From the Michigan Consumer Survey and Professional Forecaster Survey:
figure 1: Michigan’s 12-month ahead forecast error (blue) and the median forecast error of the Professional Forecaster Survey (red), in percentage points. The NBER uses shades of grey to define the peak and trough dates of the recession. Sources: Census Bureau via FRED, University of Michigan via FRED and Philadelphia Fed, NBER, and author’s calculations.
Since these are forecasts 12 months ahead, sampled at a higher frequency than the horizon—monthly (for Michigan) or quarterly (for SPF)—these errors are based on overlapping forecasts.
To see non-overlapping forecasts, I sampled each year in May:
figure 2: Michigan May forecast error 12 months ahead (blue) and the Professional Forecaster Survey median forecast error (red), expressed in percentage points. The NBER uses shades of grey to define the peak and trough dates of the recession. Source: BLS via FRED, University of Michigan by Fred and Philadelphia Fed SPFNBER, and the authors’ calculations.
From 1986M05 to 2019M05, SPF predictions show smaller errors than Michigan (18 bps vs. 43 bps) and smaller RMSE (12 bps vs. 15 bps). (See more discussion of professional economist views and different characteristics of households, see this postal.) Apparently, this pattern has reversed since the start of the pandemic – although there are only 3 non-overlapping observations, making it difficult to judge statistical significance.
All in all, professional forecasters did better (and displayed less upward bias) than consumers/households during the Great Moderation. During a pandemic, that relationship could be disrupted. However, in both cases, the prediction errors are highly correlated (adjusted R2 at 0.68).
