Tracking Economic Activity With Alternative High-Frequency Data

Monthly macroeconomic series captured the sharp fluctuations during the COVID-19 pandemic only with a lag. The use of alternative high-frequency data is promising for crisis periods, but it is difficult to extract relevant business cycle information from them. We present a Bayesian mixed-frequency dynamic factor model with stochastic volatility for measuring GDP growth at high-frequency intervals. Its novelty is an additional state-space block, in which the sparse observations in the mixed-frequency data are augmented to a balanced panel with observed and estimated latent information. The dynamic factors are then estimated conditional on the augmented data. Our model exploits the information in rich datasets of weekly, monthly, and quarterly series, including alternative high-frequency data. GDP is nowcasted timely and accurately during volatile periods.