MIDAS Modeling for Core Inflation Forecasting

Luis Libonatti

Ver

2017-12 - A forecasting exercise is presented to assess the predictive potential of a daily price index based on online prices, compiled by web scrapping by the private company PriceStats in cooperation with a finance research corporation, State Street Global Markets, as a predictor for a measure of the monthly core inflation rate in Argentina, known as “resto IPCBA” and published by the Statistics Office of the Government of the City of Buenos Aires. Mixed frequency regression models offer a convenient arrangement to accommodate variables sampled at different frequencies and hence many specifications are tested. Various classes of MIDAS models are found to produce a slight boost in terms of out-of-sample predictive performance at immediate horizons when compared to benchmark naïve models and estimators. Additionally, an analysis of intraperiod forecasts, reveals a slight trend towards increased forecast accuracy as the daily variable approaches a full month for certain horizons.