تست نظریه های امور مالی رفتاری با استفاده از روندها و ثبات در عملکرد مالی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|10794||2004||48 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Journal of Accounting and Economics, Volume 38, December 2004, Pages 3–50
Assessing the predictive ability of behavioral finance theories using out-of-sample data is important. Otherwise, the potentially boundless set of psychological biases underlying the behavioral explanations for security price behavior can lead to overfitting of theories to data. We test pricing effects attributed to two psychological biases, representativeness and conservatism, which underlie many behavioral finance theories. Using trends and consistency of accounting performance, we look for the pricing consequences of representativeness and conservatism. We find mixed evidence consistent with behavioral finance. Specifically, the theories based on representativeness are not supported, but we find some evidence of the pricing implications of conservatism.
Several studies document momentum (i.e., positive autocorrelation) in stock returns at horizons ranging from 3 to 12 months (e.g., Jegadeesh and Titman, 1993 and Jegadeesh and Titman, 2001) and return reversals at longer horizons (e.g., DeBondt and Thaler, 1985 and DeBondt and Thaler, 1987). Whether this predictability of returns, particularly at long horizons, results from time-varying discount rates in an efficient market or systematic mispricing is widely debated (e.g., Fama, 1998; Malkiel, 2003). However, the notion that it indicates market inefficiency as a result of investors’ information processing biases is rapidly gaining currency in the literature (e.g., Shleifer, 2000; Shiller, 2003). Our goal in this study is to test the predictions of market inefficiency theories (known as behavioral finance) based on investors’ biased processing of patterns in firms’ financial information. We explain that in many of the behavioral finance theories return predictability stems from investors’ over- or under-reaction to patterns, i.e., trends and consistency in recent financial information. Throughout the paper, financial performance (or financial information) refers to a firm's various operating performance measures such as sales and earnings. Trends and consistency in financial performance are identified using time-series observations of quarterly and annual operating performance data. We distinguish financial performance from the firm's share-price performance, which is measured using stock returns. Importance of the tests: Behavioral finance theories of inefficient markets have become a serious alternative to the efficient markets hypothesis, creating a need for tests to discriminate between the two. As Barberis and Thaler (2002, p. 61) observe, “There is only one scientific way to compare alternative theories, behavioral or rational, and that is with empirical tests.” In this respect, assessing the predictive ability of behavioral hypotheses using out-of-sample data is important.1 Absent such out-of-sample tests, theorists can use the potentially boundless set of psychological biases to build behavioral models and explain observed phenomena. Such attempts run the risk of over-fitting theories to observed results. Thus, by identifying pervasive psychological biases, forming empirically rejectable hypotheses, and testing for their validity, we aid behavioral theorists in isolating the fundamental behavioral phenomena, if any, influencing asset prices. In this spirit, we distill behavioral underpinnings of the theories and test for the predicted systematic mispricing. Our tests of return predictability on the basis of patterns (i.e., trends and consistency) in financial information are among the first set of out-of-sample tests of the behavioral finance theories. Previous research by Barth et al. (1999) examines whether the price-earnings ratios of firms reporting patterns of increasing earnings exceed those of other firms. However, they do not study subsequent share-price performance of such stocks to ascertain whether the stocks were overvalued. Thus, they cannot distinguish between rational valuation and excessive valuation resulting from improper extrapolation of increasing earnings patterns, which would be consistent with investors exhibiting representativeness bias.2 Background: Prompted in part by mounting evidence suggesting market inefficiency, researchers have developed behavioral finance theories that model the pricing implications of investors’ cognitive biases in information processing. These behavioral finance theories predict positive and negative autocorrelation in stock returns, which is inconsistent with market efficiency. All the theories assume that arbitrage forces are limited and therefore cannot eliminate systematic mispricing resulting from investors’ biased information processing (see Shleifer and Vishny, 1997). Notable among the attempts to construct formal behavioral models of systematic stock mispricing are Barberis et al. (1998), Daniel et al. (1998), Hong and Stein (1999), and Mullainathan (2001). While the literature describes several human information processing biases, pricing outcomes in many of the behavioral finance models are as if investors display representativeness and/or conservatism biases in processing information about stocks. Our characterization is in line with surveys of the literature on biases in human information processing and behavioral finance, which suggest the centrality of representativeness and conservatism to theories of systematic mispricing (see Fama, 1998; Barberis et al., 1998; Brav and Heaton, 2002; Daniel et al., 2002, and our discussion in Section 2 of the paper).3 The representativeness heuristic leads individuals to overestimate the probability of an event based on the similarity between its properties and the parent population's properties (see Tversky and Kahneman, 1974, and Barberis et al., 1998). That is, when assessing the probability that an object belongs to a particular category, individuals over-use that object's representativeness of the category (i.e., similarities to a typical member of the category) and under-use the base rate (i.e., the probability of the category). If someone looks like a criminal, people will overestimate the probability that he is a criminal because they over-use the similarity in looks and under-use the fact that criminals constitute only a small percentage of the population (i.e., low base rate of criminals). In the behavioral finance models, representativeness bias typically leads to initial overreaction. Therefore, representativeness predicts subsequent return reversals. Edwards (1968), among others, is credited with discovering conservatism bias, which causes people to update their beliefs slower than according to the Bayes's rule that serves as a standard of rationality in financial economics. Conservatism causes individuals to over-use the base rate and under-use the representativeness of the evidence. Therefore, Brav and Heaton (2002, p. 581) characterize conservatism as “in some sense the opposite of the representativeness heuristic.” The pricing implication of conservatism in the behavioral finance theories is that it generates underreaction. Therefore, conservatism predicts momentum in returns. Trends and consistency of performance to operationalize biases: Whether investors are predisposed to representativeness or conservatism, the information that investors process in a biased fashion is the pattern of past performance. Behavioral finance theories typically hypothesize that investors’ over- or under-use of a firm's financial information from the recent past leads to systematic mispricing. The mispricing is an outcome of investors’ decisions (e.g., buying and selling stocks) influenced by the representativeness or the conservatism bias in processing financial information. We argue that trends and consistency in financial performance operationalize two important information processing biases, namely representativeness and conservatism. We use a previously unexplored context (i.e., patterns in financial performance) to construct out-of-sample tests of behavioral theories that predict systematic mispricing. Patterns in performance or trends and sequences in performance capture not only the change in a performance measure from the start till the end of a period (e.g., five years) but also the consistency with which the change took place over the subperiods (e.g., five individual years). Summary of findings: We examine the relation between past trends and sequences in financial performance and future returns. First, the consistency in firm performance over a period, i.e., our proxy for the source of representativeness and conservatism biases in investor expectations, does not incrementally influence future price performance. Second, we fail to find evidence that investors systematically over-extrapolate trends in financial performance at long horizons. Abnormal returns in the year after 5 years of high or low growth are statistically and economically insignificant, which is inconsistent with price overreaction to performance trends. Third, we find some evidence that investors underreact to a 1-year trend in accounting performance. While this phenomenon does not appear to be distinct from post-earnings announcement drift, we interpret the evidence as lending support for conservatism bias influencing security prices. However, a strategy designed to exploit the consistency of quarterly performance within a year (i.e., a proxy for the source of conservatism bias) does not generate incremental abnormal returns. Thus, we also find evidence counter to the pricing implications of behavioral theories based on conservatism. Finally, we examine whether a firm-performance outcome that contradicts the past trend or consistency (e.g., poor performance following a string of good performance) leads to predictable return behavior. In this setting the tests fail to uncover return predictability. Overall, our evidence fails to suggest that trends and consistency in past financial growth rates generate return predictability that would be consistent with investors’ (i.e., consensus) biased expectations about future firm performance.4 Our failure to reject the null hypothesis naturally raises the question whether the tests employed in this study lack power. We believe that such a concern is not particularly warranted, but cannot be dismissed altogether. First, we perform the analysis over a fairly long time period of 36 years using annual data and 25 years using quarterly data. This is longer than that examined in most studies using financial accounting data. Second, the number of securities in most of the portfolios we study each year or quarter is in excess of 50. However, in one test the number of securities is small (i.e., less than 10 in many years and quarters). This trading strategy examines the performance of a portfolio consisting of stocks that experience a “disconfirming” earnings signal. Therefore, low power might be a reason for our failure to reject the null hypothesis in this particular trading strategy. Third, we use multiple performance measures (e.g., sales, earnings, and past stock returns) in developing trading strategies and we use several abnormal return measures in testing for mispricing according to the behavioral finance theories. To the extent these alternative measures are not perfectly correlated, the collective evidence is likely to yield more definitive conclusions. Finally, in some tests, trading strategies on the basis of past share-price performance produce significant abnormal returns, but similar strategies on the basis of financial performance do not. This reduces the likelihood that the financial performance-based trading strategies suffer from low power. Overall, our results present a challenge to the entire class of theories that predict mispricing based on investors’ representativeness or conservatism bias in processing firms’ past performance. Given past firm performance, the theories predict investors form biased expectations about future firm growth rates and commit systematic errors in setting prices. However, we find that return behavior is not easily predictable on the basis of patterns in past financial performance as suggested by the behavioral finance theories. Caveats: An important maintained hypothesis underlying behavioral theories of mispricing is that arbitrage is limited and thus it cannot eliminate the mispricing completely (see De Long et al., 1990a and De Long et al., 1990b; Shleifer and Vishny, 1997; Barberis et al., 1998). For example, the representativeness heuristic may lead the vast majority of investors to make erroneous investment decisions. However, unless arbitrage is limited, the actions of a relatively small number of rational investors seeking profitable trading opportunities will counteract these incorrect decisions and pull prices toward levels consistent with the economic fundamentals. This is the rationale underlying the efficient markets hypothesis predicting unbiased prices. Our failure to find evidence of mispricing consistent with behavioral theories does not mean that these models lack descriptive validity in the presence of limited arbitrage. That is, our failure to observe return predictability using data for securities traded on the U.S. exchanges might be due in part to abundant arbitrage that quickly eliminates systematic mispricing resulting from investors’ information-processing biases. Future research might attempt to test the predictions of these models in contexts that a priori exhibit limited arbitrage. One means of proxying for variation in the degree of arbitrage is by comparing the performance of trading strategies executed using equal- and value-weighted portfolios. Equal-weight portfolios assign more weighted to low market capitalization stocks than value-weight portfolios. Small stocks exhibit characteristics of firms with limited arbitrage: they are less closely followed by analysts, they have lower institutional ownership, and they have higher trading costs, including higher bid-ask spreads (see, for example, Bhushan, 1994). However, the similarity in the results we obtain using equal- and value-weighted portfolio returns suggests that the predictions of the behavioral hypotheses are not observed when arbitrage effectiveness is proxied for by firms’ market capitalization. Our failure to find widespread support for the (mis)pricing implications of the behavioral finance theories based on representativeness and conservatism biases says little about whether there are market inefficiencies. The implication of our evidence is only that it weakens the likelihood that systematic mispricing, if any, is a result of representativeness and conservatism as proxied for in our study. We cannot rule out another class of behavioral biases that might explain market inefficiencies if such inefficiencies were to exist. Outline of the paper: Section 2 discusses the representativeness bias and develops hypotheses about the stock-price consequences of the bias based on the behavioral finance models. Section 3 describes the data and the test methodology. Section 4 discusses results, and Section 5 concludes.
نتیجه گیری انگلیسی
Many stories about investor behavior rely on some form of the representativeness heuristic (i.e., over reliance on similarities when classifying firms) or conservatism bias. These information-processing biases can lead investors to form biased expectations of future firm performance. In a typical behavioral finance model, investors mentally misplace firms into various groups based on the past performance, and are subsequently surprised or disappointed in predictable ways. Behavioral finance theories predict that this “surprise” is reflected in returns. We use accounting data to test whether investors’ tendency to classify firms into categories influences security return behavior as modeled in the behavioral finance theories. We use trends of accounting performance to separate firms into high- and low-growth portfolios and further divide them by consistency of growth patterns. The advantage of this approach is that we use a specific source of information to model possible investor categories in a simple and straightforward way. Furthermore, our approach provides out-of-sample tests of the idea that investors under or overreact to past information. Finally, we use different horizons and growth metrics to allow for the different information investors could use. Consistent with findings in previous research, we find evidence of multi-month momentum in returns after accounting performance. However, this momentum is substantially reduced when we control for earnings-surprise effects. We fail to find support for the first hypothesis that past accounting performance trends over multi-year periods generate future return reversals. In direct tests of the behavioral finance theories, we find limited evidence that conditioning on the consistency of past growth rates improves return predictability (Hypotheses 2 and 3) as expected on the basis of representativeness bias. We observe some evidence consistent with underreaction stemming from conservatism bias in processing information. An interpretation of our findings is that, assuming widespread prevalence of representativeness and conservatism biases in investors’ processing of information, the maintained hypothesis of limited arbitrage is not particularly applicable in the U.S. context. However, notwithstanding our findings, the predictability of returns documented in the literature remains an interesting and problematic phenomenon potentially at odds with market efficiency. That is, our failure to find support for the pricing implications of representativeness and conservatism does not rule out the possibility of market inefficiencies. Our evidence simply weakens the likelihood that inefficiencies are a result of representativeness and conservatism as modeled in our study. Investors may think in categories, but using current theory as our guide, we are unable to document the stock price implications predicted in those theories. Alternatively, we failed to identify the correct categories, metrics, or horizons necessary to document the consequences of behavioral information processing biases. Our evidence poses a challenge to behavioral finance theories and therefore researchers should consider refining their models, in light of these results, to guide further empirical work.