Sales: Tests of Theories of Causality and Timing

Modern theories of sales make conflicting predictions about temporal patterns of sales, which we test using grocery scanner data. We examine both frozen (storable) and refrigerated orange juice to determine what role-if any-durability plays in the sales patterns. We start with a simple reduced-form probit analysis examining the timing of sales and who determines sales: nationally the manufacturers or locally the retailers. We turn to a vector autoregressive analysis and conduct Granger tests of temporal ordering ("causality tests") to determine whether the sale of one brand is followed in a predictable way by the sale of another brand or its own later sales. Based on the VAR estimates, we simulate impulse responses to determine the magnitude of these effects. In fact, none of the theories of sales fully describes sale patterns and price distributions. We find product durability makes little difference in sales patterns and contrary to all the existing theories, retailers rather than manufacturers determine sales. Despite sale patterns not being significantly different for national brands and private label brands, formal Granger causality analysis shows a sale of a national brand to be more likely to "cause" sales of other products than a sale of a private label product.