Nowcasting inflation indices using mixed data sampling (MIDAS) time-series models

Abstract

This study’s principal objective is to estimate and analyse an inflation nowcasting model using alternative data with mixed frequencies via a novel application of the Mixed Data Sampling (MIDAS) time series model. Through empirical application, this dissertation seeks to demonstrate the effectiveness and reliability of the MIDAS approach, offering valuable insight to market practitioners such as policymakers, investment analysts, and researchers who rely on real-time analysis of macroeconomic data. Several time-series models were built to assess the robustness of the MIDAS regressions to nowcast two inflation indices, the CPI and the Core CPI, for the United States. Furthermore, the statistical significance of the parameters employed was used to indicate which high-frequency economic indicators and weighting functions were the most significant in nowcasting the inflation indices. Accordingly, by the employment of intra-month analysis, the robustness of the MIDAS regressions to nowcast inflation indices was evaluated and compared with another inflation nowcasting model with publicly available results. The study finds that the employment of a large and varied sample of high-frequency data in models using the MIDAS approach allows for superior robustness when nowcasting inflation growth rates, especially when compared with the traditional Bridge equations used in nowcasting applications such as the inflation nowcast by the Federal Reserve Bank of Cleveland. Furthermore, the study also finds that the parsimony and parametric flexibility offered by the MIDAS approach allows the researcher to easily optimise the model regularly to realign the selection of parameters with the previous month’s official estimate and recalibrating the lag-length and weighting gradient to be given to the high-frequency regressors. This dissertation directly addresses the gap in the MIDAS model’s application in nowcasting inflation indices in the United States. While there was scarce literature on inflation nowcasting, let alone the application of the MIDAS model in this regard, this dissertation suggests that this model enjoys considerable benefits which may be employed by market practitioners as a reliable inflation nowcasting model.