at the 1% & (5%) level of significance. Mackinnon
critical value for rejection of hypothesis of a unit root has been
applied at the 1% & 5% level. Optimum lag structures are determined
by the Akaike and Schwartz information criteria.

TABLE 3. Estimation of Error Correction Model with
[DELTA]KS[P.sub.t] as the dependent variable

Regressors                         Estimated    t-ratios
                                   parameters

Constant                            0.009        (0.710)

[DELTA]KS[P.sub.t-1]                0.217        (1.224)
[DELTA]KS[P.sub.t-2]                0.375        (2.122) *
[DELTA]KS[P.sub.t-3]                0.025        (0.147)
[DELTA]KS[P.sub.t-4]                0.330        (1.920)

[DELTA]IR[S.sub.t]                 -0.035       (-2.279) *
[DELTA]IR[S.sub.t-1]                0.475       (-2.142) *
[DELTA]IR[S.sub.t-2]               -0.012       (-0.559)
[DELTA]IR[S.sub.t-3]                0.086       (-0.389)
[DELTA]IR[S.sub.t-4]               -0.667       (-1.640)

[DELTA]IN[F.sub.t]                  0.286        (0.440)
[DELTA]IN[F.sub.t-1]                0.131        (1.695) *
[DELTA]IN[F.sub.t-2]                0.147        (1.967)
[DELTA]IN[F.sub.t-3]                0.080        (1.262)
[DELTA]IN[F.sub.t-4]                0.079        (1.430)

[DELTA]EX[R.sub.t]                 -0.060       (-0.159)
[DELTA]EX[R.sub.t-1]               -0.309       (-2.074) *
[DELTA]EX[R.sub.t-2]               -0.243       (-1.419)
[DELTA]EX[R.sub.t-3]               -0.201       (-1.321)
[DELTA]EX[R.sub.t-4]               -0.142        (-.916)

[DELTA]IND[P.sub.t]                 0.035        (2.766) *
[DELTA]IND[P.sub.t-1]               0.048        (2.080) *
[DELTA]IND[P.sub.t-2]               0.010        (3.061) **
[DELTA]IND[P.sub.t-3]               0.088        (2.880) **
[DELTA]IND[P.sub.t-4]               0.004       (-0.0012)

[lambda][e.sub.t-1](E[C.sub.1-1]   -0.533       --

AFC (Dummy)                        -0.179       (-2.01) *
KS[P.sub.t-1]                      -0.532       (-0.239)
IR[S.sub.t-1]                      -0.076       (-2.089) *
EX[R.sub.t-1]                       1.055        (4.750) **
IN[F.sub.t-1]                      -0.471       (-3.663) **
IND[P.sub.t-1]                     -0.0320      (-5.085) **

[e.sub.t] = -2.302 [e.sub.t-1] + 0.620 [DELTA][e.sub.t-1]
              (-4.239) **        (2.390) *

ADF = (-6.312) ** [R.sup.2](adj) = 0.549 LM = (2.97) * RESET = 0.539

ARCH = 0.335 J-B = (1.029) * F = (2.458) * Log Likehood = 78.423

AIC = -1.709 SIC = 0.979

REFERENCES

Asprem, M. "Stock Prices, Asset Portfolios and Macroeconomic Variables in Ten European Countries," Journal of Banking and Finance 13, 1989. 589-612.

Augustine, C. A. and S. Shwiff. "Cointegration, Real Exchange Rate and Modeling the Demand for Broad Money in Japan," Applied Econometrics 25(6), 1993, 717-726.

Banerjee, A., J. J. Dalado, J. W. Galbraith, and D. F. Hendry.. "Cointegration, Error Correction and the Econometric Analysis of Non-stationary Data" Advanced Text in Econometrics, Oxford: Oxford University Press, 1993.

Box, G. E. P and G. M. Jenkins. Time Series Analysis: Forecasting and Control, San Francisco: Holden-Day, Inc. 1976.

Bulmash, S. B. and G. W. Trivoli "Time-lagges Interactions between Stock Prices and Selected Economic Variables," The Journal of Portfolio Management 17(4), 1991, 61-67.

Chen, N. F., R. Roll and S. A. Ross. "Economic Forces and the Stock Market," Journal of Business 59(3), 1986, 383-403.

Cheng, C. S. "The UK Stock Market and Economic Factors: A New Approach," Journal of Business Finance & Accounting 22(1), 1995, 129-142.

Dickey, D. A. and W. A. Fuller., "Distribution of the Estimators for Autoregressive Time Series with a Unit Root," Journal of American Statistical Association 74, 1979, 427-431.

Dwyer, G. P. and M. S. Wallace. "Co integration and Market Efficiency," Journal of International Money and Finance 11, 1992, 318-327.

Engle, R. F. and C. W. J. Granger. "Co integration and Error Correction Representation, Estimation and Testing," Econometrica 55, 1987, 251-276.

Fama, E. F. "Stock Returns, Expected Returns and Real Acivity," The Journal of Finance 45(4), 1990, 1089-1108.

Fama, E. F. "Stock Returns, Real Activity, Inflation and Money," The American Economic Review 71 (4), 1981, 545-565.

Fama, E. F. and M. R Gibbons. "Inflation, Real Returns and Capital Investment," Journal of Monetary Economics 9, 1982, 297-323.

Fama, E. F. and G. W. Schwert. "Asset Returns and Inflation," Journal of Financial Economics 5(November, 1977, 115-146.

Geske, R. and R. Roll. "The Fiscal and Monetary Linkage Between Stock Returns and Inflation," The Journal of Finance 38(1), 1983, 1-33.

Gibert, C. L. "Professor Hendry's Econometric Methodology," Oxford Bulletin of Economics and Statistics 48, 1986, 283-303.

Gjerde, O. and F, Sattem. "Causal Relations among Stock Returns and Macroeconomic Variables in a Small, Open Economy," Journal of International Financial Markets, Institution and Money 9,1999, 61-74.

Granger, C. W. J. "Developments in the Study of Cointegrated Economic Variables," Oxford Bulleting of Economics and Statistics 48, 1986, 213-228.

Hendry, D. F. "Econometric Modeling with Cointegrated Variables: An Overview," Oxford Bulletin of Economics and Statistics 48(3), 1986,201-212.

Jarque C. M. and A. K. Bera "Efficient Tests for Normality, Homoskedasticity, and Serial Independence of Regression Residuals," Economic Letters 6, 1980, 255-259.

Johansen, S. and K. Juselius. "Maximum Likelihood Estimation and Inference on Co integration with Application to the Demand of Money," Oxford Bulletin of Economics and Statistics 52, 1990, 169-210.

Kaul, G. "Stock Returns and Inflation: The Role of Monetary Sector," Journal of Financial Economics 18, 1987, 253-276.

Kaneko, T. and B. S. Lee. "Relative Importance of Economic Factors in the U.S. and Japanese Stock Markets," Journal of the Japanese and International Economics 9(3), 1995, 290-307.

Kwon, C. S., T. S. Shin, and F. W. Bacon. "the Effects of Macroeconomic Variables on Stock Market Returns of Developing Markets," Multinational Business Review (52), 1997, 63-70.

Marshall, D. A. "Inflation and Asset Returns in a Monetary Economy," The Journal of Finance 47(4), 1992, 1315-1343.

Maysami, R.C. and T. S. Koh. "A Vector Error Correction Model of the Singapore Stock Market," International Review of Economics and Finance 9, 2000, 79-96.

Maysami R.C. and S. H. Hui. "Economic Forces and Stock Returns: A General-to-Specific ECM Analysis of the Japanese and South Korean Markets," International Quarterly Journal of Finance, Vol. 1, No.1, 2001, 83-99.

Mukherjee, T.K. and A. Naka. "Dynamic Relations Between Macroeconomic Variables and the Japanese Stock Market: An Application of a Vector Error Correction Model," The Journal of Financial Research 18(2), 1995, 223-237.

Nelson, C. R. and Posser C. I. "Trends and Random Walks in Macroeconomic Time Series," Journal of Monetary Economics 10, 1982, 139-62.

Peiro, A. "Stock Prices, Production and Interest Rates: Comparison of three European Countries with the USA," Empirical Economics 21, 1996, 221-234.

Poon, S. and S. J. Taylor. "Macroeconomic Factors and the UK Stock Market," Journal of Business Finance and Accounting 18, 1991, 619-635.

Ramsey, J. B. "Tests for Specification Error in Classical Least Squares Regression Analysis," Journal of the Royal Statistical Society 31 (2), 1969, 250-271.

Schwert, G. W. "The Adjustment of Stock Prices to Information about Inflation," The Journal of Finance 36(1), 1981,. 15-29.

Dr. Mazhar M. Islam earned his M.A & Ph.D at Vanderbilt University in 1987. Currently he is an associate professor of finance at Sultan Qaboos University, Oman. He has held faculty positions at the Texas A&M University system and the SUNY system. Dr. Islam has served several years in the program committee of the AIB-US Southwest Chapter.