Quant BuffetRelax, Not Over Thinking

Machine learning Analysts’ Sentiment Industry Factor

Log in to collect

Academic paper

Artificially Intelligent Analyst Sentiment and Aggregate Market Behavior

AuthorsVidhi Chhaochharia; Alok Kumar; Ville Rantala; Alan Zhang

Institute
  • University of Miami
  • ?University of Miami - Department of Finance
  • ?University of Miami - Miami Herbert Business School
  • Florida International University
  • ?Florida International University (FIU)

Strategy in a nutshell

The strategy leverages analyst earnings forecasts and historical firm-level data to predict forecast errors using neural networks (NN). The dataset combines I/B/E/S quarterly forecasts, CRSP stock data, and Compustat firm attributes. A two-layer feed-forward NN with 15 hidden neurons is trained for each analyst to predict firm-level forecast errors based on their lagged errors. Predictions are aggregated to firm, industry, and market levels, and compared against linear regression benchmarks.

Using these predictions, the study decomposes market-level forecast errors into a predictable

Economic rationale

The strategy works because it captures systematic patterns and biases in analysts’ forecasting behavior that traditional linear models cannot. By training neural networks at the individual analyst level, the model learns both linear and non-linear relationships in past forecast errors, including analyst-specific

Backtest performance

Annualised return8.21%
Volatility15.97%
Sharpe ratio0.51