Social media and sales: Determining the predictive power of sentiment analysis towards car sales

This paper aims at exploring the use of sentiment analysis on social media as a tool for sales forecasting in the automotive industry. Previous research on this topic has presented significant results although current literature still lacks investigation on the usefulness of this technique when it comes to more expensive items. In particular, about 500,000 social media posts and eleven car models from the Dutch market are analyzed using linear models. Furthermore, the research compares these outcomes to the predictive power of search volume by using Google Trends as an indicator. Based on variables that are assessed as strong predictors of sales a prediction model using decision tree regression is built that can potentially be used by car manufacturers as an addition to traditional forecasting methods if tested and developed further. The results suggest that social media sentiments have little to no predictive power towards car sales. While search volume as well as the rate of attention a model receives on social media show significant results and can be incorporated into the prediction model, the sentiments itself only obtain weak correlations with car sales and clearly show a limitation to the technique of sentiment analysis. Although the findings cannot be generalized for other car models and markets, this research contributes to further understanding the field of sentiment analysis and explores its boundaries. It also presents a prediction model that allows to approximate car sales. It therefore has both practical and academic value. Supervisors: Dr. Fons Wijnhoven