The ability to predict future economic outcomes is a critical capability, and many people attempt to do so. Economists produce predictions of various economic variables, ranging from inflation to employment levels and gross domestic product (GDP) growth. The quality of these forecasts is often assessed in terms of their quantitative accuracy, but this is a difficult task. Quantitative accuracy, after all, is essentially a measure of the extent to which a prediction reflects an objective representation of the underlying economic process.
Unfortunately, the underlying economic processes are not easily observed or represented in the data. Even if one could accurately capture the process by observing the data, this would not necessarily result in accurate predictions. This is because the underlying economic process is non-linear and stochastic. As a consequence, the estimated values of economic variables are constantly revised and do not necessarily reflect a snapshot of reality. This explains why it is difficult to evaluate the accuracy of a single economic forecast.
Fortunately, there are other ways to assess the accuracy of economic forecasts. One such way is to study the differences between model-based and judgmental forecasts. The literature suggests that a combination of model and survey forecasts produces the most accurate forecasts, particularly at short-run horizons. Wieland and Wolters (2011), for example, use this approach to compare forecasts of U.S. GDP from the Blue Chip Indicators and the Federal Reserve’s Greenbook with those of a set of models. They find that, with the exception of the 1980-1981 recession, judgmental-based forecasts outperformed model-based ones.