Indeterminacy, Misspecification and Forecastability: Good Luck in Bad Policy?

A recent debate in the forecasting literature revolves around the inability of macroecono-metric models to improve on simple univariate predictors, since the onset of the so-called Great Moderation. This paper explores the consequences of equilibrium indeterminacy for quantitative forecasting through standard reduced form forecast models. Exploiting U.S. data on both the Great Moderation and the preceding era, we first present evidence that (i) higher (absolute) forecastability obtains in the former rather than the latter period for all models considered, and that (ii) the decline in volatility and persistence captured by a .nite-order VAR system across the two samples is not associated with inferior (absolute or relative) predictive accuracy. Then, using a small-scale New Keynesian monetary DSGE model as laboratory, we generate arti.cial datasets under either equilibrium regime and investigate numerically whether (relative) forecastability is improved in the presence of indeterminacy. It is argued that forecasting under indeterminacy with e.g. unrestricted VAR models entails misspecification issues that are generally more severe than those one typically faces under determinacy. Irrespective of the occurrence of non-fundamental (sunspot) noise, for certain values of the arbitrary parameters governing solution multiplicity, the pseudo out-of-sample VAR-based forecasts of in.ation and output growth can outperform simple univariate predictors. For other values of these parameters, by contrast, the opposite occurs. In general, it is not possible to establish a one-to-one relationship between indeterminacy and superior forecastability, even when sunspot shocks play no role in generating the data. Overall, our analysis points towards a 'good luck in bad policy' explanation of the (relative) higher forecastability of macroeconometric models prior to the Great Moderation period.