The Cost Impact of Sales Prediction Accuracy

Abstract

Managers’ expectations about future sales are at the core of managerial decision making. While prior literature examines the effects of demand expectations on cost behavior, we disentangle predicted and unpredicted sales changes. We draw on sales expectations from the joint harmonized EU program of business and consumer surveys and introduce the machine learning algorithm eXtreeme Gradient Boosting (xgboost). Consistent with our theory, both, results from subjective expectation data as well as results from the xgboost prediction model suggest that cost increases from unpredicted sales increases are significantly higher than cost increases from predicted sales increases, and costs are substantially more sticky for unpredicted than for predicted sales decreases. Thus, our study is relevant for managers and researchers to better quantify the cost impact of sales prediction accuracy. Moreover, by applying xgboost – a popular algorithm in data sciences – to predict future sales without internal data about the firms, our study also supports stakeholders in updating their beliefs about sales and costs, in particular for medium sized firms with little analyst coverage.