Abstract:
|
Forecasting of online apparel demand has many challenges since demand is both volatile and often intermittent due to out of stock periods, seasonality, peak periods, fashion trends, long lead times/supply processes, short product lifecycles, a large amount of product varieties, and nonlinearities (especially during volatile economic activity). A variety of techniques have been used in forecasting apparel demand including exponential smoothing models, ARIMA models, expert systems, fuzzy systems, neural networks, Croston's method (for intermittent demand), state space modeling, elasticity models, and hybrid approaches (combining models). We propose a hierarchical time series approach using the inherent hierarchical structure of the data. We investigate several forecasting techniques and develop an optimal way of reconciling forecasts at each level of the hierarchy to obtain revised forecasts.
|