The title of this paper, available here, is a bit misleading, since individuals who developed a screening tool for the Swiss competition authority wrote it, and the paper is devoted to describing how the method works and how successful it has been in practice.

Bid Rigging Screen

To summarise, the paper proposes a method to detect bid rigging that is particularly suited to address the problem of partial collusion, i.e. collusion that does not involve all firms and/or all contracts in a specific dataset. It explains how the authors applied mutually reinforcing screens to a Swiss road construction procurement dataset in which no prior information about collusion was available, and how this method succeeded in isolating a group of “suspicious” firms. It further describes how the screen led the Swiss Competition Commission (COMCO) to opening an investigation and sanctioning the identified “suspicious” firms for bid rigging.

The paper is structured as follows:

Section II presents literature on screening methods.

The growing literature on cartel detection can roughly be divided into two strands. Some literature discusses structural methods that try to analyse the market structure in different industries, and aims at the identification of factors that are known to sustain collusion. A second strand advances so-called behavioural methods that analyse the concrete behaviour of firms in specific markets and employ a multitude of statistical tests to this end.

Section III then describes the dataset.

The starting point for identifying the sample was the annual statistics of a Swiss canton listing all awarded contracts regarding road construction work, grouped by types of services, deliveries and construction. From 2004 to 2010, the procurement body was able to provide the official records of tenders opened for 282 of 400 contracts. For these 282 contracts, 138 firms submitted roughly 1,500 bids. Consortia submitted 228 bids and won the bid in 78 cases.

In Section IV, two simple screens are applied to the dataset.

The dataset is not well suited for all the statistical markers suggested in the literature. Given the data contained in the set, it seemed more promising to focus on markers analysing the structure of firms’ bids.

As a result, the authors applied two markers to the data set. First, the authors deployed a variance screen, which is the most comprehensively tested statistical marker to detect collusion. A variance screen relies on evidence that, in the case of collusion, prices are often less responsive to effective costs than in a competitive environment, i.e. price variability is lower in a collusive environment. The authors found that while there was no peculiar evolution of the coefficient of variation over time (i.e. no particular indication of collusion regarding the bids), there was a notable difference between invitation and open procedures. This could be interpreted as a (weak) indication that invitation procedures are more prone to bid rigging than open procedures.

A second screen focused on cover bidding. In the past few years, COMCO has uncovered several bid-rigging cartels in Switzerland. In many of these bid-rigging cases, it was found that the difference between the losing bids was systematically smaller than the difference between the winning bid and the second-best bid. Such bidding behaviour may be explained by the presence of cover bidding: bidders not intending to win a contract offer distinctly higher prices than the agreed winner does. This insight allowed the authors to construct an alternative price-related marker which considered the difference between losing bids and the difference between the two best bids for a specific contract. The results are similar to those of the variance screen: while there are no peculiar developments observable over time, the cover bidding test again suggests that collusion is more likely to be present in invitation procedures.

Each screen, employed individually, seems to indicate that all the firms in the sample were not involved in a systematic market-embracing collusive scheme. This is not surprising: COMCO’s investigations concerning bid rigging have revealed that cartels in construction markets often are partial, that is, they only involve a subset of colluding firms and/or collusion is targeted at specific contracts. Thus, excluding the presence of bid rigging would be premature without first trying to determine whether there is evidence of partial collusion.

Section V combines two screens and shows how this may help detect partial collusion.

A crucial prerequisite to detect partial collusion with a statistical marker is a sufficient degree of regular interaction between stable groups or sub-groups of firms. Irregular and selective bid-rigging agreements between loosely connected firms (for example, for special types of projects) are, however, extremely hard—if not impossible—to identify with a simple screen. The authors therefore opted for using mutually reinforcing tests, which allow conclusions as to whether collusion is likely to exist between sub-groups of firms.

This was done in four steps. First, the authors applied the variance and the cover bidding tests to the contracts of two road construction bid-rigging cartels uncovered in Switzerland. They then applied the results to isolate contracts and firms in the dataset exhibiting a suspicious bidding pattern, and identified 80 conspicuous contracts, most of which were entered into following an invitation procedure. Secondly, the authors looked at whether they could identify firms that belonged to groups of firms regularly submitting bids for the same conspicuous contracts. They sought to identify firms that had submitted a bid for at least 10% of all conspicuous contracts, in order to eliminate fringe bidders unlikely to participate in a collusive scheme – and identified 17 firms out of 180 bidders. The authors then conducted further controls, leading to the identification of sub-groups of companies which bidding behaviour seemed to correlate with others. Thirdly, the authors analysed geographical bidding behaviour, i.e. where the suspicious bidding activity is taking place and what were the local competitive constraints. This allowed the authors to identify two specific regions where the identified firms seemed to be active, and provided further indication of a possible cartel being in place. Fourthly, the authors develop a graphical method to visualise whether there was bid rotation – i.e. rotation on who wins a bid – within a group of firms, which is discussed in the next section.

Section VI discusses the final element of this screen, the bid rotation test, and how the method led to the identification of a bid rigging scheme.

The practice of bid rotation typically involves submitting cover bids for contracts. Bid rotation is likely to produce a distinct bidding pattern: whenever the designated winner submits a “low” bid, all other firms will submit a deliberately “high” bid.

The bidding behaviour in the sample seemed compatible with cover bidding: each tender was inter alia characterised by the fact that the difference between losing bids was systematically smaller than the difference between the winning and the second-best bid; each of the suspect firms had, on an average and simultaneously with another suspect firm, submitted bids for roughly ten conspicuous contracts; and the suspect firms submitted bids for fourteen contracts where 91% of all submitted bids came from the suspect group of firms.

Based on these results, COMCO opened an investigation in 2013. Dawn raids produced proof of collusion and led to the conviction and sanctioning of the involved firms in 2016. All six firms identified by the screening method were condemned by COMCO for participating in a bid-rigging cartel, together with two other firms which had made a smaller number of bids and been eliminated by the filters the authors adopted. The method did not produce any false positives, i.e. no firm was wrongly identified as a member of the cartel, and all identified firms were convicted for bid rigging.

 

Comment:

In addition to providing an example of a method developed and deployed by a competition authority to identify bid rigging, this paper provides an interesting overview of available cartel screen, and of the literature concerning them. It is not my area, but I think this paper provides a good example of the benefits of developing suitable screens to identify collusion, and of how to promote local work to the wider competition community.

Author Socials A weekly email with competition/antitrust updates. All opinions are mine

What do you think?

Note: Your email address will not be published