Kate Nardinelli
Economics Honors Thesis
Stanford University
Email: knardinelli@stanford.edu
May 2002
ABSTRACT
Investor expectations of company earnings are reflected in stock price. Thus by comparing analyst earnings forecasts to stock prices, a test can gauge analyst rationality and market efficiency. My research is based on this premise. Following Abarbanell (1991) and Lys and Sohn (1990), this paper will extend research on analyst rationality by examining how analysts revise their opinions as stock prices change. I use two sample groups consisting of quarterly earnings forecasts from 1996 to 2001, a Dow sample and an Internet sample. Analyst earnings revisions are regressed on stock and market returns since an analyst’s previous forecast, calculating “forecast response coefficients.” The Dow forecast response coefficients are positive and significant in all years and show that forecasts do move in the same direction as stock prices. However, the Internet sample forecast response coefficient is only significant in two years, moving from 1.089 in 1998 to –0.505 in 1999. Second, I run a regression of forecast error on stock and market returns for three time periods: before, during, and after the forecast announcement. The results show that analysts do not fully incorporate information available in prices in each group, so errors can be predicted based on price movements prior to an analyst’s forecast revision announcement. I find that while both Dow and Internet analyst revisions stray from informative prices, the largest departure occurred in the Internet group in 1999. Overall, evidence indicate that analysts and prices are rational at times and irrational at times, a puzzle that should motivate future research.
Introduction
The stock market plays a more visible role in contemporary American life than ever before. In particular, the country observed with much fanfare while daring entrepreneurs realized the American Dream via Internet start-ups, and later while stunned employees watched their nest eggs vanish. When the stock market bubble burst in March of 2000, investors lost more than $4 trillion (Vickers and France 2002: 42). Aside from these extreme examples, an estimated 63.7 percent of the national population owned securities in 1994, versus 61.5 in 1987 and 2.5 percent in 1929 (Poturba and Samwick 1995: 316; Kennedy 1999: 41). These participation rates show that ordinary Americans have taken part in the unprecedented prosperity of the capital markets of the 1990s. The capital markets reflected the economic growth of the decade, and this relationship motivates extensive research concerning capital market prices and efficiency. Moreover, there are economic and policy motivations for understanding and dissecting stock prices.
The Efficient Market Hypothesis (EMH) is the traditional departing point for stock market theory. The hypothesis states that security prices reflect all publicly available information regarding the underlying value of the firm at any given time; they are “rational.” It follows from this theory that no strategy can be formulated to consistently outperform market returns. Volumes of research and debate surround the EMH. For instance, the vast majority of mutual funds over time have failed to beat the market by sustaining above average returns. (This dismal performance surely contributed to the creation of index funds and their rise in popularity.) Even the financial wizards thus far have not been successful, as the spectacular failure of Long Term Capital Management demonstrates. Forgiving the professionals and academics for poor performance, Americans paid security brokerages $112.7 billion in 2000 for advice on how to beat market returns; this figure climbed from $88.7 billion in 1999 and $73.4 billion in 1998 (US Census Bureau 2001).
Investors pay because the apparent persistence of excessive stock market return anomalies suggest that increased profits are possible, challenging the efficient market hypothesis. Many individual investors thus rely on security analysts to navigate the financial markets. These analysts not only influence their institutional and private clients’ portfolios, but they also reach a wider audience by commanding airtime on television networks and coverage in the financial and mainstream press. The thriving business of security analysts suggests that they provide a valuable service, and the only valuable service in the stock market is the production of information. Stock price is the primary market signal of information, so if analysts are valuable they must offer their customers information not publicly reflected in stock prices. Yet the forecasts should not abandon price signals if they are to be rational. Indeed, the analysts’ informational function is extremely important in the market. A May 2002 Business Week cover story entitled “How Corrupt is Wall Street?” explains, “The entire economy depends on the financial system to raise and allocate capital. And that financial system, in turn, is built on the integrity of its information. Should investors lose confidence in that information, it could … easily put a damper on the economy” (Vickers and France 38). Although brokerage firms face marketing incentives to employ analysts, historically the market has respected their research and estimates.
Consensus analyst forecast estimates are widely used to represent market expectations of earnings, and research consistently reports abnormal returns surrounding forecast revisions. Yet the security analysts’ recent performance does not reflect the accuracy required to justify such a role in the market. According to the Editorial Staff of Investor Relations Business, over any four quarters analyst earnings estimates are off by an average of 40 percent for technology companies (July 2000). The website Investars.com, with the motto “Making Wall Street Transparent,” calculated the returns investors would have achieved had they followed each Wall Street firm’s investment advice since January 1997. They develop a strategy to invest funds in a stock based on the level of its initial recommendation. For instance, a “strong buy” merits a $450,000 investment versus $300,000 for a “buy.” Over time the portfolio is adjusted for revisions, adding 50% of current value for an upgrade and selling 33% of a position to respond to a downgrade. As of June 4, 2001, the returns ranged from 5.79 percent for Credit Suisse First Boston to a dismal –7.85 percent for Lehman Brothers, versus the 71.93 percent return of the S&P 500 Index, 71.70 percent of the Dow Jones Industrial Average, and 68.34 percent of the NASDAQ. Congressman Richard Baker, who chairs Congress’s capital markets sub-committee, found that in June 2000 analysts recommended that investors sell only 0.8 percent of covered American companies. At that time, nearly 74 percent were rated either “strong buy” or “buy,” verses a buy-sell split of about 50-50 in 1992 (Martinson 2001).
Moreover, with the institution of Regulation FD in October 2000, security analysts no longer have access to private information. When announcing the rule in August 2000, SEC Chairman Arthur Levitt stated, ‘Today, as Wall Street analysts play an increasingly visible role in recommending stocks, some in corporate management treat material information as a commodity– a way to gain and maintain favor with particular analysts. What’s more, as analysts become more and more dependent on the "inside word," the pressure to report favorably on a company has grown even greater, as analysts seek to protect and guarantee future access to selectively disclosed information… Simply put, these practices defy the principles of integrity and fairness… The adoption of today’s rules – appropriately named Regulation FD, for Fair Disclosure – would mean that when companies disclose material, nonpublic information, they must disseminate this information broadly. (Levitt 2000).
Beard (2001) reports that in 1998, 27 percent of companies opened their earnings announcement conference calls to individual investors, compared to 90 percent in 2001. Market conditions have become more favorable for efficiency because all investors have access to previously exclusive information. With numerous “Strong Buy” recommendations of bankrupt companies permanently on the record and Regulation FD in action, the question of analyst forecast rationality has become even more important and controversial.
Beyond battering by the financial press, security analysts and their firms are also subject to lawsuits filed by soured investors and disapproving government regulators. For instance, superstar analyst Henry Blodget and his former firm Merrill Lynch stood “accused of ‘systematic fraud … on an industry-wide basis’ for touting shares in internet stocks” without revealing the extent of the relationship between the Internet companies and Merrill’s investment banking division (Martinson 2001). The New York Attorney General has launched a full-scale investigation of Merrill Lynch research practices and relationship to its investment banking arm, and the ramifications of the case will affect the way Wall Street conducts business. My research will not directly address this influence.
Rather, recognizing that security analysts are in trouble because their recommendations influence investors, I will investigate the extent to which they departform rational forecasts, or to which they are liable. The influences faced by these professionals are innumerable, as are the influences on stock prices. However, one core influence is earnings expectations, which therefore provides a test for analyst rationality. The earnings expectations of the market are reflected in stock price, so analyst earnings forecasts can be compared to stock prices in order to gauge rationality. Their relationship to stock prices provides insight into the efficient market theory because the extent to which prices predict forecast errors informs us of the theory’s accuracy. My research is based on this premise.
My study of analyst rationality will measure how analysts use the information content of prices when making forecast revisions. If they do not fully use prices, three explanations arise. Either investors are poorly served, stock prices do not reflect all available information so the market is not efficient, or analysts have exclusive information not yet factored into prices. In fact, I find that the first case is true. Section 2 summarizes the related research that motivates my hypotheses, leading to the description of my study design and data in Section 3. Analysis of the Revision Regression and the Forecast Error Regression is presented in Sections 4 and 5, and Section 6 will state my conclusions.
Conclusion
The relationships between analyst earnings forecasts, realized earnings, and stock prices are important to both economic theory and financial market reality. On a theoretical level, an accurate model of capital markets will provide crucial insight into all economic markets. On a practical level, these relationships inform individual investors about vital investment decisions and government regulators about how to protect them. To study these issues, I compiled 25,621 Dow data points and 4,002 Internet data points covering forecasts made for the first quarter of 1996 through the first quarter of 2001. I found that the earnings forecasts in the Dow sample have a mean revision of –1.87 percent. By contrast, the Internet mean revision is 4.61 percent. In 1998, the minimum annual mean revision of –4.16 percent occurred for the Dow sample, while the maximum annual mean revision of 35.44 percent occurred for the Internet sample. The average error for the Dow sample is a 10.77 percent overestimate, whereas the Internet mean error is an overestimate of 7.46 percent. The maximum annual mean error of 22.34 percent for the Dow occurred in 2000. In 1998, the maximum annual mean error of 83.38 percent occurred for the Internet sample.
The Revision Regression measured the relationship between forecast revisions and prior stock returns. The results show that the Dow forecasts always significantly follow prices. By contrast, the Internet coefficients on price are only significant in two years, 1998 and 1999, suggesting that the link between recommendations and price was not existent in other years. The significant results show that in 1998 the coefficient of stock return on revision was 1.089 (p = 1 percent), and in 1999 it was –0.505 (p = 1 percent). Therefore, it appears that Internet analysts followed prices in 1998, the year of the highest mean revision and error, then disregarded them in 1999. The Dow analysts also follow prices more and more as announcement date approaches, whereas Internet analysts disregard them from three months to one year before the announcement. The most extreme revisions were also less likely to follow prices. Also, analysts do not consistently follow or ignore market return between their forecasts. This regression does not reveal if prices are rationally linked to earnings, so we cannot deduce whether or not analysts should have followed them. The next test addresses this point.
The Forecast Error Regression related analyst error to stock returns before, during, and after the analyst’s announcement. I find that for the Dow sample, overestimates (underestimates) correspond to falling (rising) prices preceding an analyst’s revision until 2000. Similarly, in the Internet sample, each year the errors steadily become more strongly related to observable price changes until 2000. In this year, Dow overestimates in the sample relate to rising prices, suggesting that prices strayed from their efficient levels, and the Internet prices explain error the least, meaning that analysts incorporated them or prices were less informative. The 2000 anomalies suggest the Regulation FD affected companies differently. Overall, I find evidence of both rational and irrational Internet prices. The relationship between error and return on announcement day is not sufficiently developed or significant in my study.
Finally, the Forecast Horizon coefficients for the Dow show that in all years but 2001, if a forecast turns out to be too optimistic (pessimistic), then the stock price has fallen (risen) since the announcement day. However, in 1997 and 1999 for the Internet companies, an overestimate is more likely with a rising stock price after the forecast was made. I infer that this suggests that Internet stock prices were inefficient in these years, but I am puzzled that this was not true for all years. For instance, 1998 marked the year of largest error as well as the largest correlation between revision and price, indicating inefficiency. For Dow stocks, the size and direction of revisions do not affect the relationship between error and price. For Internet stocks, however, that the larger the revision, the less the error relates to price. Also, for the Dow analysts as returns prior to forecast announcements become more positive, the relationship between error and price strengthens. The opposite applies to the Internet. This means that higher returns for these stocks contain less information, attesting to the Internet stock bubble. Finally, I compare the relationship between error and price movements before and after an announcement date. I find that half of the annual errors are more a result of subsequent price changes than of price changes observable during the revision period. I cannot explain the seemingly random split.
Overall, I find that analysts do not fully incorporate relevant price information about earnings into their forecast revisions, supporting that a semi-strong form of market efficiency exists. Analysts do not generally follow changes in earnings potential implicit in changes, and their actions do not correct market expectations, because these price changes actually predict analyst errors. I confirm that analyst and price irrationality was particularly acute for the Internet set in 1998 and 1999, although the results from both sets demonstrate rationality as well as irrationality. The main weakness of my results is that they do not incorporate a rational model the earnings-price relationship. In other words, the correct weighting that an analyst should place on the price of a Dow or an Internet company is unclear, one of the several murky issues in my analysis. Future extensions of my research should investigate a more powerful econometric test to link price rationality to analyst and investor behavior. As more evidence is uncovered regarding influences faced by analysts, pieces of this puzzle may be solved.
No comments:
Post a Comment