Probability analysis for forecasting basmati prices in Punjab: Application of advanced forecasting models

The current study focused on forecasting Basmati prices in Punjab, India, employing Trigonometric seasonality, Box-Cox transformation, ARMA errors, Trend, and Seasonal components (TBATS) model. The time series data, collected from 74 APMC markets, was aggregated into 281 monthly data points from January 2000 to May 2023. Five diagnostic tests, namely, Teraesvirta Neural Network, White test, Tsay’s test, and threshold test, were used to examine non-linearity, heteroscedastic behaviour of residuals, non-linear AR processes, and stationarity in the data. Six different models were fitted to the data, including the time series linear model (TSLM), auto-regressive integrated moving average (ARIMA), theta, neural network time series forecasting (NNETAR), seasonal and trend decomposition using loess and forecasting (STLF) and trigonometric seasonality, box-cox transformation, ARMA errors, trend, and seasonal components (TBATS) model. A sigma value of 0.18 signifies effective error control following the Box-Cox transformation in the TBATS model, indicating a strong data fit and high accuracy of forecasts. The forecasted basmati prices range from Rs. 2500-3000 to Rs. 40004500, with actual data from “agmarknet” aligning with the highest or second-highest probability categories.