Weather-Augmented Seasonal ARIMAX Modelling for Wholesale Chilli Price Forecasting in Telangana, India

Abstract

This study developed and evaluated weather-augmented seasonal autoregressive integrated moving average with exogenous variables (ARIMAX) models for short-term forecasting of wholesale dry chilli prices in Telangana, India. Monthly wholesale chilli price data and station-level weather observations from April 2002 to February 2026 were analysed. Weather variables comprised total monthly rainfall, mean relative humidity, mean maximum temperature and mean minimum temperature. Exploratory analysis used descriptive statistics, time-series plots, seasonal decomposition and autocorrelation functions. Seasonal ARIMAX models with individual weather regressors were first estimated using selected lags, followed by a mixed-lagged model including all four weather variables. A seasonal ARIMA model without exogenous regressors served as the benchmark. Forecasting performance was assessed using root mean square error, mean absolute percentage error, and training and test-set comparisons. The chilli price series showed substantial variability, with a mean of 10067.1 rupees per quintal, standard deviation of 6438.6 and coefficient of variation of 64.0 per cent. Among individual weather-augmented models, the relative humidity model performed best, with a test RMSE of 3178.6 and test MAPE of 12.316 per cent. The mixed-lagged seasonal ARIMAX model also outperformed the seasonal ARIMA benchmark, recording a test RMSE of 3344.9 and test MAPE of 13.013 per cent. The findings indicate that selected lagged weather variables, particularly relative humidity and rainfall, improve short-term wholesale chilli price forecasting relative to a univariate seasonal ARIMA model.