RBF is all we need for incorporating seasonality in MMM Models

This paper investigates methods for encoding seasonality information as features for Marketing Mix Modeling (MMM). We compare the performance of two key approaches: dummy variables, which create binary features or each time unit, and radial basis functions (RBFs), which create localized features that capture the distance to specific points in time. We focus on capturing two  significant seasonal effects: a dip in sales in February and a holiday peak in December. Our analysis includes feature engineering, exploring different RBF kernels and tuning bandwidth parameters to optimize feature representation.