Surprise Revise

A recent article in the Financial Analysts Journal® (subscription may be required) by Feng Gu and Baruch Lev titled, “Time to Change Your Investment Model”, analyzes the effectiveness of predicting earnings surprises. That is, knowing beforehand and owning stocks that meet or beat consensus estimates and shorting those that miss. Their contention is that abnormal returns to such predictions have diminished. The authors present the case that Generally Accepted Accounting Principles (“GAAP”) measurements, particularly earnings, have become less useful to analyze individual companies. Consequently, earnings surprises are less informative than they used to be, evidenced by reduced abnormal gains. As we mention often, earnings surprises, and even more so, analyst revisions, are very important to us.

The article speaks to what we call “perfect foresight”. Perfect foresight is research into two behavioral anomalies – earnings surprises and analyst behavior. We believe that human behavior is not something that changes quickly. Responses among investment professionals are often predictable because of learned behaviors, reliance on heuristics, and slow adaptability. At the core of our investment philosophy is the belief that prices reflect changing expectations of investors, and those expectations are influenced greatly by perceived experts, including sell-side analysts. Based on our research, earnings estimate revisions remain the best way to capture those changing expectations, while earnings surprises have also yielded long-term benefits, but not as great as estimate revisions. Our research into the Russell 2000 universe reveals that perfect foresight into the most significant estimate revisions would have yielded 97% annually for over 20 years, while perfect foresight into earnings surprises would have yielded 44%.

Factor performance is something we monitor regularly as part of our risk management process. Behavioral events have been shown to be serially correlated, but if expectations of future events are elevated, returns can be muted. Our models identify the attractiveness of companies exhibiting these factors in relation to past performance and future expectations.

Returns to earnings surprises have not reduced the alpha of our main model, which is dominated by our analyst behavior factors. Earnings surprises are used as one subcomponent in our systematic investment process. We aim to have “tail exposure” to the factors we believe in and we filter out “noisy” earnings that either meet or are not significantly different from consensus. In our view, a stock’s price should not produce abnormal gains if earnings results come close to expectations. Our equity portfolios have an excess exposure to companies with large positive surprises, while being under exposed to large misses. Part of our risk control is avoiding large negative surprises.

Wall Street strategists have also suggested that responses to positive earnings surprises have been more subdued, but earnings misses are getting disproportionately punished. This could be a function of market levels with elevated expectations, or that consensus earnings estimates currently do not serve as a good proxy for investor expectations. Another theory is that large inflows into passive index products since the end of the financial crisis have limited the power of certain factors. As index buyers increase, there is less emphasis on individual stock characteristics.

This quick explanation is not meant to compare or refute the authors’ conclusions. We mention this article as a basis to explain how we think about these phenomena and the role research plays within a quantitative investment framework such as ours.

Read more about Tapping Investor Behavior to Source Alpha >