Statistical Methods in Customer Relationship Management

Appendix C

Vector Autoregression Modeling

(Source: Dekimpe and Hanssens [1])

Dekimpe and Hanssens [1] introduced vector autoregression modeling under the framework of persistence modeling. Persistence modeling addresses the problem of long-run market-response quantification by combining into one measure of ‘net long-run impact’ the chain reaction of consumer response, firm feedback, and competitor response that emerges following the initial marketing action. Persistence modeling is a multi-step process. In the first step, unit-root tests are used to determine whether or not the different variables are stable or evolving. In case several of the variables are found to have a unit root, one subsequently tests for cointegration. Depending on the outcome of these two preliminary steps, one estimates a vector autoregression (VAR) model in the levels, in the differences, or in error-correction format. Finally, the parameter estimates from this VAR model are used to derive impulse-response functions (IRFs), from which various summary statistics on the short- and-long-run dynamics of the system can be derived. Each of these steps is elaborated briefly as below.