Bias Tracker

This website posts data on a non-distorted/unbiased benchmark for measuring systematic expectational errors in survey forecasts of four-quarter-ahead inflation and GDP growth. The methodology is based on the machine learning algorithm and general approach of Bianchi, Ludvigson, and Ma (2022) (BLM) using the Long Short-Term Memory recurrent neural network estimator. The machine-efficient forecast series posted below is the appropriate benchmark for evaluating systematic expectational errors of a survey respondent whose response is at or close to the median forecast in the Survey of Professional Forecasters (SPF) panel. These data are updated versions of the machine forecasts from the published paper BLM. The figure below plots the most recent time-series of the bias in the SPF median forecasts for four-quarter-ahead inflation and GDP growth, defined as the difference between the survey response and the ex ante machine forecast. Bias is thus computed as a genuine ex ante expectational error, and not via a comparison with ex post historical outcome data that could only have been known with hindsight.

Updated machine-efficient forecasts, download here; Vintages here.