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

Background and Context

Research Problem

Analysts' earnings forecasts are commonly used for asset pricing but tend to be biased upward, creating a misalignment between forecasts and actual earnings.

Novel Approach

The study develops a machine learning algorithm using random forests to create statistically optimal and unbiased earnings forecasts as a benchmark.

Methodology

The research analyzes firm fundamentals, macroeconomic variables, and analyst forecasts from 1986-2019 using random forest regression to predict earnings.

Analysts' Forecasts Show Increasing Upward Bias with Longer Horizons

  • Shows how analysts' forecast bias increases with longer forecast horizons
  • Bias ranges from 0.028 for one-quarter-ahead to 0.384 for two-year-ahead forecasts
  • Demonstrates systematic optimism in analysts' longer-term predictions

Machine Learning Model Outperforms Analysts' Forecasts

  • Compares mean squared errors between machine learning and analyst forecasts
  • Machine learning model consistently shows lower prediction errors
  • Difference in accuracy becomes more pronounced at longer horizons

Returns Decrease with Higher Conditional Bias

  • Shows monthly returns for portfolios sorted by conditional bias
  • Clear negative relationship between bias and returns
  • Highest bias quintile shows negative returns of -0.14% monthly

Stock Issuance Increases with Analyst Forecast Bias

  • Demonstrates relationship between analyst bias and company stock issuance
  • Companies with highest bias issue 6.5% more stock
  • Suggests managers exploit optimistic market expectations

Anomaly Returns Increase with Conditional Bias

  • Shows relationship between conditional bias and anomaly returns
  • Anomaly returns increase monotonically with conditional bias
  • Highest bias quintile generates 2.13% monthly anomaly returns

Contribution and Implications

  • Provides first real-time measure of analyst forecast bias using machine learning
  • Demonstrates that analyst biases affect stock prices and corporate decisions
  • Shows managers actively exploit overoptimistic market expectations through stock issuance

Data Sources

  • Bias Chart: Constructed using data from Table 2 showing analyst forecast errors across different horizons
  • Error Chart: Based on mean squared errors reported in Table 2
  • Returns Chart: Created using portfolio returns data from Table 5
  • Issuance Chart: Constructed using net stock issuance data from Table 9 Panel A
  • Anomaly Chart: Based on long-short returns data from Table 8