This study explores whether NielsenIQ scanner data from U.S. retailers contain incremental information about the GAAP revenue of corresponding manufacturers. Using retail product/store/week data from 2006 to 2018, we construct a measure of aggregated consumer purchases at the manufacturer/quarter level, and find that it strongly predicts GAAP revenues. In addition, analyst forecasts of revenues have predictable errors, which implies that analysts do not fully incorporate the information in consumer purchases. Exploring investment implications, we find that hedge portfolios that buy (sell) stocks of firms with high (low) abnormal consumer purchases generate annualized returns on the magnitude of 14%–19%, depending on specification. Overall, these findings suggest that scanner data on consumer purchases provide an information edge over GAAP revenue, shedding light on the benefits of using transactional data.

This paper reviews the empirical methods used in the accounting literature to draw causal inferences. Recent years have seen a burgeoning growth in the use of methods that seek to exploit as-if random variation in observational settings—i.e., “quasi-experiments.” We provide a synthesis of the major assumptions of these methods, discuss several practical considerations relevant to the application of these methods in the accounting literature, and provide a framework for thinking about whether and when quasi-experimental and non-experimental methods are well-suited for addressing causal questions of interest to accounting researchers. While there is growing interest in addressing causal questions within the literature, we caution against the idea that one should restrict attention to only those causal questions for which there are quasi-experiments. We offer a complementary approach for addressing causal questions that does not rely on the availability of a quasi-experiment, but rather relies on a combination of economic theory, developing and falsifying alternative explanations, triangulating results across multiple settings, measures, and research designs, and caveating results where appropriate.