Via the semiannual Livingston Survey, the Federal Reserve Bank of Philadelphia solicits forecasts for the S&P 500 index (and many other U.S. economic measures) from economists in industry, government, banking and academia. How good are their forecasts? In his June 2007 paper entitled “Predicting Stock Price Movements: Regressions versus Economists”, Paul Soderlind examines the aggregate stock return forecasting ability of surveyed experts. Using median forecasts for stock market gains during the interval 6-12 months after survey dates and associated actual data for 1952-2005, he concludes that:
- The Livingston forecasters appear to reject a random walk in stock prices, believing that they can predict stock returns based mostly on medium-term mean reversion. Their capital gain expectations tend to peak during recessions.
- In out-of-sample tests, the median survey forecasts underperform naive forecasts (the historical mean) and the most common single-variable prediction equations.
- The Livingston forecasts act like “too large” prediction models that overfit past data for a group of traditional indicators and overemphasize recent data.
The following chart, taken from the paper, plots expected excess capital gains for the U.S. stock market based on 1952-2005 Livingston survey forecasts (Survey) and actual gains for the same intervals (Ex post). Forecasts are for capital gains in excess of a risk-free rate for the six-month interval beginning approximately six months after the survey dates. The data is for the S&P Industrials up to June 1990, and the S&P 500 index thereafter. Actual data is much more volatile than the forecasts.
In summary, professional economists lose to simple models in forecasting stock market returns.
