Identifying State Dependence in Brand Choice: Evidence from Hurricanes

We analyze structural state dependence in brand choice using variation from brand switching during stockouts caused by hurricanes. We derive a simple test for structural state dependence based on the time series of choice persistence for households affected by the stockouts. Using data from the bottled water category, we show that demand increases substantially before hurricanes, causing households to purchase different brands. We find that purchase behavior reverts back to its pre-hurricane trajectory immediately after a hurricane, and we are not able to reject the null hypothesis of no structural state dependence. By contrast, the common approach of estimating structural state dependence based on temporal price variation via a discrete choice model yields a positive effect using data for the same category. We argue that our approach is better suited to identify the causal impact of past choices because it requires fewer assumptions and is based on more plausibly exogenous variation in brand switching resulting from stockouts.