The Additional Information Content of Survey Data. Evidence from Second Moments

Disagreement across forecasters is a well-established fact as widespread is the use of forecasting errors as a proxy of uncertainty. Taking stock of these facts, we aim to test whether the level of disagreement in survey-declared beliefs contains meaningful independent information on the predictive fitness (in terms of squared forecasting errors) of standard econometric models and/or vice versa. Interpreting the estimated squared errors as a proxy of uncertainty we can also shed light on the relationships between the dispersion in survey expectations and macroeconomic volatility. By examining GDP dynamics in the UK we find that the level of discord between expectations revealed in surveys helps improve the predictive power of econometric models and not vice versa. Moreover, the sign of the parameters shows that the greater the entropy in the survey, the higher the uncertainty in the market. On one hand these findings complement previous works, usually focusing
on first moments, on the significant one-way information flow running from survey to
macroeconomic data. On the other they are in line with - and enlarge our knowledge about - the links between heterogeneity in survey beliefs and aggregate volatility. Results are based on real time data and are robust both to several indicators of dispersion in survey beliefs and to a battery of predictors.