Forecasting Agricultural Prices Using Long Short Term Memory (LSTM) Models and Investigating Present Bias in the Selling Decisions of Farmers in Thirthahalli, Karnataka

This study investigates agricultural price forecasting and farmers’ selling decisions in the arecanut market of Thirthahalli, Karnataka, India. It combines a survey of 80 arecanut farmers with a time-series forecasting analysis using historical market-price data from the Shivamogga APMC market. The forecasting component compares a Long Short-Term Memory (LSTM) neural network with a five-day moving-average benchmark. Model performance is evaluated using mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and peak-detection performance. The survey component examines farmers’ reported reasons for selling, sources of price information, changes in labour costs, and selected socioeconomic characteristics. The study provides an exploratory analysis of farmers’ selling decisions alongside an empirical comparison of machine-learning and conventional forecasting methods. Its findings highlight the importance of evaluating complex forecasting models against simple benchmarks and interpreting survey-based associations within the limitations of the sample and study design. The research was conducted under the supervision of Prof. Ramappa K. B. of the Institute for Social and Economic Change (ISEC), Bengaluru, India.