Full Title:

Weak-form Efficient Market Hypothesis, Behavioural Finance and Episodic Transient Dependencies: The Case of the Kuala Lumpur Stock Exchange Kian-Ping Lima, Venus Khim-Sen Liewb and Hock-Tsen Wongc
a

Authors: Affiliation:

Labuan School of International Business and Finance Universiti Malaysia Sabah P.O.Box 80594 87015 W.P. Labuan, Malaysia Department of Economics Faculty of Economics and Management Universiti Putra Malaysia 43400 UPM Serdang Selangor, Malaysia School of Business and Economics Universiti Malaysia Sabah Locked Bag 2073 88999 Kota Kinabalu Sabah, Malaysia

Correspondence to:

Kian-Ping Lim Tel: +6087460513 Fax: +6087460477 E-Mail: [email protected]

Weak-form Efficient Market Hypothesis, Behavioural Finance and Episodic Transient Dependencies The Case of the Kuala Lumpur Stock Exchange

Abstract This study utilizes the windowed-test procedure of Hinich and Patterson (1995) to examine the data generating process of KLSE CI returns series. Unlike previous studies, the present one relates the evidence to the popular weak-form EMH and behavioural finance, with the hope of offering some plausible explanations to the controversy arises between these two camps. Our econometrics results indicate that linear and non-linear dependencies play a significant role in the underlying data generating process. However, these dependencies are not stable as the results suggest that they are episodic and transient in nature. Along the line of our interpretations, we are able to offer some plausible explanations as to why weakform EMH generally holds in KLSE, though the presence of linear and non-linear dependencies implies the potential of returns predictability. Specifically, these significant dependencies show up at random intervals for a brief period of time but then disappear again before they can be exploited by investors. Looking from a micro perspective, we are able to rationalize the co-existence of weak-form EMH and behavioural finance in KLSE when the statistical properties of random walk, linear and non-linear dependencies, which also co-exist in the time domain, are interpreted in the framework of information arrival and market reactions to that information. Keywords: Data generating process; Weak-form EMH; Behavioural finance; Kuala Lumpur Stock Exchange; Malaysia.

I. INTRODUCTION The efficient market hypothesis (EMH) introduced three decades ago was a major intellectual advance and reached its height of dominance in academic circles around the 1970s. Nowadays, it remains as one of the major building blocks of modern finance, evidenced by its inclusion in most finance and financial economics textbooks. Briefly, Fama (1970) defined efficient market in terms of a fair game where security prices fully reflect all available information that is relevant to the determination of values. In this case, all securities are correctly priced and investors will earn a normal return on their investment that is commensurate with the level of risk assumed. One important implication is that security prices will change only when new information or news arrives that was not fully considered in forming current market prices. Even so, in an efficient market, agents will process this new information efficiently by immediately incorporate it into security prices. The central question that requires careful dissection here is what constitutes relevant information. This is crucial because conclusions about market efficiency could differ according to what is included in, or omitted from the information set. Fama (1970) has outlined a standard classification for different compositions of information set, and with this, the EMH can be classified into weak form, semi-strong form and strong form. The author explained that such classification permits researchers to pinpoint the level of information at which the hypothesis breaks down. The weak-form EMH, which is the focus of this paper, asserts that the only relevant information set to the determination of current security prices is the historical prices of that particular security1. In this regard, investors cannot expect to find any patterns in the historical sequence of security prices that will provide insight into future price movements and allow them to earn abnormal rates of returns. In most of the empirical literature, the random walk behaviour of security prices is used as the basis to test for weak-form EMH. Since new information is deemed to come in a random fashion in an efficient market, changes in prices that occur as a consequence of that information will seem random. Thus, price movements in a weak-form efficient market occur randomly and successive price changes are independent of one another. Formally, the random walk model can be stated as: pt = pt-1 + t (1)

where pt is the price at time t, pt-1 is the price in the immediate preceding period and t is a random error term. A purely random process is what statisticians called independent and identical distribution (i.i.d.), such as a Gaussian with zero mean and constant variance. The price change, pt = pt pt-1, is simply t which being white noise, is unpredictable from previous price changes. Looking from a different perspective, Equation (1) states that the best forecast of the price of a security at time t+1 is the price at time t, which in turn implies that the expected gain or loss for any holding period is zero. Therefore, analysis of past prices is meaningless because patterns observed in the past occurred purely by chance. Annuar and Shamsher (1993), Yong (1993a) and
1

Fama (1970) also defined the other two versions of EMH. The semi-strong form assumes that prices fully reflect all obviously publicly available information, while the strong form is concerned with whether individual investors or groups have monopolistic access to any information relevant for price formation.

Campbell et al. (1997), amongst others, provided excellent reviews on the subject of random walk. The body of literature on weak-form EMH can be considered as one of the most voluminous. Even emerging stock market like the Kuala Lumpur Stock Exchange (KLSE) of Malaysia has received considerable attention from researchers as the testing ground for EMH in those earlier years of 1980s. Barnes (1986) examined 30 companies and 6 sector indices for the six years ended 30 June 1980. Using monthly data, the serial correlation and runs tests results exhibit a high degree of efficiency in the weak form, with little departure from the random walk hypothesis. Further spectral analysis confirms the earlier findings that the KLSE is fairly efficient. Laurence (1986) used daily closing prices adjusted for cash and stock dividends, splits and rights issues, of 16 individual stocks traded on the KLSE over the sample period of 1 June 1973 through 31 December 1978. Results from serial correlation and runs tests suggest only slightly deviation from perfect weak-form efficiency. Using data for 6 sector indices and the all-share index from 1975 to 1982, Saw and Tan (1989) found that the Malaysian stock market is inefficient in the weak form when weakly data were used, but pockets of market efficiency existed when monthly data were used. Entering the 1990s, EMH studies on KLSE continue to grow at a phenomenal rate, and it is impossible for us to review all of those published works in one single paper. However, our objective will have been accomplished if a comprehensive and accurate picture is presented in that not only the favourable results are reported, but also covering those evidences against EMH, if there is any. Annuar et al. (1991) conducted the weak-form test on 82 individual stocks which were continuously traded on KLSE from 1975 to 1989, using the unit root methodology to account for cyclicality in price series and controlling thin trading effect. Overall, about 87% of the total sample of 82 stocks possess unit root, implying that there is a 13% chance that a security price is inefficient over the fifteen-year period. Though the findings suggest the market is generally weakform efficient, pockets of inefficiency are observed for shares that suffer liquidity problem. Annuar et al. (1993) addressed similar issue but using indices data in place of individual stocks, covering sample period from January 1977 through May 1989, with weekly and monthly intervals. The results from unit root analysis, serial correlation test and Q statistics strongly suggest that the KLSE is weak form efficient, though, once again, pockets of inefficiency are reported for some indices. Kok and Lee (1994) analyzed the stock prices behaviour of 32 companies listed on the Second Board of KLSE over the period 2 January 1992 to 30 December 1994. The results from various statistical tests- runs test, serial correlation test, Ljung-Box-Pierce Q test and Von Neumanns ratio test, suggest that information based on historical prices is fully reflected in current price within a week but may not be fully impounded in current price within a day. Thus, the Second Board of KLSE is weak-form efficient with respect to weekly data. Though daily price series are serially correlated, the magnitude of their correlations is not large enough to devise any mechanical trading rules for profitable investment timing. Kok and Goh (1995) utilized daily, weekly and monthly closing prices of 7 KLSE stock indices over a period of nine years from 1984 to 1992. Using similar methodologies as Kok and Lee (1994), the authors found serial dependencies in successive price changes for all KLSE daily stock indices. However, the significant correlations found are very small that it is unlikely to have any economic value, and this led the authors to conclude the market is weak-form efficient. When weekly data were used, the efficiency of the Malaysian stock market has improved from a weak-form

inefficient market in the mid 1980s to a weak-form efficient market by the late 1980s and early 1990s. Finally, the results from monthly data provide conclusive evidence of weakform efficiency, suggesting that market efficiency improves with longer temporal aggregation of sample data. Unlike other studies, Kok and Goh (1995) proceeded to address the issue of mean reversion using long-horizon returns. Though the variance ratio test provides evidence of mean reversion, it is not statistically significant to reject the long run random walk hypothesis. The literature reported thus far is in favour of weak-form efficiency, though pockets of inefficiency are detected in some of those studies. However, empirical evidence of inefficiency cannot be suppressed. One of those is documented in Yong (1989) who conducted serial correlation and runs tests on weekly closing prices of 30 stocks of random selection over the period January 1977 to June 1988. Another significant contribution to the literature is made by Yong (1993b). To avoid being criticized of not having sample stocks that are representative of the market, Yong (1993b) used weekly closing prices of all 170 stocks traded on the KLSE from January 1977 to May 1985 inclusive. Results from various statistical tests, especially those from the runs test reinforce earlier findings of departure from weak form market efficiency. As postulated by the author, the low trading volumes in most stocks and the possible price manipulations by those investors who own majority of the stocks might help to explain the findings of the runs test. In a nutshell, the results from our limited survey on published studies reveal that a verdict of weak-form efficiency cannot be proclaimed for the KLSE. One notable similarity of all the above cited studies is the application of standard statistical testsserial correlation test, runs test, variance ratio test and unit root tests, to uncover linear dependencies in the data. However, the lack of linear dependencies does not imply that the series are random as there might be other more complex forms of dependencies which cannot be detected by these standard methodologies. Even Fama (1965: 80) admitted that linear modelling techniques have limitations, as they are not sophisticated enough to capture complicated patterns that the chartist sees in stock prices. Steurer (1995: 202) expressed similar opinion, in which he argued that there is an order to the apparent randomness of the market. This order is so complex that the random walk concept is proven by the standard linear statistical tests. Another researcher, Brooks (1996: 307) agreed that series of financial returns often appear completely random to standard linear and spectral tests. However, the author strongly believed that if a different approach, using more powerful techniques, it may be possible to uncover a more complex form of dependencies in these series. One of the possibilities that might contribute to the departure from random walk is the presence of non-linear serial dependencies in the underlying data generating process. Even the influential paper of Fama (1970: 394) acknowledged this possibility, Moreover, zero covariances are consistent with a fair game model, but as noted earlier, there are other types of nonlinear dependence that imply the existence of profitable trading systems, and yet do not imply nonzero serial covariances. In this regard, Hinich and Patterson (1985) is the first published paper reporting evidence of non-linearity in common stock returns. As recalled by Patterson and Ashley (2000), the original manuscript of Hinich and Patterson (1985) met with resistance from the finance journals because finance academics were reluctant at that time to recognize the importance of distinguishing serial correlation from non-linear serial dependencies. Subsequent

evidence documented in Scheinkman and LeBaron (1989), Hsieh (1991), Abhyankar et al. (1995, 1997), Barkoulas and Travlos (1998), Opong et al. (1999) and Ammermann and Patterson (2003) strongly suggest that non-linearity is a cross-sectionally universal phenomenon. This also explains the phenomenal growth of non-linear modelling in the literature as non-linearity is now widely accepted as a salient feature of financial returns in general, and stock returns series in particular. The evidence of non-linearity has strong implication on the weak-form EMH for it implies the potential of predictability in financial returns. Specifically, if investors could have profitably operated a trading rule (net of all transactions costs) that exploits this detected non-linearity, it would be at odds with the weak-form EMH, which postulate that even non-linear combinations of previous prices are not useful predictors of future prices (Brooks, 1996; Brooks and Hinich, 1999; McMillan and Speight, 2001). However, Hsieh (1989) argued that the standard statistical tests such as serial correlation test, runs test, variance ratio test and unit root tests may fail to detect non-linear departure from the random walk hypothesis. Motivated by this concern, Lim et al. (2003c) re-examined the random walk hypothesis as all those earlier KLSE studies in favour of EMH have implicitly disregarded the presence of non-linearity, which will have serious consequences of making incorrect inferences and policy recommendations, as highlighted by Liew et al. (2003). Using the Brock-Dechert-Scheinkman (BDS) test, which has been proven to be quite powerful in detecting departures from i.i.d. behaviour in some Monte Carlo simulations (see, for example, Brock et al., 1991; Hsieh, 1991)2, Lim et al. (2003c) found the inadequacy of random walk model to describe the price behaviour of KLSE since some cycles or patterns show up more frequently than would be expected in a true random series. As those earlier studies have ruled out the possibility of linear dependencies, the departure from random walk might be due to the presence of non-linear serial dependencies. Indeed, Lim et al. (2003b, d) and Lim and Tan (2003) provided convincing evidence that non-linearity plays a significant role in the underlying dynamics of the Malaysian stock market. In the weak-form test of EMH, one has to be cautious when interpreting the results of linear and non-linear serial dependencies. There is no doubt that the conclusion of weakform market efficiency can be made when the hypothesis of random walk cannot be rejected by a robust test such as the BDS test. However, when the hypothesis is instead rejected due to the presence of certain dependency structures, it will be a strong statement to conclude that the market is inefficient. This was highlighted by Ko and Lee (1991: 224), If the random walk hypothesis holds, the weak form of the efficient market hypothesis must hold, but not vice versa. Thus, evidence supporting the random walk model is the evidence of market efficiency. But violation of the random walk model need not be evidence of market inefficiency in the weak form. The deciding factor here is whether those detected patterns in the historical sequence of stock prices can be exploited by investors to earn abnormal rates of returns. Thus, though the evidence of linear and non-linear dependencies implies the potential of returns predictability, it has to be further demonstrated, in order to reject the weak-form EMH, that investors are able to profitably exploit those detected underlying patterns. For instance, Kok and Lee (1994) and Kok and Goh (1995) argued that though daily price series are found to be serially correlated, the magnitude of their correlations is not large enough for any mechanical
2

Specifically, the BDS test has good power to detect at least four types of non-i.i.d. behaviour: nonstationarity, linear dependencies, non-linear stochastic process and non-linear deterministic process.

trading rules to be devised for profitable investment timing. In cases like this, the presence of linear dependencies cannot be taken as evidence against the weak-form EMH. Another excellent piece of work is provided by Ammermann and Patterson (2003), in which the detected non-linear dependency structures do not seem to be persistent enough to allow improvements over the random walk for predicting stock returns. Rather, these dependencies show up at random intervals for a brief period of time but then disappear again before they can be exploited by investors. On the other hand, relating the existence of linear/ non-linear dependency structures to the concept of information arrival and market reactions to that information will prove to be enlightening. Similar attempt by Ramsey and Zhang (1997) to tie their results of localized frequency bursts to news events is able to provide useful insights into market behaviour. In this regard, the EMH asserts that if market is efficient, then new information will be reflected quickly and unbiasedly into market prices3. Conversely, in an inefficient market, there will either be over-reaction or delayed response, where it takes some time before reaching the equilibrium price. Statistical findings of random walk movements in historical price series provide evidence of weak-form market efficiency, indicating that new information is quickly impounded in current market prices. In contrast, evidence of linear/ non-linear dependency structures reflect those periods in which the speed of adjustment to new information is not immediate, due to certain underlying factors. For instance, Ammermann and Patterson (2003) found that the detected linear serial dependencies in their studies are driven, at least in part, by the price limits that were imposed on the market, which cause the instances of significant autocorrelation to be heavily clustered. This is not surprising as Ryoo and Smith (2002) also found that effective price limits prevent market prices from following a random walk and as a result, induce autocorrelation. The explanation for the existence of linear dependencies driven by price limits can be found in Fama (1970), in which the author postulated that price limits will delay the adjustment of prices to the new equilibrium value and hence result in market being inefficient. On the other hand, Brooks et al. (2000) attributed the findings of non-linear dependencies to two important events occurred during their sample period- widespread upsets in the currency markets and a change in U.S. accounting procedures that affected U.S. firms with business abroad. The authors explained that when surprises hit the market, they generally generate a pattern of non-linear price movements relative to previous movements since the traders are unsure of how to react and hence they respond slowly. Another recent work by Lim et al. (2003a) found evidence of non-linearity in the underlying data generating process of bilateral Malaysian ringgit- U.S. dollar which can be attributed to central bank interventions that stopped the exchange rates from moving freely and reacting completely and immediately to new information. This study attempts to provide deeper insights into the stock prices behaviour in KLSE, with the hope of throwing some light on the controversy arises between the two camps of weak-form EMH and behavioural finance. Briefly, behavioural finance is build on two major building blocks, that is cognitive psychology (how people think) and limits to arbitrage (when markets will be inefficient), to explain the irrational financial behaviour of investors, either due to preferences or mistaken beliefs (see, for example, De Bondt
3

Three important dimensional aspects can be abstracted from price reaction to new information, namely magnitude, direction and speed (for details, refer Annuar, 2002). Our interpretation is concerned with the speed of price adjustment to new information, which has received less attention in the literature as compared to the other two dimensions of market efficiency.

and Thaler, 1985; Shefrin, 2000; Shiller, 2000; Hirshleifer, 2001; Barberis and Thaler, 2003; Ritter, 2003). In fact, investors in general will not behave rationally all the time in their decisions as postulated by EMH; there will be periods of irrationality reflected by over- or under-reaction. As explained by Schachter et al. (1985: 324), The investor is something more than the creature the economist hypothesizes. As well as the rational, utility maximizer of the economists creation, the investor is also a social creature influenced by the opinions and actions of others as well as by his own assessment of hard, economic facts. Empirically, Schachter et al. (1985) found that the South Sea Bubble during 1720 was a time of hysteria when all common sense, let alone rationality, abandoned the aggregate of investors. On the home front, Annuar (2002) provided an excellent survey on market inefficiencies of the KLSE in the context of behavioural finance. Another recent study by Lim et al. (2004) found evidence supporting their conjecture that the 1997 Asian financial crisis contributed to the non-random behaviour of stock prices as those panic investors are not able to make a rational assessment of the market and adjust rapidly and unbiasedly to the arrival of new information. These empirical evidences of irrational behaviour, just to highlight a few, compel researchers to seek alternative explanations and hence contributed to the growth of behavioural finance, posing a direct challenge to EMH. However, this study does not side with either camp in this controversy. Instead, we attempt to shed some interesting light by statistically demonstrating that both the weakform EMH and behavioural finance can co-exist in the time domain. In this regard, due credit should be given to Annuar (2002), which has been the source of our inspiration. In a culmination of more than ten years of research on the KLSE, Annuar (2002: 15) wisely penned his concluding remarks, I conclude that, given the world and KLSE evidence both collaboratory and contradictory, market efficiency and behavioural finance co-exist just as God created us and many observations in pairs. Chaotic (irrational) and rational behaviours co-exist in any market be it efficient, moderately efficient and inefficient. At times, we may act rationally; at other times irrationally. It is a matter of degree. Using data from the KLSE and windowed-test procedure proposed by Hinich and Patterson (1995), we are able to rationalize the co-existence of weak-form EMH and behavioural finance, and hence provide empirical content to the above statement. Browsing those earlier KLSE studies, we found traces of evidence that support this postulation, such as the findings of pockets of efficiency / inefficiency documented in Saw and Tan (1989) and Annuar et al. (1991, 1993). The sub-period analysis conducted by Kok and Goh (1995) is motivated by the possibility that a significant result for the whole period could be attributed to any particular sub-period, indicating that the authors were aware the weak-form EMH might not hold all the time. This study is further motivated by the empirical findings of episodic transient dependencies detected in most financial time series. Brooks and Hinich (1998) found that the Sterling exchange rates are characterized by transient epochs of dependencies surrounded by long periods of white noise. In Ammermann and Patterson (2003), a closer examination via the windowed-testing procedure of Hinich and Patterson (1995) reveals that the detected non-linear dependencies do not appear to be cross-temporally universal in that there are few brief periods in which the dependencies are very noticeable while other periods, in fact most of the time, the returns rather closely approximating a random walk. This means that the significant full sample results for non-linearity are actually triggered by the activity within a few relatively short pockets of highly non-linear data. Unlike the above two studies, Ramsey and Zhang (1997)

utilized waveform dictionaries approach and found similar structures which they described as localized frequency bursts, representing intense activity that is local in time and in frequency. Ramsey and Zhang (1997: 364) interpreted their results in such a way: The potential implications for understanding behaviour are that even over a low frequency range, price activity can be characterized by intermittent bursts of activity involving a relatively narrow range of frequencies, separated by relative quiescent periods. This new-found feature has direct bearing on the weak-form EMH and behavioural finance when the statistical properties of random walk, linear and non-linear dependencies are interpreted in the framework of information arrival and market reactions to that information. Specifically, the evidence of episodic transient dependencies suggests that there are periods when investors react rapidly to the arrival of new information (reflected in random walk price movements and taken as evidence of EMH), while at other times information is reflected only slowly into market prices (reflected in those clusters of linear / non-linear dependencies and taken as evidence of behavioural finance). In order to justify the co-existence of weak-form EMH and behavioural finance in the time domain, the crucial task is to determine whether the price series in KLSE are also characterized by such episodic transient behaviour of dependencies. As mentioned earlier, the tool which is capable of performing this task is the windowedtest procedure of Hinich and Patterson (1995). The test is designed to detect episodes of transient serial dependencies within a data series, by breaking the full sample into smaller sub-samples or windows of data. In other words, this procedure examines whether the dependencies found in the full sample are in fact due to strong but episodic occurrences that appear only infrequently and fleetingly. Hinich and Patterson (1995), Brooks and Hinich (1998), Brooks et al. (2000), Patterson and Ashley (2000) and Ammermann and Patterson (2003) have utilized this procedure to investigate the time series properties and stability of the underlying dynamics of financial data. Unlike all these earlier works in which the contribution is methodological, the present study adopts the Hinich and Patterson (1995) methodology but relates the evidence to the popular weak-form EMH and behavioural finance, with the hope of offering some plausible explanations to the controversy arises between these two camps. In the following section, this paper reviews some features of the Malaysian stock market, the Kuala Lumpur Stock Exchange (KLSE). Section III gives a brief description of the data and the methodology used in the present study. The subsequent sections present the empirical results as well as their interpretations in the framework of market behaviour. Finally, concluding remarks and some recommendations for further research are given at the end of the paper.

II. THE KUALA LUMPUR STOCK EXCHANGE (KLSE) In Malaysia, the Kuala Lumpur Stock Exchange (KLSE) is the only body approved by the Minister of Finance, under the provisions of the Securities Industry Act, 1983, as the stock exchange in the country. The KLSE is a self-regulatory organization with its own memorandum and articles of association, as well as rules which govern the conduct of its members in securities dealings. The KLSE is also responsible for the surveillance of the market place, and for the enforcement of its listing requirements which spell out the

10

criteria for listing, disclosure requirements and standards to be maintained by listed companies. Although the history of KLSE can be traced to the 1930s, the public trading of shares in Malaysia only really began in 1960 when the Malayan Stock Exchange (MSE) was formed. When the Federation of Malaysia was formed in 1963, with Singapore as a component state, the MSE was renamed the Stock Exchange of Malaysia (SEM). With the secession of Singapore from the Federation of Malaysia in 1965, the common stock exchange continued to function but as the Stock Exchange of Malaysia and Singapore (SEMS). The year 1973 was a major turning point in the development of the local securities industry, for it saw the split of SEMS into The Kuala Lumpur Stock Exchange Berhad (KLSEB) and the Stock Exchange of Singapore (SES). The split was opportune in view of the termination of the currency interchangeability arrangements between Malaysia and Singapore. Although the KLSEB and SES were deemed to be separate exchanges, all the companies previously listed on the SEMS continued to be listed on both exchanges. When the Securities Industry Act 1973 was brought into force in 1976, a new company called the Kuala Lumpur Stock Exchange (KLSE) took over the operations of KLSEB as the stock exchange in Malaysia. Its function was to provide a central market place for buyers and sellers to transact business in shares, bonds and various other securities of Malaysian listed companies. On 1 January 1990, following the decision on the final split of the KLSE and SES, all Singaporean incorporated companies were de-listed from the KLSE and vice-versa for Malaysian companies listed on the SES. The companies listed on the KLSE are listed either on the Main Board or the Second Board. The Second Board, which complements the Main Board, was established on 11 November 1988 to enable smaller companies with strong growth potential to seek a listing on the Exchange. Each board is further classified by sectors which reflect the core business of these companies. The KLSE computes an index for each of the main sectors traded on the bourse industrial, finance, property, tin and plantation sectors - and the second board. However, the most widely followed, by far, is the KLSE Composite Index (KLSE CI). The KLSE CI was introduced in 1986 after it was found that there was a need for a stock market index which would serve as a more accurate indicator of the performance of the Malaysian stock market and the economy. At that time, there was effectively no index which represented the entire market. The KLSE CI satisfies stringent guidelines and was arrived at only after rigorous screening of the component companies that were eventually selected to compose the index. In 1995, the number of component companies was increased to 100, and will be limited to this number, although the actual component companies may change from time to time. The KLSE CI is constructed by using the weighted average method, where the weight used is the number of ordinary shares outstanding.

11

III. METHODOLOGY The Data In this study, we utilize the daily closing values of the Kuala Lumpur Stock Exchange Composite Index (KLSE CI) for the sample period of 2 January 1990 to 30 June 2002. The price series obtained are used to compute a set of continuously compounded percentage returns for the KLSE CI, using the relationship: rt = 100* ln(pt / pt-1) (2)

where pt is the closing price of the stock on day t, and pt-1 the price on the previous trading day. Hinich and Patterson (1995) Windowed-Test Procedure In the windowed-test procedure of Hinich and Patterson (1995), a correlation portmanteau test similar to the Box-Pierce Q-statistic is developed for the detection of correlation or linear serial dependencies within a window. For detecting non-linear serial dependencies within a window, the procedure uses a bicorrelation portmanteau test, which can be considered as a time-domain analog of the Hinich bispectrum test statistic (Hinich, 1982). In applying these tests, the full sample is broken down into smaller subsamples or windows of data. If the full data sample does exhibit significant linear or nonlinear serial dependencies, but there are only a few smaller windows that are significant, then this suggests the data may instead be characterized by episodes of transient dependencies. In other words, it is the activity of these few windows that is actually driving the results of the overall sample. As demonstrated in the Monte Carlo simulations of Hinich and Patterson (1995), the test performs well even with small sample sizes. In this section, we provide a brief description of the test statistics used in the windowedtest procedure4. Let the sequence {y(t)} denote the sampled data process, where the time unit, t, is an integer. The test procedure employs non-overlapped data window, thus if n is the window length, then k-th window is {y(tk), y(tk+1),.., y(tk+n-1)}. The next nonoverlapped window is {y(tk+1), y(tk+1+1),.. y(tk+1+n-1)}, where tk+1 = tk+n. The null hypothesis for each window is that y{t} are realizations of a stationary pure noise process5 that has zero bicovariance. The alternative hypothesis is that the process in the window is random with some non-zero correlations Cyy(r) = E[y(t)y(t+r)] or non-zero bicorrelations Cyyy(r, s) = E[y(t)y(t+r)y(t+s)] in the set 0 < r < s < L, where L is the number of lags.

Interested readers can refer Hinich and Patterson (1995) and Hinich (1996) for a full theoretical derivation of the test statistics and some Monte Carlo evidence regarding the size and power of the test statistics. 5 A stationary time series is called pure-noise or pure white-noise if y(n1),., nN. A white noise time series, by contrast is one for which the autocovariance function is zero for all lags. Whiteness does not imply that y(n) and y(m) are independent for m n unless the series is Gaussian.

12

Define Z(t) as the standardized observations obtained as follows: Z (t ) = y (t ) m y sy (3)

for each t = 1, 2,, n where my and sy are the sample mean and sample standard deviation of the window. The sample correlation is: CZZ (r ) = (n r )
1 nr 2 t =1

Z (t ) Z (t + r )

(4)

The C statistic, which is developed for the detection of linear serial dependencies within a window, is defined as: C = [CZZ (r )]
r =1 L 2

~ 2 (L)

(5)

The (r, s) sample bicorrelation is: CZZZ (r , s ) = (n s) 1 Z (t ) Z (t + r ) Z (t + s)

for 0 < r < s

(6)

The H statistic, which is developed for the detection of non-linear serial dependencies within a window, is defined as: H = G 2 (r , s )
s = 2 r =1 1 L s 1

~ 2 (L-1) (L/2)

(7)

where G (r , s ) = (n s ) 2 CZZZ (r , s ) In both the C and H statistics, the number of lags L is specified as L = nb with 0 < b < 0.5, where b is a parameter under the choice of the user. Based on the results of Monte Carlo simulations, Hinich and Patterson (1995) recommended the use of b = 0.4 in order to maximize the power of the test while ensuring a valid approximation to the asymptotic theory. A window is significant if either the C or H statistic rejects the null of pure noise at the specified threshold level. This study uses a threshold of 0.01. In this case, the chance of obtaining a false rejection of the null is approximately one out of every 100 windows.

13

With such a low-level threshold, it would minimize the chance of obtaining false rejections of the null hypothesis indicating the presence of dependencies where these actually do not exist. Another element that must be decided upon is the choice of the window length. In fact, there is no unique value for the window length. According to Brooks and Hinich (1998), the window length should be sufficiently long to provide adequate statistical power and yet short enough for the test to be able to pinpoint the arrival and disappearance of the transient dependencies, which is the main purpose of using a windowed-test procedure. In this study, we follow the choice of Brooks and Hinich (1998) in which the data are split into a set of non-overlapping windows of 35 observations in length, approximately seven trading weeks. In fact, it was found that the choice of the window length does not alter much the results of both test statistics.

IV. EMPIRICAL RESULTS Table 1 provides summary statistics for the KLSE CI returns series in order to get a better view of some of the important statistical features. The mean of the returns series is quite small. For the third central moment, the result indicates some degree of positive or right-skewness. The fourth central moment, on the other hand, reveals that the distribution of the series is highly leptokurtic, in which the tails of its distribution taper down to zero more gradually than do the tails of a normal distribution. Not surprisingly, given the non-zero skewness level and excess kurtosis demonstrated within these series of returns, the Jarque-Bera (JB) test strongly rejects the null of normality.

Table 1 Summary Statistics for KLSE CI Returns Series KLSE CI Returns Series Sample Period No. of observations Mean Median Maximum Minimum Std deviation Skewness Kurtosis JB normality test statistic p-value
* Denotes very small value

2/1/1990-30/6/2002 3259 0.007818 0.000000 20.81737 -24.15339 1.679596 0.457460 38.02681 166713.4 (0.000000)*

14

Table 2 presents the correlation (C) and bicorrelation (H) test statistics for the KLSE CI returns series covering the full sample period. The results reveal that both the C and H statistics are highly significant, with extremely small p-values6. This indicates the presence of strong non-linear serial dependencies, in addition to autocorrelation, in the underlying generating process of the KLSE CI returns series. The findings of nonlinearity corroborate those of Lim et al. (2003b, d) and Lim and Tan (2003) in which these authors utilized the popular BDS and Hinich bispectrum test.

Table 2 C and H Statistics for KLSE CI Returns Series KLSE CI Returns Series Sample Period No. of observations No. of lags No. of bicorrelations p-values - C Statistic - H Statistic
* denotes extremely small p-value.

2/1/1990-30/6/2002 3259 25 300 0.0000* 0.0000*

Subsequently, we examine whether the detected dependency structures in the full sample, both linear and non-linear, are in fact due to strong but episodic occurrences that appear only infrequently and fleetingly. Table 3 shows the results for the windowed testing, in which the data are split into a set of non-overlapping windows of 35 observations in length, approximately seven trading weeks. In this case, four lags are used in calculating both the C and H statistics. The fifth row in Table 3 shows the number of windows where the null of pure noise is rejected by the C statistic7. In parenthesis is the percentage of significant C windows. The results show that the null is rejected in 6 windows by the C statistic, which is equivalent to 6.45%. By using a threshold level of 0.01, we would expect the C statistic to reject 1% of the windows by random chance. However, the percentage of windows exhibiting significant linear serial dependencies is greater than the expected 1%. Similarly, the percentage of significant H windows is also larger than the 1% nominal threshold level, as displayed in the seventh row of the same table.

Instead of reporting the C and H statistics as chi-square variates, the T23 program written by Professor Melvin J. Hinich reports the statistics as p-values. Based on the appropriate chi square cumulative distribution value, the T23 program transforms the computed statistic to a p-value. 7 In this study, the threshold level has been set at 0.01. The level of significance is the bootstrapped thresholds that correspond to 0.01.

15

Since both the C and H statistics are highly significant for the full data series as reported in Table 2, one would expect these serial dependencies to be persistent throughout the data or at least many more of the windows to exhibit strong serial dependencies. Instead, these significant test results in the full data series are reflected in only a relatively few windows. In other words, it is the activity of these few windows that is actually driving the results of the overall sample. Specifically, out of the total ninety three windows, only six (6.45%) exhibit significant linear serial dependencies and four (4.30%) exhibit nonlinear serial dependencies. Taken together, there are a total of ten windows or 10.75% where the null of pure noise is rejected, as shown in the final row of Table 3. These episodic transient features detected in the data indicate that the KLSE CI returns series are not stable, with the returns during most of the time periods move along at a close approximation to random walk, while during the remaining ten time periods they are characterized by highly significant linear and non-linear serial dependencies. The windowed-test procedure has an added advantage in that it permits a closer examination of the precise time periods during which the linear and non-linear dependencies are occurring. Table 3 also reports these time periods, which are potentially useful for our future investigation into the causes of these detected episodic transient dependencies, as performed by Brooks et al. (2000) and Ammermann and Patterson (2003). This episodic and transient behaviour of dependencies can be observed graphically. The histograms in Figure 1 and Figure 2 show the percentiles (i.e. one minus the p-value) into which the C and H statistics fall in each window for the KLSE CI returns series. Thus, a very significant window is plotted as a value near 1.0. It can be observed from these figures the episodic occurrence of the dependencies that appear within the data only infrequently. Another interesting feature is the transient nature of these dependencies, in which some windows appear highly significant, but then quickly disappear, or become too weak to be detected, in subsequent windows. Figure 3 shows those significant C and H windows, which is only ten in total, suggesting that they are not stable and persistent. It is clear from this final figure that the returns move closely approximating a random walk during most of the time.

16

Table 3 Windowed-Test Results for KLSE CI Returns Series KLSE CI Returns Series Total number of windows No. of lags No. of bicorrelations Significant C windows Dates of significant C windows 93 4 6 6 (6.45%) 29/5/1990-16/7/1990 7/5/1991-24/6/1991 13/8/1991-30/9/1991 7/1/1992-24/2/1992 21/7/1992-7/9/1992 29/5/2001-16/7/2001 4 (4.30%) 17/7/1990-3/9/1990 5/3/1996-22/4/1996 4/8/1998-21/9/1998 4/9/2001-22/10/2001 10 (10.75%)

Significant H windows Dates of significant H windows

Significant C and H windows

17

Figure 1 C Statistics for KLSE CI Returns Series

0.90

0.80

0.70

0.60

Percentile

0.50

0.40

0.30

0.20

0.10

0.00 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93

Window

18

Figure 2 H Statistics for KLSE CI Returns Series

0.90

0.80

0.70

0.60

Percentile

0.50

0.40

0.30

0.20

0.10

0.00 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93

Window

19

Figure 3 Significant C and H Statistics for KLSE CI Returns Series

0.90

0.80

0.70

0.60

Percentile

0.50

C H

0.40

0.30

0.20

0.10

0.00 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93

Window

20

V. INTERPRETATIONS Given that the KLSE CI returns series are characterized by episodic transient behaviour of dependencies, how does this evidence relate to the weak-form EMH and behavioural finance? This section sheds some interesting light on the controversy and justifies the co-existence of weak-form EMH and behavioural finance. Looking at the results as a whole, the evidence of linear and non-linear dependencies indicates departure from random walk and hence implies the potential of returns predictability. However, as discussed in the earlier section, in order to reject the weakform EMH, one has to demonstrate that investors are able to profitably exploit these detected dependency structures. The episodic transient feature of these dependencies suggests that they are not stable and persistent enough for investors to benefit from it. As depicted in Figure 3, these significant dependencies show up at random intervals for a brief period of time but then disappear again before they can be exploited. Thus, it is not surprising to learn that though non-linearity is a salient feature of financial time series, researchers have thus far failed to exploit the detected non-linearity in making improved point forecasts. This issue was well posed by Diebold and Nason (1990: 317), who wrote: Why is it that while statistically significant rejections of linearity in exchange rates and many other economic and financial series routinely occur, no nonlinear model has been found that can significantly outperform even the simplest linear model in out-of-sample forecasting? The results, taken as a whole, do not constitute evidence against the weak-form EMH. When these empirical findings are put under the microscope, we are able to rationalize the co-existence of weak-form EMH and behavioural finance. The results, which are best illustrated by Figure 3, suggest that there are few brief periods in which the dependencies, linear and non-linear, are very noticeable while other periods, in fact most of the time, the returns rather closely approximating a random walk. When these statistical properties of random walk, linear and non-linear dependencies are interpreted in the framework of information arrival and market reactions to that information, then it becomes clear that the weak-form EMH and behavioural finance in fact co-exist, at least in the context of KLSE. Specifically, the long periods of pure noise reflect the movements of financial returns closely approximate a random walk. In this regard, the market adjusts rapidly to the arrival of new information and this can be taken as evidence supporting the EMH. In contrast, those brief periods of linear / non-linear dependencies suggest that information is reflected only slowly into market prices or it takes some time for market prices to reach the equilibrium value, either due to over- or under-reaction of behavioural finance. Thus, following this line of interpretation, the episodic transient feature of KLSE CI returns series empirically support the co-existence of weak-form EMH and behavioural finance, as postulated by Annuar (2002). To be more precise, during most of the time periods, the market is weak-form efficient. There are only a few brief periods in which investors are not able to make a rational assessment of the market and adjust rapidly and unbiasedly to the arrival of new information. This is not surprising as we browse through those earlier KLSE studies, most of the empirical results support weak-form EMH, with only a few exceptions. However, we recognize that these interpretations, though supported to some extent by

21

our econometrics findings, remain conjectures. We invite alternative explanations of our econometric findings.

VI. CONCLUSIONS AND RECOMMENDATIONS This study utilizes the windowed-test procedure of Hinich and Patterson (1995) to examine the data generating process of KLSE CI returns series. Unlike previous studies, the present one relates the evidence to the popular weak-form EMH and behavioural finance, with the hope of offering some plausible explanations to the controversy arises between these two camps. Our econometrics results indicate that linear and non-linear dependencies play a significant role in the underlying data generating process. However, these dependencies are not stable as the results suggest that they are episodic and transient in nature. Along the line of our interpretations, we are able to offer some plausible explanations as to why weak-form EMH generally holds in KLSE, though the presence of linear and non-linear dependencies implies the potential of returns predictability. Specifically, these significant dependencies show up at random intervals for a brief period of time but then disappear again before they can be exploited by investors. Looking from a micro perspective, we are able to rationalize the co-existence of weak-form EMH and behavioural finance in KLSE when the statistical properties of random walk, linear and non-linear dependencies, which also co-exist in the time domain, are interpreted in the framework of information arrival and market reactions to that information. Though this study does not address the issue of model adequacy, the characteristics found in these returns series can be used to construct a better econometric model. However, given the complexity of the generating mechanism, it would certainly pose a challenge to researchers in constructing a model that could adequately capture these important features of the data. For example, it was highlighted by Brooks and Hinich (1998) that these transient epochs of dependencies could not be generated by any kind of ARCH or GARCH model. Even with some modifications, such as those by Booth et al. (1994) which augment the GARCH equation with structural breaks in the mean or those by Hamilton and Susmel (1994) which allow the parameters of the GARCH equation to be drawn from one of several regimes, still could not capture these dependencies. Another related study by Ramsey and Zhang (1997) using waveform dictionaries also found similar structures which the authors describe as localized frequency bursts. These authors suggested that the relevant model is one of oscillations induced by packets of information that leave the median of changes invariant to zero. The packets are characterized in time frequency space by short bursts of activity over narrow frequency ranges. This would certainly be a stimulating task. It would also be interesting to investigate the events that trigger these irrational reactions of investors in KLSE during those brief periods. This is possible because the window test procedure of Hinich and Patterson (1995) permits a closer examination of the precise time periods during which those dependencies, linear and non-linear, are occurring, as reported in Table 3. For instance, in the work of Ammermann and Patterson (2003), the linear dependencies are found to be directly attributable to changes

22

in the Taiwan Stock Exchanges price limits that were made during 1987 and 1988. Brooks et al. (2000) found that the non-linear dependency structures in their data are due to widespread upsets in the currency markets and a change in US accounting procedures. This line of inquiry is certainly worth investigating and will be included in future research agenda.

ACKNOWLEDGEMENT The authors appreciate the generosity of Professor Melvin J. Hinich of University of Texas at Austin for sharing his T23 program and some helpful comments in conducting the test. The authors would also like to express their gratitude to Professor Chris Brooks for those inspiring discussions on this subject matter. The usual disclaimer applies to any remaining errors or omissions.

REFERENCES Abhyankar, A.H., Copeland, L.S. and Wong, W. (1995) Nonlinear dynamics in realtime equity market indices: evidence from the United Kingdom. Economic Journal, 105, 864-880. Abhyankar, A.H., Copeland, L.S. and Wong, W. (1997) Uncovering nonlinear structure in real time stock-market indexes: the S&P 500, the DAX, the Nikkei 225, and the FTSE 100. Journal of Business and Economic Statistics, 15, 1-14. Ammermann, P.A. and Patterson, D.M. (2003) The cross-sectional and cross-temporal universality of nonlinear serial dependencies: evidence from world stock indices and the Taiwan Stock Exchange. Pacific-Basin Finance Journal, 11, 175-195. Annuar, M.N. (2002) Is the KLSE efficient? Efficient market hypothesis vs behavioural finance. Inaugural speech. Serdang: Universiti Putra Malaysia. Annuar, M.N., Ariff, M. and Shamsher, M. (1991) Technical analysis, unit root and weak-form efficiency of the KLSE. Bankers Journal Malaysia, 64 (April), 55-58. Annuar, M.N., Ariff, M. and Shamsher, M. (1993) Weak-form efficiency of the Kuala Lumpur Stock Exchange: an application of unit root analysis. Pertanika Journal of Social Sciences and Humanities, 1, 57-62. Annuar, M.N. and Shamsher, M. (1993) The efficiency of the Kuala Lumpur Stock Exchange: a collection of empirical findings. Selangor, Malaysia: Universiti Putra Malaysia Press.

23

Barberis, N. and Thaler, R. (2003) A survey of behavioral finance. In G. Constantinides, M. Harris and R. Stulz (eds.) Handbook of the economics of finance. Amsterdam: North-Holland. Barkoulas, J. and Travlos, N. (1998) Chaos in an emerging capital market? The case of the Athens Stock Exchange. Applied Financial Economics, 8, 231-243. Barnes, P. (1986) Thin trading and stock market efficiency: the case of the Kuala Lumpur Stock Exchange. Journal of Banking Finance and Accounting, 13, 609-617. Reprinted in I. Ibrahim and O. Yong (eds.) Understanding the behavioural patterns of stock prices: a collection of readings on selected far eastern stock markets, pp.65-72. Kuala Lumpur: Leeds Publications. Booth, G.G., Martikainen, T., Sarkar, S.K., Virtanen, I. and Yli-Olli, P. (1994) Nonlinear dependence in Finnish stock returns. European Journal of Operational Research, 74, 273-283. Brock, W.A., Hsieh, D.A. and LeBaron, B. (1991) Nonlinear dynamics, chaos, and instability: statistical theory and economic evidence. Cambridge: MIT Press. Brooks, C. (1996) Testing for non-linearity in daily sterling exchange rates. Applied Financial Economics, 6, 307-317. Brooks, C. and Hinich, M.J. (1998) Episodic nonstationarity in exchange rates. Applied Economics Letters, 5, 719-722. Brooks, C. and Hinich, M.J. (1999) Cross-correlations and cross-bicorrelations in Sterling exchange rates. Journal of Empirical Finance, 6(4), 385-404. Brooks, C., Hinich, M.J. and Molyneux, R. (2000) Episodic nonlinear event detection: political epochs in exchange rates. In D. Richards (ed.) Political complexity: political epochs in exchange rates, pp.83-98. University of Michigan Press. Campbell, J.Y., Lo, A.W. and MacKinlay, A.C. (1997) The econometrics of financial markets. Princeton: Princeton University Press. De Bondt, W.F.M. and Thaler, R.H. (1985) Does the stock market overreact? Journal of Finance, 40, 793-805. Diebold, F.X. and Nason, J.A. (1990) Nonparametric exchange rate prediction? Journal of International Economics, 28, 315-332. Fama, E.F. (1965) The behavior of stock market prices. Journal of Business, 38, 34-105. Fama, E.F. (1970) Efficient capital markets: a review of theory and empirical work. Journal of Finance, 25, 383-417.

24

Hamilton, J.D. and Susmel, R. (1994) Autoregressive conditional heteroskedasticity and changes in regime. Journal of Econometrics, 64, 307-333. Hinich, M.J. (1982) Testing for gaussianity and linearity of a stationary time series. Journal of Time Series Analysis, 3, 169-176. Hinich, M.J. (1996) Testing for dependence in the input to a linear time series model. Journal of Nonparametric Statistics, 6, 205-221. Hinich, M.J. and Patterson, D.M (1985) Evidence of nonlinearity in daily stock returns. Journal of Business and Economic Statistics, 3, 69-77. Hinich, M.J. and Patterson, D.M. (1995) Detecting epochs of transient dependence in white noise. Mimeo. University of Texas at Austin. Hirshleifer, D. (2001) Investor psychology and asset pricing. Journal of Finance, 56, 1533-1597. Hsieh, D.A. (1989) Testing for non-linearity in daily foreign exchange rate changes. Journal of Business, 62, 339-368. Hsieh, D.A. (1991) Chaos and nonlinear dynamics: application to financial markets. Journal of Finance, 46, 1839-1877. Ko, K.S. and Lee, S.B. (1991) A comparative analysis of the daily behavior of stock returns: Japan, the U.S and the Asian NICs. Journal of Business Finance and Accounting, 18, 219-234. Kok, K.L. and Goh, K.L. (1995) Malaysian securities market. Petaling Jaya, Selangor: Pelanduk Publications. Kok, K.L. and Lee, F.F. (1994) Malaysian second board stock market and the efficient market hypothesis. Malaysian Journal of Economic Studies, 31(2), 1-13. Laurence, M. (1986) Weak-form efficiency in the Kuala Lumpur and Singapore stock markets. Journal of Banking and Finance, 10, 431-445. Liew, V.K.S., Chong, T.T.L and Lim, K.P. (2003) The inadequacy of linear autoregressive model for real exchange rates: empirical evidence from Asian economies. Applied Economics, 35, 1387-1392. Lim, K.P., Azali, M. and Habibullah, M.S. (2003a) Non-linear dynamics in bilateral Malaysian ringgit- U.S. dollar spot rate. Jurnal Analisis, forthcoming.

25

Lim, K.P., Azali, M., Habibullah, M.S. and Liew, V.K.S. (2003b) Are non-linear dynamics a universal occurrence? Further evidence from Asian stock markets. Proceedings of the 1st International Conference of the Asian Academy of Applied Business 2003, 10-12 July 2003, Kota Kinabalu, Sabah, Malaysia, pp.287-295. Lim, K.P., Habibullah, M.S. and Lee, H.A. (2003c) A BDS test of random walk in the Malaysian stock market. Labuan Bulletin of International Business and Finance, 1(1), 29-39. Lim, K.P., Habibullah, M.S. and Lee, H.A. (2003d) Non-linear dependence in Malaysias stock market. Proceedings of the National Conference on Management Science and Operations Research 2003, 24-25 June 2003, Melaka, Malaysia, pp.169-183. Lim, K.P., Habibullah, M.S. and Lee, H.A. (2004) Do Asian stock market prices follow random walk? A revisit. Jurnal Analisis, forthcoming. Lim, K.P. and Tan, H.B. (2003) Episodic non-linearity in Malaysias stock market. Proceedings of the 1st International Conference of the Asian Academy of Applied Business 2003, 10-12 July 2003, Kota Kinabalu, Sabah, Malaysia, pp.314-323. McMillan, D.G. and Speight, A.E.H. (2001) Nonlinearities in the black market zlotydollar exchange rate: some further evidence. Applied Financial Economics, 11, 209-220. Opong, K.K., Mulholland, G., Fox, A.F. and Farahmand, K. (1999) The behavior of some UK equity indices: an application of Hurst and BDS tests. Journal of Empirical Finance, 6, 267-282. Patterson, D.M. and Ashley, R.A. (2000) A nonlinear time series workshop: a toolkit for detecting and identifying nonlinear serial dependence. Boston: Kluwer Academic Publishers. Ramsey, J.B. and Zhang, Z. (1997) The analysis of foreign exchange data using waveform dictionaries. Journal of Empirical Finance, 4, 341-372. Ritter, J.R. (2003) Behavioral finance. Pacific-Basin Finance Journal, 11, 429-437. Ryoo, H.J and Smith, G. (2002) Korean stock prices under price limits: variance ratio tests of random walks. Applied Financial Economics, 12, 545-553. Saw, S.H. and Tan, K.C. (1989) Test of random walk hypothesis in the Malaysian stock market. Securities Industry Review, 15(1), 45-50.

26

Schachter, S., Gerin, W., Hood, D.C. and Andreassen, P. (1985) Was the South Sea bubble a random walk? Journal of Economic Behavior and Organization, 6, 323-329. Scheinkman, J. and LeBaron, B. (1989) Nonlinear dynamics and stock returns. Journal of Business, 62, 311-337. Shefrin, H. (2000) Beyond greed and fear. Boston: Harvard Business School Press. Shiller, R.J. (2000) Irrational exuberance. Princeton: Princeton University Press. Steurer, E. (1995) Nonlinear modeling of the DEM/USD exchange rate. In A.P. Refenes (ed.) Neural networks in the capital markets, pp.199-211. New York: John Wiley & Sons. Yong, O. (1989) The price behaviour of Malaysian stocks. Malaysian Management Review, 24(3), 23-34. Yong, O. (1993a) Introduction: a brief review on the random walk hypothesis. In I. Ibrahim and O. Yong (eds.) Understanding the behavioural patterns of stock prices: a collection of readings on selected far eastern stock markets, pp.3-10. Kuala Lumpur: Leeds Publications. Yong, O. (1993b) Market efficiency (weak-form) of the Malaysian stock exchange. In I. Ibrahim and O. Yong (eds.) Understanding the behavioural patterns of stock prices: a collection of readings on selected far eastern stock markets, pp.73-95. Kuala Lumpur: Leeds Publications.