HEURISTIC THINKING AND LIMITED ATTENTION IN THE CAR MARKET

Abstract Can heuristic information processing affect important product markets? We explore whether the tendency to focus on the left-most digit of a number affects how used car buyers incorporate odometer values in their purchase decisions. Analyzing over 22 million wholesale used-car transactions, we find substantial evidence of this left-digit bias;; there are large and discontinuous drops in sale prices at 10,000-mile thresholds in odometer mileage, along with smaller drops at 1,000-mile thresholds. We obtain structural estimates of the inattention parameter from a simple model of this left-digit bias. We also investigate whether this heuristic behavior is primarily attributable to the final used-car customers or the used-car salesmen who buy cars in the wholesale market. The evidence is most consistent with partial inattention by final customers. We discuss the significance of these results for the literature on inattention and point to other market settings where this type of heuristic thinking may be important. Our results suggest that information-processing heuristics may be important even in markets with large stakes and where information is easy to observe. Although economic models are usually based on the assumption that agents are unconstrained in their ability to process information, economists have long recognized that individuals have limited cognitive abilities (Herbert A. Simon, 1955). An extensive literature on heuristics and biases, originating primarily in psychology, has shown that people often use simple cognitive shortcuts when processing information, leading to systematic biases in decision making. 1 There is large evidence on the nature of these heuristics from surveys and laboratory experiments, but there has been much less research exploring whether these cognitive limitations impact important market settings. In this paper, we study the effects of heuristic information processing in the used-car market. We investigate whether the market is affected by consumers exhibiting a heuristic known as left-digit bias when they incorporate odometer mileages into their decision process. odometer values between 80,000 and 80,100 miles, but for only $10 less than cars with odometer readings between 79,800 and 79,899. Regression analyses show significant price discontinuities at each 10,000-mile threshold from 10,000 to 100,000 miles. The size of the discontinuities is similar across each threshold, consistently on the order of $150 to $200. Consistent with our model, we also find smaller price discontinuities at 1,000-mile thresholds. The left-digit bias we identify in this paper not only influences wholesale prices but also affects supply decisions. If sellers are savvy and aware of threshold effects, they will have an incentive to bring . Indeed we show that there are large volume spikes in cars before 10,000-mile thresholds. 1 See Gilovich, Griffin, and Kahneman (2002) for a review. 2 These volume spikes, however, also make the task of identifying unbiased estimates of the price drops at thresholds more difficult. Because of the seller response to threshold effects, it is necessary to account for potential selection in our analysis, and we do so in several different ways. First, we present our findings after controlling for selection on observables, including fixed effects for the combination of make, model, model year, body style of a car, and auction year. In our most restrictive specification, we are able to identify the impact of crossing a 10,000-mile threshold by comparing cars of the same make, model, model year, body style, and that are brought to auction by the same seller in a given year. We also run our analyses separately for different types of sellers at the auctions. All of the buyers at the wholesale auctions are licensed used-car dealers, but sellers can be both car dealers and companies with fleets of cars, such as leasing companies and rental-car companies. We show that the selection varies considerably across these seller types and yet we find similar price discontinuities for both types. We also discuss additional selection issues and present a range of evidence suggesting that unobserved heterogeneity is unlikely to affect our findings. We perform further checks in order to allay concerns that the observed threshold effects might be a result of institutional features related to the used car market. The results are robust to considering a number of alternative explanations, such as the potential for odometer tampering and the structure of car warranties. We also test a secondary prediction of our model;; because inattention leads to discontinuous changes in perceived mileage around thresholds, the price discontinuities at these thresholds should be larger for cars that are depreciating at a faster rate (i.e., those more affected by mileage changes). Indeed we find larger price discontinuities for cars that depreciate quickly (e.g., Hummers) than for cars that depreciate slowly (e.g., Honda Accords). Finally, we use a smaller sample from Canadian data to construct a type of placebo test. We find price discontinuities in Canadian used-car auctions at the 10,000-kilometer marks, but not at the 10,000-mile marks. The particular setting of our study the wholesale used-car market allows us to at least partially investigate the influence of heuristic information processing on different economic agents. The price discontinuities in the wholesale market may arise because used-car dealers who buy at the auctions recognize that their final customers will exhibit the left-digit bias and purchase cars at the auction accordingly. It is also possible, however, that it is the used car dealers themselves who are subject to the left-digit bias. It is not easy to disentangle these cases because there is little observational difference between the two. However, we can address whether inattention seems to be driven primarily by used-car dealers or final customers. A range of evidence including volume patterns, purchase patterns for experienced versus inexperienced dealers at the auctions, pricing 3 dynamics right before thresholds, and data from an online retail used-car market are all suggestive that our findings are driven by limited attention of the final used-car customers