One of the major tasks of the modern firms is the development of new products. Forecasting the sales of these new products is of critical importance to diffusion models make forecasts or present a description of the diffusion process but few of these models deal with the multilevel substitution phenomenon. This thesis focuses on developing a new approach to estimating the parameters of diffusion models for describing multilevel substitution and comparing the predictive ability of selected models with that of a new one. Two sets of sales data, Dynamic Random Access Memory (DRAM) and Tire Cord fabric, have been selected for testing.;The findings of this study suggest that (1) new models built by new approach have better predictive ability than other selected models, (2) the internal influence of diffusion process decreases and external influence of diffusion process increases over product generations. |