Detecting Bid-Rigging in Public Procurement. a Cluster Analysis Approach

administrative sciences

Article Detecting Bid-Rigging in Public Procurement. A Cluster Analysis Approach

Mihail Busu 1,* and Cristian Busu 2

1 Faculty of Business Administration in Foreign Languages, Bucharest University of Economic Studies, 010374 Bucharest, Romania 2 Faculty of Management, Bucharest University of Economic Studies, 010374 Bucharest, Romania; [email protected] * Correspondence: [email protected]

Abstract: This paper analyses the public procurement auctions for snow removal contracts to find out whether bid-rigging occurred. Due to the limited participation in the auction processes, detection of anticompetitive agreements was possible. The econometric analysis used in our study supported the findings of a cartel agreement. Cluster analysis, statistical hypothesis, normality and symmetry and nonparametric tests reveal two types of auctions: competitive and noncompetitive bids. The aim of this paper is to analyze the public procurement auctions with nonparametric statistical methods. Our findings are in line with the literature in the field.

Keywords: bid-rigging; public procurement; hypothesis testing; cluster analysis; public administration

1. Introduction Bid rigging in public procurements is one of the main problematic aspects targeted by the governments and the local public authorities. Government procurements usually Citation: Busu, Mihail, and Cristian account for 10–15% of gross domestic product (GDP) in developing countries (Global Trade Busu. 2021. Detecting Bid-Rigging in Negotiations 2006) and up to 25% of GDP in the developed countries (Blume and Heidhues Public Procurement. A Cluster 2006). Local public authorities outsource their utilitarian projects and services through Analysis Approach. Administrative open-bid auctions, direct negotiation, or competitive biddings (Menezes and Monteiro Sciences 11: 13. https://doi.org/ 2006). Due to the public procurement legislation, in most cases, public procurements 10.3390/admsci11010013 procedures are conducted through competitive auctions (Bolotova et al. 2008). Collusion and corruption are the main factors of concern in the auctions (Porter Received: 29 December 2020 2005; Pesendorfer 2000; Connor 2001). Other studies (Harrington 2005; Carayannis and Accepted: 5 February 2021 Popescu 2005; Porrini 2015; Liao et al. 2003) conﬁrm that bidders concluded anticompetitive Published: 10 February 2021 agreements to increase the bidding price and get better contracts from the public authorities. Public procurements organized by the governmental bodies and the local authorities’ Publisher’s Note: MDPI stays neutral departments have many weaknesses, depicted in the economic literature, as follows: with regard to jurisdictional claims in published maps and institutional afﬁl- 1. Excessive state intervention, discrimination in awarding contracts, favoritism for local iations. contractors create substantial problems in the awarding process (Iri¸sand˙ Santos-Pinto 2013; Clancy 2018); 2. Lack of transparency in the public procurement contracts is a common practice within the government departments (McAfee and McMillan 1992; Stempel et al. 2018;

Copyright: © 2021 by the authors. Marshall and Marx 2007); Licensee MDPI, Basel, Switzerland. 3. Abuse of the state prerogatives and corruption in the public auctions (Previtali and This article is an open access article Cerchiello 2018; Kim et al. 2017); distributed under the terms and 4. Bidders form cartels to increase the auction price and get contracts at the auctioneers’ conditions of the Creative Commons costs (Banerjee et al. 2018; Olaussen et al. 2018; Otamendi et al. 2018). Attribution (CC BY) license (https:// Nevertheless, there are few articles reviewed in the empirical literature on bid-rigging creativecommons.org/licenses/by/ for public procurement auctions. Zona(1986) proposes and analyzes the bid-rigging in 4.0/).

Adm. Sci. 2021, 11, 13. https://doi.org/10.3390/admsci11010013 https://www.mdpi.com/journal/admsci Adm. Sci. 2021, 11, 13 2 of 14

the construction of the highways tendered in the public auctions, based on a multilinear regression model. McMillan(1991) describes the agreement between bidders of public works in Japan, while Howard and Kaserman(1989) evaluate the damages in bid-rigging cases with regression methods. Finally, Commander and Schankerman(1997) examined the scheme to submit identical bids within the public procurement auctions. In general, the detection of rigged tenders is based on two main components: structural screening (which addresses issues related to the specific features of the market and the tender) and behavioral screening (which addresses issues related to the behavior of tenderers). Structural screening is a predominantly preventive measure, but it can be applied both ex-ante and ex-post to the auction. Thus, the analysis of the market structure and the specificity of certain forms of auction organization can provide important information and indications about those markets or those types of purchases that are prone, by their specificity, to facilitate the auctioning of auctions. In developing the proposal for a set of indicators that could be considered for structural screening, both the indicators suggested by the OECD guide to combating auction fraud and those selected to build the aggregate competitive pressure index were considered. It is appreciated that the latter is relevant in assessing the ability of a market to facilitate or not the occurrence of such anticompetitive behavior. The role of structural screening is to act as a filter, which can lead to the identification of those tenders in industries or areas more exposed to the conclusion of agreements between competitors. Behavioral screening is performed ex-post during the auction and has the role of observing certain aspects that may suggest agreements between the participants in the auctions. Thus, certain behavioral patterns of bidders can be captured, related either to the way of bidding or to the subsequent development of contracts, which may be the result of agreements. Some of the indicators proposed by the OECD in the document entitled: “Guidelines for detection of bid-rigging in public procurement” were used for behavioral issues. Unlike structural screening, behavioral screening can lead to the identification of tenders with a high probability of fraud, the indicators considered may be, initially, indications of initiating investigations and, subsequently, even evidence, as appropriate. Traditionally, public procurement had only to be economically efficient, with little regard for other objectives than the purely economic ones. In recent times, however, due to a more general ascension of the sustainable development concept, governments have been put in the position to “lead by example” and use their purchasing power to advance the goals of sustainable development; as a specific development, sustainable public procurement has been slowly creeping in. From “secondary considerations” in the 2004 Directives (Arnould 2004; Caranta et al. 2013; Caranta 2016), the need to include social and environmental considerations in public tendering procedures has led to the coining of new terms, much more powerful and all-encompassing, such as “horizontal policies” (Kunzlik and Arrowsmith 2009; Arrowsmith and Kunzlik 2009; Comba 2010), “sustainable procurement” or even “strategic procurement”. We can state that with the new 2014 Directives, the sustainability paradigm is almost taking over the realm of public procurement, and it is marketed as a major “selling point” of the new legislation. With strict reference to Romania, the special law is dedicated to the regulation of public procurement, which—among others—compels the elaboration of a national plan in this respect, with concrete objectives, but also to “the introduction in the process of public procurement of environmental protection criteria that would allow the improvement of services’ quality and optimization of costs with public procurement in the short, medium and long term” (Government of Romania 2006). However, based on the studies undertaken, we can say that a rethinking of the legal norms pertaining to the public procurement field, in accordance with the best European practices, is meant to lead to the attainment of some higher performance thresholds to increase achievements. The national legal framework related to the GPP system was extremely limited at that time, consisting of (i) the national strategy for sustainable development, horizons 2013– Adm. Sci. 2021, 11, 13 3 of 14

2020–2030 (Government of Romania 2008) and the emergency ordinance, which regulated the awarding of public procurement contracts (Government of Romania 2014). Before stating the expected changes, one should mention that the chapter “Environment” of the government program 2013–2016 (Parliament of Romania 2016) contains some commitments regarding public procurement, among which their encouragement by adopting a specific action plan, aimed at “promoting the models of sustainable production and consumption”, the development of clean and environmentally friendly technologies, setting efficient criteria for public procurement, recalling on the need to inform/raise the awareness of the authorities in the same field. This article presents a case study of an anticompetitive agreement put in place by the snow removal operators in Romania. We evaluate the bids during 2015–2017 and make statistical analysis between the starting prices and the bidding prices. The structure of the paper is the following. First, we introduce the statistical methods used in our analysis. Then, we make a description of the bids and test the hypothesis. In the Results section, we make a split of the data in two clusters, from the perspective of competition. The Discussion section presents conclusions of the analyses and underlines limitations of the study and future research.

2. Materials and Methods To analyze the existence of a cartel on a given market, an adequate quantitative method is represented by the statistical analysis of the auction time-series from that market (Connor 2001; Tenorio 1993). In this case, it is necessary to use more variables, not just starting and bid awarding price. Examples of such variables are sales capacities, transportation, experience, etc. The analytical method to detect bid-rigging was the cluster method, mainly used when we do not have a priori information above the existence of a cartel (Smith 1993; Imhof et al. 2018; Abrantes-Metz et al. 2012; Bu¸su 2012; Andrei and Busu 2014). The first step was to divide bids into several categories, which differ significantly from one another. Thus, it was assumed that there were only two ways to bid—collusive and competitive—which means we were likely to have 2 clusters that differ significantly from each other. The first cluster with the auctions resulting from the struggle between firms to obtain the respective good, and the other involved auctions in which the participants had concluded various agreements between themselves (so-called rigged bids). To perform this analysis, we started from the ratio between the bidding price and starting price of an auction. Detecting possible anticompetitive agreements was achieved in the following steps: 1. Performing a cluster analysis. This was done by dividing the data into two groups according to the ratio defined above. The first cluster included those auctions with a high ratio (when the sale price was close to the starting price), while the second cluster included auctions with a low ratio (sale price was significantly lower than the starting price) likely to be the result of competitive behavior. 2. Applying nonparametric tests. These tests helped us check whether the two clusters were significantly different in terms of distribution from a statistical point of view. It also tested the statistical assumptions that there were significant differences between the two bidding modalities. Thus, if there were collusive behaviors, two different bidding models should have resulted. In addition, we needed to do a box-plot test to analyze the difference between the average of the two data series. 3. Testing for normality and symmetry. Normality and symmetry series data of the two groups was tested by Kolmogorov–Smirnov test. This test compared the probabilistic distribution of the two clusters with a certain theoretical distribution. In general, the normal distribution was selected for theoretical distribution. Prior, considering the selection of the two clusters, we expected the first cluster (for which the ratio was high) to have a normal distribution, while the second cluster (for which the ratio was low) had an asymmetric distribution. This result would confirm the hypothesis that Adm. Sci. 2021, 11, x FOR PEER REVIEW 4 of 14

Adm. Sci. 2021, 11, 13 selection of the two clusters, we expected the first cluster (for which the ratio4 was of 14 high) to have a normal distribution, while the second cluster (for which the ratio was low) had an asymmetric distribution. This result would confirm the hypothesis that thethe cluster cluster for for which which the the ratio ratio was was low low corre correspondssponds to to the the situation of competition, whilewhile the the second second cluster cluster would would correspond correspond with with the thesituation situation where, where, among among the par- the ticipants,participants, there there was was a collusive a collusive agreemen agreementt (because (because anticompetitive anticompetitive agreements maintainmaintain an an extremely extremely low low bidding bidding level level close close to to the the starting starting price). price). 4. TestingTesting statistical hypotheses.hypotheses. TestingTesting statisticalstatistical assumptions assumptions were were performed performed to to see see if ifthere there were were significant significant differences differences between between different different data data categories.categories. InIn ourour case, we testedtested if if there there were were significant significant difference differencess between between the the bids bids in consecutive years. National Company forfor HighwaysHighways and and National National Roads Roads of of Romania Romania (CNADNR) (CNADNR) has has the theresponsibilities responsibilities of exploitation, of exploitation, permanent perman maintenance,ent maintenance, modernization modernization and and development develop- mentof national of national road networkroad network and highway and highway day on day the on territory the territory of Romania. of Romania. CNADNR CNADNR has its hasstructure its structure 7 subunits 7 subunits without without legal personality, legal pers calledonality, the called Regional the DirectorateRegional Directorate of Roads and of RoadsBridges and (DRDP), Bridges located (DRDP), in Bucharest, located in Craiova, Bucharest, Iasi, Craiova, Cluj, Timisoara, Iasi, Cluj, Constanta, Timisoara, Brasov. Con- stanta,Hereinafter, Brasov. we Hereinafter, refer to them we as refer SDN1, to them SDN2, as etc. SDN1, SDN2, etc. The authors focused on the snow removal sect sectoror in Romania as it is one of the most importantimportant sectors sectors with with a a high high percentage percentage impact impact on on national national GDP. GDP. The data were were collected collected fromfrom the CNADNR’s website. The The data data repres representent all snow removal bids during 2015 and 2017. During thethe analyzedanalyzed period, period, respectively respectively the the winter winter seasons seasons 2015–2016 2015–2016 and 2016–2017and 2016–, 2017,modalities modalities of awarding of awarding public public procurement procurement contracts contracts used used by by DRDP DRDP werewere tendering procedures and negotiation without prior publ publicationication of a contract notice, and the award criterion was the lowest price. DRDP has concluded framework agreements, subsequent agreements, service service contracts contracts and and additional acts, as their object was services routine maintenance in winter of the nati nationalonal roads under their management. From the the information information provided provided by by CNADNR, CNADNR, a comparative a comparative analysis analysis between between contrac- con- tualtractual and actual and actual prices prices per km per removed km removed snow wa snows performed was performed (see Figure (see 1). FigureIt was 1found). It was that therefound are that significant there are differences significant between differences them between both at them the sa bothme season at the sameas well season between as well the twobetween seasons. the two seasons.

9 12 7.50 8 10.13 10 7 6.26 6.41 6 5.56 8 4.90 5 4.36 4.52 6.03 6 4.94 4.93 euro 4 euro 4.58 4.55 2.54 3.78 3 2.19 4 3.08 2.89 2.0 2.03 1.84 2.57 2.53 2 1.52 1.60 2.27 1.95 2.26 2 1 0 0 SDN 1 SDN 2 SDN 3 SDN 4 SDN 5 SDN 6 SDN 7 SDN 1 SDN 2 SDN 3 SDN 4 SDN 5 SDN 6 SDN 7 Contractual Price 2015-2016 Effective Price 2015-2016 Contractual Price 2016-2017 Effective Price 2016-2017

Figure 1. Comparison betweenbetween contractualcontractual price price and and effective effective price price per per kilometer, kilometer, in euro,in euro, during during 2015–2016 2015–2016 and and 2016–2017. 2016– Source:2017. Source: own processingown processing using using Statistical Statistical Package Packag for thee for Social the Social Sciences Sciences (SPSS) (SPSS) 23 software. 23 software.

The data revealed that the large differencesdifferences between the values of contracts for the provision of the current winter maintenancemaintenance of national roads and motorways and the amounts actually actually paid paid for for them them were were because because the the value value of the of thecontracts contracts is established is established con- sideringconsidering the theduration duration of the of thewinter winter season season of approximately of approximately 5 months 5 months (15 (15 October–31 October– March);31 March); and and uniform uniform weather weather conditions conditions throughout throughout this this period. period. Statistical analysis was approached in two steps. In the ﬁrst stage, the increases were analyzed in statistical terms, i.e., percentage of the effective price paid/km, and the differences between the contract price and that actual price paid by the authorities for each

Adm. Sci. 2021, 11, 13 5 of 14

lot in the 2015–2017 period. The second step is an analysis of the percentage increase of the price actually paid per kilometer from year-to-year using the cluster method to identify a discontinuity of data on the procedures for awarding public procurement contracts. The result of the statistical analyses is presented in the Results section.

3. Results 3.1. Testing the Statistical Hypotheses The first test was to verify whether there are statistically significant differences between the prices awarded in 2015–2016 and those from 2016–2017, calculated as price/km. The statistical assumptions that were tested were as follows: H0 : µ1 = µ2 (there are no statistically significant differences between the 2015–2016 and 2016–2017 of the auctions in terms of the actual price paid per kilometer); Ha : µ1 6= µ2 (there are statistically significant differences between the 2015–2016 and 2016–2017 auctions in terms of the price actually paid per kilometer). Testing these statistical assumptions was performed using Student’s t-test. The results of this test are shown in Table1.

Table 1. Paired test.

95% Conﬁdence Interval Mean Std. Dev. Std. Error Mean of the Difference t df Sig. Lower Upper 2015–2016 vs. −2839.8 3988.25 608.2 −4067.26 −1612.5 −4.67 42 0 2016–2017 Source: own processing using Statistical Package for the Social Sciences (SPSS) 23 software.

From this table, it can be noticed that there are significant differences in bid data during the period 2015–2016 to the period 2016–2017 (t0.05,42 = −4.67, p-value = 0.000). This shows that the price paid per kilometer of removed snow has significantly increased between the two analyzed periods. To test from a statistical point of view, the differences between the two periods, regarding the contract price per skimmed and the actual price paid, an analysis was made both for the period 2015–2016 and for 2016–2017. The statistical assumptions that were tested were as follows: H0 : µ1 = µ2, µ3 = µ4 (there are no statistically significant differences between the contract price per kilometer paid for the 2015–2016 or 2016–2017 bids) Ha : µ1 6= µ2, µ3 6= µ4 (there are statistically significant differences between the contract price per kilometer paid for the 2015–2016 and 2016–2017 bids) Testing of these statistical hypotheses was carried out by Student’s t-test. The results of this test are shown in Table2.

Table 2. Paired sample test.

95% Conﬁdence Interval Mean Std. Dev. Std. Error Mean of the Difference t df Sig. Lower Upper Estimate_2015– 16,757.86 8621.47 1314.76 14,104.56 19,411.16 12.75 42 0.000 Effective_2016 Estimate_2016– 10,746.66 4838.09 684.21 9371.69 12,121.63 15.71 49 0.000 Effectiv_2017 Source: own processing using Statistical Package for the Social Sciences (SPSS) 23 software.

The above table indicates that there are statistically signiﬁcant differences between the contract price and the price actually paid for each kilometer of snow removed, both for the period 2015–2016 (t = 12.75, p-value = 0.000) and 2016–2017 (t = 15.71, p-value = 0.000). Adm. Sci. 2021, 11, x FOR PEER REVIEW 6 of 14

Adm. Sci. 2021, 11, 13 6 of 14

The above table indicates that there are statistically significant differences between the contract price and the price actually paid for each kilometer of snow removed, both for the periodIt 2015–2016 should be (t noted = 12.75, that p-value for SDN = 0.000) 5, the and contract 2016–2017 for (t the = 15.71, period p-value 2016–2017 = 0.000). was di- videdIt intoshould smaller be noted batches, that andfor SDN thus, 5, the the number contract of for degrees the period of freedom 2016–2017 (No. was of contracts divided intoawarded—1) smaller batches, differs 2016–2017 and thus, (df the = 49)number compared of degrees to 2015 toof 2016freedom (df = (No. 42). of contracts awarded—1) differs 2016–2017 (df = 49) compared to 2015 to 2016 (df = 42). 3.2. Cluster Analysis 3.2. ClusterAn appropriate Analysis method to analyze market discontinuities on a given auction market is the statisticalAn appropriate analysis method of the biddingto analyze time market series discontinuities on that market on (Hazak a given et al.auction 2016; Vadmarketász iset the al. 2016statistical). analysis of the bidding time series on that market (Hazak et al. 2016; Va- dász Anet al. analytical 2016). method for detecting such behaviors is the cluster method, predomi- nantlyAn used analytical when method there is for no detecting a priori information such behaviors on theis the existence cluster method, of collusive predominantly behaviors (Caldiero et al. 2010). The first step is to divide auctions into two categories with sig- used when there is no a priori information on the existence of collusive behaviors (Caldiero et nificant differences between them. Thus, it is assumed that there are only two ways to al. 2010). The first step is to divide auctions into two categories with significant differences bid—collusive and competitive—which means we are likely to have 2 clusters that differ between them. Thus, it is assumed that there are only two ways to bid—collusive and com- significantly from each other. petitive—which means we are likely to have 2 clusters that differ significantly from each other. To carry out this analysis, we start with the calculation of the percentage increases To carry out this analysis, we start with the calculation of the percentage increases between the prices which were actually paid by the contracting authority for snow removal between the prices which were actually paid by the contracting authority for snow re- kilometer in the analyzed period (see Figure2). moval kilometer in the analyzed period (see Figure 2).

20 17 15

10 6 55 4 5 22

Number of contracts of Number 0 <0% [0,20]% [20,40]% [40,60]% [60,80]% [80,100]% >100% Distribution of the percentage increases per km of snow

Figure 2. Distribution of the percentage increases per km of snow, 2016–2017 compared to 2015–2016 (actual price paid). Source: own processing using SPSS 23 software. Source: own processing using SPSS 23 software.

From the chart above, we observe that most snow removal contracts have an increase of approximately 20–40%.20–40%. SuchSuch (moderate)(moderate) increases increases could could be be justified justified by by the the tightening tightening of ofweather weather conditions conditions in winterin winter 2016–2017 2016–2017 compared compared to 2015–2016.to 2015–2016. Detection of discontinuities in auction data is accomplished by dividing the data into twotwo groups groups according to the percentage increase of the paid amount per snowy kilometer from the two time periodsperiods defineddefined above.above. The The first first cluster contains those auctions for which the percentage increase is significant, significant, while the second includes auctions for which thisthis ratio is low or even negative (see FigureFigure3 3)(the) (the bidbid priceprice inin 2016–20172016–2017 isis lowerlower thanthan inin 2015–2016). A cluster analysisanalysis waswas performed performed in in SPSS, SPSS, splitting splitting the the auctions auctions that that took took place place in two in twoclusters. clusters. This wasThis donewas todone test to whether test whether there is athere group is ofa group auctions offor auctions which thefor percentagewhich the percentageincreases in increases contract valuesin contract are substantiallyvalues are substantially higher in 2016–2017 higher in 2016–2017 than in 2015–2016 than in 2015– (over 201650%). (over 50%). First, some data for which the analysis would have been distorted was was removed. removed. These correspond to certain changes in the contractualcontractual conditions in SDL countycounty 33 andand 4.4.

Adm. Sci. 2021, 11, 13 7 of 14

Figure 3

Figure 3. Box-plot cluster. Source: own processing using SPSS software.

From this figure, it can be clearly seen, splitting into two clusters of auctions. The first cluster, located on the left side, includes 14 auctions (34% of the total) with high percentages previously defined, ranging from 50% to 126%. Thus, the price per kilometer of snow removal is significantly higher in the 2016–2017 period compared to 2015–2016. Moreover, the standard deviation of the data series on these prices is 0.32. Regarding the second cluster, comprising 27 auctions (13% of total auctions), the percentage increase was in this case from −63% to 37%. The value of the percentage increments for the second cluster has a much greater variation (the distribution normality test is done at the next step). A descriptive statistics of the two clusters could be seen in Table3.

Table 3. Descriptive statistics.

N Range Minimum Maximum Mean St. Dev. Cluster 1 14 1.26 0.49 1.75 0.7736 0.3227 Cluster 2 27 1 −0.63 0.37 0.1526 0.26107 Source: own processing using SPSS 23 software. 1

Now we apply the nonparametric test. To test whether the distribution of each cluster is normal, the Kolmogorov–Smirnov and Shapiro–Wilk tests were performed to test the normality of a series of data. This test was applied to each cluster, and the results can be seen in Table4.

Table 4. Test of normality.

Kolmogorov–Smirnov a Shapiro–Wilk Statistic df Sig. Statistic df Sig. Cluster1 0.28 14 0.004 0.741 14 0.001 Cluster2 0.156 14 0.2 0.905 14 0.132 a Lilliefors signiﬁcance correction.

The conclusion of these tests is that the data of the ﬁrst cluster is not normally distributed (Sig = 0.004 < 0.05, Sig = 0.001 < 0.05), while the data of the second cluster have a normal distribution (Sig = 0.200 > 0.05, Sig = 0.132 > 0.05). This indicates a discontinuity Adm. Sci. 2021, 11, 13 8 of 14

Adm. Sci. 2021, 11, x FOR PEER REVIEW 8 of 14

of data by the existence of a “gap” between the first and the second cluster. Auctions that correspond to the first cluster, those whose increase in the actual price paid in winter 2016–2017Now, we compared test the to normality 2015–2016 and was symmetry. between 49% This and step 175%, is closely would related require to a the more previous one.in-depth We analysis.tested the symmetry of the two distributions to check for differences between comparedNow, weand test theoretical the normality models. and symmetry.In theory, Thisthe steppercentage is closely increases related to due the previousto collusive be- haviorsone. We follow tested an the asymmetric symmetry of distribution, the two distributions while the to distribution check for differences of symmetric between percentage compared and theoretical models. In theory, the percentage increases due to collusive be- increases is related to symmetric distributions. For this, we calculated the coefficient of haviors follow an asymmetric distribution, while the distribution of symmetric percentage symmetryincreases is (skewness) related to symmetric for the two distributions. clusters and For the this, kurtosis we calculated of distribution. the coefficient The results of can besymmetry seen in Table (skewness) 5. for the two clusters and the kurtosis of distribution. The results can be seen in Table5. Table 5. Descriptive statistics. Table 5. Descriptive statistics. Skewness Kurtosis Statistic Skewness Std. Error Statistic Kurtosis Std. Error Cluster 1 2.365Statistic Std. 0.597 Error Statistic 6.655 Std. Error 1.154 ClusterCluster 2 1−1.95 2.365 0.5970.448 6.655 3.384 1.154 0.872 Cluster 2 −1.95 0.448 3.384 0.872 Source: own processing using SPSS 23 software. Source: own processing using SPSS 23 software.

FromFrom thethe tabletable above, above, we we observe observe that that data data from from the ﬁrstthe clusterfirst cluster is asymmetrical, is asymmetrical, meaningmeaning that thisthis cluster cluster auction auction tends tends to concentrateto concentrate at the at maximum the maximum value, value, which maywhich may correspondcorrespond to a collusive behavior,behavior, while while data data at at the the second second cluster cluster has has a symmetrica symmetric distribution,distribution, corresponding corresponding to to a a possible possible non- affectedaffected auction. auction. 3.3. Box-Plot Analysis 3.3. Box-PlotThe data Analysis discontinuity described above can be noted in Figure A2 (AppendixA). The clusterThe data analysis discontinuity in SPSS 23 described shows the above following can be distribution noted in Figure in the two A2 (Appendix clusters (see A). The clusterFigure 4analysis). in SPSS 23 shows the following distribution in the two clusters (see Figure 4).

FigureFigure 4. Box-plot cluster.cluster. Source: Source: own own processing processing data data in SPSS in SPSS 23. 23. Normality tests for the two clusters can be seen in Table6. NormalityThe conclusion tests of for these the tests two is clusters that the data can ofbe the seen first in cluster Table are 6. normally distributed (Sig = 0.000 < 0.05), while data for the second cluster do not have a normal distribution Table(Sig = 6. 0.061 Test >of 0.05 normality.—Test Shapiro–Wilk, Sig. = 0.073 > 0.05—Kolmogorov–Smirnov test). This confirms that the difference between the two clusters is statistically significant. Kolmogorov–Smirnov a Shapiro–Wilk Statistic df Sig. Statistic df Sig. Cluster 1 0.145 17 0.00 0.897 17 0.00 Cluster 2 0.198 25 0.07 0.791 25 0.06 a Lilieforce significance correction. Source: own computation in SPSS 23.

Adm. Sci. 2021, 11, 13 9 of 14

Adm. Sci. 2021, 11, x FOR PEER REVIEW Now we focus on auctions for winter 2016–2017. The data discontinuity described 9 of 14 above can be seen in Figure A3 (AppendixA). The cluster analysis in SPSS shows the next split between the two clusters (see Figure5).

TableThe 6. Test conclusion of normality. of these tests is that the data of the first cluster are normally distributed (Sig = 0.000 < 0.05), while data for the second cluster do not have a normal distribu- a tion (Sig = 0.061 > 0.05—TestKolmogorov–Smirnov Shapiro–Wilk, Sig. = 0.073 >Shapiro–Wilk 0.05—Kolmogorov–Smirnov test). This confirmsStatistic that the difference df between Sig. the Statistic two clusters dfis statistically Sig. significant. ClusterNow 1 we focus 0.145 on auctions 17 for winter 0.00 2016–2017. 0.897 The data 17 discontinuity 0.00 described aboveCluster can 2 be seen 0.198 in Figure A3 25 (Appendix 0.07 A). The cluster 0.791 analysis 25 in SPSS shows 0.06 the next a Lilieforce signiﬁcance correction. Source: own computation in SPSS 23. split between the two clusters (see Figure 5).

Figure 5. Box-plot cluster. Source: own processing data in SPSS 23. Figure 5. Box-plot cluster. Source: own processing data in SPSS 23. Normality tests for the two clusters can be seen in Table7. Normality tests for the two clusters can be seen in Table 7. Table 7. Test of Normality. Table 7. Test of Normality. Kolmogorov–Smirnov a Shapiro–Wilk a Statistic dfKolmogorov–Smirnov Sig. Statistic dfShapiro–Wilk Sig. Cluster 1 0.15Statistic 40 df 0.025 0.939Sig. 40Statistic 0.033df Sig. ClusterCluster 2 1 0.1920.15 10 40 0.200 0.025 0.945 100.939 0.60640 0.033 a LilieforceCluster significance 2 correction.0.192 Source: own processing10 using SPSS0.200 23 software. 0.945 10 0.606 a Lilieforce significance correction. Source: own processing using SPSS 23 software. The conclusion of these tests is that the data for the second cluster are normally distributedThe conclusion (Sig. = 0.60 of > 0.05—Shapiro–Wilkthese tests is that the test, da Sigta = for 0.200 the > 0.05—Kolmogorov–Smirnovsecond cluster are normally dis- test) of the first cluster did not have a normal distribution (Sig. = 0.033 < 0.05—Shapiro– tributed (Sig. = 0.60 > 0.05—Shapiro–Wilk test, Sig = 0.200 > 0.05—Kolmogorov–Smirnov Wilk test; Sig. = 0.025 < 0.05—Kolmogorov–Smirnov test). This statistically confirms the test)discontinuity of the first in datacluster from did the not two have clusters. a normal distribution (Sig. = 0.033 < 0.05—Shapiro– Wilk Nowtest; Sig. we perform = 0.025 a< cluster 0.05—Kolmogorov–Smir analysis on the percentagenov test). increases This between statistically the price confirms per the discontinuitycontract kilometer in data in 2016–2017 from the comparedtwo clusters. to 2015–2016. The data discontinuity described aboveNow can we be seenperform in Figure a cluster A4 (Appendix analysisA on). Thethe clusterpercentage analysis increases in SPSS between shows the the next price per contractdivision kilometer in the two clustersin 2016–2017 (Figure compared6). to 2015–2016. The data discontinuity described above can be seen in Figure A4 (Appendix A). The cluster analysis in SPSS shows the next division in the two clusters (Figure 6).

Adm. Sci. 2021, 11, 13 10 of 14

Figure 6

Figure 6. Box-plot cluster. Source: own processing using SPSS software.

Normality tests for the two clusters can be seen in Table8.

Table 8. Test of Normality.

Kolmogorov–Smirnov a Shapiro–Wilk Statistic df Sig. Statistic df Sig. Cluster 1 0.234 28 0.000 0.834 28 0.000 Cluster 2 0.305 13 0.062 0.744 13 0.096 a This is a lower bound of the true signiﬁcance. Lilieforce signiﬁcance correction. Source: own computation in SPSS.

The conclusion of these tests is that the data of the second cluster is normally distributed (Sig. = 0.096 > 0.05—Shapiro–Wilk test; Sig. = 0.062 > 0.05—Kolmogorov–Smirnov), while the of the ﬁrst cluster do not have a normal distribution (Sig. = 0.000 < 0.05—Shapiro– Wilk test, Sig = 0.000 > 0.05—Kolmogorov–Smirnov test). The statistical tests performed in this section are proving the existence of two types of bid-rigging during the analyzed period: competitive and anticompetitive. 2 4. Discussion of the Results The existence of the two clusters indicates a certain “gap” or discontinuity in data. This discontinuity occurs at auctions where price growth is over 15% in 2016–2017 compared to 2015–2016. Most maintenance contracts for snow removal have a 20–40% increase in value in the 2016–2017 winter season compared to that of 2015–2016. Regarding the actual price paid per km of snow removal, the increases of percentages were set between 49% to 175% (in the analyzed period), which may not be economically reasonable. From the cluster analysis on the percentage increases between the actual price and the contractual price per designated km in the two analyzed seasons (winter 2015–2016 and winter 2016–2017), it follows that there is a category of contracts with a normal (Gaussian) value and another category of contracts that do not follow such data distribution. This indicates that auctions in the second cluster could raise some suspicions about the organization/performance of these award procedures. The lots/SDNs in this category are found in Figure A2 of AppendixA. Adm. Sci. 2021, 11, 13 11 of 14

We also note a discontinuity in the data as far as it concerns the contract prices per km of snow in the 2016–2017 season, compared to 2015–2016. This discontinuity appears in contracts whose value has increased by over 15% (with a maximum increase of 104%) (Figure A1 in AppendixA). This is a confirmation through the statistical analysis that there have been certain award procedures where the increase in the price paid per kilometer of snow can raise some suspicions about the organization/running of these award procedures. Lots/SDNs in this category can be found in Figures A1 and A2 of AppendixA. We may conclude, based on the testing of the statistical assumptions, that there was a significant difference between the contractual price and the actual per km of snow, both for the winter 2015–2016 and for winter 2016–2017 (Table2). In addition, there are significant differences in terms of both statistics and in terms of the actual price paid per km over the 2016–2017 season compared to 2015–2016 (Figures A3 and A4 from AppendixA). These results show that we may be in the presence of an anticompetitive practice/agreement/bid- rigging regarding the results of the procedures for the snow removal auctions for the period 2016–2017. The econometric analyses used in our study (Sections 3.1–3.3) supported the findings of a cartel agreement. Cluster analysis, statistical hypothesis, normality and symmetry and nonparametric tests reveal two types of auctions during the analyzed period: competitive and noncompetitive bids.

5. Conclusions This scientific research confirms the need to pay maximum attention to the procurement problem, for the reasons we referred to in the paper, in line with our wide review of the specialized literature. Incidentally, a clear institutional framework was adopted at the EU level, containing norms meant to bring significant improvements to the above-mentioned plan, which were predominantly transposed at the Member State level. However, as we have found by studying the reports of the relevant institutions, the margins of expectation are still high regarding the actual achievements. Second, when we discuss the issues related to the elaboration and implementation of national legal instruments aimed at stimulating public procurement, we come up with a whole series of critical issues. In a nutshell, we find that these instruments have not demonstrated enough efficiency in stimulating green procurement in the public sector. The statistical analyses underline the high probability of stating for prerequisites for alleged anticompetitive agreements between the undertakings which participated in public procurement auctions in the analyzed period. The analytical methods for detecting anticompetitive behaviors are often used by worldwide competition authorities in dealing with anticompetitive cases. The enterprises could claim compensation whenever they have been harmed by the existence of a cartel on their operational market. The use of analytical methods based on statistical data could be a method for observing certain anticompetitive behaviors on the market. By utilizing these methods, we are not able to directly prove the collusive behavior of the analyzed enterprises, but we could highlight the improbable results, which would require more careful attention. These methods aim primarily to avoid false-positive and false-negative results. A false-positive result states that there is an anticompetitive agreement on a given market, although it does not actually exist. False-negative results are those which state that there is not an anticompetitive agreement on a certain market, although this cartel really exists. Moreover, the use of these analytical methods: should have empirical support, be easily applied and not too costly to implement. There is a limitation of this study, which comes from the fact that only five regions are analyzed, and therefore further research should be extended to other regions. Another limitation of the research could be related to the fact that the number of companies that can provide the service object of the contract was slightly different between the two analyzed periods. Thus, further research should focus on other similar analyses to other types of bid-rigging in public procurements. Adm. Sci. 2021, 11, 13 12 of 14

Author Contributions: Conceptualization, M.B.; Data curation, M.B.; Formal analysis, M.B.; Funding Adm. Sci. 2021, 11, x FOR PEER REVIEW 12 of 14

Adm. Sci. 2021, 11, x FOR PEER REVIEWacquisition, C.B.; Methodology, M.B.; Project administration, C.B. and M.B.; Resources, C.B.;12 of Software,14 Adm. Sci. 2021, 11, x FOR PEER REVIEWM.B.; Supervision, C.B.; Validation, C.B. and M.B.; Writing—original draft, M.B.; Writing—review12 of 14 &

editing, C.B. All authors have read and agreed to the published version of the manuscript. Software, M.B.; Supervision, C.B.; Validation, C.B. and M.B.; Writing—original draft, M.B.; Writing— Funding:reviewSoftware, &This editing,M.B.; research Supervision, C.B. All received authors C.B.; nohaveValidation, external read and C.B. funding. agreed and M.B.; to the Writing—original published version draft, of the M.B.; manuscript. Writing— reviewSoftware, & editing,M.B.; Supervision, C.B. All authors C.B.; Validation,have read and C.B. agreed and M.B.; to the Writing—original published version draft, of the M.B.; manuscript. Writing— InstitutionalFunding:review & Thisediting, Review research C.B. Board All received authors Statement: no have external readNot andfunding. applicable. agreed to the published version of the manuscript. Funding: This research received no external funding. InformedInstitutionalFunding: Consent This Review research Statement: Board received Statement:Not no applicable.external Not applicable.funding. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. DataInstitutional Availability Review Statement: Board Statement:The data presentedNot applicable. in this study are available within the article. Informed Consent Statement: Not applicable. DataInformed Availability Consent Statement: Statement: The Not data applicable. presented in this study are available within the article. ConﬂictsData Availability of Interest: Statement:The authors The data declare presented no conﬂict in this of study interest. are available within the article. ConflictsData Availability of Interest: Statement: The authors The declaredata presented no conflicts in this of interest.study are available within the article. Conflicts of Interest: The authors declare no conflicts of interest. AppendixConflicts Aof Interest: The authors declare no conflicts of interest. Appendix A Appendix A Appendix A

Cluster 1 Cluster 2 Cluster 1 Cluster 2 Cluster 1 Cluster 2

-100% -50% 0% 50% 100% 150% 200% -100% -50% 0% 50% 100% 150% 200% -100% -50% 0% 50% 100% 150% 200% Figure A1. Percentage increase in actual prices paid 2016–2017 compared to 2015–2016. FigureFigure A1. A1.Percentage Percentage increase increase in in actual actual pricesprices paid paid 2016–2017 2016–2017 compared compared to 2015–2016. to 2015–2016. Figure A1. Percentage increase in actual prices paid 2016–2017 compared to 2015–2016.

Cluster 1 Cluster 2 Cluster 1 Cluster 2 Cluster 1 Cluster 2

0% 100% 200% 300% 400% 500% 600% 0% 100% 200% 300% 400% 500% 600% 0% 100% 200% 300% 400% 500% 600% Figure A2. Percentage increase in estimated prices per kilometer in 2015–2016 compared to prices actually paid. FigureFigure A2. Percentage A2. Percentage increase increase in in estimated estimated prices prices perper kilometerkilometer in in 2015–2016 2015–2016 compared compared to prices to prices actually actually paid. paid. Figure A2. Percentage increase in estimated prices per kilometer in 2015–2016 compared to prices actually paid.

Cluster 1 Cluster 2 Cluster 1 Cluster 2 Cluster 1 Cluster 2

0% 50% 100% 150% 200% 250% 300% 350% 400% 450% 0% 50% 100% 150% 200% 250% 300% 350% 400% 450% 0% 50% 100% 150% 200% 250% 300% 350% 400% 450% Figure A3. Percentage increase in estimated prices per kilometer of snow in 2016–2017 compared to prices actually paid. Figure A3. Percentage increase in estimated prices per kilometer of snow in 2016–2017 compared to prices actually paid. FigureFigure A3. Percentage A3. Percentage increase increase in in estimated estimated prices prices perper kilometerkilometer of of snow snow inin 2016–2017 2016–2017 compared compared to prices to prices actually actually paid. paid.

Adm. Sci. 11 Adm.2021 Sci., 2021, 13, 11, x FOR PEER REVIEW 13 of 1413 of 14

Cluster 1 Cluster 2

-70% -50% -30% -10% 10% 30% 50% 70% 90% 110%

FigureFigure A4. Percentage A4. Percentage increase increase in in estimated estimated prices prices perper kilometerkilometer of of snow snow inin 2016–2017 2016–2017 as compared as compared to 2015–2016. to 2015–2016.

ReferencesReferences Abrantes-Metz,(Abrantes-Metz Rosa M.,et al. Michael 2012) Abrantes-Metz, Kraten, Albert Rosa D. Metz,M., Michael and Gim Krat S.en, Seow. Albert 2012. D. Metz, Libor and manipulation? Gim S. Seow. 2012.Journal Libor of Banking manipulation? & Finance 36: 136–50.Journal of Banking & Finance 36: 136–50. Andrei,(Andrei Florin, and and Busu Mihail 2014) Busu. Andrei, 2014. Florin, Detecting and Mihail cartels Busu. through 2014. analyticalDetecting ca methods.rtels throughRomanian analytical Competition methods. Journal Romanian24. Competition Available online: https://heinonline.org/HOL/LandingPage?handle=hein.journals/rocmpj2014&div=7&id=&page=Journal 24. Available online: https://heinonline.org/HOL/LandingPage?handle=hein.journals/rocmpj2014&div=7&id=&page= (accessed on 15 December 2020).(accessed on 15 December 2020). (Arnould 2004) Arnould, Joel. 2004. Secondary policies in public procurement: The innovations of the new directives. Public Arnould, Joel. 2004. Secondary policies in public procurement: The innovations of the new directives. Public Procurement Law Review Procurement Law Review 13: 187–97. 13:(Arrowsmith 187–97. and Kunzlik 2009) Arrowsmith, Sue, and Peter Kunzlik, eds. 2009. Social and Environmental Policies in EC Procurement Arrowsmith,Law: Sue, New and Directives Peter and Kunzlik, New Directions eds. 2009.. Cambridge:Social and Cambridge Environmental University Policies Press. in EC Procurement Law: New Directives and New Directions(Banerjee. et Cambridge: al. 2018) Banerjee, Cambridge Priyodorshi, University Shashwat Press. Khare, and Priyodorshi Srikant. 2018. Learning to Set the Reserve Price Banerjee, Priyodorshi,Optimally in Shashwat Laboratory Khare, First Price and PriyodorshiAuctions. Games Srikant. 9: 79. 2018. Learning to Set the Reserve Price Optimally in Laboratory First Price(Blume Auctions. and HeidhuesGames 2006)9: 79. Blume, [CrossRef Andreas,] and Paul Heidhues. 2006. Private monitoring in auctions. Journal of Economic Theory Blume, Andreas,131: 179–211. and Paul Heidhues. 2006. Private monitoring in auctions. Journal of Economic Theory 131: 179–211. [CrossRef] Bolotova,(Bolotova Yuliya, et John al. 2008) M. Connor, Bolotova, and Yuliya, Douglas John J.M. Miller. Connor, 2008. and The Doug impactlas J. Miller. of collusion 2008. The on priceimpact behavior: of collusion Empirical on price results behavior from: two recentEmpirical cases. International results from Journal two recent of Industrial cases. International Organization Journal26: of 1290–307. Industrial Organization [CrossRef] 26: 1290–307. (Buşu 2012) Buşu, Mihail. 2012. An economic analysis on the degree of market concentration: Competition Indicators. Paper Bu¸su,Mihail. 2012. An economic analysis on the degree of market concentration: Competition Indicators. Paper presented at 6th presented at 6th International Management Conference, Bucharest, Romania, November 15–16, pp. 529–37. International(Caldiero et al. Management 2010) Caldiero, Conference, Christopher, Bucharest, Maureen Taylor, Romania, and Lia November Ungureanu. 15–16; 2010. pp. Organizatio 529–37. nal and media use of technology Caldiero, Christopher,during fraud Maureen crises. The Taylor,Handbook and of Crisis Lia Ungureanu. Communication 2010. 396–409. Organizational doi:10.1002/9781444314885. and media use of technology during fraud crises. The(Caranta Handbook 2016) of Caranta, Crisis Communication Roberto. 2016. ,Public 396–409. procurement [CrossRef and] award criteria. In Research Handbook on EU Public Procurement Law. Caranta, Roberto.Cheltenham: 2016. Edward Public Elgar procurement Publishing. and award criteria. In Research Handbook on EU Public Procurement Law. Cheltenham: Edward(Caranta Elgar et al. Publishing.2013) Caranta, Roberto, Gunilla Edelstam, and Martin Trybus, eds. 2013. EU Public Contract Law: Public Procurement Caranta, Roberto,and Beyond Gunilla. Bruxelles: Edelstam, Primento. and Martin Trybus, eds. 2013. EU Public Contract Law: Public Procurement and Beyond. Bruxelles: Primento.(Carayannis and Popescu 2005) Carayannis, Elias G., and Denisa Popescu. 2005. Profiling a methodology for economic growth and Carayannis,convergence: Elias G., and Learning Denisa from Popescu. the EU 2005.e-procurement Profiling experience a methodology for central for and economic eastern European growth and countries. convergence: Technovation Learning 25: 1–15. from the (Clancy 2018) Clancy, Timothy. 2018. Application of Emerging-State Actor Theory: Analysis of Intervention and Containment EU e-procurement experience for central and eastern European countries. Technovation 25: 1–15. [CrossRef] Policies. Systems 6: 17. Clancy,(Comba Timothy. 2010) 2018. Comba, Application Mario E. of Emerging-State2010. XI. Green and Actor Social Theory: Considerations Analysis of in Intervention Public Procurement and Containment Contracts: Policies.A ComparativeSystems 6: 17. [CrossRefApproach.] The Law of Green and Social Procurement in Europe. 2: 299–319. Comba,(Commander Mario E. 2010. and XI.Schankerman Green and 1997) Social Commander, Considerations Simon, in Publicand Mark Procurement Schankerman. Contracts: 1997. Enterprise A Comparative restructuring Approach. and socialThe Law of Green andbenefits. Social Economics Procurement of Transition in Europe. 5: 1–24.2: 299–319. Commander,(Connor Simon, 2001) Connor, and Mark John M. Schankerman. 2001. Global Price 1997. Fixing Enterprise our Customers restructuring are the Enemy and. Norwell: social benefits.Kluwer AcademicEconomics Publishers. of Transition 5: 1–24. [CrossRef(Global Trade] Negotiations 2006) Global Trade Negotiations. 2006. Available online: https://stateinnovation.org/ (accessed on 30 Connor, JohnMarch M. 2001. 2020).Global Price Fixing our Customers are the Enemy. Norwell: Kluwer Academic Publishers. Global(Government Trade Negotiations. of Romania 2006. 2006) Available Government online: of Romania. https://stateinnovation.org/ 2006. Ordonanța de Urgen (accessedță No. 34 Privind on 30 Atribuirea March 2020). Contractelor de Achiziţie Publică, Bucharest, Romania. Government of Romania. 2006. Ordonanta de Urgentă No. 34 Privind Atribuirea Contractelor de Achizi¸tiePublică; Bucharest: Government (Government of Romania 2008) Government, of, Romania. 2008. Hotărârea No. 1460 Pentru Aprobarea Strategiei Naţionale Pentru of Romania.Dezvoltare Durabilă—Orizonturi 2013-20-2030. Bucharest: Guvernul României. Government(Government of Romania. of Romania) 2008. GovernmentHotărârea of No.Romania. 1460 2014. Pentru Expunere Aprobarea de Motive Strategiei la Legea Na¸tionalePentru Privind Achiziţ Dezvoltareiile Publice Verzi. Durabil Availableă—Orizonturi 2013-20-2030online: ;http://mmediu.ro/new/wp-content/uploads/2014/08/2014- Bucharest: Guvernul României. 08-26_Expun_motiv_Lege-APV.pdf (accessed on 12 January Government2021). of Romania. 2014. Expunere de Motive la Legea Privind Achizi¸tiilePublice Verzi. Available online: http://mmediu.ro/ new/wp-content/uploads/2014/08/2014-08-26_Expun_motiv_Lege-APV.pdf (accessed on 12 January 2021). Harrington, Joseph. 2005. Detecting Cartels, Advances in the Economics of Competition Law. Baltimore: Department of Economics, Johns Hopkins University. Hazak, Aaro, Tanel Kerikmäe, and Evelin Pärn-Lee. 2016. Efficiency and Bid Rigging in Simultaneous Procurement Auctions under an Oligopoly. International Journal of Procurement Management 13: 199–213. [CrossRef] Howard, Jeffrey H., and David Kaserman. 1989. Proof of damages in construction industry bid-rigging cases. Antitrust Bull 34: 359. Adm. Sci. 2021, 11, 13 14 of 14

Imhof, David, Yavuz Karagök, and Samuel Rutz. 2018. Screening For Bid Rigging—Does It Work? Journal of Competition Law & Economics 14: 235–61. Iri¸s,Doruk,˙ and Luís Santos-Pinto. 2013. Tacit collusion under fairness and reciprocity. Games 4: 50–65. [CrossRef] Kim, Eunji, Yoonhee Ha, and Sangheon Kim. 2017. Public Debt, Corruption and Sustainable Economic Growth. Sustainability 9: 433. [CrossRef] Kunzlik, Peter, and Sue Arrowsmith. 2009. Public Procurement & Horizontal Policies in EC Law–General Principles. Cambridge: Cambridge University Press. Liao, Shu-Hsien, Ching-Hsue Cheng, Wei-Bang Liao, and I-Liang Chen. 2003. A web-based architecture for implementing electronic procurement in military organizations. Technovation 23: 521–32. [CrossRef] Marshall, Robert C., and Leslie M. Marx. 2007. Bidder collusion. Journal of Economic Theory 133: 374–402. [CrossRef] McAfee, R. Preston, and John McMillan. 1992. Bidding rings. American Economic Review 82: 579–99. McMillan, John. 1991. Dango: JAPAN’S price-fixing conspiracies. Economics & Politics 3: 201–18. Menezes, Flavio M., and Paulo Klinger Monteiro. 2006. Corruption and auctions. Journal of Mathematical Economics 42: 97–108. [CrossRef] Olaussen, Jon Olaf, Are Oust, and Ole Jakob Sønstebø. 2018. Bidding Behavior in the Housing Market under Different Market Regimes. Journal of Risk and Financial Management 11: 41. [CrossRef] Otamendi, F. Javier, Isabelle Brocas, and Juan D. Carrillo. 2018. Sequential auctions with capacity constraints: An experimental investigation. Games 9: 15. [CrossRef] Parliament of Romania. 2016. Law No. 69 on Public Procurement. Available online: http://www.mmediu.ro/app/webroot/uploads/ files/2016-08-04_LEGE_Nr_69_din_2016.pdf (accessed on 5 July 2019). Pesendorfer, Martin. 2000. A study of collusion in first-price auctions. Review of Economic Studies 67: 381–411. [CrossRef] Porrini, Donatella. 2015. Risk classification efficiency and the insurance market regulation. Risks 3: 445–54. [CrossRef] Porter, Robert H. 2005. Detecting collusion. Review of Industrial Organization 26: 147–67. [CrossRef] Previtali, Pietro, and Paola Cerchiello. 2018. The Prevention of Corruption as an Unavoidable Way to Ensure Healthcare System Sustainability. Sustainability 10: 3071. [CrossRef] Smith, Anthony A., Jr. 1993. Estimating nonlinear time-series models using simulated vector autoregressions. Journal of Applied Econometrics 8: S63–S84. [CrossRef] Stempel, Peter, Isaac Ginis, David Ullman, Austin Becker, and Robert Witkop. 2018. Real-Time Chronological Hazard Impact Modeling. Journal of Marine Science and Engineering 6: 134. [CrossRef] Tenorio, Rafael. 1993. Revenue equivalence and bidding behavior in a multi-unit auction market: An empirical analysis. The Review of Economics and Statistics 75: 302–14. [CrossRef] Vadász, Pál, András Benczúr, Géza Füzesi, and Sándor Munk. 2016. Identifying Illegal Cartel Activities from Open Sources. In Open Source Intelligence Investigation. Cham: Springer, pp. 251–73. Zona, John Douglas. 1986. Bid-Rigging and the Competitive Bidding Process: Theory and Evidence. Ph.D. dissertation, State University of New York at Stony Brook, New York, NY, USA.