Modeling Advertising Expenditures and Spillover Effects Applied to the U.S. Non-Alcoholic Beverage Industry: Vector Autoregression (VAR) and Polynomial Distributed Lag (PDL) Approaches

The non-alcoholic beverage market in the U.S. is a multi-billion dollar industry growing steadily over the past decade. Also, non-alcoholic beverages are among the most heavily advertised food and beverage groups in the United States. Several studies pertaining to non-alcoholic beverages including the incorporation of advertising effects have been conducted, but most of these have centered attention on milk consumption. Some studies have considered demand interrelationships for several beverages including advertising effects in systems-wide analyses. In our analysis, we develop and employ a unique monthly time-series data set derived from Nielsen Homsescan panels for household purchases of non-alcoholic beverages over the period from January 1998 through December 2009. This data set is subjected to the use of vector autoregression (VAR) and polynomial distributed lags (PDL) to examine own-advertising and cross-advertising effects for non-alcoholic beverages. Contemporaneous causal structures among advertising expenditures, quantities and prices of various non-alcoholic beverages also are studied using artificial intelligence approaches such as directed acyclic graphs (DAGs). Once the VAR and PDL (through augmenting the AIDS model) are estimated, we can ascertain own- and cross-advertising effects of various beverages. Impulse response functions gleaned from the use of the VAR approach would show us the impacts of advertising expenditures of various beverages to a one-time-only shock. The direction and the strength of the impulse would speak to the effect of the adverting expenditure of one beverage on others. Error variance decompositions help us determine the magnitude of the contribution of advertising expenditures of several beverages on a given beverage. Similarly, own- and cross-adverting expenditure elasticities generated through the PDL help us determine the negative and positive spillover effects of advertising expenditure of one beverage on several others. The contemporaneous causal structure pertaining to advertising expenditures, quantities and prices of various beverages help identify endogenous, exogenous and/or weakly exogenous factors. Ultimately, the better model to ascertain the effect of advertising expenditures of various non-alcoholic beverages is identified through the study of out-of-sample forecasts generated though both models.