DP8604 Incorporating theoretical restrictions into forecasting by projection methods

Author(s): Raffaella Giacomini, Giuseppe Ragusa
Publication Date: October 2011
Keyword(s): Bayesian VAR, Euler conditions, Exponential tilting, Forecast comparisons
JEL(s): C53
Programme Areas: International Macroeconomics
Link to this Page: cepr.org/active/publications/discussion_papers/dp.php?dpno=8604

We propose a method for modifying a given density forecast in a way that incorporates the information contained in theory-based moment conditions. An example is "improving" the forecasts from atheoretical econometric models, such as factor models or Bayesian VARs, by ensuring that they satisfy theoretical restrictions given for example by Euler equations or Taylor rules. The method yields a new density (and thus point-) forecast which has a simple and convenient analytical expression and which by construction satisfies the theoretical restrictions. The method is flexible and can be used in the realistic situation in which economic theory does not specify a likelihood for the variables of interest, and thus cannot be readily used for forecasting.