Evaluating the use of internet search volumes for time series modeling of sales in the video game industry

Abstract

Internet search volumes have been successfully adopted for time series analysis of different phenomena. This empirical paper evaluates the feasibility of search volumes in modeling of weekly video game sales. Building on the theoretical concepts of product life cycle, diffusion, and electronic word-of-mouth advertisement, the empirical analysis concentrates on the hypothesized Granger causality between sales and search volumes. By using a bivariate vector autoregression model with a dataset of nearly a hundred video games, only a few games exhibit such causality to either direction. When correlations are present, these rather occur instantaneously; the current weekly amount of sales tends to mirror the current weekly amount of searches. According to the results, search volumes contribute only a limited additional statistical power for forecasting, however. Besides this statistical limitation, the presented evaluation reveals a number of other limitations for use in practical marketing and advertisement foresight. Internet search volumes continue to provide a valuable empirical instrument, but the value should not be exaggerated for time series modeling of video game sales.