Journal of Scientific Research and Reports

Cotton (Gossypium spp.) price forecasting is important for market decision-making in Andhra Pradesh, where price instability affects farmers, traders and related stakeholders. This study evaluated statistical, machine-learning and wavelet decomposition models for predicting monthly wholesale cotton prices in the Adoni market using secondary modal price data collected monthly from January 2010 to December 2024, comprising 180 observations. The period from January 2010 to December 2023 was used for model training, while January to December 2024 was retained for testing. The models compared were ARIMA (0,1,0), GARCH (1,1), ANN, Wavelet-ARIMA, Wavelet-ANN and Wavelet-GARCH. Forecasting performance was assessed using RMSE, MSE and MAPE, and residual diagnostics were used to examine model adequacy. The descriptive results indicated substantial variability in prices, with right-skewed price behaviour. Among the fitted models, GARCH (1,1) and ARIMA (0,1,0) produced the lowest training errors. In the test set, WANN recorded the lowest RMSE and MAPE, indicating strong out-of-sample predictive accuracy. However, the wavelet-based models showed residual autocorrelation. ARIMA demonstrated comparatively stable forecasting performance across training and testing, whereas ANN produced residuals without significant autocorrelation or nonlinear dependence. The findings indicate that model selection should consider both forecast accuracy and residual adequacy when forecasting cotton prices in the Adoni market.