Evaluation of Mixed-Frequency Regressions in Forecasting Seasonal Inflation Rate of Iran

This paper evaluates the predictive ability of mixed-frequency autoregressive models in forecasting seasonal inflation rate of IRAN economy. For this, forecasting accuracy of models with monthly lags of inflation rate are compared with a benchmark model which uses seasonal lags. Results indicate incorporating monthly lags, instead of seasonal lags, improves the accuracy of seasonal forecasts, especially for 1-step ahead forecasts. Among mixed-frequency models, MIDAS regressions are more accurate than the benchmark model in 1-step, 3-step and 4-step ahead forecasts. Step-weighting model which features proliferation of parameters outperforms the MIDAS regression which is non-linear and parameter efficient in estimation.