Purposes

During the period of earnings announcements, the releases of financial information are publicly available in the market, and thus investors perceive the financial information to be either good or bad. However, investors with limited attention have an incomplete understanding about the content of the financial information. For example, investors are attracted by one salient performance of the company, while in turn ignoring other important information, and therefore exhibiting representativeness bias. On the other hand, if investors are inattentive to the earnings information and are slow in updating the status of a company’s performance, then they exhibit conservatism bias.

Barberis, Shleifer, and Vishny (1998) attribute underreaction to the psychological finding of conservatism bias. Thus, underreaction means that the average return on the company’s stock in the period following an announcement of good news is higher than the average return in the period following bad news. The authors also attribute overreaction to the psychological finding of representativeness bias, and they define overreaction as occurring when the average return following a series of good news announcements is lower than the average return following a series of bad news announcements.

Since representativeness and conservatism biases are probably an indirect consequence of limited investor attention (Hirshleifer and Teoh, 2003), this dissertation contributes to the exiting empirical work on investors’ inattention and heterogeneous beliefs to public information, as well as representativeness heuristics.

Thus, the general purpose of this dissertation is to consider psychological explanations in the search for a better understanding about the behavior of financial markets. More importantly, the dissemination of information in the market and the assimilation of information by investors are now that much more appreciated and explored.

As proposed by Simon (1956), when analyzing economic problems it is important to study the human cognitive limitation in conjunction with the structure of the information environment. Thus, the first purpose aims to examine the market reaction at different stages of a sequential release of annual reports in Taiwan. The evidence herein supports the hypothesis of limited attention.

Investors with limited attention are more easily attracted by salient information, invoking representativeness bias. The second purpose of this dissertation thus pertains to the tests of conservatism and representativeness biases. In the model of Barberis, Shleifer, and Vishny (1998), investors under certain circumstances tend to over-extrapolate to a string of earnings based on their pattern’s similarity, but they do not specify what kind of earnings leads to such a market misreaction. Accordingly, I choose the three most salient and available accounting performances in the income statement (revenue, operating income, and net income), Tobin’s Q, and stock return as proxies for a firm’s performances. For comparative purposes, accounting data are backward-looking measures, whereas Tobin’s Q is a forward-looking measure, and the stock return is purely a performance from the market.

The third purpose of this dissertation is to provide an explanation of behavioral biases. Griffin and Tversky (1992) argue that people update their beliefs based on the

strength of the evidence and its weight.¹ However, extensive psychological studies document that people generally do not combine strength and weight in accordance with the probability theory. One of the most important findings is that people tend to evaluate an uncertain event by the degree to which it: (i) is similar in essential properties to its parent population; and (ii) reflects the heuristics salient features of the process by which it is generated, which is known as representativeness heuristic (Kahneman and Tversky, 1972).

If people focus primarily on the strength of the evidence, they tend to neglect its weight and manifest representativeness bias. On the other hand, when people are unimpressed by the strength of the evidence, they focus too much on its weight and exhibit conservatism bias. Thus, the representativeness bias is in some sense the opposite of the conservatism bias (Brav and Heaton, 2002).

Griffin and Tversky’s (1992) theory predicts that, holding the weight of information constant, news with more strength will generate a bigger reaction from investors. Barberis, Shleifer, and Vishny (1998) provide reasonable, and empirically supportable, assumptions about the strength and weight of different pieces of evidence and derive empirical implications from these assumptions. Thus, Chapter 6 attempts to examine several measures of performance that have different strengths and weights of information features, leading investors to invoke the representativeness heuristic.

Despite the notion that empirical tests in Chapters 5 and 6 are closely related to Chan, Frankel, and Kothari (2004), the tests herein offer different perspectives. The major difference is in the methodology, in which I reserve the spontaneities of trend and sequence of a firm’s performances, whereas Chan, Frankel, and Kothari (2004)

1 The evaluation of a recommendation letter for a graduate student written by a former teacher can illustrate this concept. The strength indicates how positive or warm the letter is, whereas the weight measures how credible or knowledgeable the teacher is (Griffin and Tversky, 1992).

define 5 years as being long term and 1 year as the medium term to capture investors’

reaction. To this end, I distinguish the sequences of a firm’s performance between global and local sequences. The empirical results in Chapter 5 show that in the global representativeness test, there is a conservatism bias in revenue, operating income, and Tobin’s Q, while there is a representativeness bias in the stock return performance, which means that there is return predictability following the above performance trends. Furthermore, when I decompose those global sequences into local sequences, I find that there is no evidence of local representativeness bias.²

The dissertation’s fourth purpose is to shed light on how investors are sensitive to the streak length of performance. The empirical results in Chapter 5 show that among a firm’s five performances, only the sequences of net income and Tobin’s Q have significant impact on the stock return, in which Tobin’s Q has a significant reaction at a longer length of performance sequences. Furthermore, the empirical evidence in Chapter 6 indicates that investors exhibit the gambler’s fallacy to the trend of performance.

The fifth purpose of this dissertation is to provide empirical tests on the behavioral theories. Barberis and Thaler (2003) posit that there is really only one scientific way to compare alternative theories, behavioral or rational, and that is with empirical tests. Since direct empirical tests run the risk of over-fitting theories for each misreaction pattern observed in the market, I construct operational definitions of behavioral biases based on psychological evidence and apply out-of-sample tests that

2 The major difference between Chan, Frankel, and Kothari (2004) and my test is in the methodology, as there is a distinct component of stock market investors in the U.S. and Taiwan. The U.S. stock market is dominated by institutional investors and may have abundant arbitrage that quickly eliminates systematic mispricing resulting from investors’ information processing biases (Chan, Frankel, and Kothari, 2004). On the other hand, Taiwan’s stock market is mainly dominated by domestic individual investors who constituted about 92% of market volume in 1998 and who still made up 76% by 2004. The trading costs are extremely low for small investors.

can potentially reduce such risk (e.g., Campbell, Lo, and MacKinlay, 1997; Chan, Frankel, and Kothari, 2004; Hong and Stein, 1999).