A nonlinear time series framework for forecasting weekly tomato prices in Chikkaballapura Market, Karnataka

Abstract

Frequent price fluctuations in perishable agricultural commodities complicate decision-making for farmers, traders, and policymakers. This study develops and compares linear and nonlinear time series models for predicting weekly tomato prices in the Chikkaballapura market of Karnataka. Weekly data covering January 2010 to November 2018 were analysed. Stationarity was evaluated using the Augmented Dickey–Fuller procedure. Since the original series exhibited non-Stationarity, first-order differencing was applied. A linear ARIMA structure was estimated to model the conditional mean, while conditional variance dynamics were captured through a GARCH specification. Residual diagnostics from the ARIMA model revealed significant volatility clustering, justifying nonlinear modelling. Comparative forecast evaluation based on RMSE, MAE, and MAPE demonstrated that the GARCH (1, 1) formulation produced more accurate short-term predictions than the ARIMA (2, 1, 1) model. The results underline the importance of accounting for time-varying volatility when forecasting highly unstable agricultural prices.