Are foreign investors noise traders? Evidence from Thailand. Sinclair Davidson and Gallayanee Piriyapant * Abstract It is plausible to believe that the entry of foreign investors may distort asset pricing in a partially segmented market. Sellin (1996) provides evidence that foreign investors are noise traders on the Swedish stock market. In this paper we apply his method to the Thai stock market. While it appears that foreign investors are noise traders, we provide evidence that foreign investors are able to time the Thai market. JEL: G10, G15 Corresponding Author: Sinclair Davidson School of Economics and Finance Royal Melbourne Institute of Technology GPO Box 2476V Melbourne 3001 Australia * School of Economics and Finance, Royal Melbourne Institute of Technology, GPO Box 2476V, Melbourne, 3001, Australia. We would like to thank, without implication, Robert Brooks and Robert Faff for helpful improvements on a previous version of this paper.
In an efficient market we anticipate that shares should reflect, on average, their fundamental value. Most asset pricing theories, such as the capital asset pricing model, would also predict, in equilibrium, that all shares are perfect substitutes for each other. These truisms, however, are complicated in an international setting. Once we consider integration and segmentation, some basic issues arise. For example, what exactly are fundamentals? Do we consider covariance risk relative to a local market portfolio or a world market portfolio? More importantly, from the perspective of local decision makers (in both the private and public sectors) what impact do foreign investors have on domestic market pricing? It is easy to imagine a scenario where foreigners do distort domestic pricing. Imagine a small economy opens its capital market to foreigners who subsequently invest. Imagine also that a number of the country s largest listed firms are included in an emerging market index. Foreign investors simply buy the indexed firms. They are not particularly interested in domestic issues of corporate governance and the like. Given that their investment opportunity set and decisions are based on the global portfolio (or multiple risk factors in an APT framework) their behaviour in the market may distort pricing in the local market relative to that market s fundamentals. This is especially the case when domestic investors are unable to legally invest offshore, ie. the market is partially segmented. The effect of this is to induce downwardly sloped demand curves for some stock, thus making them imperfect substitutes. Thailand has the potential to be such a country. Restrictions of foreign ownership create the potential for partial segmentation (see Agtmael (1993) and Bailey and Jagtiani (1994)). In addition, foreign trading statistics are collected by the Stock Exchange of Thailand, allowing us to evaluate the impact of foreign trading on that market. Noise Traders and Information Traders In an efficient market, prices will contain all available information about the value of a particular share. Unlike the standard assumption of homogeneous expectations made in asset pricing theories, however, market participants in reality are faced with having to 1
determine what information is. So, for example, is a particular news item information that will affect the market or not? Will all investors react to the same news item in the same way? The answer clearly must be no. Investors do not react to all information (valid or invalid) in similar ways and do not interpret the same information in similar ways. Investors may misinterpret valid information or even react to non-valid information. This can be called noise trading. Noise trading was first discussed by Black (1986). He argues that but for noise there would be no trading on the market. This phenomena that facilitates trading, however, makes the market inefficient. The point about markets and information is that if everyone has the same information and everyone is agreed to the effect it will have on prices, no one will be prepared to trade individual securities. Noise trading overcomes this problem. If there are people trading on noise, (even if they think the noise to be information) then there will exist incentives for those investors with information to trade. Black (1986: 531) states that as prices get noisier so it will be more profitable to trade on information. Thus we will see more information traders taking positions in the market. This does not imply that share prices will become more efficient. Share prices will tend to move away from their true value as more noise trading occurs. Countering this phenomena will be the fact that more information-traders will enter the market and take larger positions in the mispriced share. Thus there will be a tendency for the share price to return to its true value over time. Restrictions on short selling may impede this process (or, at least, slow it down) when shares are overvalued. Sellin (1996) investigates whether foreign investors are noise traders on the Swedish stock market. He differentiates between two hypotheses: An information trading hypothesis that posits that foreign investors are information traders and their previous buying activity is reflected in current prices. A noise trading hypothesis posits that foreign investors are not information traders and that their activity moves domestic prices away from fundamentals. Local investors, who are information traders, then enter the market and drive prices back to fundamentals. This hypothesis predicts that knowledge of past foreign buying behaviour can predict current returns. Sellin (1996) employs Granger causality analysis to demonstrate that trading causes returns, but that returns do not Granger cause trading. He then estimates the following equation: 2
r t = α 0 + α 1 r t-1 + α 2 r t-2 + β 0 z t + β 1 z t-1 + β 2 z t-2 + ε t (1) where r t = return on the Swedish stock index at time t and z t is a measure of trading activity at time t (net foreign purchases/(purchases plus sales)). He makes the argument that restrictions on the β estimates differentiate between his hypothesis. If the information trading hypothesis is true, we should fail to reject β 1 + β 2 = 0. If the noise trading hypothesis is true, we should reject the information trading restriction but fail to reject β 0 + β 1 + β 2 = 0. These restrictions, however, are not entirely convincing. The argument that Sellin (1996) is making assumes that foreigners as a group, on average, are noise traders. Domestic investors know this and when they discover that foreigners have been active in the market, they enter and trade on the basis of (superior?) information to drive prices to their correct levels. In addition, we should anticipate a sign constraint on, at least, one of the β coefficients. Specifically, if foreigners misprice in the domestic market and domestic investors profit from this mispricing, we expect β i < 0. Consistent with this view, Sellin (1996) reports significant and negative coefficients for β 2 when z = net purchases and gross purchases. Now either the Swedish authorities release foreign purchases and sales data with a two month lag, or foreigners are able to forecast Swedish stocks two months in advance (Sellin argues that this is unlikely), or Swedish investors wait two months to correct prices and earn profits. This last explanation is itself inconsistent with market information efficiency. Data and method. Data for the purchases and sales of foreign investors, local retail investors and local institutions are captured from the Factbook of the Stock Exchange of Thailand and various issues of the Monthly Review (issued by the Stock Exchange of Thailand) for the period 1995-1998. Net purchases for foreign investors, local (retail) investors and local institutions are calculated and standardised by trading activity (purchases plus sales). In addition the price series for the Stock Exchange of Thailand is collected from the Datastream database denominated in Baht. All data are monthly. Continuously 3
compounded returns are calculated as r t = ln(p t /P t-1 ). Summary statistics are shown in table one. TABLE ONE ABOUT HERE Sellin s (1996) study is replicated, ie. equation (1) is estimated using Thai data. We extend his analysis, however, to include local retail investors and local institutions. Results are shown in table two. TABLE TWO ABOUT HERE The results indicate that none of the three classes of investor are information traders. This can be described as an unexpected result. We would anticipate that local institutional investors would be, on average, information (or at least, informed) traders. It is plausible, however, to believe that information traders are not uniformly distributed to any one investor group and that the test procedure cannot detect their trading activity. In this sense, the Sellin (1996) test is weak in that it assumes that a particular class of investors is uniformly composed of a particular type of trader - either information or noise traders. The behaviour of foreign traders, however, while not meeting Sellin s test, is consistent with information trading. The coefficient β 1 is negative, indicating that prices are falling when foreign investors are buying, while the β 2 coefficient is positive indicating that Thai share prices rise in the month after foreign investors have been buying. This is consistent with the notion that foreign investors are able to time the Thai market. Given that the Thai authorities release trading information with a one month lag, it is also consistent with Thai investors responding to improved foreign confidence. If this latter explanation is true, then the Thai investors clearly do not believe that foreigners are noise traders, on average. 4
Conclusion This paper has investigated the impact that foreign investors have on the Stock Exchange of Thailand. Employing Sellin s (1996) method, we show that foreign investors appear to be noise traders. This method, however, also shows that local retail investors and institutional investors are noise traders. Either the test method is weak, or the Thai market is not highly efficient. Given that foreign investors appear to be able to time the market, informational inefficiency cannot be lightly discarded as a possible cause of our results. 5
Table one: Summary statistics. Data are shown for the monthly returns on SET and three classes of investors, Foreign, Local (retail) and institutional investors. This data are net purchases standardised by total trading ((purchases less sales)/(purchases plus sales)). ADF is the augmented Dickey-Fuller test and PP is the Phillips-Perron test for stationarity. All series are stationary for p = 0.05. Foreign Investors Institutional Investors Local Investors Returns on SET Mean 0.0337-0.0554-0.0132-0.0278 Median 0.0151-0.0301-0.0055-0.0179 Max 0.2939 0.1881 0.1279 0.4209 Min -0.2379-0.3673-0.2224-0.3312 Std.Dev 0.1164 0.1226 0.0735 0.1338 Skew 0.0957-0.4857-0.6671 0.7344 Kurt 2.6555 3.0631 3.3219 4.6488 JB 0.3105 1.8954 3.7669 9.7513 (0.8562) (0.3876) (0.1521) (0.0076) ADF -3.5015-4.5776-3.2077-4.8619 PP -5.1507-5.5646-4.7556-7.4231 N 48 48 48 48 6
Table two: Sellin method replicated for Thailand. Panel A shows results from equation (1) for foreign investors, Panel B shows results for Local investors and Panel C shows results for Institutional investors. White s (1980) test indicates heteroskedasticity for the Foreign investors equation, but not the local or institutional investors equations. p- values in parentheses in panel A are robust to heteroskedasticity, but not in panels B and C. The table also contains results of Wald tests on the coefficients with p-values in parentheses. Panel A α 0 α 1 α 2 β 0 β 1 β 2 Adj-R 2 F Foreign -0.0467-0.0296-0.0193-0.3218 0.8994-0.1662 0.3428 5.6940 p-value (0.0188) (0.8765) (0.9092) (0.0306) (0.0000) (0.4796) (0.0005) Info.Trading 13.5256 β 1 = β 2 = 0 (0.0000) Noise Trading 1.9432 β 0 + β 1 + β 2 = 0 (0.1710) Panel B α 0 α 1 α 2 β 0 β 1 β 2 Adj-R 2 F Local -0.0432-0.1128-0.0413 0.3107-1.0945 0.0745 0.2029 3.2916 p-value (0.0416) (0.4828) (0.7718) (0.2639) (0.0005) (0.8126) (0.0139) Info.Trading 7.8053 β 1 = β 2 = 0 (0.0014) Noise Trading 4.2219 β 0 + β 1 + β 2 = 0 (0.0465) Panel C α 0 α 1 α 2 β 0 β 1 β 2 Adj-R 2 F Institution -0.0332-0.0632-0.0244 0.5142-0.5256-0.0096 0.2833 4.5578 p-value (0.1609) (0.6986) (0.8671) (0.0015) (0.0037) (0.9569) (0.0022) Info.Trading 5.1344 β 1 = β 2 = 0 (0.0104) Noise Trading 0.0063 β 0 + β 1 + β 2 = 0 (0.9372) 7
Bibliography Agtmael, A. 1993. Thailand: From mini-market to leading emerging market. In, K. Park and A. Agtmael (eds). The world s emerging stock markets: Structure, developments, regulations and opportunities. Oxford: Heinemann Asia. Bailey, W. and J. Jagtiani. 1994. Foreign ownership restrictions and stock market prices in the Thai capital market. Journal of Financial Economics. 36: 57-87. Black, F. 1986. Noise. The Journal of Finance. 41: 529-543. SET. (1998). Factbook of the Stock Exchange of Thailand. Bangkok. Sellin, P. 1996. Inviting excess volatility? Opening up a small stock market to international investors. Scandinavian Journal of Economics. Reproduced in T. Andersen and K. Moene (eds). Financial liberalization and macroeconomic stability. Oxford: Blackwell. White, H. 1980. A heteroscedasticity-consistent covariance matrix estimator and a direct test for heteroscedasticity, Econometrica. 48: 817-838. 8