Combining survey long-run forecasts and nowcasts with BVAR forecasts using relative entropy

Abstract

This paper constructs hybrid forecasts that combine both short-and longterm conditioning information from external surveys with forecasts from a standard fi xed-coeffi cient vector autoregression (VAR) model. Specifi cally, we use relative entropy to tilt one-step ahead and long-horizon VAR forecasts to match the nowcast and long-horizon forecast from the Survey of Professional Forecasters. The results indicate meaningful gains in multi-horizon forecast accuracy relative to model forecasts that do not incorporate long-term survey conditions. The accuracy gains are achieved for a range of variables, including those that are not directly tilted but are affected through spillover effects from tilted variables. The forecast accuracy gains for infl ation are substantial, statistically signifi cant, and are competitive with the forecast accuracy from both time-varying VARs and univariate benchmarks. We view our proposal as an indirect approach to accommodating structural change and moving end points.

Structural scenario analysis with SVARs

“…systems and uncertainty about the paths of the conditioning variables . The use of conditional forecasts has increased dramatically after the 2008-09 financial crisis, with a variety of applications including assessing structural change (Aastveit et al, 2017;Giannone et al, 2019), evaluating of non-standard monetary policies (Giannone et al, 2012;Altavilla et al, 2016), nowcasting and forecasting (Giannone et al, 2014;Tallman and Zaman, 2018), and evaluating of macro-prudential policies (Conti et al, 2018). We stress that all of this recent wave of research is reduced-form in nature, conditioning exclusively on the path of observable variables, even when a structural interpretation is at times intended.…”

Does judgment improve macroeconomic density forecasts?

“…We do this across a range of forecast horizons. Exponential tilting has also been used to add external information, including judgment-based survey forecasts, to model-based forecasts by , Cogley et al (2005), Giacomini and Ragusa (2014), , Krüger et al (2017) and Tallman and Zaman (2020).…”

Real-Time Density Nowcasts of US Inflation: A Model-Combination Approach

“…Density nowcasts that are well calibrated have PITs that are uniformly distributed across observations t; in a large sample, the actual realizations would be expected to span the entire region of the density nowcast with a probability matching the probability implied by the density nowcast. Therefore, a visual assessment for calibration can be Interval forecasts are another popular metric to gauge the calibration of the density forecasts (e.g., Clark, 2011;Carriero, Clark, and Marcellino, 2015;Tallman and Zaman, 2020).…”

“…Building on the literature that finds that the accuracy of multistep point forecasts can be improved by conditioning on high-quality point nowcasts, Krüger, Clark, and Ravazzolo (2017) and Tallman and Zaman (2020) document that conditioning quarterly macroeconomic models with both nowcast means and nowcast densities leads to improvements in the accuracy of multistep point and density forecasts, especially for inflation. 3 Realizing these gains in practice requires relatively accurate nowcast means and nowcast densities for inflation.…”