Agricultural performance of a country, generally, depends to a large extent on the quantum and distribution of rainfall. So its accurate forecasting is vital for planning and policy purposes. An attempt is made here for modelling and forecasting of Indian monsoon rainfall time-series data by using the promising nonparametric methodology of 'Wavelet analysis in frequency domain'. Maximal overlap discrete wavelet transform (MODWT) which, unlike discrete wavelet transform (DWT), does not require the number of data points to be a power of two is employed. Haar wavelet filter is used for computing the same in order to analyze the behavior of time-series data in terms of different times and scales. Wavelet methodology in frequency domain and Autoregressive integrated moving average (ARIMA) methodologies are applied for describing the data and for computing one-step ahead forecasts for hold-out data. Relevant computer programs are developed in SAS, Ver. 9.3 and R, Ver. 2.15.0 software packages and are appended as an Annexure. Comparative study of performance of the two methodologies is carried out from the viewpoint of one-step ahead forecasts on the basis of Root mean square prediction error (RMSPE), Mean absolute prediction error (MAPE) and Relative mean absolute prediction error (RMAPE). It is concluded that, for the data under consideration, Wavelet analysis in frequency domain approach is superior to ARIMA approach.
展开▼
