Umeå University's logo

Purpose: In this paper, the author tests five econometric methods for cartel detection suggested in the aforementioned papers on a data set covering the so-called Swedish asphalt cartel, where collusive bidding was detected by the Swedish Competition Authority (SCA) in 2001 and later verified by court in 2003. This paper also elaborates on the econometric specifications suggested by Lundberg (2017) and Bergman et al. (2020). The analysis concentrates on collusive bidding on the same contract. Finally, this paper relates the author’s findings to the analysis by Aaltio et al. (2025), who applied the distributional regression approach suggested by Clark et al. (2025) and the machine learning approach suggested by Huber and Imhof (2019) to the Swedish asphalt cartel.

Design/methodology/approach: The author’s data set contains information on procurements of paving and asphalt works (surfacing works) during the period 1995–2009 and the identities of convicted firms, making it possible to analyze bidding behavior before and after the 2003 court order. The empirical strategy is to first test if the econometric methods suggested by Porter and Zona (1993, 1999) and Bajari and Ye (2003) are able to verify the asphalt cartel, as did the methods suggested by Lundberg (2017) and Bergman et al. (2020). This paper also elaborates on the econometric specifications suggested by Lundberg and Bergman et al.

Findings: In short, when applied to the Swedish asphalt cartel and in contrast to the findings by Porter and Zona (1993) for highway construction contracts on Long Island in the early 1980s, this paper does not find any systematic differences in bidding behavior between competitive and collusive bidders or evidence supporting the hypothesis that the ranking of high cartel bids is related to different determinants than the ranking of competitive or high bids by noncartel firms. Moreover, this paper finds no clear evidence that the probability of submitting bids differs between competitive and cartel bidders, as Porter and Zona (1999) found for school milk procurements by Cincinnati school districts in the 1980s. What this paper do find is a positive correlation between backlog and participation during the cartel period, suggesting that even if the cartel firms are (over)loaded with jobs, they tend to submit bids to create the impression of competition. Besides coordinating their bids, larger cartel members, in terms of the number of bids on different contracts each year, tended to bid less aggressively against each other during the cartel period. The pairwise test of conditional independence by Bajari and Ye (2003) indicates collusive bidding behavior among the convicted firms both during the cartel period and for some of the larger convicted firms after the court order in 2003. One potential explanation for this is that the testing procedure by Bajari and Ye does not discriminate between bids by the designated winner and phony bids by the other cartel members. Adding backlog, the spatial econometric approaches by Lundberg (2017) and Bergman et al. (2020) still verify collusive bidding behavior during the cartel period and reject such behavior for the postcartel period. This result is important as it suggests that the spatial econometric approach is robust to potential omitted variable bias. The result also provides more information to the findings of the bimodal distribution of bids by Aaltio et al. (2025); the phony or predetermined losing bids by members of the cartel are also correlated. That is, leaving a gap to the designated winning bid, the phony bids are not random but coordinated.

Practical implications: From the perspective of a competition authority, some considerations should be made before the authority initiates an investigation into a suspected cartel or conducts a more general scan of several different markets. Firstly, a competition authority often has limited resources and must therefore assess in most cases whether the monetary damage caused by the cartel is substantial enough to justify an investigation and the collection of further information. This should involve the direct economic damage that the cartel inflicts on the contracting authority, but also concern the extent to which the cartel affects the functioning of the market and thereby causes harm to competing companies and consumers. Secondly, are there sufficient procurements to enable an econometric analysis of the companies’ bidding behavior? If not, further analysis must be based on other methods and the collection of additional information to substantiate any suspicions. Thirdly, if there is enough data for an econometric analysis, decide on which econometric method is best suited to detect the suspected behavior or for scanning.

Social implications: Coordinated bids on public contracts are probably more common than this paper realizes. As Harrington (2008) pointed out, …cartels are among us. Their attempts to coordinate the prices they set and the quantities they produce are often effective. […] When we find them, we ought to prosecute and penalize them. But how do we find them? Different econometric methods to detect coordinated bids have been proposed by, among others, Porter and Zona (1993, 1999), Bajari and Ye (2003), Lundberg (2017) and Bergman et al. (2020). But which one should we use? And do they reach the same conclusions? To what extent do they complement each other in our understanding of how the cartel operates?

Originality/value: This paper adds to the existing literature in at least three ways. First, this paper applies and evaluates the different methods on the same data. To our knowledge, this has not been done before with these specific methods. Such evaluation is important for national competition authorities in their choice of econometric method to detect and test for suspicious bid-rigging. Second, by introducing backlog in the econometric specifications used by Lundberg (2017) and Bergman et al. (2020), this paper tests the sensitivity and robustness of their results. Third, this paper complements recent and ongoing research on the evaluation of different econometric methods in this area (Huber and Imhof, 2019; Aaltio et al., 2025). It is possible to obtain a bimodal distribution without the nonwinning cartel bids correlating with each other. It may also be the case that the participating companies rotate in different procurements, which does not create a correlation between the nonwinning bidders (as they do not participate in many procurements due to this rotation), while still achieving a bimodal distribution of the cartel’s bids.