The purpose of this study is to investigate the lead-lag relationship between spot and futures markets in Malaysia during Asian financial crisis based on non-linear causality test. We use the daily spot and futures prices from 15 December 1995 to 21 August 2009. Our result based on Cross Correlation Function (CCF) approach indicates that the Malaysia stock markets are slow in absorbing new information to predict the future return movement. Based on our findings, we have made suggestion of implications into two perspectives. First perspective is the investors should not rely on historical return movement to predict future stock performance in order to hedge against the risk. Second perspective is academicians able to know the speed of transmission of information flow to identify the market efficiency.
In this chapter, we are studying the spot and futures markets in Kuala Lumpur Options and Financial Futures Exchange (KLOFFE). We will explain on the background of study in the first section, followed by the next section which is the problem statement of our study on how the asymmetric information would affect the investors. The subsequent section explains the research objective during the three sub-periods of the Asian financial crisis. It is followed by section four which studies on how the contribution will benefit the managers and policy makers. Section five is the chapter layout that explains the overview of the five chapters in this study. The last section is the summary of this chapter on the study from the previous section.
1.1 Background of study
.
In July 1980, Malaysia's capital market products were expanded by establishing the Kuala Lumpur Commodity Exchange (KLCE) and the KLCE was the first futures exchange in Southeast Asia. Products of capital market products in Malaysia have been expanded further when Malaysia has launched Kuala Lumpur Options and Financial Futures Exchange (KLOFFE) on 15 December 1995, such as launching the Stock Index Futures (SIF) contract. The contract was popular and has impressive growth in trading volume but has substantially declined in trading volume when the government implemented the capital controls in September 1998. KLCE merged with the MME in December 1998 and became the Commodity and Monetary Exchange of Malaysia (COMMEX). In June 2001, KLOFFE and COMMEX merged to become Malaysia Derivatives Exchange Berhad (MDEX). Nowadays, the MDEX becomes part of Kuala Lumpur Stock Exchange (KLSE).
The speed of information flow determines the movement of spot and futures prices. Now, when the spot and futures prices receive simultaneous information flow, the movement of prices becomes contemporaneous. There is no causal effect on the prices relationship. But this situation will change when the information flows faster in the spot or futures market. As a result, there appears to be a lead lag relationship between the prices. The price which receives faster information will lead other markets.
There were a lot of the researchers interested in examining the relationship between spot and futures prices since the emergence of the derivatives market in April 1982 but all the researchers were still concerned to investigate the relationship between spot and futures markets. Some researchers, for instances Silvapulle and Moosa (1999), Floros and Vougas (2007) find that futures price lead to the spot price while some researchers such as Bekiros and Diks (2008) and Bu (2011), find that spot prices lead to futures prices. Some researchers even find that there was a bi-directional relationship between spot and futures price.
Our paper intends to study the causal relationship between spot prices and futures price when the Asian Financial Crisis happened. The Asian Financial Crisis in 1997-1998 was caused by currency depreciation and stock market crash in all Asian countries. Furthermore, this crisis had affected the performance of spot and futures prices, especially in Malaysia. During the crisis, prices in both markets were highly fluctuated and traded at high volume. Furthermore in the post-crisis period, the volatility in prices became a moderate level and fairly traded.
1.2 Problem Statement
Asymmetric information exists among market participants which are due to each of them accessing different information about the next market price movement. However, the market price movement would bring the indirect effect such as profit or loss to the investors which are hedgers, speculators, and arbitragers. Due to asymmetric information, the investors have a lack of knowledge about market price movement, so they might face losses if they have made any wrong decisions in investments. The purpose of our study is to investigate which price movement will react faster between the spot and futures prices in KLOFFE in order to investigate asymmetric information issues.
Furthermore, we study spot and futures markets due to Malaysian stock market having a short history which started from 1995. Hence, we attempt to examine the issues of asymmetric information and market inefficiency during the structural break.
1.3 Research Objective
General objective
Our objective for this research is to determine the relationship and causality of the market which acts as an indicator to predict the stock price movement during the 3 sub-periods which is before, during and after the financial crisis. In addition, the study also aims to determine which market will react faster during the 3 sub-periods to the information.
Specific objective
In order to achieve our objective, we have made three specific objectives which are to explain the characteristics of spot and futures markets during the 3 periods of financial crisis. Next, to examine the direction of causal effect between spot and future markets during the 3 sub-periods of financial crisis, whether the causal directional is unidirectional, bidirectional or has no effect across different periods. Lastly, to examine the impacts of stock prices on lead-lag relationship between spot and future markets in Malaysia.
1.4 Significance of study
There are two expected contributions in our study. In terms of portfolio stock investment, it can provide to make better decision based on the behavior of price volatility. For instance, investors can use the knowledge of causal relationship between spot and futures returns in the past to predict price movement in the future. As such, this will provide proper guidance for investors to manage or diversify their investment risk effectively during a crisis.
In terms of economic growth, it can also facilitate policymakers to intervene in the stock market in order to stabilize the stock market performance during a crisis. For example, it can provide information to regulators regarding which periods of crisis in a country are most likely to affect the stock market. This will help them to revise or impose new regulations to recover stock market performance through monetary policy such as appreciation of currency. Subsequently, this can stimulate and enhance economic growth indirectly.
1.5 Chapter Layout
In our study, Chapter One presents the background of KLOFFE, problem statement, and expected contribution of study towards investors. It is followed by Chapter Two that reviews on spot and futures prices relationship. The motivation of studying this chapter is to reveal what the previous studies have done in their research. The next chapter is to describe data, variable and methodology that we use for study relationship and the results we get are whether they are consistent with the previous studies. The subsequent chapter is to present and interpret the results and findings. The last chapter is to summarize the findings from the previous chapter and propose the policy implication, limitation of studies and recommendations to the future researcher.
1.6 Conclusion
In a nutshell, the derivative market in Malaysia of spot and futures prices under KLOFFE started on 15 December 1995 and it was first established in Malaysia. The causal effect we will look through is the relationship between spot and futures market during three sub-periods, so we proceed to the next chapter to study the causal relationship on the past studies.
CHAPTER 2: LITERATURE REVIEW
2.0 Overview
In this chapter, we are studying the relationship between spot and futures markets in the previous studies. We will explain the correlation between spot and futures markets in the first section; followed by the next section which is bi-directional relationship between spot and futures markets. The subsequent section explains the uni-directional relationship between spot and futures markets in terms of spot lead futures. It is followed by section four which studies on the uni-directional relationship in terms of futures lead spot. The last section is the summary of this chapter on the study from the previous section.
2.1 Correlation between spot and futures market
Peroni and McNown (1998) studied three energy futures markets which were Heating Oil, West Texas Intermediate (WTI) and Unleaded Gasoline. The sample period covered from January 1984 to March 1996 for WTI, November 1979 to August 1995 for heating oil while January 1985 to March 1996 was for unleaded gasoline. Two tests included non-informative and informative had been adopted to study the market efficiency of these several energy markets. In terms of non-informative test, there was market inefficiency of futures market for heating oil, unleaded gasoline and WTI crude oil due to bias estimation. Besides, two informative tests yielded evidence which were sustained to the market efficiency. This implied that these three markets were co-integrated. Moreover, two of the markets were weak evidence of serial correlation. This indicated that the market was either inefficiency or autocorrelation of risk premium.
Chuang (2003) investigated on how the information flow affected the performance of spot and futures prices. They studied on Taiwan stock Exchange capitalization weighted index (TAISEX) and Morgan Stanley capital International (MSCI) of the daily data from July 21, 1998 to September 20, 1999. They employed multivariate error correction EGARCH (EC-EGARCH) model to study on volatility and long run relationship. The results indicated that the volatility spillover effects between spot and futures in MSCI, however there was volatility spillover from spot to futures in TAISEX. Therefore, volatility of futures market could be clarified the growth rate of futures in trading activities.
Maghyereh and Kandari (2007) were examining the relationship between oil prices and stock market in Gulf Cooperation Council (GCC) countries. This research conducted was based on the data from 1 January 1996 to 31 December 2003. They adopted new innovate method - rank tests of nonlinear co-integration to detect the co-integration when the error-correction mechanism were found to be nonlinear. They found that there was no relationship between oil prices and the GCC stocks market returns. Furthermore, stock price indices were affected by oil prices in GCC countries which was in a nonlinear modeling of the relationship in the market.
Switzer and Khoury (2007) examined the efficiency of market for New York Mercantile Exchange (NYMEX) of crude oil futures market during the extreme conditional volatility which was from January 1986 to April 2005. Fama regression had been implemented and proved that the basis, the premium and the change in future spot price were stationary and thus this regression was well specified. In addition, random walk predictor's model and futures contract were used to test futures prices of NYMEX using daily data. The result showed that both were significant. Besides, crude oil futures prices were found to be co-integrated with spot price and unbiased predictors of futures prices. Both prices presented asymmetric volatility characteristics during these periods.
Li (2008) examined the relationship between spot and futures market volatility in three markets, which were S&P 500, FTSE 100 and DAX, and Brazilian BOVESPA and Hungarian BSI with daily data from April 3, 1995 to December 12, 2005. The Vector Error Correction Model (VECM) was adopted to investigate the long run relationship between spot and futures markets. In terms of low volatility, the changes in price occurred in spot market. In terms of high volatility, the change in prices does depend on the futures market. The results showed there was a high variance change on spot and futures market, however it was weak correlation between both markets due to the risk aversion of investors who were sensitive to the price of disequilibrium.
Kao and Wan (2009) compared price discovery relationship between spot and futures price for United States and United Kingdom's gas markets. For the sample period of June 26, 1998 to December 31, 2007, they employed Vector Error Correction Model (VECM) to test on their hypothesis. Based on their findings, U.S futures market led the price discovery process in terms of higher trading volume and lower transaction cost. Furthermore, there was a positive correlation between volatility in U.S futures market and its returns due to presence of asymmetric volatility spillover effects. As a conclusion, U.S futures market is the main dominant for price discovery and leads spot price.
Maslyuk and Smyth (2009) have studied the co-integrating relationship between crude oil spot and futures markets. They selected the US WTI and UK Brent daily spot and futures prices which covered the period from January 1991 to November 2008. They employed residual-based co-integration test to examine on long run relationship of spot and futures with the structural break. The results indicated there was a long run relationship between both markets. This implied that the structural break change has affected the hedger on world oil market. They concluded that the future research could be used as panel co-integration techniques to further investigation of these relationships.
Choi and Hammoudeh (2010) studied the correlation between WTI oil, Brent oil, gold, silver, copper and S&P index in U.S with the weekly data from January 2, 1990 to May 1, 2006. Markov-switching GARCH models were applied to investigate the high or low volatility of return and correlation between spot and futures markets. The results show that the low volatility return is higher than the high volatility return between all the commodities and stock market, excluding the Gold market. They claimed that these findings on the low volatility are more suitable for risk-averse investors among these commodities with the spot and futures market. Besides, the correlation changes in the commodities due to reducing their hedging in the portfolios. Hence, it is easier to use monetary policy for policy makers in order to prevent inflation.
Lean, McAleer, and Wong (2010) examined the market efficiency of oil spot and futures prices. WTI crude spot prices covered the period from January 1, 1989 to June 30, 2008. They employed mean-variance (MV) and stochastic dominance (SD) to examine this market efficiency. In terms of MV, the results showed no strong evidence to support the market efficiency. In terms of SD, there was significance in spot and futures market efficiency. This implied that there is no arbitrage opportunity between both markets. Thus, they concluded that SD was more sensible to the investors in decision making.
Ye, Zyren, Shore and Lee (2010) examined the changing relationship between spot and futures in short and long term maturity crude oil markets with monthly data from January 2000 to the middle of 2009. The two theories were adopted to support the evidence, which were head and tail futures contract. They claimed that expected futures prices increased in spot price on physical and financial asset with the longer time period, vice versa. The results indicated the significance in spot price and volume while responding to head and tail markets. However, they examined the market position of spot markets, the spot prices expected to be decreased in the future. In short, the short or long position does depend on the spot and futures market information.
Jawadi and Bellalah (2011) studied the relationships on whether the oil was affected by the stock price fluctuations in France, Mexico, and USA. The monthly data of stock indexes were used in considering data from December 1987 to March 2008. While the methodology used was non-linear econometric modeling, the results indicated that the stock return had an impact on the oil market and non-linear relationship among these three countries significantly. They concluded that there were non-linearity and non-linear mean reversion between oil and stock price due to moving with the different direction.
Liu and Wan (2011) examined the correlations between return of WTI spot and futures prices. They used daily spot and futures prices of West Texas Intermediate (WTI) crude oil from Jan 2, 1990 to Dec 31, 2009. Furthermore, they used the rolling sample test and found that external factors such as the Gulf War and financial crisis effects were affected by these correlations. They also employed Ljung-Box test and found that that the cross-correlations were significant and implied that there were long-range cross-related between spot and futures returns.
Lee and Zeng (2011) examined the relationship between spot and futures oil prices of West Texas Intermediate (WTI) and covered the daily data from January 2, 1986 to July 6, 2009. They employed quantile cointegrating regression and found the long-run relationships between spot and futures oil prices. This relationship has significant differentials among futures maturities and the performance of spot markets. From the Granger-causalities test, they found spot prices have leaded the futures prices and implied that market participants place more focus on spot markets than futures oil market.
Liu, Chen, and Su (2011), examine the nonlinear relationship between spot and futures oil market in WTI. The daily data covered from 1 January, 2004 to 30 September, 2009. They employed bivariate threshold error-correction model (TECM) and GJR-GARCH model to examine on short and long run relationship between spot and futures market. The results indicated that there was long run relationship between both markets. This implied that there was interaction effect between these two markets. Thus, the hedgers, speculators, and financial managers could be applied in their investing and hedging on futures market.
Wang, Wei and Wu (2011) examined cross-correlation and auto-correlation between spot and futures markets of West Texas Intermediate (WTI) crude oil daily data from 2 January 1990 to 9 March 2010. They applied Multifractal Detrended cross correlation analysis (MF-DCCA) to determine the trends and variance of spot and futures. The results showed the cross correlation between spot and futures price which were generally higher than the autocorrelation for each series. These indicated there was significance in cross-correlation with spot and futures market. Hence, cross-correlation and auto-correlation were multi-fractal and showed that the cross-correlation was larger than the auto-correlation in short term but lower in the long term. In short, the past information could be used for forecasting in futures markets.
Lei and Yong (2011) have examined the properties of Brent crude oil spot and futures prices with daily data covered from Jan 1, 2002 to Dec 31, 2009. They employed the stochastic unit root (STUR) and stochastic co-integration to study the long run relationship during the financial crisis. They found that both oil spot and futures with STUR and changes of time auto-regressive coefficients were significant, and implied that it was useful for forecasting and risk management. In addition, they found that there was long run relationship between both markets; therefore it could be hedging the oil market risk. In short, they claimed that the STUR and co-integration are the significant tools to study the series model were on strong volatility.
Huyghebaert and Wang (2010) examined the co-integration and causality of the seven major seven stock exchanges for the pre, during and post Asian financial crisis on 1997-1998 which were Shanghai SE Composite, Shenzhen SE Composite, Hong Kong Hang Seng, Taiwanese SE Weighted, Singaporean Strait Times, South Korean SE Composite, Japanese Nikkei 225 Stock Average and US S&P 500 Composite Index. The daily data covered from 1 July 1992 to 30 June 2003. They adopted the VAR model which studied the long run relationship. Their findings of this study were crisis in Hong Kong and Singapore which significantly led to shocks compared to other East Asian countries. After the crisis, Singapore and Hong Kong largely affected other East Asian countries whereas USA is the country which strongly influenced the stock return in East Asian countries. Thus, it was the macroeconomics factors affected in globalization.
2.2 Bi-directional relationship between spot and futures market
Tse (1995) examined lead-lag relationship between spot and future price of the Nikkei Stock Exchange from December 1988 to April 1993. Error correction model was used to analyze this relationship and it was found that both prices moved simultaneously, whereby, the short term adjustment in the spot index had been affected by a lagged change in the futures price, not vice versa. In addition, vector autoregressive (VAR) and uni-variate time series methods were used to forecast the spot price and make a comparison with the error correction model. Conclusions showed that error correction model was the best for forecasting while the vector autoregressive model was found better than the martingale model, whereas the uni-variate time series method presented the worst results.
Kavussanos and Visvikis (2004) examined the lead-lag relationships between and spot and futures markets in Atlantic and Pacific route markets. They employed the Vector Equilibrium (VECM) model and used sample period from 16 January 1997 to 31 July 2000.Findings implied that futures price is an important source of information and leads spot price in price discovery. In addition, spot price also has higher transaction cost than futures prices and contributes to higher volatility. Therefore, there is bidirectional causality existing between futures and spot prices in the Atlantic and Pacific routes.
Hasan (2005) examined the lead-lag relationship between spot and futures market in United States and United Kingdom. For the sample period of July 22, 1983 to August 11, 1992, results implied that there was a high correlation between spot and futures markets for United Kingdom market via linear granger causality test. Their finding for United States market also indicated a bi-directional relationship between spot and futures return. Based on these results, they concluded that feedback relationship existed between spot and futures return for both countries. Further, this relationship implication is important for efficient market hypothesis, price discovery and prediction of cost-of-carry model.
Hseu, Chung and Sun (2007) examined the intra-day price of S&P 500, Nasdaq-100 and DJIA from April 1, 1998 to March 31, 2002 on New York Stock Exchange (NYSE). They applied Vector Error Correction Model (VECM) and co-integration to examine the long run relationship between spot and futures market. The results indicated existing bi-directional and long run relationship between the three spot and futures market. Therefore, they concluded that the stock return had changed in intrinsic value.
Chang and Lee (2008) investigated the causal relationship between spot and futures market. The daily data covered from January 2001 to May 2005 of Taiwan Stock Exchange. The study was based on the threshold error-correction model (TECM) to study the long run relationship. Their findings indicated the bi-directional causality relationship between the spot and futures market in the short run. However, there were significantly negative deviations correlated between the spot and futures market which were being affected by negative deviations while there were no correlated positive deviations on the spot market. They concluded that they should be obtainable to the investors and financial institution in order to invest in long term portfolio investments.
Ozen, Bozdogan, and Zugul (2009) investigated the causality relationship between spot and futures prices of Izmir Derivatives Exchange (VOB) and Istanbul Stock Exchange Index 30 (Ä°MKB30) from the Turkey stock exchange market. The sample period covered from 4 February 2005 to 27 February 2009. They employed Error correction models (VECM) which were used to analyse whether there occurred a causality relation between spot and IMKB 30 futures. The result indicated the long term bidirectional causality relationship occurred among spot and IMKB 30 futures prices in the long run. However, the unilateral causality relationship only exists in the short run.
Srinivasan and Bhat (2009) examined on the relationship between spot and futures of twenty one selected banking stocks in National Stock Exchange (NSE). The sample period covered from 27 May, 2005 to 29 May, 2008. They adopted the co-integration and VECM model studied the lead-lag relationship between these markets. The results indicated there was futures lead spot on nine selected stocks, spot lead futures on another six stocks and feedback relationship on the rest. These indicated that bi-directional relationship exists on these selected stocks. It is also implied that it acted as a price discovery and received rapid information on the market efficiency. Thus, they concluded that price discovery responds to each other, however, it spreads in bank activities due to transaction costs, initial margin, and leverage position.
Athanasios (2010) examined the lead-lag relationship between futures market and spot market in Athens stock exchange (ASE) from the period from 2 January 2000 to 30 May 2008. The serial dependence of volatility or conditional variance had been forecasted using GARCH. The purpose of adopting GARCH was that it could offer a trustable estimation for volatility, capture the tendency in asset returns for volatility clustering and equalize the potential negative correlation between future volatility and spot returns. Besides, granger causality test was applied in structural equation model (SEM) to study the dynamic effects of spot and future returns in terms of volatility. Finally, he concluded that spot and futures markets were fluctuated but well characterized by GARCH process. With SEM, it can strengthen the granger causality test which could be occurred when there were unidirectional and bidirectional causal effects between the market indices and volatilities.
Apergis (2010) studied the relationship between oil spot and futures market of NYNEX and conditional volatilities from January 2000 to April 2009. VECM was used to study the non-linearity of the relationship. The results showed that nonlinearities existed both in means and conditional volatilities between oil spot and future markets. Non-linear causality in volatilities implied there was bi-directional relationship, thus it was the significant tool to policy makers, hedge fund managers, portfolio managers in order to make precise decisions.
Ahmad, Shah, and Shah (2010) studied the relationship between spot and futures stock market in Pakistan from sample period of July, 2001 to January, 2010. Findings from GARCH model have provided the existence of volatility clustering. This is due to high volatility for spot and futures market from past period which will also lead to high volatility in the current period. Furthermore, results also exhibit bi-directional relationship between the two markets. However, they found that probabilities of negative returns are higher than positive returns for both spot and futures market during the sample period. The reason behind this is due to the 2008 stock crisis and political instability which caused the stock exchange market to crash. Hence, high downside risk from stock market returns during that period worsened the spot and futures stock market return in Pakistan.
Anoruo (2011) examined the linearity and nonlinearity causal relationships between spot and futures in S&P crude oil market. The period of data covered from February 1974 through December 2009. They employed standard VAR and the Bivariate Mackey- Glass (M-G) model studied the non linear causal relationship between spot and futures crude oil market. The results indicated that there was bidirectional relationship between spot and futures crude oil market. This implied that the volatility of spot and futures return was non-linear. In conclusion, the past information cannot be used for forecasting due to market inefficiency.
Mariam (2011) studied the causal relationship, long run relationship and market efficiency of natural gas in NYMEX spot prices from November 1996 to December 2004 and futures prices from 1999 to 2004. Granger causality and co-integration was analyzed for long run relationship between spot and futures markets. Results showed that spot lead futures in the short run and futures lead spot in the long run. At the same time, examining on co-integration test has found long run relationship between spot and futures market. This implied that bi-directional relationship exists on both markets, therefore spot and futures move with the same direction due to transportation and transaction costs in the long run equilibrium.
Diaw and Olivero (2011) investigated the dynamics relationship of CAC 40 between spot and futures markets. There are separated, pre and during crisis, into two sub periods which were September 2006 and October 2008. They employed three models which were GJR-GARCH, EAR-GARCH and EC-EGARCH examined on the short run and long run dynamic relationship. The results indicated that there was low volatility during pre crisis, however, high volatility during crisis tested on three models. This implied that they could generate abnormal returns and futures lead spot during pre crisis, however, spot lead futures during crisis. Therefore, there was existing the bi-directional relationship between spot and futures markets. In short, the volatility effects could help the investors in hedging during crisis.
Choudhary and Bajaj (2012) examined the lead lag relationship among spot and futures markets for selected thirty one securities of S&P CNX Nifty on National Stock Exchange (NSE). The daily data covered from April 2010 to March 2011. Johanson test and VECM were adopted to study the long run relationship among these individual securities. The results indicated that futures lead spot on 12 securities and spot lead futures for the rest 19 securities. These implied that bi-directional relationship exists between these securities. Thus, they concluded that spot and futures are important in price discovery due to consist of useful information to the investors.
2.3 Uni-directional relationship between spot and futures market (spot lead futures)
Bekiros and Diks (2008) examined the causal relationship between spot and futures market in West Texas Intermediate (WTI) crude oil and NYMEX futures market for two sample periods. The sample periods covered from October 21, 1991 to October 29 for the first period and November 1, 1999 to October 30, 2007 for the second period. They applied the Granger Causality test for linear and non-linear of the spot and futures prices. They found that there is uni-directional causal linear relationship due to more volatility on spot and futures prices of crude oil in the second period, spot tend to lead futures prices. At the same time, non-linear causal relationship in both sample periods was tested but results showed that it was only uni-directional in the first period and had vanished in the second period. Therefore, they concluded that the speculators would form their expectation based on the spot price and forecast the expected futures prices due to its importance to test the market efficiency.
Srinivasan and Deo (2009) examined on relationship between spot and futures markets in Multi Commodity Exchange of India (MCX) and National Commodity Derivatives Exchange (NCDEX) from January 1, 2005 to December 31, 2008. Johanson Co-integration test and Vector Error Correction Mechanism (VECM) were adopted to study the long run relationship between spot and Mini Gold futures market. The results indicated there existed the long run relationship and Mini Gold spot lead futures. Therefore, they concluded that investors prefer spot rather than futures market due to liquidity in gold market.
Bu (2011) studies the impact of speculator's trading activities to the volatility of crude oil futures return in China. He used weekly data from WTI (NYMEX) futures prices over the period from June 13, 2006 to December 28, 2010 and employed Granger causality method based on GARCH (1, 1) model to examine the hypothesis. Findings have shown a unidirectional Granger causality relationship, whereby crude oil futures return will lead to speculator's trading position. This result implied that speculators are feedback traders that followed the trends of futures return in the market. On the other hand, he also found that there is a bi-directional relationship between futures return and net long position held by speculators. This also implied that changes of position held by speculators will give impact to changes of crude oil price volatility. Furthermore, there was a large volatility change in crude oil futures return during financial crisis. Subsequently, this effect had caused futures prices to decline instantly with evidence of high volatility clustering.
Ghalayini (2011) studied the causal relationship between volatility of spot oil market and changes of futures market trading activity. They used West Texas Intermediate (WTI) weekly data from January of 2000 to December of 2010. They employed Granger Causality test examined the causal relationship between spot and futures market. The results showed there was uni-directional causality spot lead futures and concluded that spot carried the information to futures for forecasting. Thus, the risks of futures could be determined by the spot market.
Srinivasan (2009) studied the lead lag relationship between spot and futures prices of nine selected oil and gas stocks on National Stock Exchange (NSE) of India. The daily data sample period covered from May 12, 2005 to January 29, 2009. Johansen co-integration test and Vector Error Correction Model (VECM) used to conduct lead lag relationship among the selected stocks. The results indicated that there was a long run relationship and bidirectional relationship among spot and futures prices of four selected stocks, however futures lead spot price for three selected stocks and spot lead futures price in two of the selected stocks in India.
2.4 Uni-directional relationship between spot and futures market (futures lead spot)
Schwarz and Szakmary (1994) investigated directional relationship between spot and futures price in crude oil market which are traded on the New York Mercantile Exchange (NYMEX). For the sample period of January 1, 1984, to May 15, 1991, they used granger causality test to examine the relationship. Finding had shown that spot prices followed futures prices of crude oil. This finding points out that crude oil futures price is the price leadership and investors tend to invest in spot market based on information they obtain from futures returns. In conclusion, futures price of crude oil significantly gives positive impact to growth and success of the energy futures contracts traded at the NYMEX.
Silvapulle and Moosa (1999) investigated the lead-lag relationship between futures price and spot price. As such, they compared the linear and nonlinear causality testing methods for West Texas Intermediate (WTI) crude oil in the United States. Based on their linear causality results, they found that futures prices lead spot prices. In contrast, opposite finding from the nonlinear causality approach suggests that futures and spot prices have a bi-directional causal relationship. The difference between these results is because of the changes of investor's reaction towards new information regarding returns.
Zhong, Darrat and Otero (2004) examined the price discovery and volatility of spillover effects on the Mexican stock market with the daily data covered from 15 April 1999 to 24 July 2002. They employed bivariate EC-GARCH model which studied the spillover effect with conditional mean and variance between spot and futures market. The results indicated that there was futures lead spot in the short run and occurred arbitrage opportunity in the long run. This implied that the futures prices were more responsive than the spot prices. Thus, the information of futures market in Mexico does depend on past information.
Lee, Chiu, and Lee (2007) analyzed the lead-lag relationship between spot and futures return traded in CME-Nikkei 225 and SIMMEX-Nikkei 225 in Singapore and the United States respectively. They employed Granger causality test for the period of January 5, 1994 to December 31, 2003. Findings implied that there is a unidirectional relationship from futures returns to spot returns due to low overall transaction costs and high leverage in the futures market.
Batchelor, Alizadeh, and Visvikis (2007) examined the relationship between spot and futures freight rates in Atlantic and Pacific routes. They used several time series model such as Vector Equilibrium (VECM) model, ARIMA model, and Vector Auto-regression (VAR) and claimed that spot and futures prices are co-integrated from sample period of 16 January 1997 to 31 July 2000. Also, findings from these three models implied that forward returns would forecast spot returns. They further argued that ARIMA model was a better forecasting tool to predict spot and futures rates when the underlying market structure was growing.
Floros and Vougas (2007) examined the lead lag relationship among spot and futures of Athens Derivatives Exchange (ADEX) in Greece. There were separated into two sub periods which were FTSE/ASE 20 (August 1999-August 2001) and FTSE/ASE Mid 40 (January 2000-August 2001). The Bivariate-GARCH methods were used to study these relationships. From their findings, futures lead spot due to futures markets which were lower transaction costs and more liquid than the spot market. Thus, it was useful to speculators, traders and financial managers in trading activities.
Kavussanos, Visvikis and Alexakis (2008) studied on linear and non-linear relationship in spot and futures markets in FTSE/ATHEX 20 and FTSE/ATHEX 40 on Greece's derivative market from February 2000 to June 2003, and FTSE/ATHEX market from July 2000 to June 2003 in these two sample period. They implemented Johanson cointegration test and Generalized Autoregressive Heteroskedasticity (GARCH) model to determine the long run relationship. The results indicated the spillover effects from futures were significant but spot lead futures was insignificant due to futures market which was less expensive compared to spot markets. They recommended that investors should invest in futures contracts due to the futures market more liquid than the spot markets. Finally, they claimed that futures prices were useful for risk management, portfolio management and budget planning decisions.
Huang, Yang and Hwang (2009) examined on the dynamic relationship between spot and futures prices, whether in linear or non-linear causal relationship in WTI. The sample period of the WTI data are spread to three periods, which were period 1 (01/02/1986-02/28/1991), period-2 (03/31/1991-08/31/2001), and period-3 (09/01/2001-04/30/2007). They applied cost-of-carry hypothesis which studied the spot and futures responding to the arbitrageurs. The results indicated that the spot prices were higher than futures prices. They claimed that significantly between spot and futures, futures lead spot prices changed. This implied that there was higher transaction cost in spot markets, however, they found that there was no relationship between spot and futures with no arbitrage opportunity. Thus, the arbitrageurs sold the futures contracts and purchased the spot contracts.
Luengo (2009) examined the causal relationship between spot and futures market S&P 500 volatility with daily data covered from January 17, 2000 to November 26, 2002. Vector autoregressive (VAR) model was adopted to study the volatility of interaction effects of spot and futures markets. The result indicated that futures lead spot due to spot market was more costly while trading in the market. Therefore, the researcher concluded that futures market is a leader with availability of new information to spot markets. In other words, current spot markets do depend on past information of futures market.
Kaufmann and Ullman (2009) examined the causal relationship between crude oil prices in North America, Europe, Africa, and the Middle East. They compiled a series of daily spot and futures price data for crude oil for the sample period of January 1986 to March 2007. Findings from Vector Error Correction Model (VECM) indicated a unidirectional causal relationship between spot and futures return. Accordingly, it is found that changes in futures price affect spot price of crude oil. Furthermore, this result produced an upward pressure to oil prices because of high demand for speculation. However, increase in price would also lead to slow economic growth. Therefore, researchers suggest that market fundamentality is an important factor to determine both returns.
Tse and Chan (2010) studied the lead lag relationship of S&P 500 spot and futures markets from 5 March 2004 to 1 July 2004. They employed the Threshold Regression Model (TRM) to investigate the causal effect of non-linear between spot and futures markets. They found that when the market condition was good, the futures markets were more effectively served as price discovery function, vice versa. They also observed that when the market was in bull or bear, the effect of futures leading the spot market would be more obvious due to more market information.
Mall, Bal and Mishra (2012) studied causal relationship between spot and futures market in India. They used daily data of National Stock Exchange (NSE) from June 12, 2000 to May 2011. They employed VECM which examined the short run relationship and co-integration test which studied the long run relationship between these two markets. The results indicated futures lead the spot during long periods but no relationship during short periods. They claimed that it was helpful for investors to forecast since the market was efficient.
2.5 Conclusion
In a nutshell, we study the correlation of spot and futures market in order to identify the volatility of stock market performance during the normal and crisis period. In our outcome, it indicates the high volatility during crisis and low volatility before and after crisis. The market is efficient when there is more market information on spot and futures, and acts as a price leadership to each other. However, the previous studies showed that they were helpful to investors to forecast the futures prices by using existing market information to generate the abnormal profit. In short, there are bi-directional or uni-directional relationships before, during and after crisis.
CHAPTER 3: DATA DESCRIPTION AND METHODOLOGY
3.0 Overview
In this chapter, we emphasize on how we conduct this study in order to achieve the objectives of our study. For the first section, we will explain the source and the type of data used. Furthermore, we will explain the variable that will be used in our study. Then, we will explain regarding the application of the estimated model and the test that we will carry out to investigate on the relationship between spot and future during, pre and post financial crisis in Malaysia.
3.1 Data
We conduct this study by using the daily closing data of spot and futures prices of Kuala Lumpur options and financial futures exchange (KLOFFE). We obtain the data from Thomson Reuters Data Stream with 3272 observations, the sample period covered from 15th December 1995 to 21th August 2009. According to Huyghebaert and Wang (2010), the pre Asian financial crisis sample period covered before June 30, 1997, during crisis from July 1, 1997 to June 30, 1998 and post crisis after June 30, 1998. In order to examine the relationship of spot and futures prices towards Asian financial crisis, the sample period is divided into three sub periods which are pre-Asian financial crisis (15th December 1995 to 3rd July 1997), during Asian financial crisis (4th July 1997 to 1st September 1998) and post-Asian financial crisis (2nd September 1998 to 21st August 2009).
3.1.1 Variables
In this study, we emphasize on two series only, which are the daily closing spot and futures prices. Both series will be transformed into natural logarithms form also known as (lnP) measure in Ringgit Malaysia in order to reduce the variation of series times.
) (1)
) (2)
Where Sr denotes as the spot return and sp denotes as the spot prices of KLOFFE. On the other hand, fr denotes as the future return and fp denotes as the future prices of KLOFFE.
3.2 Methodology
3.2.1 Unit root test
According to Granger and Newbold (1974), an existence of non-stationary series in regression model would lead to spurious estimation and invalid hypothesis testing inferences. Therefore, this study attempts to conduct unit root tests to investigate the stationary or non-stationary of each variable in each time series to avoid obtaining spurious estimated results. As such, two unit root tests were applied in this study, namely augmented Dickey-Fuller, ADF test (1981) and Phillips-Perron, PP test (1988). This study will test the stationary for daily closing stock price in the natural logarithms form in stock market with whole period and three sub periods.
3.2.1.1 Augmented Dickey-Fuller (ADF) test
Augmented Dickey-Fuller (ADF) is an extended procedure of DF test by adding lagged value of dependent variables which is developed by Dickey and Fuller (1981) in order to eliminate autocorrelation. This test will add the lagged value of dependent variable, in the equation in order to accept the case when is correlated. Therefore, these models can be shown in equation (3) and (4) respectively as below which include intercept and intercept and trend which is in level form.
With intercept:
With intercept and trend:
where is a constant, is the coefficient on a trend and is the lagged order of the autoregressive process, is the pure white noise error term. In both equation (3) and (4), is denoted as ln daily price (lnP), is level form for lnP, and where, when , it indicates unity and unit root. The lag length for dependent variable is based on Schwarz Information Criterion, SIC. In order to test stationary of and , the null hypothesis is and alternative hypothesis is . The null hypothesis can be rejected if DF statistic is more negative than critical value. In this regards, it can be concluded that the series is stationary.
3.2.1.2 Phillips-Perron (PP) test
PP test is developed by Phillips and Perron (1988) which generalized the ADF test by allowing the assumption on the distribution of errors. To cope with the serial correlation of the error term without adding lagged difference terms is by using nonparametric statistical methods. The PP test applies the same asymptotic distribution as ADF test. Therefore, these models can be shown equation (5) and (6) respectively as below which include intercept and intercept and trend which are in level form.
With intercept:
(5)
With intercept and trend:
Where is a constant, us the coefficient on a time trend and is the pure white noise error term. In both equation (5) and (6), is denoted as ln daily price (lnP). is the level form for lnP, and where when , it indicates unity and unit root. In order to test stationary of and , the null hypothesis is while the alternative hypothesis is . The null hypothesis can be rejected if the test statistic is less than the critical value. In this regards, it can conclude that the series is stationary.
3.2.2 Cross-correlation Function (CCF) approach
Cross-correlation function (CCF) test is used to investigate the nonlinear causality between spot and futures markets. This approach is developed by Cheung and Ng (1996). This approach is useful in the study with the large number of time series and predicts the future value from existing information. Before this approach is used, we should conduct two analyses which are univariate and augmented analysis. Univariate analysis is to determine the causal effect between spot and futures return in conditional mean and conditional variance without taking spillover effect into account. Apart from that, augmented analysis is to investigate the interaction and robust changes with the spillover effects between spot and futures return.
3.2.2.1 Univariate Analysis
Univariate analysis without spillover effects explains the move with the same direction between spot and futures market. We separated the whole period and post crisis by adding the dummy variable due to structural change, pre and during without adding dummy variable into our equation. Therefore, these models can be shown below which show includes dummy and without dummy variable in the different periods.
Without Dummy variable
Without dummy, where is constant, represent the spot return and represent the futures return in mean equation and represent the variance of spot and futures return. is the lagged order of spot and futures return. The equation shown in (7) and (8) which include ARMA, however, the equation (9) and (10) included ARCH and GARCH in spot and futures return, and sum up is less than one represent the persistency of the volatility. With dummy, the equation has shown in (13) and (14), is the coefficient of dummy variable. and assume to be close to zero or in white noise error term.
With Dummy Variable
Dum 1= structural change, 0= otherwise
3.2.2.2 Augmented Analysis
Augmented analysis with the spillover effects between spot and futures by improving our model with getting higher log likelihood and lower SIC, therefore can reduce the autocorrelation problem. These models can be shown below which includes the interaction effect between spot and futures with dummy and without dummy variables.
Without dummy variable
Without dummy, where is constant, and take the interaction effect into account between spot and futures return in equation (15) and (16). and represent the variance of spot and futures return with the interaction effect show in equation (17) and (18). With dummy, the mean equation has shown in (19) and (20) while the variance equation has shown in (21) and (22). is the coefficient of dummy variable. and assume to be close to zero or in white noise error term.
With dummy variable
As in the univariate and augmented analysis, CCF has tested on standardized and square of standardized residuals with mean and variance concurrently. The advantage of CCF is not simultaneously involved and it is easy to study. ARMA is commonly used to identify the series of the model autocorrelation (ACF) and partial autocorrelation (PACF) responding to the lag length of the variables. On the other hand, GARCH model is the conditional variance of the disturbance of the sequence of ARMA process. If our model is adequate, the ACF and PACF indicate that there is white noise process. As such, the adequacy can be fitted with the correlogram by using the Ljung-Box statistic from the square of standardized residuals shown in equation (23) and (24) to determine the autocorrelation problem.
(24)
The test statistics equation are given below
TS = (25)
The null hypothesis of CCF can be rejected if the test statistics is greater than the critical value shown in (25), so we conclude that there exists the causal relationship between spot and futures return in mean and variance.
CHAPTER 4: EMPIRICAL RESULTS
4.0 Introduction
This chapter reports and discusses the results of the relationship between spot and futures markets during whole and three sub periods. Firstly, descriptive statistics and unit root tests are carried out in order to examine the characteristics of spot and futures return. Besides, the findings from Cross Correlation Function (CCF) approach based on univariate and augmented analyses are to find out the causal effect between spot and futures returns. The lag length of CCF presenting the direction effect in spot and futures return will be presented. Based on second and third objective, the non-linearity test based on CCF approach is implemented in each period to examine the information flow between both markets. The non-linearity test based on univariate and augmented analyses will be presented and interpret in last section for this chapter.
4.1 Descriptive Statistics
Table 1 Descriptive Statistics
Whole Period
Pre Crisis
During Crisis
Post Crisis
S
F
S
F
S
F
S
Mean
4.77E-05
4.31E-05
0.0003
0.0002
-0.0039
-0.0040
0.0003
Maximum
0.2143
0.4880
0.0472
0.0486
0.2143
0.2019
0.1713
Minimum
-0.1846
-0.2391
-0.0424
-0.0422
-0.1845
-0.1333
-0.1168
Standard
Deviation
0.0178
0.0199
0.0101
0.0101
0.0390
0.0386
0.0155
Skewness
0.1923
3.9522
-0.0972
0.0642
0.2037
0.4225
0.4005
Kurtosis
20.4241
125.8380
5.9515
6.5284
8.5111
7.1742
14.7078
Jarque-Bera
41398.46
2065044
112.2814
159.9776
278.6574
165.5113
15733.72
Table 1 shows the descriptive statistics of spot and futures return for whole and three sub periods. The kurtosis of the spot and futures return indicates the kurtosis is greater than 3.0 of all periods, so it is known as leptokurtosis. According to Vosvrda and Zikes (2004), there is very important leptokurtosis for further studies on the characteristics and estimation on spot and futures market. The Jacque-bera test indicates the normal assumption of the spot and futures return. Result from Jacque-Bera statistic shows that we reject the null hypothesis of normality assumption. Therefore we conclude that both returns are not normally distributed.
4.2 Unit root tests
Table 2 Unit root tests results
Order Difference ADF PP
Intercept Trend and Intercept Trend and
Variables Intercept Intercept
Whole period S -59.1154 *** -59.1183 *** -59.1217 *** -59.1314 ***
F -35.9015 *** -35.9096 *** -63.3907 *** -63.4101 ***
Pre-crisis S -15.4738 *** -15.6458 *** -15.3599 *** -15.5540 ***
F -14.3979 *** -14.5638 *** -14.2079 *** -14.3306 ***
During crisis S -15.0074 *** -14.9749 *** -15.0554 *** -15.0209 ***
F -14.5165 *** -14.4858 *** -14.5586 *** -14.5252 ***
Post crisis S -55.2438 *** -55.2384 *** -55.1648 *** -55.1597 ***
F -14.5108 *** -14.5212 *** -61.0158 *** -61.0133 ***
Note *** indicate the rejection of the null hypothesis at 1% of significance level, respectively.
The rejection of null hypothesis for ADF and PP is test is based on Mackinnon (1994) critical value. The number of lags was selected based on Schwarz Information Criteria.
= the series has a unit root. = the series is stationary.
Table 2 presents the results of ADF and PP in the level form. According to the findings, Augmented Dickey Fuller (ADF) indicates all periods statistically significant to reject null hypothesis at 1% significance level. In order to obtain more robust results, Philips-Peron (PP) is used as a complementary test for ADF test to confirm the stationary of all the variables. Likewise, PP test results show the spot and futures return reject null hypothesis at 1% significance level. Therefore, the spot and futures return are stationary at I (0) process.
4.3 Diagnostic testing
Table 3 presents the diagnostic testing on univariate and augmented analysis of spot and futures returns for whole period and three sub periods which are pre crisis, during crisis and post crisis. The result indicates the ARMA and GARCH process that we employ to study the non-causality of mean and variance. We employed Maximum Likelihood (ML) estimation on the spot and futures return, the log likelihood indicates logarithm of the likelihood ratio that has been used. The estimation coefficient of ARCH and GARCH indicates the volatility of whole and three sub periods. According to Vosvdra and Zikes (2004), the coefficients of ARCH and GARCH close to one. This implied that the volatility is constant. The result indicates the coefficient is less than one that implies the volatility is persistent during whole and three sub periods.
The log likelihood shows there is improvement in our model based on augmented analysis when taking the spillover effects into account between spot and futures returns. Besides, Ljung-Box Q test is to examine the autocorrelation problem using the standardized square residuals and ARCH LM test is used to check on heteroskedasticity problem. The ARCH LM test indicates there is a failure to reject the null hypothesis at 10% significant level. Therefore we conclude that there is no heteroskedasticity problem. The results of failed to reject the null hypothesis even at 10% significance level. This implies that there is no autocorrelation problem. The two periods which are whole period and post crisis are due to structural change in the spot and futures returns, therefore we apply Threshold-GARCH with including dummy variables into the model. The outcome shows that autocorrelation problem has been removed as compared to without adding dummy variables in two periods. This is because the dummy variable is important to capture the structural change.
Table 3 Diagnostic testing results based on univariate and augmented analysis
Whole period
Pre-Crisis
During Crisis
Post Crisis
S
F
S
F
S
F
S
Univariate
Log likelihood
9591.424
9450.45
996.0868
1001.181
418.8646
399.9726
8162.702
ARCH LM
0.696043
1.903867
0.073126
0.102584
0.766974
0.538324
0.089555
(0.4041)
(0.386)
(0.7868)
(0.7488)
(0.3812)
(0.4631)
(0.7647)
Ljung-Box Q statistics
Q² (10)
4.7636
4.366
3.4191
3.2087
4.2854
5.7183
5.9089
(0.906)
(0.929)
(0.97)
(0.976)
(0.934)
(0.838)
(0.823)
Q² (30)
6.9999
5.9396
7.9807
5.5781
19.083
12.731
11.154
(0.958)
(0.981)
(0.925)
(0.986)
(0.210)
(0.623)
(0.742)
Coefficient
0.914882
0.893405
0.967976
0.976996
0.996094
0.980775
0.943165
Log likelihood
14477.18
14357.23
1379.771
1386.906
736.8513
673.3629
12141.89
Augmented
ARCH LM
0.146323
1.172121
0.109933
0.234996
0.016289
0.432707
0.086633
(0.7021)
(0.5565)
(0.7402)
(0.6278)
(0.8984)
(0.5107)
(0.7685)
Ljung-Box Q statistics
Q² (10)
8.5436
7.9747
6.9958
6.3406
5.3295
6.998
13.782
(0.576)
(0.631)
(0.726)
(0.778)
(0.868)
(0.726)
(0.183)
Q² (30)
13.148
13.01
14.944
7.3153
11.399
14.154
20.493
(0.591)
(0.602)
(0.455)
(0.948)
(0.724)
(0.514)
(0.154)
Coefficient
0.966204
0.749351
0.718692
0.557061
0.750106
0.901649
0.911003
Note: whole period indicates that S used ARMA (12,12), T-GARCH (1,1) and F used ARMA (1,1), T-GARCH (2,2). Pre crisis indicates that S used ARMA (1,1), GARCH (1,1) and F used ARMA (2,2), GARCH (1,1). During crisis indicates that S used ARMA (3,3), GARCH(2,1) and F used ARMA(9,3), GARCH (1,1). Post crisis indicates that S used AR (2), T-GARCH (1,1) and F used ARMA (1,1), T-GARCH (1,1).
Note *** ,**,* indicate the rejection of the null hypothesis at 1%,5%, 10% of significance levels, respectively. indicates Ljung-Box statistics of the order of standardized square of residuals. ** ( ) indicates that is the probability of Ljung-Box. The rejection of null hypothesis for ARCH is test is based on Engle (1982) critical value. The rejection of null hypothesis for autocorrelation is test is based on Ljung-Box statistics critical value.
= the series has no heteroskedasticity. = the series has heteroskedasticity.
= the series has no autocorrelation. = the series has autocorrelation.
Table 4 causality in mean based on univariate analysis
Whole Period
Pre crisis
During crisis
Post crisis
Mean Return
Mean Return
Mean Return
Mean Return
lead
lag
lead
lag
lead
lag
lead
Lag
S ïƒ F
F ïƒ S
S ïƒ F
F ïƒ S
S ïƒ F
F ïƒ S
S ïƒ F
0
0.9280 ***
0.9280 ***
0.9046 ***
0.9046 ***
0.8953 ***
0.8953 ***
0.9361 ***
1
0.0422 **
0.0293
0.0808
0.0996
0.0354
-0.0771
0.0324
2
0.0182
0.0129
0.0267
0.0213
0.0263
-0.0300
0.0303
3
0.0197
0.0072
0.0234
-0.0194
-0.0294
-0.0041
0.0357
4
0.0091
0.0117
-0.1205 **
-0.0643
0.0118
0.0517
0.0249
5
-0.0022
-0.0106
0.0138
-0.0045
-0.0582
-0.0036
0.0041
6
-0.0095
0.0044
-0.02
