Macroeconomic forecasting with mixed frequencies

The forecasting of macroeconomic time series frequently encounters the problem that the employed indicators and the target time series are often in different time frequencies. GDP, for example, is available only on a quarterly basis while most indicators, for example the Ifo Business Climate, appear monthly. This article examines two models in the more recent literature that make possible estimates with mixed frequencies without aggregation-conditioned losses: MIDAS models and VAR state space models. These procedures have the advantage that high-frequency information that is added within a period can be included in the forecast. By means of a case study for German GDP, the article shows that the new model classes supply more precise forecasts than the usual time-series models. The new procedures are used at the Ifo Institute for short-term forecasts.