An Empirical Study Of Financial Analysts Earnings Forecast Accuracy Andrew Stotz

Over the past 12 years, financial analysts across the world have been optimistically wrong with their 12-month earnings forecasts by 25.3%. This study may be the first of its kind to assess analyst earnings forecast accuracy at all listed companies across the globe, covering 70 countries.A review of prior research shows little uniformity in the preparation of the data set, yet differences in how outliers are treated, for example, can create substantially different results. This research lays out six specific steps to prepare the data set before any analysis is done.Three main conclusions come from this research:First, analyst earnings forecasts globally were 25.3% optimistically wrong, meaning on average, analysts started each year forecasting company profits of US$125, but 12 months later the company they were forecasting actually reports profits of US$100. Second, analysts had a harder time forecasting earnings for companies in emerging markets, where they were 35% optimistically wrong. Third, that analyst optimism mainly occurred when the companies they forecasted experienced very low levels of actual earnings growth, analysts did not make an equal, but opposite error for fast growth companies.The uniqueness of this research is most likely the first that it is global-including all analysts, covering all stocks, across all countries. Furthermore, it is not US-centric. Second, it covers the complete population of data, not a sample or an unrealistically small number of companies. Third, it is long term in nature, covering a total of 12 years from 2003 through 2014, a time period that spans more than just one portion of the business cycle. Fourth, this research includes China. This is one of the first papers on analyst earnings forecast accuracy to fully incorporate all Chinese companies that are large and liquid. Fifth, this research is based on the mean earnings forecasts of all analysts, or what is commonly referred to as the "consensus" forecasts, not on the performance of individual analysts. Sixth, this research is comparable. Unlike most prior research in this area, this research scales earnings forecast error by actual earnings, rather than by share price, hence it does not distort results by including the highly variable, random share price factor into the equation. This makes the results of this research comparable across time, geography, sectors, and both developed and emerging markets. Finally, this research is actionable. This study focuses on a 12-month, consensus forecast, rather than a much shorter-term forecast time horizon, something that investors can actually follow and profit from.After properly preparing the data set, the methodology employed to test the significance of various factors was group compare. The factor focused on, which has not previously been tested, was the level of actual earnings that a company reported. This research grouped companies into four groups based upon the actual earnings they produced each year:very fast, moderately fast, moderately slow, and very slow. Though a regression was performed to test the association between analyst forecast accuracy and the group that the company fell into, such a test was less reliable since the main measure tested, whether the actual earnings of a company were high or low, was also a component of the dependent variable, scaled forecast error. In addition, this research clearly shows that the distribution is skewed, rather than a normal distribution, making regression less applicable. Instead of regression this research used a group compare methodology and showed that almost all of the analyst earnings optimism appears at companies which exhibited very slow growth. And that analysts were not equally inaccurate when a company had very fast earnings growth.From this work a few areas for further research stand out. This research showed that as the number of analysts increased, earnings forecast accuracy improved. However, at a level of coverage above 30 analysts, accuracy actually worsened. It would be interesting to ascertain the source of this difference.Of the Emerging Countries South Korea, China and Brazil are all highly skewed; in fact, the level of skewness in South Korea is above all others and provides for an excellent area of further research.A further question to answer in future research is whether analysts are more successful during certain periods of market movements or of the earnings cycle.Lastly, is to consider research on whether a profitable trading strategy could be adopted from this deeper understanding of analyst earnings forecast error.