Modelling and Forecasting Seasonality in Indian Macroeconomic Time Series

This paper models the univariate dynamics of seasonally unadjusted quarterly macroeconomic time series for the Indian economy including industrial production, money supply (broad and narrow measures) and consumer price index. The seasonal integration-cointegration and the periodic models are employed. The `best' model is selected on the basis of a battery of econometric tests including comparison of out-of-sample forecast performance. The results suggest that a periodically integrated process with one unit root best captures the movements in industrial production. The other variables do not exhibit periodically varying dynamics, though narrow money and consumer price index exhibit nonstationary seasonality. For the index of industrial production, the periodic model yields the best out-of-sample forecasts, while for broad money, the model in first differences performs best. On the other hand, for narrow money and the consumer price index, incorporating nonstationary seasonality does not lead to significant gains in forecast accuracy. Finally, we find significant conditional heteroskedasticity in industrial production, with error variance in the first two quarters (highest and lowest economic activity quarters, respectively) almost three times that in the other two quarters.