New models and methodologies for making marketing predictions

Almost all disciplines of management are concerned with forecasting. Within marketing forecasting plays an important role in many areas (e.g., new product development, competitive actions, choice). The aim of this dissertation is to present new models and methodologies for making predictions in three marketing contexts: marketing applications of random sums, incorporating explanatory variables in stochastic models, and combining forecasts. These three contexts represent distinctive and important decision-making situations in marketing, which lend themselves to substantial refinement and improvement of the existing models in each context. Marketing applications of random sums deal with pervasive economic processes defined as Y = x$\sb1$+ ... + x$\sb{\rm n}$, where both N and x$\sb{\rm i}$\u27s are random variables. Our major concern for this part is to develop a general model of random sums with which one can make the predictions of the future value of Y conditioned on the present information of N and $\rm \{x\sb{i}\}$. The second topic involves incorporating the effects of marketing mix or other purchase influence variables in NBD models. Here we propose a generalized NBD model by introducing nonstationarity for both parameters of the gamma mixture distribution in NBD framework. The third part of the dissertation deals with the problem of combining forecasts. In making marketing decisions, information is obtained from multiple sources such as models with different specifications and expert opinions. When a decision maker has multiple sources of forecasts and the statistical properties (i.e., variances, covariances) of the errors of the forecasts, the use of the optimal method for integrating this information is expected to lead to better decisions. In this part we propose a heuristic which works well in selecting the optimal combined forecasts from any number of sources