Testing for predictability in emerging equity markets

Transcription

1 Emerging Markets Review 5 (2004) Testing for predictability in emerging equity markets Eui Jung Chang, Eduardo José Araújo Lima, Benjamin Miranda Tabak* Banco Central do Brasil, SBS Quadra 3, Bloco B., Ed. Sede, 9 andar, Brasilia, DF , Brazil Received 1 November 2003; received in revised form 1 March 2004; accepted 1 March 2004 Available online 4 August 2004 Abstract In this paper we test whether returns for emerging stock markets are predictable. We analyze predictability by means of multivariate variance ratios using heteroscedastic robust bootstrap procedures. Empirical results suggest that emerging equity indices do not resemble a random walk while for developed country indices (US and Japan) we are not able to reject this hypothesis. Furthermore, by employing variable moving average (VMA) and trading range break (TRB) technical trading rules we show that there is some evidence of forecasting power. However, when we take into account trading costs and a buy and hold strategy, only a few rules generate positive excess returns. We check for robustness by analyzing returns from 1559 different trading rules, testing different sub-samples, analyzing returns in bear and bull markets, and also comparing results found for emerging markets to the US and Japan. Furthermore, for the US the Variable Moving Average trading rules suggested in Brock et al. [J. Finance 47 (1992) 1731] do not seem to have forecasting power for the recent sample used, which could be due to the fact that these rules have been widely employed by market participants having the potential abnormal gains from them disappear. D 2004 Elsevier B.V. All rights reserved. JEL classification: G12; G14; G15 Keywords: Technical analysis; Emerging stock markets; Predictability 1. Introduction Emerging markets have received massive inflows of capital in the past and have become interesting alternatives for investors seeking diversification. Indeed, Harvey (1995) shows that emerging markets provide investment opportunities for world * Corresponding author. Tel.: ; fax: address: benjamin.tabak@bcb.gov.br (B.M. Tabak) /$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi: /j.ememar

2 296 E.J. Chang et al. / Emerging Markets Review 5 (2004) investors. In general, emerging markets offer high expected returns with an associated high risk. The aim of this paper is to assess whether emerging stock markets in Latin America and Asia exhibit predictability by testing the random walk hypothesis (RWH) using a multivariate version of the variance ratio (VR) test with a heteroscedastic robust bootstrap procedure and by testing technical trading rules such as the variable moving average (VMA) and trading range break (TRB) levels. We study Argentina, Brazil, Chile and Mexico in Latin America, and India, Indonesia, Korea, Malaysia, the Philippines, Thailand, and Taiwan in Asia. We also present results for Japan and the US for comparison purposes. 1 Most papers use variable and fixed moving average trading rules to assess whether there are significant profits to be made by technical analysis, following the seminal paper of Brock et al. (1992). The current paper analyzes whether technical analysis possesses forecasting power for price changes by studying the forecast power of 1559 different trading rules using a bootstrap procedure. The VMA technical trading rules proposed by Brock et al. (1992) are used as a benchmark. 2 We also test trading rules using TRB levels. Furthermore, the analysis is performed on a very recent sample the period from January 1991 through January 2004 in which most countries have liberalized their current accounts and received massive foreign portfolio inflows. We compare sub-samples (before and after the Asian crisis) and check whether the results are robust to the period used in the analysis. We also study the results of trading rules in bear and bull sub-samples. 3 The remainder of the paper is organized as follows. Section 2 presents a brief literature review. The empirical methodology is discussed in Section 3. In Section 4 we describe the data and the sample used in this paper. Section 5 shows empirical results. Finally, Section 6 concludes the paper. 2. Brief literature review If a stock price does satisfy the RWH, it follows that future equity prices are not predictable based on past prices. This has important implications for asset pricing modeling, especially for traders and practitioners that are searching for patterns in prices and betting on the markets using these patterns. Lo and MacKinlay (1988), in a seminal paper, present evidence that the RWH is strongly rejected by using VR statistics for the sample period and for different subperiods. 4 Since their seminal work, a variety of papers have found mixed evidence for a number of countries and sample periods. 5 1 International comparisons such as those presented in Harvey (1995) provide evidence that emerging markets returns are generally more predictable than the developed market returns. 2 Brock et al. (1992) propose the use of 1_50, 1_150, 1_200, 5_150 and 2_200 VMA. The first observation is the length of the short-term moving average while the second one is the length of the long-term moving average. 3 Although these bull and bear markets are defined on an ex-post basis they are useful in understanding why even if trading rules do not perform well for the full sample, they still are widely employed around the world by investors and traders. There could be substantial differences in the excess returns in these periods, which could explain their relative success. 4 Lo and MacKinlay (1989) show that these VR statistics are more reliable than traditional unit root tests. 5 See Frennberg and Hansson (1993) and Ayadi and Pyun (1994) for applications of this methodology for the Swedish and Korean equity markets, respectively.

3 E.J. Chang et al. / Emerging Markets Review 5 (2004) Urrutia (1995) tests the RWH for Latin American emerging equity markets and rejects it for some of these countries, suggesting that there is predictability. In another paper, Huang (1995) also shows that we can reject the RWH for Korea, Malaysia, Hong Kong, Singapore and Thailand. 6 Another branch of the literature focuses on the predictability of price changes using technical trading rules and determines whether one can earn abnormal profits from them, even when taking into account trading costs relative to a buy and hold strategy. In a seminal paper, Brock et al. (1992) test technical trading rules for the U.S. equity market. Their results are consistent with technical rules having predictive power. The authors suggest that technical rules may pick up some of the hidden patterns that are not detected by simple linear models. 7 They find that the returns during buy periods are statistically larger than those during sell periods, and that the former are less volatile. 8 Ratner and Leal (1999) analyze technical trading rules for 10 emerging equity markets from January 1982 through April Similarly to Bessembinder and Chan (1995), these authors have found using 10 VMA rules that for Taiwan, Thailand, and Mexico there were substantial profits to be made. 9 Additionally, Gencß ay (1998), Fernandez- Rodriguez et al. (2000) and Harvey et al. (2000) employ neural nets to test for nonlinear predictability for stock market returns for different stock markets and find strong evidence of predictability. Hence, for different countries the literature has found mixed empirical evidence depending on the time frequency, country, and time span among other characteristics. This could be due to differences in market microstructure or to other characteristics of these countries. 3. Methodology 3.1. VR tests Since the seminal works of Lo and MacKinlay (1988, 1989) and Poterba and Summers (1988), the most commonly employed statistic to test the RWH for stocks and exchange rates has been the VR statistic. 6 The reader is referred to other studies such as Blasco et al. (1997), Karamera et al. (1999), Darrat and Zhong (2000) and Grieb and Reyes (1999) that have found mixed results for a set of different countries using VR statistics. 7 The authors also suggest that transaction costs should be carefully considered before implementing such strategies. 8 See also Bessembinder and Chan (1995) who have found that technical trading rules appear to be successful in Asian countries, and Hudson et al. (1996) who show that technical trading rules for UK data have predictive ability. However, Hudson et al. (1996) suggest that investors would not benefit from them in the presence of costly trading. 9 See also Parisi and Vasquez (2000) who evaluate the performance of technical rules for the Chilean stock market, Gunesekarage and Power (2001), which analyze South Asian capital markets and Kwon and Kish (2002), who analyze the NYSE and NASDAQ indices. All these studies have found evidence of predictability for trading rules.

4 298 E.J. Chang et al. / Emerging Markets Review 5 (2004) If a time series of returns follows a random walk then in a finite sample the increments in the variance are linear in the observation interval, i.e., the variance of returns should be proportional to the sample interval. Thus the variance of monthly returns should be four times the variance of weekly returns. In order to test the RWH we can use the VR defined as VRðqÞ ¼ r2 q qr 2 ; ð1þ where the r 2 q and r 2 are the variance for q and one-period increments, respectively. The test relies on checking whether the VR is statistically indistinguishable from one. The VR( q) satisfies the relation VRðqÞi1 þ 2 Xq 1 1 k q k¼1 ˆqðkÞ; where qˆ(k) is the estimated kth order autocorrelation coefficient of returns. The statistics proposed to test this hypothesis have an asymptotically standard normal distribution. Nonetheless, statistical inference based on the asymptotic distribution could be misleading in finite samples. Furthermore, Chow and Denning (1993) have demonstrated that in testing for multivariate VR it is important to control the joint size test statistic. Additionally, the authors show that the multivariate test statistic has a Studentized Maximum Modulus distribution with degrees of freedom equal to the sample size. In order to circumvent the problems inherent to the use of a finite sample we propose a bootstrap procedure to derive the empirical distribution of these statistics. The bootstrap procedure implemented is robust to heteroscedasticity and can be found in Malliaropulos and Priestley (1999). Basically, we normalize returns by multiplying each observation of actual returns for each one of the time series of returns by a corresponding random factor and resample from these normalized returns. We then employ a multivariate version of the VR statistic due to Ceccheti and Lam (1994) to test the RWH null, in order to control for the investment horizon Technical trading rules The VMA rule states that one should take a long position if the short-term VMA is above the long-term VMA and stay short otherwise I t ¼ 1 if 1 X S P t s z 1 X >< L P t l S L s¼1 l¼1 ; ð3þ >: 0 otherwise where S and L stand for short and long-term, respectively. Brock et al. (1992) argue that the most common VMA rules are 1_50, 1_150, 5_150, 1_200 and 2_200, where 1, 2 and 5 represent the number of days in the short-term moving 10 By short is meant that one is out of the market and not short selling. Short selling can be costly and many emerging markets have restrictions to short selling. Thus our trading rules do not generate short sell signals. If a sell signal is generated then the analyst stays out of the market or sells, if he has previously bought the index. ð2þ

5 E.J. Chang et al. / Emerging Markets Review 5 (2004) average and 50, 150 and 200 the number of days in the long-term moving average. Thus, most of the literature has used these VMA rules to perform tests of whether technical analysis helps to correctly forecast price changes. The other technical rule that is used in this paper is a TRB trading rule. One receives a buy signal if prices penetrate the resistance level, i.e., go above a local maximum and a sell signal is given if prices fall below a local minimum (support level). If prices remain in the intermediate range then one maintains the original position. This rule can be defined as I t ¼ 1ifP t 1 ¼ Max½P t 1 ; P t 2 ;...; P t H Š; ð4aþ and I t ¼ 0ifP t 1 ¼ Min½P t 1 ; P t 2 ;...; P t H Š; ð4bþ and I t ¼ I t 1 if P t 1 aðmin½p t 1 ; P t 2 ;...; P t H Š; Max½P t 1 ; P t 2 ;...; P t H ŠÞ; ð4cþ where H stands for the number of days that is used in the TRB trading rule. The return for these strategies can be given by P tþ1 P tþ1 R tþ1 ¼ I t 1 I t 1 ði t 1Þ 1 CTAI t I t 1 A: P t P t ð5þ Transaction costs were imputed to the first buy and sell signals. If traders stay out of the market then the return is null. The returns of this active trading rule are compared to a buyand-hold strategy. In order to test whether buy signals generate significant returns one has to use a t- statistic, which is given by l B ¼ b l q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ð6þ r 2 =n þ r 2 b =n b where l b and n b are the mean return and number of signals for the buys and l, r 2 and n are the unconditional mean, variance and number of observations for the entire sample. Furthermore, l b ¼ 1 P nþ1 n b t¼0 R tþ1i t, and r 2 b ¼ 1 P nþ1 n b t¼0 ðr tþ1 l b Þ 2 I t, where the indicator function is defined in Eqs. (3) (4c). Analogously, the t-statistic for sell signals is given by l S ¼ s l p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi r 2 =n þ r 2 s =n ; ð7þ s where l s and n s are the mean return and number of signals for the sells. The t-statistic for the buy sell difference is l BS ¼ b l q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi s : ð8þ r 2 b =n b þ r 2 s =n s Since these t-ratios assume normality, stationarity, and time-independent distributions, one has to perform a bootstrap procedure to calculate critical values. Thus 500 new time

6 300 E.J. Chang et al. / Emerging Markets Review 5 (2004) series of returns were generated by randomly drawing from the original series with replacement. 11 The proportion of statistics as large as that from the original series was then used to calculate a simulated p-value. The bootstrapped returns are then exponentiated back to a bootstrapped price index with initial value equal to 100 and the technical trading rules are applied to this index. 4. The data The sample employed in this study consists of 11 emerging markets and indices for the United States and Japan, which are included for comparison purposes. We have collected daily closing prices for Argentina, Brazil, Chile, India, Indonesia, Malaysia, Mexico, the Philippines, South Korea, Taiwan, Thailand, Japan and the US. A problem often encountered is of low liquidity and infrequent trading in emerging stock markets. Therefore, in order to disentangle the effects of autocorrelation and infrequent trading we employ Morgan Stanley Capitalization Indices (MSCI), which mitigates these problems as it focuses on the largest capitalization stocks. Table 1 presents market information for the stock exchanges for the selected countries above. The first column presents information on trading costs, which were used to compute net returns of trading costs for these stock markets. 12 In order to mitigate the inherent problem of data snooping biases, we study a relatively long data series, from January 1991 through January Morgan Stanley provided the indices used in this study. Due to differences in holidays across countries, the total number of observations ranges from 2812 for India to 3395 for Japan. 14 To assess the robustness of the results, the analysis is performed for nonoverlapping subperiods. Two different time periods were selected for all return series, which represent bear and bull markets, in order to assess whether there is evidence of superiority of technical trading rules under specific market conditions. From Table 2 we can see that for most series the first order autocorrelation is significant at the 5% level. Only for the US are the first order autocorrelations insignificant. For the US, the autocorrelations at lags five and six are significant at the 10% confidence level but they are not significant at 5%. However, for the US, the autocorrelations for higher orders tend to increase substantially in absolute value. 15 The Jarque Bera (JB) statistic suggests non-normality for all return series. This motivates the use of the VR statistic, which is nonparametric, to assess whether time series of returns for equity indices have predictable components. These results are in line with those of Bekaert et al. (1998), which have found significant skewness and kurtosis for emerging markets, markedly different from those usually found for developed markets, for lower frequency data. Emerging markets equity 11 Ratner and Leal (1999) argue that this is a sufficient number of replications. 12 For emerging markets these costs tend to be high and it is important to account for them. 13 The beginning of the series was selected by availability of data for these indices. 14 For India the data series begin in January These results are not reported in the table to conserve space.

7 E.J. Chang et al. / Emerging Markets Review 5 (2004) Table 1 Market information Average trading costs Number of listed companies Market capitalization (in US$ millions) Developed economies US ,451, Japan ,495, Latin America Argentina , Brazil , Chile , Mexico , Market turn-over (%) Asia India , Indonesia , South Korea , Malaysia , The Philippines , Thailand , Taiwan , Column one shows average trading costs in basis points and accounts for fees commission and market impact (the average cost to trade versus the average price) for the fourth quarter of Column two presents the number of listed companies in major stock exchanges within specific countries in the end of 1998, and column three the Market Capitalization based on end-1998 total markets of listed companies in US dollars. The last column presents the market turn-over ratio (1998 total value traded by the average market capitalization for 1997 and 1998). Trading costs for Japan were calculated by averaging sell and buy costs. returns exhibit substantial departures from normality even when sampled on a lower frequency. 5. Empirical results 5.1. VR tests Table 3 presents results for the VR statistics for all return series. In the first row we present the country index that is used in the first column and the VR statistics for investment horizons up to 512 days. Instead of analyzing the single VR statistics we focus our analysis on the Wald statistics to control the joint size test statistic. From the Wald statistics we see that we can reject the RWH for all emerging markets (with the exception of Taiwan, which is in the boundary with a p-value of 10.7%). Furthermore, we are not able to reject the RWH for the US and Japan as previous studies have done (using lower frequency data). From the VR tests we can conclude that there seems to be some opportunities to exploit serial dependence in equity indices, especially for emerging equity markets. Since emerging equity indices seem to be predictable the next natural step would be to

8 302 Table 2 Descriptive statistics for daily returns in US dollars 01/02/ /31/2004 Equity exchange indices Argentina Brazil Chile Mexico India Indonesia South Korea Malaysia The Philippines Thailand Taiwan US Japan N Mean S.D Skewness * * * * * * * * * * * Kurtosis 18.95* 9.15* 7.15* 17.46* 5.61* 40.32* 24.00* 87.00* 20.05* * 6.47* 6.67* JB 34,881 5,221 2,406 29, ,845 60, ,813 40,785 14, ,661 1,968 p-value p < p < p < p < p < p < p < p < p < p < p < p < p < q * 0.108* 0.241* 0.089* 0.147* 0.128* 0.115* 0.089* 0.201* 0.2* 0.038** *** q * 0.029* 0.029* 0.021* 0.007* 0.022* 0.099* 0.051* 0.009* 0.028* 0.029** ** q * 0.004* 0.011* 0.006* 0.025* 0.02* 0.105* 0.006* 0.032* 0.019* 0.032** ** q * 0.004* 0.044* 0.002* 0.017* 0.092* 0.09* 0.077* 0.007* 0.003* 0.037* ** q * 0.02* 0.046* 0.002* 0.008* 0.027* 0.042* 0.042* 0.013* 0.039* 0.021* 0.042*** 0.023** q * 0.051* 0.027* 0.01* 0.042* 0.066* 0.068* 0.046* 0.008* 0.031* 0.007* 0.018*** 0.04* Q(6) The kth order autocorrelations are given by q k, JB is the Jarque Bera test for normality, Q(6) is the Ljung Box Q statistic for the null hypothesis of no autocorrelation up to lag 6. Returns were calculated as R t = ln( P t /P t 1 ), where P t is the price of the index at instant t. The *, ** and *** denote statistical significance at the 1%, 5% and 10% level, respectively. Average standard errors for skewness and kurtosis are and 0.086, respectively. E.J. Chang et al. / Emerging Markets Review 5 (2004)

9 E.J. Chang et al. / Emerging Markets Review 5 (2004) Table 3 VR of equity returns in US dollars (daily data from January 1991 to January 2004) VR( q) Investment horizon ( q) Wald Argentina p-value (0.008) Brazil p-value (0.003) Chile p-value (0.000) Mexico p-value (0.094) India p-value (0.002) Indonesia p-value (0.006) South Korea p-value (0.021) Malaysia p-value (0.049) The Philippines p-value (0.000) Thailand p-value (0.000) Taiwan p-value (0.107) US p-value (0.355) Japan p-value (0.225) Row 1 reports the VR( q) and row 2 reports one-sided p-value. The final column shows the Wald statistic and the bootstrapped p-value is given in parenthesis. investigate the source of this predictability. In the next subsection we will perform tests to check whether we are able to earn positive profits from technical trading rules through the use of bootstrap techniques as in Brock et al. (1992) and Ratner and Leal (1999) Technical trading rules Tables 4 and 5 present results for the 1_50, 1_150, 5_150, 1_200 and 2_200 VMA rules. A rule is significant when it satisfies two conditions: a p-value less than 0.10 and a positive excess return. For emerging markets we have 11 countries and 5 rules. In Table 4, 16 out of these 55 rules are statistically significant at least at the 10% level. If we only consider rules that are statistically significant at the 5% level, there are only 6 rules with explanatory power. From Table 5 we can see that if we average returns we find an average annualized excess return of 3.35% for emerging markets with the VMA rules. In Tables 6 and 7, we present results for the TRB trading rules. From Table 6 we see that out of the 55 rules we have nine significant rules. However, most of these significant strategies are concentrated in trading rules that use a greater number of observations. Rules

10 Table 4 Test results from VMA rules Argentina Brazil Chile Mexico India Indonesia South Korea Malaysia The Philippines Thailand Taiwan US Japan 1_50 Buy Sell p-value *** *** _150 Buy Sell p-value *** ** 0.02** ** _150 Buy Sell p-value ** *** 0.01** 0.05*** ** _200 Buy Sell p-value *** 0.05*** *** _200 Buy Sell p-value *** *** The first and second rows present annualized returns conditional on buy and sell signals. The third row presents the bootstrapped p-value for the Buy Sell statistic. If the Buy Sell statistic is significant at the 5% level then its value is greater than the Buy Sell statistic for more than 95% of the bootstrapped series. We only consider as statistically significant rules that have Buy Sell p-values less than 10% and also positive excess return (see Table 5). The *, ** and *** denote statistical significance at the 1%, 5% and 10% level, respectively. The Buy Sell statistic is given by: l BS ¼ b l q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi s r 2 b =n b þ r 2 s =n s 304 E.J. Chang et al. / Emerging Markets Review 5 (2004)

11 Table 5 Test results from VMA rules Argentina Brazil Chile Mexico India Indonesia South Korea Malaysia The Philippines Thailand Taiwan US Japan 1_50 N(Buy) N(Sell) Excess Return * * p-value _150 N(Buy) N(Sell) Excess Return ** ** 2.92** ** p-value _150 N(Buy) N(Sell) Excess Return 0.80*** ** *** 6.05** 4.27*** ** p-value _200 N(Buy) N(Sell) Excess Return *** 0.93*** *** p-value _200 N(Buy) N(Sell) Excess Return ** 2.68*** *** p-value The first and second rows present the number of buy (N(Buy)) and sell signals (N(Sell)) generated by a specific rule. The third row presents the annualized excess return for the strategy (considering transaction costs and the buy and hold strategy). If the excess return is greater than 95% of the excess return of the bootstrapped series, then it is significant at the 5% level. The *, ** and *** denote statistical significance at the 1%, 5% and 10% level, respectively. E.J. Chang et al. / Emerging Markets Review 5 (2004)

12 Table 6 Test results from TRB rules Argentina Brazil Chile Mexico India Indonesia South Korea Malaysia The Philippines Thailand Taiwan US Japan 2 Buy Sell p-value Buy Sell p-value Buy Sell p-value * *** Buy Sell p-value * *** ** 0.00* *** Buy Sell p-value ** 0.00* The first and second rows present annualized returns conditional on buy and sell signals. The third row presents the bootstrapped p-value for the Buy Sell statistic. If the Buy Sell statistic is significant at the 5% level then its value is greater than the Buy Sell statistic for more than 95% of the bootstrapped series and one can infer that there is forecasting power for these technical rules. We only consider as statistically significant rules that have Buy Sell p-values less than 10% and also positive excess return (see Table 7). The *, ** and *** denote statistically significance at the 1%, 5% and 10% level, respectively. The Buy Sell statistic is given by: l BS ¼ b l q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi s r 2 b =n b þ r 2 s =n s 306 E.J. Chang et al. / Emerging Markets Review 5 (2004)

13 Table 7 Test results from TRB rules Argentina Brazil Chile Mexico India Indonesia South Korea Malaysia The Philippines Thailand Taiwan US Japan 2 N(Buy) N(Sell) Excess Return p-value N(Buy) N(Sell) Excess Return p-value N(Buy) N(Sell) Excess Return * ** p-value N(Buy) N(Sell) Excess Return 0.30*** * *** ** 9.14** 7.20*** 1.52*** p-value N(Buy) N(Sell) Excess Return ** *** 8.87** p-value The first and second rows present the number of buy (N(Buy)) and sell signals (N(Sell)) generated by a specific rule. The third row presents the annualized excess return for the strategy (considering transaction costs and the buy and hold strategy). If the excess return is greater than 95% of the excess return of the bootstrapped series, then it is significant at the 5% level. The *, ** and *** denote statistical significance at the 1%, 5% and 10% level, respectively. E.J. Chang et al. / Emerging Markets Review 5 (2004)

14 308 E.J. Chang et al. / Emerging Markets Review 5 (2004) with 2 and 5 days are hardly significant and, as we can see in Table 7, most of them have a negative excess return. In Table 7, we present the number of buy and sell signals and the excess returns of TRB rules net of trading costs and a buy and hold strategy. This excess return is on average 17.99% for TRB rules. We can see from Tables 4 and 6 that for emerging markets all VMA trading rules have returns conditional on buy signals greater than returns conditional on sell signals, while the same holds true for 53 of the TRB trading rules. This suggests that these rules have forecast power for price movements in these countries. An interesting finding is that for the US and Japan we have the same qualitative results: 9 out of 10 rules generate returns conditional on buy signals greater than from sell signals for TRB rules and all VMA rules share this characteristic. However, even if these rules have forecasting power they are not able to produce significant excess returns when compared to a buy and hold strategy and when faced with transaction costs. From Tables 5 and 7 we see that the number of buy signals is higher than the number of sell signals for all countries from Latin America, Malaysia and the US, which indicates an upward trend for this group of countries and a downward trend for the remainder group of countries as pointed out by Hudson et al. (1996). Empirical results suggest that although there seems to be evidence of predictability, which is in line with Harvey (1995) and Ferson and Harvey (1991, 1993), it is difficult to exploit this predictability and the benchmark technical rules proposed in Ratner and Leal (1999) do not provide investors with a high return as one would expect after taking into account transaction costs and a buy and hold strategy. Equity returns for the US seem to be less predictable than returns for the rest of the countries in the sample. Furthermore, it seems that equity returns for countries from Latin America possess less predictive power than those from Asian equity markets. An interesting question is to ask why these rules are so appealing to investors. One could guess that these technical rules may generate profits in good market conditions. Therefore, we test whether these results hold in both bear and bull subperiods. In Table 8 we define bull and bear sample subperiods for all indices to test whether the excess return from technical trading rules behave differently within these subperiods. 16 In Fig. 1, we present the behavior of these indices and the window for the subperiods defined in Table 8. The bull and bear markets were defined arbitrarily using visual inspection of these indices. All indices are in log scale. The percentage of technical rules (both VMA and TRB) that have returns generated conditioned on buy signals greater than conditioned on sell signals in both bull and bear markets is presented in the left corner for each country. It is striking that for almost all technical rules there is a huge difference in the performance of these rules across countries. These rules seem to perform much better in bull markets than in bear markets, i.e., they seem to have an increase in the forecasting power. However, it is important to notice that still not much is significant. These results are qualitatively the same for developed markets. 16 It is important to note that even if one finds that there is a significant difference it cannot be exploited as we do not know a priori whether we are in a bull or bear market. However, significant differences across these different market conditions could explain the relative success of such rules among both traders and investors.

15 E.J. Chang et al. / Emerging Markets Review 5 (2004) Table 8 Bull and bear subperiods Bull subperiod Bear subperiod Argentina 02/24/95 09/04/97 02/17/00 11/05/01 Brazil 02/20/95 07/07/97 07/08/97 01/19/99 Chile 05/10/93 11/07/94 06/04/97 09/16/98 Mexico 03/21/95 09/10/97 02/22/94 03/20/95 India 11/20/98 02/18/00 02/21/00 07/23/01 Indonesia 02/04/93 02/03/94 07/30/97 09/28/98 South Korea 08/05/98 12/13/99 12/14/99 08/30/01 Malaysia 10/01/98 03/02/00 02/11/97 09/30/98 The Philippines 01/02/91 11/08/93 02/13/97 10/08/98 Thailand 12/04/91 12/16/93 06/20/96 12/29/97 Taiwan 11/24/95 07/31/97 05/01/00 09/13/01 US 04/16/97 04/12/00 04/13/00 11/11/02 Japan 08/17/98 01/27/00 01/28/00 02/11/02 The bull and bear subperiods have been defined a priori from visual inspection. The beginning and ending for each one of these subperiods are presented in the first and second columns, respectively. From the results presented so far, it seems that although technical trading rules have some forecasting power they seem to perform much better in bull rather than in bear markets and that potential profits seem to be much higher in emerging markets than in developed markets. The next step in our testing strategy is to analyze the behavior of different combinations of the VMA and the TRB rules. We allow the short-term moving average to range from 1 to 10 days while the long-term moving average ranges from 50 to 200 days. By studying all these combinations we have 1510 trading rules. For the TRB we analyze rules that use 2 to 50 days (49 technical rules). In Fig. 1, we also present the results for all trading rules inside the graph box for each country. In the first line we show the proportion of the 1559 trading rules that have had a significant buy sell statistic. Furthermore, we have analyzed two subperiods; we study these rules from January 1991 to December 1997 and from January 1998 to January Therefore, we have used the Asian crisis to split the sample. We can see that for the US none of these rules are significant and that predictability is much greater for emerging economies than for the US. For Malaysia and the Philippines, the proportion of rules that possess forecasting power is certainly high. The average proportion for the full sample of significant buy sell statistics is 30.29% for emerging markets, 7.02% for Latin America and is 43.59% for Asian economies. The numbers are 3.59% and 15.65% for Latin America during the first and second subperiods, respectively, and 27.84% and 20.27% for Asian economies. These results suggest that the degree of forecasting power in Asian economies has decreased over time while those of Latin America have increased. 17 We also present, in Table 9, the best rules for the full sample and for the same subperiods. We fix the best rule for the first subperiod and test its performance in the latter 17 It is important to notice that some of these Asian economies changed from a currency peg exchange rate regime to a floating exchange rate regime. As we are employing returns denominated in US dollars the exchange rate component could lower the predictive power.

16 310 E.J. Chang et al. / Emerging Markets Review 5 (2004) Fig. 1. The indices are presented in log levels with bull and bear subperiods chosen by visual inspection. In the left corner we present the proportion of rules that have returns conditional on buy signals greater than returns conditional on sell signals. We also present the same information inside the figures for the full sample and also for the January 1991 December 1997 and January 1998 January 2004 subperiods.

17 E.J. Chang et al. / Emerging Markets Review 5 (2004) Fig. 1 (continued ).

18 312 E.J. Chang et al. / Emerging Markets Review 5 (2004) Fig. 1 (continued ).

19 E.J. Chang et al. / Emerging Markets Review 5 (2004) Fig. 1 (continued). subperiod. The best rule is defined as the one that generates the highest excess return (regardless of its statistical significance). In the first row we present the name of the rules and in the second and third rows we show the excess return and the p-value for the buy sell statistic. For example, TRB_17 stands for a TRB with 17 days and 4_57 a VMA with a short moving average of 4 days and a long-term moving average of 57 days. From Table 9 we can see that only for Chile, India, Indonesia and the Philippines we have significant rules for both subperiods. These results are in line with previous findings. Although technical rules possess some forecasting power not much is statistically significant and this is robust when searching through the best rules available from different combinations. 6. Concluding remarks From multivariate VR statistics evidence suggests that emerging equity market indices do not resemble a random walk. On the other hand, for the US and Japan we are not able to reject the RWH using the same methodology, which is in line with the findings of Chow and Denning (1993) for the US using another multivariate VR statistic. These findings suggest that there is predictability in emerging equity returns. Our empirical results for trading rules show some evidence of predictability, which is in line with those found with the use of VR. However, the empirical results suggest that, in general, trading rules do not generate statistically significant profits after taking into account both transaction costs and a buy and hold strategy. Furthermore, on average TRB rules tend to perform worse than moving average trading rules. The technical trading rules employed in this study do not seem to have forecasting power for the US. This could be due to the immense literature that has focused on this

20 314 Table 9 First and second best rules from 1559 trading rules (1510 VMA rules and 49 TRB rules) for the full sample and subperiods Argentina Brazil Chile Mexico India Indonesia South Korea Malaysia The Philippines Thailand Taiwan US Japan 1st Best Rule Full sample 7_190 TRB_34 TRB_17 TRB_42 7_156 5_51 10_175 10_95 5_63 TRB_25 10_137 9_196 5_69 Return Buy Sell *** 0.00* 0.04** 0.04** 0.06*** 0.06*** 0.00* 0.00* ** *** 1st subperiod 3_149 6_124 TRB_28 9_61 8_152 1_79 7_144 10_57 TRB_49 TRB_30 7_107 9_197 10_60 Return Buy Sell ** 0.03** 0.06*** 0.09*** * 0.02** nd subperiod 3_149 6_124 TRB_28 9_61 8_152 1_79 7_144 10_57 TRB_49 TRB_30 7_107 9_197 10_60 Return Buy Sell *** 0.02** *** 0.01** *** 0.03** *** nd Best Rule Full sample 7_189 5_70 TRB_20 TRB_43 7_155 5_50 6_171 10_94 5_62 TRB_23 10_140 9_194 5_70 Return Buy Sell * 0.03** 0.03** 0.06*** 0.07*** 0.00* 0.00* ** *** 1st subperiod 1_108 6_123 3_54 10_62 7_152 1_81 7_145 7_53 TRB_50 TRB_29 10_110 9_198 5_69 Return Buy Sell *** 0.06*** 0.07*** 0.09*** * 0.01** nd subperiod 1_108 6_123 3_54 10_62 7_152 1_81 7_145 7_53 TRB_50 TRB_29 10_110 9_198 5_69 Return Buy Sell *** 0.08*** ** 0.07*** *** 0.10 We have analyzed 1559 trading rules using different combinations of VMA and TRB rules. We present the 1st and 2nd best rules for the full sample. We also divide the sample in two subperiods, from January 1991 to December 1997 and from January 1998 to January We test whether the 1st and 2nd best rules found for the 1st subperiod remain significant in the 2nd subperiod. Comparing the actual Buy Sell statistic with the Buy Sell statistics of the bootstrapped series we assess the significance of the Buy Sell statistics. The *, ** and *** denote statistical significance at the 1%, 5% and 10% level. Returns are annualized excess returns net of trading costs and a buy and hold strategy. TRB_ j stands for a Trading Range Break strategy (TRB) with j days and i_j for a variable moving average (VMA) with short and long moving averages, of i and j days, respectively. E.J. Chang et al. / Emerging Markets Review 5 (2004)

21 E.J. Chang et al. / Emerging Markets Review 5 (2004) market for the past decade. If traders and analysts read financial papers, since the literature has pointed out that these trading rules have significant forecasting power, the widespread use of these rules could have wiped out the potential for gains from them. In Chile s case, Parisi and Vasquez (2000) find evidence providing strong support for technical strategies. Our paper suggests that these opportunities seem to have been decreased substantially. Nonetheless, in this paper we have used a more recent data set and compared the returns for trading rules with a buy and hold strategy, which could explain the difference in the results. With regard to Ratner and Leal s (1999) results, which point out that Taiwan, Thailand, and Mexico emerged as markets with trading opportunities, this paper has found that evidence for Mexico has disappeared while there is renewed evidence for other Asian emerging markets. In general, trading rules seem to perform better for both Malaysia and the Philippines. When we investigate a combination of 1559 rules we find that there is some forecasting power, but not much is significant. These results suggest that the usual trading rules that have been used in most of the literature stemming from Brock et al. (1992) do not possess forecasting power for these markets but other rules could have to some extent. This could be due to the fact that these rules are widely known and have been broadly employed, which would minimize the potential gains that one could obtain from them. The general conclusion from the studies of Brock et al. (1992) and Hudson et al. (1996) is that technical trading rules have predictive ability if sufficiently long series of data are considered. In contrast to their findings, our empirical results suggest that for emerging markets, if we take into account both transaction costs and a buy and hold strategy we find some evidence of predictability using technical trading rules but not much is statistically significant. The results in this paper indicate that it is worthwhile to investigate more elaborate rules since simple technical trading rules can yield positive excess returns even when accounting for transaction costs and a buy and hold strategy. An interesting alternative would be to test for the forecasting power of nonlinear rules. Acknowledgements The authors gratefully acknowledge Donald Piantto and Nathalia Souza for helpful comments. We are especially indebted to an anonymous referee whose comments led to a considerable improvement of the paper. We also thank participants at the 3rd Brazilian Finance Association Meeting and XXV Brazilian Econometric Society Meeting where a preliminary version of the paper was presented. References Ayadi, O.F., Pyun, C.S., An application of the variance ratio test to Korean securities market. Journal of Banking and Finance 18, Bekaert, G., Erb, C.B., Harvey, C.R., Viskantas, T.E., Distributional characteristics of emerging markets returns and asset allocation. Journal of Portfolio Management, (Winter volume). Bessembinder, H., Chan, K., The profitability of technical trading rules in the Asian stock markets. Pacific- Basin Finance Journal 3,

22 316 E.J. Chang et al. / Emerging Markets Review 5 (2004) Blasco, N., Rio, C.D., Santamaría, R., The random walk hypothesis in the Spanish stock market: Journal of Business Finance and Accounting 24, Brock, W., Lakonishok, J., LeBaron, B., Simple technical trading rules and the stochastic properties of stock returns. Journal of Finance 47, Ceccheti, S.G., Lam, P.-S., Variance-ratio tests: small-sample properties with an application to international output data. Journal of Business & Economic Statistics 12, Chow, K.V., Denning, K.C., A simple multiple variance ratio test. Journal of Econometrics 58, Darrat, A.F., Zhong, M., On testing the random-walk hypothesis: a model comparison approach. The Financial Review 35, Fernandez-Rodriguez, F., González-Martel, C., Sosvilla-Rivero, S., On the profitability of technical trading rules based on artificial neural networks: evidence from the Madrid stock market. Economics Letters 69, Ferson, W.E., Harvey, C.R., Sources of predictability in portfolio returns. Financial Analysts Journal, (May/June volume). Ferson, W.E., Harvey, C.R., Explaining the predictability of asset returns. Research in Finance 11, Frennberg, P., Hansson, B., Testing the random walk hypothesis on Swedish stock prices: Journal of Banking and Finance 17, Gencßay, R., The predictability of security returns with simple technical trading rules. Journal of Empirical Finance 5, Grieb, T., Reyes, M.G., Random walk tests for Latin American equity indexes and individual firms. The Journal of Financial Research 22, Gunesekarage, A., Power, D.M., The profitability of moving average trading rules in South Asian stock markets. Emerging Markets Review 2, Harvey, C.R., Predictable risk and returns in emerging markets. The Review of Financial Studies 8, Harvey, C.R., Travers, K.E., Costa, M.J., Forecasting emerging markets returns using neural networks. Emerging Markets Quarterly 4, Huang, B.-N., Do Asian stock market prices follow random walks? Evidence from the variance ratio test. Applied Financial Economics 5, Hudson, R., Dempsey, M., Keasey, K., A note on the weak form efficiency of capital markets: the application of simple technical trading rules to UK stock prices 1935 to Journal of Banking and Finance 20, Karamera, D., Ojah, K., Cole, J.A., Random walks and market efficiency tests: evidence from emerging equity markets. Review of Quantitative Finance and Accounting 13, Kwon, K.-Y., Kish, R.J., A comparative study of technical trading strategies and return predictability: an extension of Brock, Lakonishok, and LeBaron (1992) using NYSE and NASDAQ indices. The Quarterly Review of Economics and Finance 42, Lo, A.W., MacKinlay, A.C., Stock market prices do not follow random walks: evidence from a simple specification test. The Review of Financial Studies 1, Lo, A.W., MacKinlay, A.C., The size and power of the variance ratio test in finite samples: a Monte Carlo investigation. Journal of Econometrics 40, Malliaropulos, D., Priestley, R., Mean reversion in Southeast Asian stock markets. Journal of Empirical Finance 6, Parisi, F., Vasquez, A., Simple technical trading rules of stock returns: evidence from 1987 to 1998 in Chile. Emerging Markets Review 1, Poterba, J., Summers, L., Mean reversion in stock returns: evidence and implications. Journal of Financial Economics 22, Ratner, M., Leal, R.P.C., Tests of technical trading strategies in the emerging equity markets of Latin America and Asia. Journal of Banking and Finance 23, Urrutia, J.L., Test of random walk and market efficiency for Latin American emerging equity markets. The Journal of Financial Research 18,