This article was originally published in the Fall 2012 edition of OnAnalytics, published by the Institute for Business Analytics at Indiana University’s Kelley School of Business.
This article focuses on insights from James M. Whalen, the chairperson of the Kelley School of Business department of Accounting. Find more information about this research in the article, “Can financial statement analysis beat consensus analysts’ recommendations?”
Analysts’ consensus recommendations are a tool used by investors, but do they really reflect the full information available from financial statement reports? With this study, Jim Wahlen and Matt Wieland developed and tested a rating system that predicts future changes in earnings based on financial statement information. When these future earnings predictions are used to design an investment strategy, the researchers’ stock portfolios significantly outperformed portfolios formed using consensus recommendations, suggesting that consensus recommendations fail to incorporate information that could more reliably predict future earnings changes and identify under- or overpriced stocks.
Statement of the problem
Consensus recommendations from financial analysts do not follow a normal distribution but instead skew toward “strong buy,” “buy,” and “hold” categories, with only a tiny percentage of stocks identified as “underperform” or “sell.” Additionally, the “hold” category is amorphous; it could be seen as implying neutrality, ambivalence, uncertainty, or a veiled low opinion of the stock’s potential. These observations and prior research indicate that consensus recommendations are not reliably informative and may even be misleading. A more useful predictive tool would transparently process information from financial statements to offer a reliable indication of future changes in earnings, which in turn provide a fundamental basis on which to identify stocks to buy or sell.
Data Sources Used
The researchers gathered consensus recommendations for firm-years 1994 through 2005 from the Institutional Brokers Estimate System (I/B/E/S). They also collected financial statement information for the same period from Compustat, and stock prices, returns, and market capitalization data from the Center for Research in Security Prices (CRSP). The sample, containing intersections from the three sources, consisted of 25,168 firm-year observations.
Abnormal returns for each firm-year were calculated by compounding monthly raw returns over a one-year holding period and then subtracting the compounded returns on the CRSP sizebased decile portfolio. The researchers calculated descriptive statistics on 1-year ahead cumulative and abnormal returns, financial statement analysis variables, market multiples, stock returns, and analysts’ earning forecasts across all consensus recommendation levels.
To predict future earnings increases, the researchers used a previously developed six-signal scoring model known as the predicted earnings increase score (PEIS). Firms were ranked into quintiles for each of the six criteria:
- Return on net operating assets (RNOA). Extreme returns tend to mean revert over time; a firm generating extremely high RNOA will likely encounter competition while extremely low RNOA must implement improvements in profit margins and operating efficiency.
- Gross margin signal (GM). The growth of gross margins relative to growth in sales reflects the firm’s control of production costs relative to product prices.
- Selling, general, and administration signal (SG&A). This signal reflects changes in operating costs relative to sales. In the presence of sales growth, an increase in the SG&A percentage indicates poor control of overhead whereas a decrease implies cost control and operating leverage. If sales decline, the signals reverse, with an SG&A percentage increase indicating managerial optimism and a decrease suggesting that managers are pessimistic about future earnings.
- Asset turnover ratio (ATO). Changes in the efficiency of the firm’s assets tend to predict the direction of future profitability.
- Growth in net operating assets (GNOA). GNOA interacts negatively with RNOA; low GNOA relative to RNOA indicates increasing efficiency while high GNOA relative to RNOA implies inefficient buildup of operating assets.
- Accruals (ACC). Measured as the difference between operating income and cash flow while controlling for RNOA, accrual levels are negatively associated with future earnings predictions.
For each of the six criteria, firms in the lowest and highest quintiles received a score of -1 or +1 according to the direction of the prediction, while firms in the middle three quintiles received scores of 0. The researchers summed the scores of all six signals for each firm-year to compute PEIS. Firms received scores between -6 and +6, with higher scores indicating a likelihood of future earnings increases.
The researchers computed descriptive statistics of 1-year ahead RNOA for the PEIS quintiles for the full sample and each consensus recommendation level. Next, the researchers tested the abnormal returns of portfolios formed based on PEIS. Their primary trading strategy, labeled the fundamental strategy, took equally weighted long positions on firms in the highest PEIS quintiles and short positions on firms in the lowest PEIS quintiles. Their benchmark strategy, labeled the buy/sell strategy, captured the effects of following consensus recommendations by taking long positions on the strong buy and buy recommendations and short positions on the sell recommendations. Within the stocks labeled “hold” in the consensus recommendations, the researchers tested a conditional hold strategy taking long (short) positions on the highest (lowest) PEIS quintile, and compared it with an unconditional hold strategy taking long positions for all “hold” stocks. The researchers also implemented a regression equation to control for the annual effects of market-to-book ratio, earnings yield, and beta.
The descriptive statistics revealed that the mean recommendation value is 2 (buy), indicating that consensus analysts tend to advise investors to purchase shares of the firms they follow. The mean (median) size-adjusted abnormal return generated by the full sample was 2.1% (-7.9%). Moreover, 1-year ahead mean and median cumulative and abnormal returns tended to vary inversely with recommendation levels, suggesting that consensus recommendations were not reliable indicators of future stock return performance.
After computing the PEIS scores, the researchers observed striking differences in the financial performance measures between firms in the top and bottom quintiles, with firms in the lowest quintile exhibiting 0.4% RNOA and almost no sales growth and those in the highest quintile exhibiting 29.6% RNOA and nearly 40% sales growth. 63.9% of firms in the highest quintile reported an earnings increase compared to only 51.3% of firms in the lowest quintile.
The scoring system was most effective in predicting performance within the “hold” category, with 66.3% of firm-years in the highest PEIS quintile generating an earnings increase the following year in contrast to only 42.5% of those in the lowest quintile. Applying PEIS with the fundamental strategy yielded positive changes in the abnormal returns across each of the consensus recommendation categories. Within the hold category, conditioning on PEIS increased abnormal returns from 6.8% to 22.4%.
For the full sample, the fundamental strategy using PEIS significantly outperformed the buy/sell strategy, with average annual abnormal returns of 9.8% using PEIS in contrast to -9.0% following consensus recommendations. Use of the regression showed that after controlling for other predictors of future returns, the buy/ sell strategy yielded positive abnormal returns in 8 out of 12 years with an average abnormal return of 0.9%, while the fundamental strategy yielded positive abnormal returns in all but one year with an average abnormal return of 10.9%.
The unconditional hold strategy yielded an 8.5% average abnormal return, which was higher than expected given the ambiguity of the recommendation. The conditional hold strategy, however, yielded an average abnormal return of 19.3% and positive abnormal returns in all 12 years, outperforming the unconditional hold strategy by an average of 10.8 percentage points per year.
The strong performance of the PEIS-reliant strategies as compared to consensus recommendations suggests that analysts are failing to fully incorporate the information available in financial statements into their recommendations.
The results of this study could be useful to analysts in improving their strategies for forming recommendations. PEIS could also be used directly by investors to improve their returns over a strategy of following consensus recommendations. Another potential application is to use the conditional hold strategy to fill in the information gap created by the unclear “hold” category.
By testing different means of predicting earnings increases, this study not only demonstrates ways to apply financial statement data to investment strategies but also challenges assumptions about the advice currently provided to investors.