Please rotate your device to landscape mode to view the charts.

Man versus Machine Learning: The Term Structure of Earnings Expectations and Conditional Biases

Journal: Review of Financial Studies

Date: 20230601

Author: van Binsbergen, Jules H.; Han, Xiao; Lopez-Lira, Alejandro

Abstract:
We introduce a real-time measure of conditional biases to firms' earnings forecasts. The measure is defined as the difference between analysts' expectations and a statistically optimal unbiased machine-learning benchmark. Analysts' conditional expectations are, on average, biased upward, a bias that increases in the forecast horizon. These biases are associated with negative cross-sectional return predictability, and the short legs of many anomalies contain firms with excessively optimistic earnings forecasts. Further, managers of companies with the greatest upward-biased earnings forecasts are more likely to issue stocks. Commonly used linear earnings models do not work out-of-sample and are inferior to those analysts provide.

Link: Google Scholar

Key Findings

Superior Machine Learning Forecasts

A novel machine learning algorithm produces more accurate earnings forecasts than traditional analyst predictions, with forecasting errors near zero compared to consistent analyst overoptimism.

Return Predictability from Forecast Bias

Stocks with overly optimistic analyst forecasts relative to machine learning predictions earn significantly lower future returns, generating -1.46% monthly return spread.

Market Timing by Managers

Companies with the most upward-biased analyst forecasts issue significantly more stocks, suggesting managers capitalize on overoptimistic market expectations.

Forecast Accuracy Comparison

  • Machine learning forecasts show minimal bias across all horizons
  • Analyst forecast bias increases with forecast horizon
  • Two-year ahead analyst forecasts show largest bias of 0.384

Portfolio Returns Based on Forecast Bias

  • Monthly returns decrease monotonically with forecast bias
  • Lowest bias quintile earns 1.32% monthly return
  • Highest bias quintile earns -0.14% monthly return

Net Stock Issuance by Forecast Bias

  • Stock issuance increases with analyst forecast bias
  • Highest bias quintile issues 6.5% more shares annually
  • Relationship remains significant after controlling for firm characteristics

Contribution and Implications

  • First study to develop a real-time measure of analyst forecast bias using machine learning
  • Demonstrates that machine learning can improve upon analyst forecasts by optimally combining public information
  • Provides evidence that managers actively exploit upward-biased market expectations through stock issuance
  • Results suggest significant market inefficiency in processing earnings expectations

Data Sources

  • Forecast accuracy comparison based on Table 2 showing term structure of earnings forecasts
  • Portfolio returns visualization derived from Table 5 showing returns on portfolios sorted by conditional bias
  • Stock issuance chart constructed using data from Table 9 Panel A showing net stock issuances by bias quintile