Time series and forecasting · insight-0250

Set low expectations and strict out-of-sample tests for financial return forecasts

Situation
A model with a strong in-sample fit is proposed to forecast monthly equity returns for allocation decisions.
Action
Compare against the historical-mean forecast in a strict rolling out-of-sample test using only data available at each date, include transaction costs, and report how stable results are across subperiods.
Reason
A comprehensive study found that proposed equity-premium predictors performed poorly both in-sample and out-of-sample and were unstable.
Boundary
Asset returns, FX, crypto prices, and other competitive markets.

Symptom or risk

Return predictability is weak and unstable, so in-sample fit and a single backtest often overstate what a real-time investor could achieve.

Volumes, volatility and operational finance series (cash demand, call volumes) can be much more forecastable.

Strict testing rejects most ideas; that is the point.

Using revised macro data that were not available at the time.

Welch and Goyal found that proposed equity-premium predictors performed poorly both in-sample and out-of-sample over about 30 years and appeared unstable.
A Comprehensive Look at the Empirical Performance of Equity Premium Prediction