Forecasting Inflation using Online Daily Prices : A MIDAS Approach for Brazil

Usually, inflation is optimally forecasted using simple time series models or a Phillips' curve process. However, as more people become online shoppers, ``online inflation'' turns out to be a good predictor of official inflation too. Online prices can be obtained at a higher frequency than inflation is released, so we can use contemporaneous online inflation to forecast offline price variation. In this work, we propose forecasting Brazilian inflation rate, released at monthly basis, using web-scrapped daily prices in a mixed-frequency approach (MIDAS models). By not aggregating online inflation to match official inflation frequency, we obtain better forecasts than the single frequency benchmarks (ARIMA, VAR, and Bridge Equation), at least for the short term. The results are improved in periods of stable inflation and robust to changes in the sample size, in the forecast horizon, in the data origin, in the benchmarks, and in the forecasts evaluation