Predicting automotive sales using pre-purchase online search data

Abstract

Sales forecasting is an essential element for implementing sustainable business strategies in the automotive industry. Accurate sales forecasts enhance the competitive edge of car manufacturers in the effort to optimize their production planning processes. We propose a forecasting technique that combines keyword-specific customer online search data with economic variables to predict monthly car sales. To isolate online search data related to pre-purchase information search, we follow a backward induction approach and identify those keywords that are frequently applied by search engine users. In a set of experiments using real-world sales data and Google Trends, we find that our keyword-specific forecasting technique reduces the out-of-sample error by 5% as compared to existing techniques without systematic keyword selection. We also find that our regression models outperform the benchmark model by an out-of-sample prediction accuracy of up to 27%.