This presentation by the Swiss Competition Commission was made during a workshop on “Cartel screening in the digital era” held by the OECD in Paris on 30 January 2018. More papers and presentations on the topic can be found out at oe.cd/wcsde.

1. Wettbewerbskommission WEKO
Commission de la concurrence COMCO
Commissione della concorrenza COMCO
Swiss Competition Commission COMCO
OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
SCREENING FOR BID RIGGING
OECD Workshop on Cartel Screens
30th January 2018, Paris

2. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
CONTENT
I. Introduction
II. Data and Case
III. Variance Screen
IV. Relative Distance Screen
V. Screening for Partial Collusion
VI. Cover-Bidding Screen
VII. Discussion & Conclusion
VIII. Literature and Further Readings
2

3. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
The Swiss Competition Commission (COMCO) has opened several investigations
in the past years. Despite all these procedures, it seems that bid-rigging cartels
are still a recurrent issue. Why?
 Prosecution:
 The prosecution of bid-rigging cartels is difficult because COMCO generally
needs inside information to open an investigation.
 Firms submit leniency applications after the opening of an investigation and
not before.
 Therefore, to fight more effectively against these harmful practices, COMCO
needs to reduce the dependency on these external sources.
 Question:
 Is there any method that can detect bid-rigging cartels without requiring
inside information or a leniency application? Which characteristics should
such a detection method have?
3
I. Introduction

4. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
In 2008, COMCO decided to initiate a screening project to reinforce the fight
against bid rigging.
 Objective:
 To elaborate a screening tool based on data that is available without the
collaboration of possible cartel participants (no prior leniency).
 Required Characteristics of the Detection Method:
 It should provide evidence of reasonable grounds for suspicion,
convincing non-economists (particularly judges) to launch an investigation
ex officio.
 It should solely use publicly available data, because the method should
run in secrecy without raising the suspicion of potential cartel members.
 It should be a simple process to analyze large datasets; no systematic
in-depth analysis of individual tenders.
 Conclusion:
 The detection method should be simple and reliable.
 Topic: Presentation of COMCO’s screening project.
4
I. Introduction

5. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
5
 Therefore: COMCO’s detection project explicitly and consciously focused on
“simpler” behavioural detection methods.
Screening
Structural methods
(Analysis of industries)
Behavioural methods
(Analysis of bidding behaviour of firms)
“complex” behavioural
methods
“simple” behavioural
methods (statistical
markers)
I. Introduction

6. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
Chronology and current state of play of the screening project
 Starting the screening project (2008).
 Data gathering process (data obtained in September 2011).
 Elaborating and applying screening tool:
 indicated a high probability for the existence of a bid-rigging
cartel in one region (subject of the following slides).
 Opening of an investigation (15th April 2013).
 Dawn raids (15th April 2013).
 Closing the investigation (8th July 2016):
 The investigation confirmed the results of screening tool.
 COMCO fined the members of the uncovered bid-rigging
cartel.
 Discussion Paper of Imhof, Karagök and Rutz (2017):
http://crese.univ-fcomte.fr/WP-2017-09.pdf
6
II. Data and Case

7. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
The data used and the development of the screening project:
 Data
 Data relating to road construction in the canton of St. Gallen (no prior
investigation).
 282 Tenders from 2004 to 2010, about 1500 bids and 138 firms
 Only bids of the tender participants.
 Available to procurement authorities.
 Characteristics of the Canton of St. Gallen
 No prior investigation.
 No prior leniency application.
 No complaint.
 Ex-ante analysis with no prior information.
 Benchmark data base
 Data relating to the canton of Ticino (closed investigation, April 2005-Nov. 2007).
 Data relating to the canton of Aargau (closed investigation, June 2009-Dec. 2011).
7
II. Data and Case

8. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
First experience: In tenders where bids are set collusively the
variance is lower than in non collusive tenders.
 Investigation «Canton Ticino» (see next slide).
 Empirical and theoretical evidence: Abrantez-Metz et al.
(2006), Jimenez / Perdiguero (2012), Athey et al. (2004),
Harrington / Chen (2006).
Operationalization:
 The coefficient of variation (CVj = sj /j) captures the
variance from the distribution of the bids.
 Because the CV is scale invariant, it allows comparisons
between tenders.
Conclusion: The smaller the CV the more likely is the collusion.
8
III. The Variance Screen

9. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
9
III. The Variance Screen
Figure 1: Coefficient of Variation, Canton of Ticino
(Source: Imhof, 2017b)
 Clear-cut findings!

10. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
10
III. The Variance Screen
Figure 2: Coefficient of Variation, Canton of St. Gallen
(Source: Imhof, Karagök and Rutz, 2017)
 No findings!

11. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
Second experience:
 Bid riggers manipulate the auction by creating a notable difference between
the bid of the designated winner and the “cover bids”. We observe a notable
gap between the first and the second best bid (see black arrow in figure 3).
 However, the gaps between the cover bids are very small (see red arrows
in figure 3).
Figure 3: Typical Pattern
11
IV. Relative Distance Screen
510'000
520'000
530'000
540'000
550'000
560'000
570'000
580'000
0 1 2 3 4 5 6 7
CHF
Bids
We calculate the relative distance
test with the following formula:
Difference between the two lowest bids
Standard deviation of the cover bids
 If the “relative distance” > 1, the gap
between the first and second best bid is
greater than the gaps between the cover
bids indicating potential bid-rigging
activities.
 For figure 3, the relative distance is 4.97.

12. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
Why is this pattern representative of bid rigging?
 Price is not the only criterion in the tendering process.
 When bids are close to each other non-price criteria (technical
solution, quality or environmental aspects) may influence the award
of a contract and undermine bid-rigging conspiracy. But, in practice,
price is an essential criterion to award contracts.
 Witnesses in bid rigging cases have reported that members of bid
rigging cartels regularly make sure that the designated winning bid is
3-5% lower than the next best bid.
 Gap between the first and the second best bid is notable.
 Losing bids are close to each other because bidders do not want to
appear too expensive for procurement authorities.
 They do not want to tarnish their reputation by sending a negative
signal concerning their price level.
 Gaps between cover bids are small.
12
IV. Relative Distance Screen

13. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
Figure 4: The Relative Distance, Canton of Ticino
13
IV. Relative Distance Screen
(source: Imhof, 2017b)
 Clear-cut findings!

14. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
14
IV. Relative Distance Screen
Figure 5: The Relative Distance, Canton of St. Gallen
(source: Imhof, Karagök and Rutz, 2017)
 No findings!

15. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
V. Screening for Partial Collusion
 Our analysis so far indicates no systematic market embracing
collusion in the canton of St. Gallen (no all-inclusive cartel) but
Third experience:
 COMCO’s investigations involving bid rigging have revealed that bid-
rigging cartels may concern only a subset of contracts: firms collude
solely with regards to some tenders and not all tenders (partial
collusion).
Questions
1. (How) Can we identify the conspicuousness of tenders?
2. (How) Can we identify a subgroup / subgroups of firms from the
tenders identified as conspicuous?
3. (How) Can we identify a common denominator for the identified
subgroup / subgroups?
4. (How) Can we identify a collusive behaviour for the identified subgroup
/ subgroups?
15

16. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
16
V. Screening for Partial Collusion
Table 1: Identification of Conspicuous Contracts
Scenarios
Coefficient of
Variation
Relative
Distance
Measure
Number of
Projects
% of
Total
Sample
1 ≤ 0.03 > 1.30 38 13.5%
2 ≤ 0.05 > 1.15 65 23.1%
3 ≤ 0.06 > 1.00 80 28.4%
First Question: (How) Can we identify the conspicuousness of tenders?
Answer: Combination of both screens
 We target contracts which simultaneously exhibit a low coefficient of
variation and a high relative distance measure.
 Results from previous investigations (Ticino and Aargau) serve to choose
the threshold values for both screens.
 Both screens capture different aspects of the price setting behaviour of
colluding firms and are not correlated.

17. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
V. Screening for Partial Collusion
Second Question: (How) Can we identify a subgroup / subgroups of
firms from the tenders identified as conspicuous?
Answer: We use two criteria:
 First criterion: Frequent participation in conspicuous tenders
 Second criterion: Regular simultaneous participation of the same
firms in conspicuous tenders (see table below)
17
Table 2: Interaction between Firms in Conspicuous tenders
Firm 1 2 3 4 5 6
1 15 2 8 5 1 4
2 17 14 16 9 15
3 45 18 11 17
4 23 12 19
5 14 12
6 20
Results:
 First criterion: from 138 to 17 firms
 Second criterion: from 17 to 6 firms;
Identification of a homogeneous
subgroup

18. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
V. Screening for Partial Collusion
Results:
 We find that the six firms identified in the subgroup are jointly active in
region A and E (neighbouring regions). Region A is the common
denominator of the identified subgroup.
18
Table 3: Regional Analysis (number of tenders won)
Firm
Region:
A B C D E F G H
1 5 (1) 1 (0) 2 (1) 1 (0) -- -- 4 (0) 3 (2)
2 13 (3) -- -- -- 4 (0) -- -- --
3 17 (3) 8 (3) 4 (0) -- 6 (2) 2 (0) 10 (4) 3 (1)
4 18 (5) -- -- -- 5 (0) -- -- --
5 13 (5) 1 (0) -- -- 2 (0) -- -- --
6 16 (5) -- -- -- 4 (0) -- -- --
Third Question: (How) Can we identify a common denominator for the
identified subgroup / subgroups?
Answer:
 We analyze the geographic activities of the six firms identified in the
subgroup. The analysis contains only the conspicuous tenders.

19. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
V. Screening for Partial Collusion
19
 To sum up, the geographical analysis largely validates the results from
the identification of the homogenous subgroup.
 It raises suspicions concerning a local bid rigging cartel operating in
region A.
 Additional results:
 Few firms active in region A for road construction.
 Geographic distance between region A and the other regions:
transportation costs play a major role in the construction industry.
 Political foreclosure: region A borders several other cantons and
such political frontiers may limit market access for potential
competitors.
 Analyses of structural screens (number of bidders per tender and
size of contract) reveal no difference between region A and the
other regions.

20. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
Fourth question: (How) Can we identify a collusive behaviour in
the identified subgroup?
Answer:
 Cover-Bidding Screen (next slides)
20
V. Screening for Partial Collusion

21. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
21
VI. Cover-Bidding Screen
Background:
 Cover bids are typically calculated to ensure that the designated
winner really wins the contract: The designated winner submits a
“low” bid, all other firms submit a deliberately “high” bid.
 In the absence of side payments, a cover bidding behaviour
typically involves a rotation element.
 To test for cover bidding, we analyse the (pairwise) bidding
behaviour of the suspect firms and the interaction of the suspect
firms within the identified subgroup.
 To ensure comparability, bids need to be normalized:
bij = the bid of firm i in submission j
bj,min (bj,max) = the lowest (highest) bid in submission j
 For each tender, the winning bid always gets assigned the value
0 while the highest bid gets assigned the value 1.
 1,0
min,max,
min,
; 



jj
jij
ijnorm
bb
bb
b

22. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
22
VI. Cover-Bidding Screen
Figure 6: Theoretical Illustration Pairwise: Two firms
Three zones:
• Grey zones: Non-competitive; One
wins while the other keeps distance
• Red zone: Non-competitive; both keep
distance favouring an other winner
• White zone: competitive; either one or
both are in rivalry with winning bid
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
Firm2
Firm 1
”non-competitive areas”
Take the point A (0.8;0)
Winner bid (firm 2) = 0
Bid firm 1= 0.8 ==>
Distance to winner = 0.8 high! (max. = 1)
A
Take the point B (0.7;0.9)
Winner bid = 0 (firm x, x ≠ 1, 2)
Bid firm 1 = 0.7
Bid firm 2 = 0.9 ==>
Distance to winner = 0.7 and 0.9 high!
B
Hypothesis: Accumulation of points in the red and grey areas suggests
cover bidding.

23. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
23
VI. Cover-Bidding Screen
Figure 7: Cover Bidding Screen for the Cartel Period, Ticino
(source: Imhof; 2017b)

24. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
24
VI. Cover-Bidding Screen
Figure 8: Cover-Bidding Screen for the Post-Cartel Period, Ticino
(source: Imhof; 2017b)

25. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
25
VI. Cover-Bidding Screen
Figure 9: Cover-Bidding Screen for the Suspect Group of Firms, St. Gallen
(source: Imhof, Karagök and Rutz; 2017)
 The results indicate that the identified subgroup operates
like a bid-rigging cartel with cover-bidding mechanism.

26. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
26
VII Discussion & Conclusion
Are simple screens appropriate to detect bid-rigging cartels?
 Simple screens produce good results for the cases of Ticino and of
St. Gallen (see Imhof, Karagök and Rutz, 2017; Imhof, 2017b).
 Imhof and Huber (2018) combine the screening method with
machine learning. Using four bid-rigging cases (among them Ticino
and St. Gallen), they show that simple screens perform well to
detect bid-rigging cartels. The detection method based on simple
screens exhibits a very high prediction rate for collusion: they are
able to detect 91% of the collusive tenders.
Conclusion:
 Simple screens produce reliable results at least for the construction
industry in Switzerland.

27. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
27
VII Discussion & Conclusion
Are simple screens suitable as detection method for competition
agencies?
 Low data requirement:
 Simple screens use only data on bids (winning and losing bids),
which is available at procurement agency or statistical office, if
they are not publicly available.
 However, the size of the sample must be large enough.
 Easy to understand even for non-economists
 Easy to implement:
 Simple screens solely require the calculation of descriptive
statistics per tender.
Conclusion:
 Simple screens are suitable for competition agencies.

28. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
28
VII Discussion & Conclusion
Are there other methods to detect bid-rigging cartels?
 Yes. The method of Bajari and Ye (2003) is the most widely used
detection method applied to bid-rigging cartels and proposed two
econometric tests.
 BUT it produces too many false negative results when applied to
the Ticino case. Imhof (2017a) finds that 68% to 89% of the pairs
pass the tests even though they should fail.
 AND the method proposed by Bajari and Ye (2003) needs
information on costs, which is difficult, if not impossible, to obtain
when trying to detect bid-rigging cartels.
Conclusion:
 The method of Bajari and Ye (2003) is surely more data-intensive
and time-consuming than simple screens.
 For the Ticino case, it produces less reliable results than the simple
screens.

29. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
29
VII Discussion & Conclusion
How far can we generalize the use of simple screens: Can we
apply simple screens to other markets?
 Generalisation is possible if the context is equivalent:
 First-price sealed-bid auction where price is the most important
criterion to award contracts;
 Absence of side payments;
 Repeated and sustainable bid-rigging cartels.
 Five different simple screens:
 The use of different screens help to capture different possible
forms of bid manipulation (see Imhof, 2017b; Imhof and Huber,
2018).
 Use of prior information from competition agencies:
 Competition agencies can examine the validity of the simple
screens with former cases in other markets, and adapt them if
necessary.

30. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
30
VII Discussion & Conclusion
Can firms beat the simple screens?
 The risk exists that firms try to beat the simple screens.
 However!
 Beating the simple screens implies more coordination from firms;
more coordination means competition agencies can find more
hard evidence.
 As soon as competition agencies know how firms try to beat the
simple screens, they can adapt them.
 Deterrence effect caused by the use of the simple screens
destabilizes existing bid-rigging cartels.
Conclusion:
 Even though it is theoretically possible that some bid-rigging cartels
beat the simple screens, we do not observe it in practice. We believe
that a greater share of bid-rigging cartels would be detected or
deterred by the use of simple screens.

31. OECD Workshop on Cartel Screens, Paris, 30th January 2018
David Imhof and Yavuz Karagök, Swiss Competition Commission
31
VIII Literature and Further Readings
1. Athey, S., K. Bagwell and C. Sanchirico (2004) Collusion and Price Rigidity, Review of Economic Studies, Vol. 71, pp.
317-349.
2. Abrantes-Metz R., L. Froeb, J. Geweke and C. Taylor (2006) A Variance Screen for Collusion, International Journal of
Industrial Organization, Vol. 24, pp. 467-486.
3. Bajari, P. and L. Ye (2003) Deciding Between Competition and Collusion, The Review of Economic and Statistics, Vol. 85,
Nr. 4, pp. 971-989.
4. Harrington J.E. and J. Chen (2006) Cartel Pricing Dynamics with Cost Variability and Endogenous Buyer Detection,
International Journal of Industrial Economics, Vol. 24, pp. 1185-1212.
5. Harrington, J.E. (2008) Detecting Cartels. In: Buccirossi P (ed) Handbook of Antitrust Economics, MIT Press.
6. Imhof D., Y. Karagök and S. Rutz (2017) Screening for Bid-rigging: Does it Work? Working Paper, available at:
(http://crese.univ-fcomte.fr/WP-2017-09.pdf).
7. Imhof D. (2017a) Econometric Tests to Detect Bid-rigging Cartels: Does it Work? Working Papers SES 483, Faculty of
Economics and Social Sciences, University of Fribourg (Switzerland), available at:
(http://doc.rero.ch/record/288965/files/WP_SES_483.pdf).
8. Imhof D. (2017b) Simple Statistical Screens to Detect Bid Rigging, Working Papers SES 484, Faculty of Economics and
Social Sciences, University of Fribourg (Switzerland), available at:
(http://doc.rero.ch/record/289133/files/WP_SES_484.pdf).
9. Imhof D., M. Huber (2018), Combining Screening and Machine Learning to Detect Bid-rigging Cartels, Forthcoming as
Working Paper SES.
10. Jiménez, J.L. and J. Perdiguero (2012) Does Rigidity of Prices Hide Collusion?, Review of Industrial Organization, Vol.
41(3), pp. 223-248.
11. OECD (2014), Roundtable on ex officio Cartel Investigations and the use of screens to detect cartels, Background Note
by the Secretariat.
12. Porter, R.H. and J.D. Zona (1993) Detection of Bid Rigging in Procurement Auctions, Journal of Political Economy, Vol.
101, Nr. 3, pp. 518-538.
13. Porter, R.H. and J.D. Zona (1999) Ohio School Milk Markets: An Analysis of Bidding, Rand Journal of Economics, Vol. 30,
Nr. 2, pp. 263-288.