Ever since Stigler (1961), a growing number of papers have analysed empirically the association between consumer information and price dispersion. Studying gasoline demand, Marvel (1976) finds that both price dispersion at a point in time and price variability over time depend on a set of proxy variables that represent the benefits and costs to consumers of acquiring information. In a review of the literature, Baye et al. (2006) conclude that much of the observed price dispersion in online and offline markets stems from consumers’ costs of acquiring information about firms, and firms’ costs of transmitting information to consumers. The question that arises is: How do consumers acquire information? The theoretical literature offers two models: clearinghouse and sequential search models. In the former, (some) consumers enter an information clearinghouse and become fully informed through, for instance, newspaper and all-or-nothing non-sequential searches. In the latter, consumers assemble information by sequentially visiting the stores in the market and stopping at some point without becoming perfectly informed. The main feature in all-or-nothing non-sequential search models is that they do not allow for searching more or less, but for a discrete decision only: consumers do not search and remain uninformed, or do search and become fully informed.
During the last decade or so, the introduction of transparency initiatives in the form of mandatory price posting and the advent of vendor websites have led to the study of the effect of information disclosure mechanisms on market outcomes such as price levels and price dispersion; see, inter alia, Hong (2014), Jang (2015), Rossi and Chintagunta (2016) and Luco (2019). At the same time, a number of authors have started to use the number of daily visits to these websites as a proxy for consumer search, with the purpose of studying the factors that guide consumers to search for better prices, reversing, to some extent, the price dispersion-consumer search link studied in (almost all) the previous research.
In this paper, we examine the effect of price changes and price dispersion on consumer search in retail diesel markets in France. We take advantage of a piece of French legislation requiring that all stations that sell more than 500 m3 of fuel per year must report all fuel price changes to the Ministry of Economy, which are then made freely available to the public through a governmental website. The motivation behind the website is to provide motorists with accurate and up to date price information to help guide their purchase decisions. From theoretical models of price dispersion, the French government website serves as an information clearinghouse that fully informs consumers who visit the site. It means sequential search models are less likely to apply to this situation.Footnote 1 We use these data to obtain an accurate measure of market-level cross-sectional price dispersion and their changes from day to day. In addition, the main feature of this research is that for measuring consumer search we use novel data on daily consultation measures from this government-run site.
While Lewis and Marvel (2011), Byrne et al. (2015), Noel (2018) and Noel and Qiang (2019) have already used retail price data from websites, i.e. GasBuddy.com for the US and Canada, the prices in such websites are provided by voluntary observers, and so there is the possibility of sample selection bias; see e.g. Atkinson (2008). In contrast, the fact that French legislation mandates that stations must report their prices implies that geographical coverage is maximised, while sample selection biases, data inaccuracies and report delays are minimised. In this respect, we follow closely Byrne and de Roos (2017) inasmuch as we also have a measure of search intensity given by the number of unique daily visits to the website. Unlike Byrne and de Roos (2017), the distinctive aspect of our analysis is that we use for the first time a measure of time spent per search. As in all models of search (see e.g. Varian 1980, Stahl 1989, Yang and Ye 2008, Lewis and Marvel 2011), consumers make decisions on whether to search to become informed about prices or not to search and stay uninformed. In these models, informed consumers are those that observe the prices charged by all firms, and then purchase from the firm that offers the lowest price, while uninformed consumers shop randomly and only observe the price at which they purchase. A consumer’s decision to search depends on whether the expected benefit of searching outweighs individual search cost. Thus, the information disclosure policy enforced by the French government affects consumer search decisions by reducing the costs of obtaining accurate and reliable price information.
In this paper, we provide two consumer search intensity measures: the number of visitors to the respective website and the time spent there.Footnote 2 The availability of a free site where fuel prices are posted raises a number of questions of interest: How can the disclosure of daily prices affect consumer behaviour? What does the time spent on this government website tell us? Who is expected to spend more time searching the market? Answering these questions involves the consideration of first- and second-order effects. Indeed, one potential first-order effect is that the government website pushes some part of the mass of uninformed consumers to become informed, since it reduces search costs and implicitly increases search benefits. Then, assuming consumers behave rationally, as in Tappata (2009), one would expect that the number of daily visits to the website are positively associated with both price dispersion and price variability over time. However, increased search benefits occur in models that only deal with the consumer side. A firm would find it optimal to compete for the larger share of informed consumers with a low price. Hence, the mass in the price distribution would shift downwards, giving less dispersion. The benefits to searching would go down because a consumer is more likely to already observe a low-price firm at random; see e.g., Petrikaitė (2016) for a theoretical analysis of increased transparency and the pricing decision by firms.
The second-order effects comprise the easiness to search the web relative to the gains from searching.Footnote 3 Some people who are efficient at using computers and surfing the Internet can retrieve information rapidly from the website and therefore have low or zero search costs. These individuals will use the website even if the gains from search are small (for example, because price dispersion is small). Other consumers are less efficient and only search if they perceive significant benefits. The time spent on the website reflects the efficiency of the search process. Therefore, to the extent that less and less efficient consumers start searching the market, the gains from search increase. Two effects emerge: First, the number of visits increases and, second, the average time spent on the website increases. Thus, the second-order effect arises from the successive entrance of less proficient consumers in using computers and surfing the Internet. Conceptually and to some extent, one can think of the customers with a high ability to navigate the government website in the spirit of the heterogeneous consumers in Tappata’s model. To say, the portion of consumers he calls shoppers with negative or low search costs.
Another salient aspect that distinguishes our work from that in Byrne and de Roos (2017) is that these authors use data from a single city where, for reasons not specified, price changes always occur on Thursdays, allowing consumers to anticipate variations and increase their searches in the prior days. In contrast, in the French database at our disposal price changes do not exhibit any regular behaviour that can be anticipated.
The paper is organised as follows. Section 2 describes the data. Section 3 presents the results of the econometric analysis on the determinants of search activity, as measured by the number of website visits and time spent per visit. Section 4 concludes.