This paper is part of the growing literature that uses micro-data to distinguish among alternative price behavior models and provide some measurements essential for calibration exercises. We use weekly grocery US scanner data and attempt to distinguish among three types of models: State dependent menu cost models, time dependent sticky price models and sequential trade models. Our main findings are: 1. The probability that a store will change its price is increasing in the distance from the mean price charged by other stores and is decreasing in the time since the last price change; 2. The time between price changes is negatively autocorrelated; 3. There is a positive relationship between the standard deviation of transaction prices (across units) and the surprise in sales. State dependent menu cost models say that only the level of the real price matters. Our first finding is that the probability of repricing does depend on the relative price but unlike the prediction of the theory other variables also matter. State dependent models suggest that stores whose (S,s) band is relatively wide will make large nominal price jumps relatively infrequently. This suggests that the time between price changes should exhibit positive serial correlation. We actually find a negative autocorrelation. Time dependent models assume that a store change its price every N periods. This suggests that the probability of making a price change should increase with the time since the last change. We find that the probability actually decreases with the time since the last change. The uncertain and sequential trade model says that when the realization of demand is low only low priced goods are sold, but when the realization of demand is high both low priced and high priced goods are sold. This suggests a positive relationship between the standard deviation of transaction prices (across units) and the surprise in the number of units sold. The correlations in the data are consistent with this prediction.
