This paper systematically studies the use of mixed-frequency data sets and suggests that the use of high frequency data in forecasting economic aggregates can improve forecast accuracy. The best way of using this information is to build a single model, for example, an ARMA model with missing observations, that relates data of all frequencies. The implementation of such an approach, however, poses serious practical problems in all but the simplest cases. As a feasible and consistent alternative, we propose a two-stage procedure to obtain pseudo high frequency data and to subsequently use these artificial values as proxies for macroeconomic or financial models. This alternative method yields a sub-optimal forecast in general but avoids the computational problems of a full-blown single model. Our approach differs from classical interpolation since we only use past and current information to get the pseudo series. A proxy, which is constructed by classical interpolation, may fit very well in sample, but it is not useful for out-of-sample forecasts. As applications of linking series generated at different frequencies, we show that the use of monthly proxies of GDP improves the predictability of absolute stock returns and the unemployment rate compared to the use of industrial production as an alternative proxy.
Linked ARMA Kalman Filter Interpolation Temporal Transformation High Frequency Data Virtual Reality Variable