Stochastic Demand Forecast of Novel and Short life Products by using Markov based Algorithm

Abstract

In case of novel products with short shelf life, sales data was either unavailable or scarcely available. The available methods for the estimation of demand of such products were direct survey methods, collection of opinion or indirect survey methods, comparison with established products and limited market trial. From literature review it was concluded that existing literature for predicting the demand of novel and short life products were scarce. This led to identification of problem namely demand forecast of relatively novel and short life products. Initially conventional methods like naive, exponential smoothing and moving average methods were used to predict the demand. Markov based model was then applied to forecast errors of the conventional methods. This model or algorithm requires only demand data of two consecutive months and hence is suited for demand forecast of novel products.This algorithm was then applied to two novel baked products, one of relatively large quantity and another of relatively small quantity. Naive, exponential smoothing and moving average methods were applied to this data and the forecasts as well as error for all the working days of two consecutive months were estimated. Markov based algorithm was then applied for these errors and the steady state probability was determined for each state of demand. A state of a system is where the system was at a point of time. The demand corresponding to the state with maximum probability was selected and the corresponding profit was estimated. The obtained profits were then compared and the combination with maximum profit was identified and the method is validated by estimating the annual savings that this method will bring to the firm when compared to existing methods in case of products A and BThe suitability of the model was validated by the fact that its implementation on product A and product B fetched more annual savings when compared to existing practice. Return on investment increased for product A and product B when compared to existing methods. Thus it was concluded that a firm can further enhance its profit by implementing this model or algorithm for more number of products. Further the model can be generalized by applying it to more types of novel products with short shelf life. The forecasting of novel and short life products was not much explored in previous research works. This model can act as the benchmark for future researches in forecasting of novel and short life products