Demand Elasticity Prediction in Retail Using MOMENT: A T5 - Based Time Series Model

Manish Reddy Bendhi

Abstract: Demand elasticity prediction is critical for optimizing retail pricing, promotions, and inventory. Traditional econometric models struggle with non - linearities and high - dimensional data while existing deep learning methods focus on numeric forecasting rather than elasticity estimation. We propose MOMENT, a T5 - based sequence - to - sequence model that converts structured retail time - series data (e. g., price, promotions) into tokenized sequences to predict elasticity coefficients. Experiments on Kaggle?s Corporation Favorita Grocery Sales Dataset and synthetic data demonstrate MOMENT?s superiority over statistical (ARIMA, Prophet) and deep learning (LSTM, Transformer) baselines, achieving a 9.1% MAPE (15?20% improvement). Practical guidelines for data preparation, hyperparameter tuning, and deployment are provided, alongside a case study showing a 15% reduction in overstocking for a European retail chain.

Keywords: Demand Elasticity, Retail, Time Series Forecasting, T5, MOMENT, Transformer