Forecasting fashion store reservations: Booking horizon forecasting with dynamic updating

Abstract

A highly accurate demand forecast is fundamental to the success of any booking management model. As often required in practice and theory, we aim to forecast the accumulated booking curve as well as the number of expected reservations for each day in the booking horizon. To reduce the high dimensionality of this problem, we apply singular value decomposition on the historical booking profiles. The forecast of the remaining part of the booking horizon is dynamically adjusted to the earlier observations using the penalized least squares and the historical proportion method. Our proposed updating procedure considers the correlation and dynamics of bookings within the booking horizon and between successive product instances. The approach is tested on simulated demand data and shows a significant improvement in forecast accuracy.