Optimal forecast accuracy achieved with 5-10 economists

A new working paper examines the benefits of diversifying macroeconomic survey forecasts. It finds that optimal forecast accuracy is largely achieved with the inclusion of 5-10 professional forecasters, with greater gains observed for inflation than for growth predictions.

The power of diversified forecasts

The study, focusing on the U.S. Survey of Professional Forecasters (SPF), investigates how the accuracy of macroeconomic forecasts improves as more individual predictions are averaged.

Researchers developed 'crowd size signature plots' to summarize forecasting performance as a function of the number of forecasters included in the average.

A key finding is that the benefits of this diversification are largely exhausted once 5 to 10 representative forecasters are included in the average.

Beyond this threshold, the marginal gains in accuracy diminish sharply.

This pattern suggests that a relatively small, well-chosen group of experts can capture most of the 'wisdom of the crowd' effect in economic forecasting.

The paper also highlights a notable difference in diversification benefits between variables, with inflation forecasts showing considerably larger gains from averaging compared to growth forecasts.

This distinction is attributed to the underlying variance-covariance structure of individual forecast errors, which the study characterizes theoretically.

Equicorrelation and survey design

The research builds on the long-standing concept of the 'wisdom of crowds' in economics and finance, applying it to professional macroeconomic forecasts.

The authors propose comparing 'direct signature plots,' estimated from actual U.S. Survey of Professional Forecasters (SPF) data, with 'model-based signature plots,' derived from an equicorrelation model of forecast errors.

This approach allows for a robust characterization of diversification benefits.

The SPF, maintained by the Federal Reserve Bank of Philadelphia since 1990, typically includes about 40 participants and is crucial for guiding real-time policy and investment decisions.

The study's sample period spans from 1968Q4 to 2023Q2. The equicorrelation model fit for both growth and inflation forecast errors was found to be near-perfect, providing a strong theoretical underpinning for the observed empirical patterns.

Efficiency in expert consensus

This study offers valuable insights for central banks and other institutions relying on expert surveys.

It suggests that expanding the number of forecasters beyond a small core group yields diminishing returns, potentially optimizing resource allocation for survey design.

While the findings confirm the power of collective intelligence, they also underscore the importance of selecting a representative, rather than merely large, pool of experts.