Parameter Estimation and Model Testing
When the causal ordering, functional form, and lag structure are believed known, all that needs to be done is to estimate the model. Nonetheless, any assumptions underlying a model should be tested. In what follows, our discussion is primarily on a conceptual level. Technical details can be found in most leading econometric texts; Judge et al. (1985) is especially helpful in providing guidance for applied work.1
KeywordsGeneralize Little Square Specification Error Nonlinear Little Square Usual Test Statistic Sales Territory
Unable to display preview. Download preview PDF.
- 1.Other econometric texts include Theil (1971), Maddala (1977), Harvey (1981), Pindyck and Rubinfeld (1981), Chow (1983), Johnston (1984), and Amemiya (1985).Google Scholar
- 2.Weinberg and Weiss (1986) have proposed an alternative estimation approach to the Blattberg and Jeuland model.Google Scholar
- 3.Naert and Weverbergh (198la, p. 149) concisely explained the reason for using generalized least squares in market share models: When one of the market shares is overestimated, the error must be compensated for by an underestimation of one or more of the other brands. It follows that the error terms… will be contemporaneously correlated, that is, error terms in different equations but with respect to the same time period will not be independent. Estimation efficiency therefore can be improved by explicitly accounting for this correlation in the estimation process.Google Scholar
- 5.A major problem is the arbitrary nature of units of area. Openshaw and Taylor (1979) emphasize that “since the area over which data is collected is continuous, it follows that there will be numerous alternative ways in which it can be partitioned to form areal units for reporting the data.”Google Scholar