Analyst Forecast Accuracy Weighting and Equity Alpha Factors
Summary
The report examines Chinese analyst earnings forecasts, ratings, and target prices as inputs to equity selection. It compares six ways to combine analyst estimates and finds that weighting forecasts by analysts’ past accuracy performs better than several alternatives; its accuracy is statistically similar to a commercial consensus method, while offering greater transparency. The report also describes which company and analyst characteristics are associated with more accurate forecasts, including company size and stability, analyst experience, and prior forecast accuracy.
Using accuracy-weighted consensus estimates, the authors construct earnings, valuation, rating, and implied-return signals. They report meaningful stock-selection results across tested universes and find that weighting each analyst’s revisions relative to their own previous forecast improves the information ratio versus changes in consensus earnings. Most signals retain significance after controlling for other factors, although forecast disagreement has limited coverage and loses significance in one test. The authors caution that analyst coverage is selective and recommend risk neutralization; they also flag model failure and extreme market conditions.
Key ideas
- Analyst coverage is associated with stronger future performance, but its signal should be risk neutralized because coverage is selective.
- Large, stable companies and forecasts from experienced, historically accurate analysts tend to have smaller forecast errors.
- Weighting estimates by analyst forecast accuracy is a transparent consensus method with accuracy comparable to a commercial alternative.
- Analyst-level forecast revisions weighted by past accuracy outperform simple changes in consensus earnings in the reported information-ratio comparison.
- Forecast disagreement has limited coverage, and all reported factor results remain subject to model and extreme-market risks.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.