Forecasting and Signal Extraction with Regularised Multivariate Direct Filter Approach

The paper studies regularised direct filter approach as a tool for high-dimensional filtering and real-time signal extraction. It is shown that the regularised filter is able to process high-dimensional data sets by controlling for effective degrees of freedom and that it is computationally fast. The paper illustrates the features of the filter by tracking the medium-to-long-run component in GDP growth for the euro area, including replication of Eurocoin-type behavior as well as producing more timely indicators. A further robustness check is performed on a less homogeneous dataset for Latvia. The resulting real-time indicators are found to track economic activity in a timely and robust manner. The regularised direct filter approach can thus be considered a promising tool for both concurrent estimation and forecasting using high-dimensional datasets and a decent alternative to the dynamic factor methodology.