The Editors selected papers on the basis of both the importance of the contribution to the area of model uncertainty and macroeconomics, and the themes of the workshop. Each of the papers contained here has undergone the usual rigorous Journal of Applied Econometrics refereeing process.
The workshop and this Special Issue focus on arguably the most exciting area of macroeconomic research: the role of model uncertainty in empirical macroeconomics. This literature treats the 'true' model as an unobservable; this admission has implications for many areas of macroeconomic analysis and has generated several distinct research programs. One program represents a renewed interest in model comparison and selection. A second program is based on accounting for model uncertainty explicitly in the construction of statistical inferences and in policy evaluation; this can be both frequentist and Bayesian. A third program focuses on the evaluation of models when model misspecification is explicitly recognized.
The first part of the Special Issue focuses on issues of model evaluation and combination in the presence of model misspecification, with particular attention paid to model selection and forecasting. The paper by Todd Clark and Michael McCracken examines the forecasting performance of vector autoregression (VAR) models using real-time US data for US output, prices, and interest rates in the presence of model instabilities which cause unknown misspecification errors. This paper shows that simple point forecast combinations from VAR models perform well, making the case for forecast combination as an effective hedge against the risk of model instability.
Oleg Korenok, Stanislav Radchenko and Norman Swanson advocate a comprehensive approach to model comparisons as a response to model uncertainty. They consider a number of popular sticky price and sticky information models, and study how well they capture the sample dynamics of inflation across 13 OECD countries. Drawing conclusions from a number of measures of inflationary pressure, and samples, the authors show that a hybrid model performs