A Simple Method for Assessing Linear Trend and Seasonality Components in Database Models

The majority of database models are built with only cross-sectional data – present information from a random sample of the relevant database. A database model provides estimates of an individual’s performance (response or contributed profit) to a forthcoming solicitation based on present values of the predictor variables that define the model. However, a database holds more than an isolated “snapshot” of information. It invariably retains past information for the same people over time - time-series data - that offers potential for better models by using both present and past information. Unfortunately, times-series data are not commonplace in database models. Deterrents to their use are the well-known problems (without a simple solution) of autocorrelation and multicollinearity to the modeling process. The purpose of this article is to stimulate the use of time-series data in database models by way of presenting a simple method for assessing the intrinsic linear trend and seasonality components of time-series data.