Ebook The Efficiency of Market Reactions to Earnings News
The efficient market hypothesis (EMH), which maintains that market prices fully incorporate all information, is one of the most widely debated topics in the finance and accounting literatures. The EMH implies that future security prices should not be predictable conditional on the current information set. To date, however, empirical research has documented a large number of market anomalies that seem inconsistent with the EMH (see Fama 1998, Schwert 2003 for a review of this research), raising considerable doubt about the rationality of security prices and the efficiency of financial markets.
Amidst this debate, critiques of extant studies of market (in)efficiency are abundant. Aside from discussions related to data and methodology, a growing number of researchers have voiced concern about the research question itself. For example, at a 2001 roundtable discussion of capital market rationality, Kenneth French argued that the question of market efficiency should be framed as a continuum instead of as a yes/no dichotomy as in most current studies (Doukas et al. 2002). This view is based on the observation that due to various market imperfections, it is impossible for perfectly efficient markets as described by the EMH to exist in reality (Grossman and Stiglitz 1980). Therefore, rather than test whether or not the market is efficient, it may be more fruitful to investigate the degree to which a market is informationally efficient (Campbell et al. 1997, Lo 2007).
In this study we empirically investigate the cross-sectional variation in the degree of market efficiency and its determinants within the continuum framework, focusing on the context of corporate earnings announcements. We focus on earnings announcements for two reasons. First, prior research shows that the market does not fully incorporate the value-implications of earnings news in a manner consistent with the EMH (post- earnings announcement drift, hereafter PEAD). Second, the market under-reacts to earnings announcements, a feature that is of particular interest for our research methodology as we discuss presently (Bernard and Thomas 1990, Kothari 2001, among many others).
Our continuous market efficiency measure is based on a stochastic frontier model (Aigner, Lovell, and Schmidt 1977 and Meeusen and van den Broeck 1977), which assumes that inefficiency always exists in a production system and hence that the realized output will always be less than the theoretical maximum value represented by production functions. In a capital market setting, the outcome is the price of the stock and the inputs to the production function include information and other firm characteristics. If the market is perfectly efficient, then share prices should adjust completely and instantaneously after an information event. However, due to market frictions (such as high transaction costs) or investors’ behavioral biases, inefficiencies creep into prices and the market under-reacts to information Thus, the price-information relation under inefficiency makes earnings announcements a very suitable information event to study using the stochastic frontier approach (see Section 3 for further details). Note that the stochastic frontier model can be estimated at the individual stock level or the portfolio level. We estimate it at the portfolio level, which reduces the noise generated by liquidity-based trading around individual earnings announcements (Lee 1992). This approach yields one efficiency measure for each portfolio.
Using a sample that consists of 1,251 portfolios containing 110,881 quarterly earnings announcements from 1985 to 2005, in our first analysis we estimate the earnings-return stochastic frontier model separately for good and bad news sub-samples to allow for variation in parameter values between these two types of information events. We find that the mean (median) estimate of the efficiency of the market’s reaction to earnings news is 50.5% (50.2%) for positive unexpected earnings, and slightly lower at 47.4% (45.3%) for negative unexpected earnings. This suggests that, during the two decades covered by our sample, on average investors have been able to recognize less than half of the entire price implication at the time the earnings news was first made public. These results imply that investors’ initial under-reaction to earnings news predicts abnormal returns of as much as 11.2% in the post-announcement period, assuming that the temporary mis-pricing will finally be corrected in the long run. In addition, we find the portfolio-specific market efficiency estimates display considerable variation, ranging from as low as 0.6% to as high as 98.8%, with a standard deviation of 26.6%.
Next, we investigate the determinants of the cross-sectional variation documented above. Using several factors whose potential impact on market efficiency has been highlighted by previous research, we find that efficiency is generally higher for firms that have larger market capitalization, more analyst following, lower transaction costs, higher liquidity, less information uncertainty, lower arbitrage risk, and lower book-to-market.
Finally, we document a strong and positive correlation between our estimates of under-reaction at the time of announcement and post-announcement abnormal return. Perhaps more interestingly, we find that this correlation largely subsumes the ability of unexpected earnings to predict future returns. This finding provides further support to the hypothesis that the PEAD is due to investors’ initial under-reaction to earnings news. In addition, it also implies that our measure is a more accurate proxy for the magnitude of under-reaction than the SUE measure that has been widely used in literature.
Our study makes important contributions to the literature by developing a new measure of the degree of market efficiency under the continuum framework and providing direct estimates of the magnitude of investors’ under-reaction to earnings announcements. Extant studies have mostly relied on the post-event abnormal return as an indirect proxy for the magnitude of initial under reaction (e.g. Frazzini 2006, Vega 2006, and most other PEAD studies). However, this method is subject to several problems that could significantly reduce the accuracy of estimation.
First, the well-known expected return model specification problems cause estimated abnormal return to be biased, especially when they are cumulated over relatively long periods as in the PEAD studies. For example, Ball et al. (1988) find that betas tend to shift upwards after good earnings news and vice versa. Although the limited magnitude of beta shifts is not enough to entirely explain away the post-announcement drift (Bernard and Thomas 1989), it likely results in overstated abnormal returns.
Second, the market’s initial under-reaction to earnings news may not always lead to post-announcement price drift, a phenomenon that occurs when the gap between a security’s price and fundamental value is not large enough to cover the transaction costs required to profit from the mispricing. In this case, even if prices were set by irrational market participants at the time of an earnings announcement, the market remains Minimally Rational (Rubinstein 2001) and post-announcement drift will not be observed.
Therefore, PEAD may not always be a manifestation of the market's self correction process, and in some cases, ex post observed drift may not be related to investors’ initial under-reaction to earnings news at all.
The last issue in using abnormal announcement returns to estimate the magnitude of initial under-reaction is that neither the timing nor the duration of PEAD can be easily determined. For instance, while theories tell us why PEAD could occur, they make few predictions about when the drift will occur and how long it will last. Moreover, many behavioral finance models suggest return reversals in the long run. As a consequence of the absence of theoretical guidance, past studies of PEAD have used various announcement periods, ranging from 60 trading days to three years after earnings announcement. Not surprisingly, the reported magnitudes of drift have also been vastly diverse (Doyle et al. 2006).
In contrast to the traditional approach, the method we use to estimate the magnitude of under-reaction does not require post-announcement market information. Instead, it uses only unexpected earnings and short-window (2-day) announcement returns as inputs. Our method is therefore free of the problems discussed above.
The rest of the paper proceeds as follows. Section 2 reviews the related literature and Section 3 introduces the methodology. Section 4 describes the sample and defines our variables. Results are discussed in Section 5. The last section, Section 6, concludes.
Download
PDF Ebook The Efficiency of Market Reactions to Earnings News
Posted in :
