Simulated Annealing Sales Combining Forecast in FMCG

Fast Moving Consumer Goods (FMCG) industry is faced with a wide range of consumers, high frequency of consuming, quick change of demands, low loyalty, and high demand for convenience. Those characteristics determine sales demands as the largest uncertainty. To facilitate this situation, companies in FMCG require subjectivity-free and accurate sales forecasts. Many quantitative time-series models are brought out to achieve this goal, but single forecast selection still relies on the subjective judgment of operators and results may be biased. An approach, therefore, is proposed to try to employ the concept, Combining Forecast to solve the single-selection problem by leveraging Simulated Annealing Algorithm. Basic forecasting models in potential set generate their own predicted series relying on historical sales data. And Simulated Annealing Algorithm trains respective weights of all basic models. Calculating weighted average produces the ultimate forecast series. Finally, the experimental results show that the optimized approach is able to reduce the Means Absolute Percentage Error (MAPE) value by up to 16.9%, to allow multi-selection on models, and to bring scalability and adjustability into forecast.