Often the unintentionally biased forecasts by Analysts can prove detrimental for the stock market.

In the stock market, forecasts are key to investments. If the forecasts go wrong, then the whole outcome becomes detrimental. Companies lost money, and the global economy becomes shabby. Remember the 1929 stock market crash? Wrong predictions led to the loss of billions of dollars. In fact, some famous personalities like the then Prime Minister of UK Winston Churchill became lost a major share in that crash.

Fortunately, humans haven't encountered any such incidents again. But such incidents are precedented. While forecasting analysts are often rendered to make unintentionally biased predictions, due to the high expectations of the firms and the approaching deadlines. As analysts forecasts are heavily dependent on the public information available, often limited or redundant information ensues the wrong prediction.

To thwart such incidents, the researchers at The Wharton Business School in collaboration with the University of Edinburgh and Norwegian Business School have formulated a model that can optimize the unbiased benchmark for firms' earnings expectations. In the paper titled, "Man v/s Machine Learning: The Term Structure of Earnings Expectations and Conditional Biases", the researchers explain that the analyst expectations are on average biased upwards and that this bias exhibits substantial time-series and cross-sectional variation. On average, the bias increases in the forecast horizon, and analysts revise their expectations downwards as earnings announcement dates approach.

The researchers indicate that analysts' biases are associated with negative cross-sectional return predictability, and the many anomalies consist of firms for which the analysts' forecasts are excessively optimistic relative to the benchmark. The researchers have used the cross-sectional information of firms' balance sheets, macroeconomic variables, and analysts' predictions, to provide conditional expectations in real-time.

Random Forecast Regression

In the model, the researchers have utilized the Random forest regression approach for the principal analysis of the research. The paper cites that it has two significant advantages. First, it naturally allows for nonlinear relationships. Second, it is designed for high-dimensional data and is therefore robust to overfitting. The researchers have constructed one-year- and two-years-ahead forecasts for annual earnings, and one-quarter, two-quarters, and three-quarters-ahead forecasts for quarterly earnings. This approach is taken as the analyst's forecasts for other horizons' have fewer observations.

Through the benchmark expectations provided by the machine learning algorithms, the researchers were able to calculate the bias in expectations as the difference between the analysts' forecasts and the machine learning forecasts. The researchers were able to identify the windows of opportunity that it opens, and the continuous mistakes made by the analysts.

Through this process, the researchers were able to determine that analysts' biases increase in the forecast horizon. The paper states that on average, analysts revise their expectations downwards as the earnings announcement day approaches. These revisions induce negative cross-sectional stock predictability: stocks with more optimistic expectations earn lower subsequent returns. Moreover, managers of those companies with the largest biases seem to take apparent advantage of the over-optimistic expectations by issuing stocks.

Earnings Forecasts via Machine Learning

The researchers conclude that the earnings forecasts via machine learning are closer to the realized earnings as compared to the analysts' forecasts. Moreover, this new measure is useful not only as an input to asset-pricing exercises but also as an available real-time benchmark against which other forecasts can be compared. Additionally, this machine learning model enables profitable trading strategies.