گرایش محافظه کاری در بازار سهام در حال ظهور: شواهد از تایوان

Behavioral theories predict that investors underreact to earnings announcements stemming from the conservatism bias and overreact to a string of earnings news due to representativeness heuristic. This paper thus examines trading strategies of buying past high EPS growth stocks and selling past low EPS growth stocks over 4 to 20 quarters. The results generally support conservative reactions in the medium-term horizon, but provide little support for the over-use of representativeness heuristic on the long-term horizon. Moreover, we find that investors react differently to the consistency sequences of the two extreme earnings growth portfolios.

In recent years the finance literature has documented two market anomalies: underreaction and overreaction. Barberis et al. (1998) attribute these two findings to human psychological biases — namely, conservatism and representativeness heuristic — in which investors misreact to a string of news such as earnings announcements. Empirical and experimental evidence unfortunately shows mixed results. Chan et al. (2004) and Durham et al. (2005) provide little support for the parsimonious model of Barberis et al. (1998), while some studies are supportive (e.g. Bloomfield and Hales, 2002 and Frieder, 2004). A common feature of these studies is that they all use U.S. market data.1 In this paper we provide empirical tests using data from the Taiwan Stock Exchange (TSE), uncovering additional insights into the heuristic simplification of investors. In the model hypothesized by Barberis et al. (1998), investors do not realize that earnings follow a random walk. They misperceive that earnings belong to either a mean-reverting regime (in which investors react slowly to earnings announcements and exhibit conservatism) or a trending regime (in which investors extrapolate past earnings trends and show representativeness). These two behavioral biases accommodate the predictability of momentum and reversal in returns. However, Chan et al. (2004) fail to provide evidence of returns' predictability using U.S. market data. They identify the possibility of considerable arbitrage in the U.S. market, which quickly removes systematic mispricing resulting from investors' information processing biases. Taiwan's stock market is mainly dominated by domestic individual investors who constituted about 92% of market volume in 1990, but who still made up 71% by 2006. Chui and Wei (1998) find that Taiwan has the largest standard deviation of monthly excess returns among the five Pacific-Basin emerging markets.2Titman and Wei (1999) attribute this high volatility phenomenon probably to the investor sentiment story, because of the pervasive low level of sophistication encountered in this market. There is on average about one brokerage account per family,3 and trading costs are extremely low for small investors. This peculiar characteristic in Taiwan's stock market enables us to test the prevalence of investors' behavioral biases. According to Barberis et al. (1998), the most relevant variables in their model should be the sequences of earnings. We characterize earnings as earnings per share (EPS) due to its salience and availability, whereby what analysts or investors are most concerned about is the improvement or deterioration of EPS.4 Consequently, to formalize the sequential arrival of earnings information, we use quarterly EPS data to track the trends and sequences of listed firms' earnings.5 Most listed firms in Taiwan have fiscal years ending in December and are required to publish their quarterly financial statements to the market. Thus, investors regularly receive the same monthly extent of accounting performance reports for nearly all listed firms during the same period of time. While the fiscal periods of U.S. companies are clustered in June, September, and December fiscal year-ends (Asthana and Balsam, 2001), the sequences of financial statement releases are non-synchronous in terms of monthly performance measures. In addition, not all countries require listed firms to prepare quarterly financial reports. Taking the United Kingdom as an example, the London Stock Exchange only requires listed companies to submit interim (semiannual) and annual reports. Realizing these environmental differences will help increase our understanding of how earnings sequences have impacts on investors' behavior. When investors overreact or underreact to the trends and sequences of EPS growth, then implementing trading strategies of buying past high EPS growth stocks and selling past low EPS growth stocks generates either negative or positive results. Thus, we examine different time horizons from 4 to 20 quarters of past EPS growth in conjunction with holding periods ranging from 3 to 12 months. Doing so gives a set of 20 trading strategies. We find that the raw return behavior is predictable in the medium-term horizon. In other words, there is an underreaction to the high and low growth trends which represents an implication of the conservatism bias. After controlling for the Carhart four-factors, the underreaction still exists in the medium-term horizon. The low growth portfolios are somewhat riskier than high growth portfolios in terms of the market beta, size, and book-to-market factors. Once investors perceive a firm as extremely high (low) growth, then consistent sequences of EPS growth lead investors to fall into the trending regime, and they overreact. On the other hand, inconsistent sequences of EPS growth lead investors to stay in the mean-reverting regime, and they underreact. The results show an asymmetric reaction in the consistency tests of the two separately high and low growth portfolios. The differences in returns between consistent and inconsistent sequences are insignificant. Few exceptions occur at a 20-quarter horizon which shows return reversals with an implication of representativeness heuristic for the high growth portfolios, and a delayed reaction occurs at a 16-quarter horizon for the low growth portfolios. This paper tests the trend and sequences using the adjusted EPS to preclude stock dividends and/or stock splits effects since they are very common in Taiwan.6 The adjusted EPS is calculated as multiplying the EPS with the weighted average number of shares outstanding of the current quarter and then dividing it by the weighted average number of shares outstanding of the preceding quarter. Overall, our results generally support underreaction stemming from the conservatism bias in the medium-term horizons. However, we find little support for the over-use of reprensentativeness heuristic as described in behavioral theories. The paper is organized as follows. Section 2 provides some psychological evidence and the link to the operational definitions. Section 3 describes the data. Section 4 presents the empirical results. Section 5 concludes.

Behavioral theories predict that people assess the outcome which reflects the salient features of the process by which it is generated. The announcement of an EPS represents a salient feature of a firm's earnings power in which investors tend to form biased expectations. Thus, this paper provides an empirical test of behavioral theories using data from the Taiwan stock market. Prior empirical research has found underreaction over the medium term horizon and overreaction over the long term horizon in the stock market. An alternative behavioral view seeks to explain that investors underreact to earnings news stemming from the conservatism bias and overreact to a string of earnings news due to representativeness heuristic. However, the time horizons over which behavioral biases come into play are unspecified. Thus, we construct trading strategies that comprise 4 to 20 quarters for formation periods and 3 to 12 months for holding periods. The trading strategies that buy past high EPS growth stocks and sell past low EPS growth stocks yield significant cumulative raw returns in the medium-term horizon. After controlling for the Carhart four-factors, the profitability of such trading strategies still exists in the medium-term horizon. In addition, we find that the low growth portfolios are riskier than the high growth portfolios to the three Fama–French factors. We further examine the trading strategies based on the consistency sequences in the two separate high and low growth portfolios. The evidence shows that there is a reversal in returns for the high growth portfolios in the long-term horizons, though marginally significant over the 3 months. While we still find slow recognition of consistent low growth sequences, this suggests that investors react differently to the consistency sequences of the two extreme earnings growth portfolios. This paper examines various time horizons in which we can observe when and how behavioral biases arise. Overall, we find some evidence of conservatism bias in the medium-term horizons, but little support for the over-use of representativeness heuristic. The fact that investors react differently to the consistently high growth sequences versus the consistently low growth sequences invites further research.