Modeling Seasonality in New Product Diffusion

Although high frequency diffusion data is nowadays available, common practice is still to only use yearly figures in order to get rid of seasonality. This paper proposes a diffusion model that captures seasonality in a way that naturally matches the overall S-shaped pattern. The model is based on the assumption that additional sales at seasonal peaks are drawn from previous or future periods. This implies that the seasonal pattern does not influence the underlying diffusion pattern. The model is compared with alternative approaches through simulations and empirical examples. As alternatives we consider the standard Generalized Bass Model and ignoring seasonality by using the basic Bass model. One of our main findings is that modeling seasonality in a Generalized Bass Model does generate good predictions, but gives biased estimates. In particular, the market potential parameter will be underestimated. Ignoring seasonality gives the true parameter estimates if the data is available of the entire diffusion period. However, when only part of the diffusion period is available estimates and predictions become biased. Our model gives correct estimates and predictions even if the full diffusion process is not yet available