Fashion Sales Forecasting With a Panel Data-Based Particle-Filter Model

In this paper, we propose and explore a novel panel data-based particle-filter (PDPF) model to conduct fashion sales forecasting. We evaluate the performance of proposed model by using real data collected from the fashion industry. The experimental results indicate that the proposed panel data models outperform both the traditional statistical and intelligent methods, which provide strong evidence on the importance of employing the panel-data approach. Further analysis reveals that: 1) our proposed PDPF method yields a better forecasting result in item-based sales forecasting than in color-based sales forecasting; 2) a larger degree of Granger causality relationship between sales and price will imply a better sales forecasting result of the PDPF model; 3) increasing the amount of historical data does not necessarily improve forecasting accuracy; and 4) the PDPF method is suitable for conducting fashion sales forecasting with limited data. These findings provide novel insights on the use of panel data for conducting fashion sales forecasting.