DISTRIBUTED LAGS AND MULTIPLE TIME SERIES ANALYSIS

THOMAS L PILON, Purdue University

Abstract

The primary objective of this dissertation was to demonstrate the usefulness of a relatively new technique, Multiple Time Series Analysis, as an alternative method for specifying distributed lag models. While classical distributed lag techniques have been somewhat successful, they usually impose unwarranted restrictions on the shape of the lag distribution. Also, they are often confounded by the peculiar properties of time series data. Multiple Time Series Analysis is an extension of univariate time series analysis and a more general case of a procedure known as Transfer Function Analysis. The above mentioned technique was applied to a data base on a widely used service. Unfortunately, it was necessary for the author to sign a non-disclosure agreement as a prerequisite for obtaining the data. Therefore, the industry and the company are described only in general terms. Data on sales volume, price, employment, customers, and advertising were available for the analysis. Although thirteen models were developed: eight state A models, three national models, and two state B models, only two models are discussed in any detail. The results were unexpected. Once the data series were made stationary, as required by the Multiple Time Series Analysis procedure, it was found that there were practically no relationships between any of the variables. The only two relationships that were found were the only relationships that were known to exist by management since these relationships involved decision variables that were manipulated by management. Although none of the relationships that were found were interesting from a managerial perspective, the fact that these particular relationships were identified demonstrates the validity of the technique.

Degree

Ph.D.

Subject Area

Marketing

COinS