A Nonparametric Algorithm to Estimate True Demand Using Historical Sales Data

Abstract— Demand forecasting is an important part of every inventory control system. The forecasted demand fed in to the optimization module for taking decisions about optimal inventory levels. In most cases, especially when quantitative forecasting methods used, the historical sales data which are stored in the information system’s databases, are used as a data source. This historical data are usually incomplete and censored because of some phenomena such as stock-out in retail or reaching the booking limit in airlines. So they do not show the true demand and could not be a good input to the forecasting module. We proposed a nonparametric algorithm in a multi period multi product environment. The proposed algorithm is capable to estimate true demand using historical sales data with a reasonable level of accuracy and fast convergence time. We have evaluated the algorithm with simulated datasets and the results are promising so the proposed method could be a good replacement for current demand estimation methods in inventory control systems. In contrast to current state of the art methods, our algorithm improved the root mean square error between true simulated demand and true estimated demand by about 21%.