Masked macroeconomic nowcasting

Nowcasting consists of estimating a low-frequency unobserved target by tracking the flow of timely available high-frequency input features. The essential characteristics of a data set for macroeconomic nowcasting are differing sampling frequencies and the ragged-edge, i.e. the pattern of predictor availability due to non-synchronicity of data releases. The mask mirrors the ragged-edge data availability pattern at a specific nowcast occasion. During a sequence of nowcast occasions, models are traditionally re-estimated on an expanding data set as more and more data becomes available. We propose to first mask the historical data and re-estimate the parameters based on the masked historical data. The re-estimation on dynamically masked data at each consecutive nowcast juncture thus tailors the model to the specific data availability patterns. We analyze the parameter restrictions induced by the masking operator and show that nowcasts are only being revised to the extent that the incoming data release is different from its model-consistent value. Finally, an empirical exercise of nowcasting US GDP shows the performance of the masking operator to handle missing observations.