ADVERSE EFFECTS OF CORPORATE LENIENCY PROGRAMS IN VIEW OF INDUSTRY ASYMMETRY

Journal of Applied Econoic Sciences ADVERSE EFFECTS OF CORPORATE LENIENCY PROGRAMS IN VIEW OF INDUSTRY ASYMMETRY Evgenia MOTCHENKOVA, Daniel LELIEFELD VU University Asterda, NETHERLANDS, Dutch Antitrust Authority (NMa), NETHERLANDS eotchenkova@feweb.vu.nl, d.b.leliefeld@nanet.nl Abstract This paper studies the effects, leniency progras have on cartel stability and the subsequent abuse of arket power. A gae-theoretical odel, which allows for asyetry and retaliation, is eployed to analyze this proble. We find that a leniency progra does not always lead to a breach of trust; in certain industries leniency progras are unable to break collusion. Moreover, they ay have the adverse effect in the sense that they strengthen cartel stability or lead to abuse of arket power. A relatively large fir can use coercion to reove the option to a saller fir to self-report. In industries characterized by a certain degree of asyetry in arket shares and high exit costs this is an even ore likely scenario. In view of this liitation, policies aied at the reoval of the threat of retaliation need to be considered. This paper s ephasis is placed on leniency progras for cartels, but the line of reasoning ay also be extended to corporate whistle-blowing progras. Keywords: antitrust policy, antitrust law, self-reporting, leniency progras JEL Classification: K21, L41 1. Introduction Leniency progras for a relatively new feature of antitrust law enforceent. Its ain objective is to reove trust between cartel ebers. Trust is an essential eleent of every conspiracy. A siilar approach is successfully eployed in the prosecution of the afia (the so called witness protection progra ). In practice, contact is established between a eber of a conspiracy and the justice departent with a proposition to serve as a witness against its co-conspirators. As a reward the witness receives (partial) anesty fro its own isconduct and protection fro punishent by the other ebers of the (forer) crie syndicate. In the 80 s and 90 s of the previous century the witness protection progra proved to be a great success. Even though the trust between (faily) ebers of the afia was relatively strong the progra enabled the successful conviction of a great nuber of criinals. No wonder a siilar approach was adopted by the antitrust authorities in the fight against organized infringeents of copetition law. This paper exaines the consequences of the introduction and use of leniency progras in their attept to reove trust between cartel ebers. The ephasis of this paper is on leniency progras for cartels and the prosecution of price-fixing agreeents. The reasoning can, however, also be applied to corporate whistle blowing progras and witness protection progras. Several econoists have previously expressed their doubts as to the success of leniency progra s application in this new industrial setting. This literature is reviewed in Spagnolo (2008) and includes such contributions like Buccorossi and Spagnolo (2006) or Ellis and Wilson (2001). The ain intuition they provide in order to explain the existence of counterproductive effects of leniency progras is that in the fraework of repeated interactions leniency ay increase the firs ability to punish deviations (e.g. threatening to report to the antitrust authority after cheating occurs), thereby stabilizing the cartel agreeents by reducing gains fro defecting. In this way leniency can reduce the incentives to cheat and ake cartels ore stable. The current paper adds to these considerations by addressing another issue that has been underestiated in the design of leniency progras. An iportant eleent of the witness protection progra, which is issing in the design of leniency progras for cartels, is to safeguard the witness fro punishent by its forer partners in crie. We show that, using the realistic assuption of industry asyetry, the current design of leniency progras can t prevent larger firs fro using a threat of punishent as a eans of coercion, effectively not allowing saller firs to appeal for leniency. When the antitrust authority is unable to credibly protect leniency applicants fro retaliation by convicted cartel ebers the progra is abused by cartels. It actually serves to strengthen trust between its ebers. The progra ay have the adverse effect in the sense that it ay facilitate cartel stability. 114

Volue V/ Issue 2(12)/ Suer 2010 It should be stressed that any type of leniency progra should contain the following crucial eleents. Firstly, the criinal should provide sufficient evidence on the isconduct by its forer partners to the authorities. Secondly, the criinal acts as a witness and receives a, previously agreed to, lenient treatent with regard to the criinal s punishent. This ranges fro a reduction to a fine to cleency fro a prison sentence and can even entail a reward (see e.g. Kovacic 2006). A discretionary approach ensures the incentive is set according to the constraint faced by the proposed witness and no resources are wasted. Thirdly, the witness is protected fro punishent by its forer co-conspirators. When all policy paraeters of these eleents are custoized to fit the typical case, the proposed witness accepts the offer, the crie syndicate is terinated, its ebers are convicted, no resources are wasted and, perhaps ost iportantly, an exaple is set for those firs or individuals considering the prospect to conspire. Leniency progras for cartels differ fro witness protection progras in two ain aspects, first of all the protection aspect and secondly the fact that leniency progras are less custoized. Both parts of the policy instruent have, however, proven to be detriental to its success. In its new design and new environent these paraeters have unfortunately reained underexposed. It can be shown that, unless the necessity to protect the reporter fro punishent is acknowledged by the authorities, cartels ay strengthen their ties by eans of the leniency progra. Custoization of the progra with respect to the size of the fine and protection after self-reporting can partially overcoe this deficiency in the progra. More generally, the leniency progra will always need to be accopanied by the traditional law enforceent efforts of the copetition authority. A typical punishent strategy a fir ight consider eploying involves duping or aggressive pricing. Not only can a larger (ore efficient) fir usually establish lower arginal costs, the arket share asyetry also enables such a fir to establish a larger buffer than the saller fir. When the larger fir eploys its aggressive punishent strategy both firs incur losses. These eat away at both firs buffers. Since the larger fir s buffer is generally greater it will be able to sustain these losses for a longer period of tie. Eploying aggressive strategy for a sufficient period of tie will eventually push the saller fir into bankruptcy. This set-up resebles the long-purse story explanation of predatory pricing (see e.g. Stigler 1964 or Tirole 1988). The next point we want to address is the role of asyetry for cartel stability. The bigger part of the literature on leniency progras uses undiversified copanies and industry syetry as a starting point. The ain reason behind this is the general perception that asyetry reduces cartel strength. Leading textbooks such as Tirole (1988) state that, based on work by e.g. Chaberlin (1929) and Stigler (1964), the coordination towards a focal price under differentiated costs and products is ore difficult. Motta and Polo (2003) argue that asyetries between large and sall firs represent an obstacle for industry wide collusion. In reality firs are rarely truly syetrical in their cost functions, products or arket presence. Asyetry is the rule rather than the exception. Syetry would iply that all colluding firs apply for leniency at the sae tie. This is rarely the case. In general though syetry increases the incentive to collude, but this scenario is not realistic and the introduction of a leniency progra by an antitrust authority ight change the incentives for firs. Asyetry aong firs in products, arkets and cost functions is the subject of this paper. We odel asyetry by considering firs that have different arket shares due to e.g. historic reasons. The fact that a cartel can be stable under asyetrical conditions was recently proved by the EU Copetition Directorate General that uncovered a beer cartel in Holland and (parts of) Belgiu in which sall players coexisted with larger firs (see EC decision 2007, reference IP/07/509). An illustration of coercion aintained by the threat of retaliation can be found in the leniency application of British Petrol (BP) in the Bituen Cartel (see EC decision 2006, reference MEMO/06/324). During its existence the colluders anaged to increase trust between the cartel s ebers by designing a collective punishent strategy. Each tie a cartel eber violated the cartel s agreeents the other ebers agreed to punish the perpetrator. The cartel anaged to create a threat of retaliation by joining forces, using an asyetry of power, sustained by foral trust. The literature on analysis of self-reporting schees in antitrust starts with the paper by Motta and Polo (2003). They study a two-stage gae in which the AA first chooses once and for all its antitrust policy followed by the copetition phase in which the firs copete with each other, which is odelled as an infinitely-repeated oligopoly gae. The cartel adopts gri-trigger strategies in which cheating on the cartel by either setting a different price or applying for leniency triggers copetitive 115

Journal of Applied Econoic Sciences behaviour forever, while the cartel continues collusion as usual each tie it is caught by the AA. Under the optial antitrust policy, introduction of ex-post leniency progras will increase the chance of the cartel being captured, but ex-ante leniency progras that grant reduced fines are ineffective. As later shown in Spagnolo (2004) and Rey (2003), effective ex-ante leniency progras require substantial rewards. This is also recoended in Kovacic (2006). Spagnolo (2004) concludes that courageous leniency progras are closest to the optial. He uses a gae theoretical odel to relate a first best courageous leniency schee and a oderate leniency schee to a benchark case of traditional law enforceent. The courageous progra is one in which the reporting party is actually rewarded with a part of the fine paid by the other parties besides receiving anesty. In this way a first best solution is established according to Spagnolo. The above arguents are closely related to the ore general analysis of the optial structure and design of leniency progras that has been extensively discussed in the literature. See, for exaple, Rey (2003), Hinloopen (2003), Feess and Walzl (2004), Motchenkova (2004), Buccorosi and Spagnolo (2006), Spagnolo (2008), Harrington and Chen (2007), Chen and Rey (2007), Harrington (2008), and Houba et al. (2009). The question of optial design of leniency progras has two ain debatable coponents. They are the nuber of fine reductions and the size of fine reductions. The research norally coes to the conclusion that the currently applied oderate leniency progra could be ade ore effective by liiting a fine reduction to the first fir to report. Adding ore than one possibility to a fine reduction reduces the deterring effect of the schee. Most of the analysis also concludes that the first self-reporter should be fully exepted fro the fine. The current paper analyzes the effects of leniency progras on the survival of cartels fored by firs of different size (i.e. different arket shares). We analyze a setup where the bigger fir can threaten to force the saller fir to exit the arket by eploying aggressive strategy in case the saller fir self-reports. 1 The innovative aspect of the paper relates to considering how cartel ebers can react to the introduction of leniency progras in order to preserve the continuation of collusion. The punishent strategy for self-reporters is intended to counterbalance the incentives to reveal inforation to the antitrust authority. We find that a leniency progra does not always lead to a breach of trust; in certain industries leniency progras are unable to break collusion. Moreover, they ay have the adverse effect in the sense that they strengthen cartel stability or lead to abuse of arket power. A relatively large fir can use coercion to reove the option to a saller fir to self-report. In view of this liitation, policies aied at the reoval of the threat of retaliation need to be considered. The structure of the paper is as follows. Section 2 contains a foral description of the odel. In Section 3, we solve the odel and find sub-gae perfect equilibrius of the gae. Finally, in Section 4 the policy iplications are discussed and the analysis is concluded. In the appendix a coparative analysis of the approach to leniency progras used in the United States and in several European countries is provided. 2. The Model (Foral Analysis) We consider two asyetric firs, which ay for a cartel, taking into account the enforceent activity of the antitrust authority. The asyetry is related only to the size of the firs or their arket shares, while it is assued that firs have identical arginal costs. The antitrust authority coits to a certain enforceent policy, which uses leniency progras. Leniency progras grant either coplete or partial exeption fro fines to the firs, which reveal the existence of a cartel to the antitrust authority and coe up with sufficient evidence. The ain innovation of this odel is that we consider firs that have different arket shares. This iplies different accuulated profits during the period of collusive pricing. Hence, unless the antitrust authority (AA) is able to reove any asyetry in the accuulated profits (buffer) of each individual eber of cartel, soe bigger ebers enjoy a strategic advantage. A fir with a relatively large buffer will be able to eploy the difference in buffer size as a eans of coercion, such as the threat of punishent though aggressive pricing in case the rival deviates fro cartel agreeent by self-reporting. Essentially, K denotes the costs of aggressive behaviour (through e.g. setting price below arginal costs for soe periods) for the bigger fir. These costs are high when firs are ore syetric and, vise versa, these costs are low when asyetries are high. 1 In this paper we adopt the legal definition of aggressive (predatory) pricing, which is characterized as setting price below the arginal cost. 116

Volue V/ Issue 2(12)/ Suer 2010 The credibility and ipact of this type of pricing strategy depends on the asyetry in size between firs, such as the difference in arket shares. Market shares are denoted by β for bigger fir and by 1- β for sall fir, with β >1- β and 0< β <1. First, we describe the policy choices of the antitrust authority. Second, we describe the tiing of the gae. And, finally, we specify the firs strategies. Enforceent policy: The ain goal of the antitrust authority is to prevent the foration of cartels in the first place. However, if the cartel has already been fored, the antitrust authority ais to break the trust at the lowest possible cost. Here, following the reasoning in Section 1, we restrict the nuber of fine reductions in case of ultiple applications for leniency to one. Only the first reporter gets coplete exeption fro the fine. This, as explained above, reduces trust aong cartel ebers. This set-up is also otivated by the fact that the structure of leniency progras eployed in US allows only for one fine reduction. Moreover, the US schee also has a longer history than its European counterpart and has proven to be ore successful. 2 Following Becker (1968), we distinguish two ain paraeters of enforceent policy: penalty and probability of detection. Hence, the antitrust policy in the presence of leniency progras can be described by the following paraeters: The full fines F=απ, which are proportional to illegal gains for firs that were proven guilty and have not cooperated with the antitrust authority, or are not the first to coe forward with inforation about cartel. 3 Here α is the coefficient of proportional fine. π denotes per period illegal profits fro cartel foration. Copetitive profits (π n ) are assued to be zero for siplicity. So that π can also be viewed as pure illegal gains. Note also that π is axial per period payoff for each fir in case of full collusion (i.e. when firs are able to charge onopoly prices); The reduced fine f specified by the US leniency progra is equal to zero. 4 This set-up allows for the strictest adherence to the leniency rules; The probability of law enforceent by the antitrust authority equals pє(0,1]. This variable can be thought of as an instantaneous probability that the fir is checked by antitrust authority and found guilty. Contrary to Motta and Polo (2003), we assue that whenever the antitrust authority checks the guilty fir, the violation is successfully discovered. Moreover, we assue that p is deterined by e.g. an exogenous budget of the antitrust authority financed by the governent that can be used to proote enforceent, so that p reflects the costs of efforts of antitrust authority put into law enforceent activities. Tiing of the gae: Two asyetrical firs play the two stage gae in the presence of antitrust enforceent which incorporates leniency progras. At tie t=0 the antitrust authority sets paraeters of the enforceent policy: F = απ and p and paraeters of the leniency progra (which allows for only one fine reduction and reduced fine f equals 0). So, self-reporting becoes an attractive option at this stage. Prior to this stage t<0 firs ay decide to for a collusive agreeent. As conventional analysis of super-gaes (see Tirole 1988) iplies, in the absence of the antitrust enforceent, collusion can arise in equilibriu only when the discount factor is large enough, naely, δ β = δ c. 5 So, for further analysis we will direct our attention to the values of the discount factor δ δ c, which ensures that cartels are stable in the absence of antitrust enforceent and, hence, the first stage of the revelation-retaliation gae is reached. Next, the gae between the two asyetrical firs is played. At tie t=1 (stage 1 of the gae) the sall fir oves. 6 It can choose between two actions: self-report or keep cartel secret. 2 See historical overview and overview of structures of leniency progras in Appendix. 3 The otivation for this type of structure and exaple of application of proportional penalty schees in copetition law enforceent was studied in Motchenkova and Kort (2006). 4 See also Appendix. 5 In the absence of any antitrust enforceent, i.e. when neither fines nor rate of law enforceent can be used, collusion can be sustained only when the short run gain fro an unilateral deviation fro collusive agreeent by undercutting in prices together with copetitive profits thereafter is saller than the payoff fro sustaining collusive strategy forever: βπ /(1-δ)>π +δπ n /(1-δ) for i=1,2. Hence, with copetitive profits π n noralized to 0, we have δ 1-β for bigger fir and δ β for sall fir. The second constraint is binding since β>1-β. Hence, we have δ c = β. 6 We assue here that incentives for the bigger fir to keep the cartel secret are always higher since it gets higher expected gains fro continuation. So, the big fir would either need stronger incentives or will self-report only later in tie than the sall fir. 117

Journal of Applied Econoic Sciences Further, at tie t=2 (stage 2 of the gae) the big fir responds to the action of the sall by choosing whether to punish the sall fir for reporting the cartel or to abstain fro punishent. Note that the antitrust authority does not take an active part in the gae. It only sets policy paraeters, F, f, α, and the rules of leniency progras. This coplies with the current one size fits all setting of the antitrust policy paraeters. Payoffs of both players in each of the four possible cases are described in the following subsection. Each tie we refer with Sall to the saller fir and with Big to the other player. Strategies and Payoffs: 1. Sall has decided to report and Big responds by setting a predatory price: Big receives its current share βπ of collusive profits and the onopoly profits forever after ( i.e. π (δ/(1-δ))). But it has to overcoe a loss of size K (due to aggressive pricing) and a one tie fine of size αβπ and there is the risk of a fine when setting a predatory price of pαπ (δ/(1-δ)). The latter is the net present value of the expected fine Big ight have to pay, because of the abuse of its doinant position. Sall receives its current share (1-β)π but loses an aount of S (exit cost), since it has to leave the arket. Sall cooperated with the antitrust authority, so it is exepted fro a fine. 2. Sall has decided to self-report and Big decides not to retaliate and siply oves to arginal costs pricing (copetitive equilibriu) 7 : Big receives its current share of collusive profits βπ and is fined αβπ. Sall receives its current share (1-β)π, but doesn t ake any econoic profit forever after. Since it reported to the antitrust authority it isn t fined. 3. Sall has decided not to report and Big is inclined to use aggressive strategy. This eans that aggressive pricing was attractive strategy already before the antitrust enforceent and leniency progras were introduced (t=0). We will rule out this possibility later on when we discuss the solution of the gae (it just iposes additional constraint on discount factor (δ δ**, see section below). In this case Big receives its current share of onopoly profits βπ less a loss due to the aggressive pricing (K), but after sall leaves it will receive the entire (discounted) onopoly profit forever after π (δ/(1-δ)), though it also faces a risk of detection during the transition stage over its share of profits pαβπ and there is the chance of a fine in every period thereafter, which results in pαπ (δ/(1-δ)). Sall receives its current share (1-β)π but loses its exit cost S. Since the fir is bankrupt the authorities cannot levy a fine on the fir for its isconduct. 4. Sall has decided not to report and Big is inclined to continue the collusive price setting: Big receives its share of collusive profits forever βπ (1/(1-δ)), but faces the risk of being fined in every period there after. This results in pαβπ (1/(1-δ)). Siilarly, Sall receives its current share forever (1- β)π (1/(1-δ)) but faces the risk of prosecution in every period after pα(1-β)π (1/(1-δ)). It should be stressed that for any t > 2, decisions of both players do not change, and payoffs obtained at t=2 will be discounted. This is due to the fact that the coefficient of the proportional penalty and the rate of law enforceent are fixed and, hence, the environent does not change. Moreover, we assue that in case of self-reporting trust is broken and firs do not go back to collusion ever again. Therefore, outcoes (1) and (2) are stable by assuption. We suarize the above description of the gae as follows: Stage 0: The Antitrust Authority announces the paraeters of the penalty schee: p and F, and the paraeters of leniency progra: f=0 and the nuber of fine reductions. Stage 1: The saller fir decide whether to reveal inforation about the existence of the cartel to the antitrust authority or not (once and for all decision). Stage 2: The bigger fir observes the decision of the saller fir and decides whether to punish it for self-reporting or not (once and for all decision). If no self-reporting is chosen by the saller fir and the bigger fir decides to continue collusion, then the repeated gae, between authority and firs, where authority can discover violation with probability p in each period, is played till infinity (under assuption that even in case violation is discovered by antitrust authority, firs go back to collusion). 8 The discount factor is denoted by δ=1/(1+r), where r is the interest rate. The gae tree and players payoffs are suarized in Figure 1. 7 We assue here that in case one of the firs self-reported, trust will be broken forever and firs will not go back to collusion anyore. 8 Siilar assuption is eployed in Motta and Polo (2003). 118

Volue V/ Issue 2(12)/ Suer 2010 Stage 1 Stage 2 Predate (1) ( 1 β ) π S πδ βπ+ 1 δ pαπδ K αβπ ( ) ( 1 δ) Report Not Predate (2) ( 1 β)π βπ αβπ Not report Predate (3) ( 1 β ) π S πδ pαπδ βπ + K pαβπ ( 1 δ) ( 1 δ) Not Predate ( β) π ( 1 δ) ( 1 β) ( 1 δ) 1 p α π βπ p αβπ ( 1 δ ) ( 1 δ ) Figure 1. Gae tree and players payoffs. We now proceed to establish the sub-gae perfect equilibrius of the two-stage repeated gae, which is described in Figure 1, played by both firs once the policy paraeters are set. 3. Solution of the Gae To find the sub-gae perfect equilibrius of the gae we eploy backward induction. First we consider the decision of the bigger fir which is taken in Stage 2, and next the decision of the saller fir which is taken in Stage 1. Three different parts of the gae in Figure 1 can be distinguished as subgaes. First, Sall has a choice whether to report or not. It will base its decision on the reaction to its choice it expects fro the bigger player. Therefore Big faces two possibilities. If Sall has reported their coon illegal conduct, Big has to decide whether to retaliate or not. In case of no retaliation, Big sets its price equal to arginal or variable costs so that sall is punished for reporting to the antitrust authority, but keeps operating. In case of retaliation Sall fir is driven out of the arket. If Sall decides not to report, Big faces the sae decision. Clearly Sall s initial action depends on the Big s reaction. Since it knows Big s position and is aware of the value of all other paraeters (e.g. law enforceent and discount value) it will choose to play accordingly. 3.1. Collusion is the preferred strategy before leniency is introduced We start by considering the choice of the bigger fir between retaliation and collusion (i.e. by coparing outcoes (3) and described in previous section). Note that if outcoe is preferred over outcoe (3) by the bigger fir, collusion is also the preferred strategy before leniency is introduced (under traditional antitrust enforceent). This happens when βπ (1/(1-δ))-pαβπ (1/(1-δ)) > βπ -K+π (δ/(1-δ))-pαβπ -pαπ (δ/(1-δ)). This inequality iplies that retaliation is ore attractive than collusion for the bigger fir in both situations (with or without the availability of a leniency progra) when the discount factor is greater than the following threshold: K * * δ > = δ ( K, α). K+ π (1 β )(1 pα) (1) So, for any values of the discount rate above the threshold δ**, there is no collusion before the leniency progra is introduced and the gae doesn t apply. In the reainder of this paper all values of the discount rate over threshold δ** are regarded as values for the paraeter δ, in which equilibriu (3) 119

Journal of Applied Econoic Sciences (no self-reporting, retaliation) arises. This equilibriu isn t of any interest to answering the questions posed above and will therefore be left out of the analysis. Expression (1) gives us the first incentive copatibility constraint. It is represented in Figures 2 and 3 by the line δ**, which plots δ(k) as a function of K in the (δ,k)-plane. In addition, as discussed above, coparative statics of the behaviour of δ**(k,α) with respect to the ain paraeters of the odel shows that ** ** ** ** δ ( K, α) δ ( K, α) δ ( K, α) δ ( K, α) > 0 if pα< 1, < 0 if pα> 1, > 0, > 0. (2) K K p α These inequalities show that the likelihood of collusion is increased further and predation is less likely when higher values of p and α apply before the introduction of the leniency progra. The first derivatives of δ** with respect to K, p and α are positive when pα < 1 (i.e. in the setting with low expected penalty). Meaning that raising either of these paraeters will increase the height of this threshold, thus increasing the likelihood of the situation in which collusion is sustainable. Intuitively this akes sense, since retaliation is also illegal and increasing K iplies ore syetry. This coplies with general theory on collusion and syetry (see Motta and Polo 2003). Finally, in order to ensure consistent behaviour (eaning that collusion is sustainable and there are no incentives to retaliate in the absence of the possibility of self-reporting and subsequent cleency) we will consider only interval δ c < δ < δ**, so that outcoe (3) is ruled out and collusion is sustainable before the revelation gae starts. Recall fro Section 2 (footnote 5) that δ c = β. Taking into account (1), this iplies that interval [δ c, δ**] is not epty when δ** > β, i.e. when K > π β(1-pα) with pα < 1 or when K > π β(1-pα) with pα > 1. This iplies that the issue we are considering becoes especially sharp in industries characterized by relatively low asyetry (i.e. where K is high). 3.2. Deterination of other thresholds for equilibriu intervals Big s choice between aggressive strategy and siple copetitive strategy is based on a coparison of the outcoes (1) and (2). The outcoe (2) in the odel is the situation in which strategies (report, not retaliate) are used by the saller fir and the bigger fir respectively. Outcoe (1) in the odel is the situation in which strategies (report, retaliate) are used. Big is not inclined to retaliate in case of reporting by Sall when Big considers its payoff in equilibriu (2) to be higher than its payoff in equilibriu (1). The condition for equilibriu (report, not retaliate) to arise holds when the following inequality is satisfied: βπ -αβπ > βπ K+π δ/(1-δ)-αβπ -pαπ δ/(1-δ). This inequality iplies that copetitive pricing is ore attractive for the bigger fir than retaliation after the saller fir applied for leniency if the discount factor is less than the following threshold: K * δ < = δ ( K, α ). K+ π (1 pα) (3) Differentiating this expression with respect to K iplies that * δ ( K, α ) K > 0 if pα < 1. This iplies that when pα<1 (i.e. expected penalty is low) the equilibriu (2) is less likely to occur the saller the size of K. Recall that K is the size of the buffer of Sall, since it equals the cost of e.g. driving the saller fir out of the arket. After Sall loses its buffer it can t sustain the losses associated with the aggressive behaviour by Big. Intuitively this eans that the greater the size difference (asyetry), the lower K and therefore threshold δ* will be lower when asyetry is greater. It also iplies that raising the risk of being fined will increase δ*. Intuitively it eans that the saller the asyetry and the higher the chance of detection and substantial fine, the ore likely the perceived discount rate is below the threshold δ*. 120

Volue V/ Issue 2(12)/ Suer 2010 Next, we ove to Stage 1 and consider the decision of the saller fir given no retaliation is chosen by Big in the second stage of the gae. Outcoe (2) is preferred over outcoe by Sall if the following inequality is satisfied: (1-β)π > (1-β)π /(1-δ)-pα(1-β)π /(1-δ). This inequality iplies that self-reporting is ore attractive for Sall when the discount factor is lower than the following threshold: *** δ < p α = δ ( K, α). (5) This is a clear indication that raising the probability of detection and the proportional fine will ake the saller fir to choose equilibriu (2) over the payoff fro equilibriu, and will therefore decide to self-report instead of continuing to collude. Finally, we also have to copare the payoffs for Sall in case outcoe (1) arises and in case outcoe arises. Equilibriu (1) in the odel is the situation in which strategies (report, retaliate) are eployed by the saller fir and the bigger fir respectively. Equilibriu in the odel is the situation in which strategies (not report, not retaliate) are used. Now the saller player is confronted with a choice between being predated on by Big (and the associated exit cost) or going along with Big in the collusive price setting. The latter iplies that Sall prefers to choose a strategy leading to the collusive price setting over a strategy leading to bankruptcy. This occurs when the payoff of equilibriu is higher than the payoff in equilibriu (1) for Sall. I.e. the following inequality is satisfied: (1- β)π (1/(1-δ))-pα(1-β)π (1/(1-δ)) > (1-β)π -S. This inequality iplies that collusion is ore attractive for the sall fir if discount factor is deterined by the following inequalities: pαπ (1 β ) S **** δ > = δ ( K, α), when S < π (1 β ) π (1 β ) S S pαπ (1 β ) **** δ < = δ ( K, α), when S > π (1 β ) S π (1 β ) (6) (7) Closer analysis of expressions (6), (7) and (5) shows the following regularities: δ**** > δ*** when pα > 1 and (1-β)π > S, or when pα < 1 and S (1-β)π. In this case we have also that δ**** > 1. δ**** < δ*** when pα > 1 and S (1-β)π, or when pα < 1 and (1-β)π > S. In this case we have also that δ**** < 1. 3.3. Derivation of Equilibriu Solutions Next, once we have deterined all the thresholds in ters of the discount factor, we can ove to the description of equilibriu outcoes for each possible cobination of the paraeter values. Firstly, two cases need to be distinguished: when pα>1 and when pα<1. Inequality pα>1 corresponds to the case when the expected penalty is already high enough to prevent any isconduct (in a static setting) in the absence of leniency progras. The other inequality corresponds to the situation when traditional antitrust enforceent is not strong enough. We start our analysis with the discussion of a sufficiently high penalty (the case where pα>1). In this setting two sub-cases depending on the size of δ****(k,s,α) and on the size of exit costs can arise. When pα>1 and (1-β)π >S, we obtain that δ****>1, and the distribution of outcoes can be described as is done in the left hand side of Figure 2. When pα>1 and (1-β)π <S, we obtain that δ****<1. Hence, the distribution of outcoes is given by the right hand side of Figure 2. 121

Journal of Applied Econoic Sciences δ pα > 1, δ**** > 1 low exit cost δ pα > 1, δ**** < 1 high exit cost (3) (1) (2) (3) (1) (2) δ**** δ c (3) (2) δ c K Figure 2. Equilibriu outcoes when pα>1. K Figure 2 presents the locuses δ*, δ**, δ***, and δ**** (derived in previous subsection and given by (3), (1), (5), and (6) respectively) in (K,δ)-space. The left panel of Figure 2 iplies that, in industries with low exit costs and relatively strong antitrust enforceent, depending on the degree of asyetry the following three outcoes can arise. When there is high asyetry (K < π β(1-pα) ), outcoe (3) will arise in equilibriu. 9 This eans that in this industry retaliation is the ost attractive strategy even before a leniency progra is introduced. With an interediate degree of asyetry (i.e. π β(1-pα) < K < π (1-pα) ), outcoe (1) arises in equilibriu. 10 In this case Big (strong) fir will choose to retaliate on a saller fir after the latter chooses to self-report. This is the outcoe the antitrust authority wants to avoid. In these types of industries a greater ephasis needs to be put on the protection part of a leniency progra. Perhaps through stricter onitoring after a fir reported to the AA. Finally, when there are low asyetries (i.e. K> π (1-pα) ), outcoe (2) will arise in equilibriu. 11 This iplies that with high penalties in the industries with high K (or low asyetries) there is no danger of retaliation or collusion. The first best outcoe with self-reporting and copetitive pricing afterwards arises. In this setting leniency progras appear to be effective. The right panel of Figure 2 represents the results of the analysis for industries with relatively high exit costs and relatively strong antitrust enforceent. Here, again depending on the degree of asyetry between firs, the following outcoes can arise. When there are high asyetries (K< π β(1-pα) ), outcoe (3) will arise in equilibriu. When asyetries are low (i.e. K> π (1-pα) ), outcoe (2) arises in equilibriu. For an interediate level of asyetry both outcoe (1) and outcoe can arise in equilibriu. So in addition to the possibility of retaliation, there is a sall danger of collusion, when exit costs for sall fir are too high. This result is quite intuitive, since with high exit costs the threat of a possible retaliation can force sall fir to keep the cartel secret and not to apply for leniency. The above analysis can be suarized in the following proposition. Proposition 1: When traditional antitrust enforceent is strong (i.e. pα>1), after introduction of leniency progras there exists a threat of retaliation and of even stronger collusion in the industries with an interediate level of asyetry (i.e. π β(1-pα) <K< π (1-pα) ). 9 Proof: since δ**<0 and δ*<0, any δ in the interval (0,1) is higher than δ**, hence (3) is played in equilibriu. 10 Proof: since δ**>1 and δ*<0, any δ < δ** and any δ > δ*. This in turn iplies that outcoe is preferred over (3) by Big in case Sall chooses not to self-report and outcoe (1) is preferred over (2) by Big in case Sall chooses to reveal inforation. Next, since any δ < δ****, when δ****>1, taking into account best response of Big, Sall will prefer outcoe (1) over. Hence, (1) is played in equilibriu. 11 Proof: since δ**>1 and δ*>1, any δ < δ** and any δ < δ*. This in turn iplies that outcoe is preferred over (3) by Big in case Sall chooses not to self-report and outcoe (2) is preferred over (1) by Big in case Sall chooses to reveal inforation about the cartel. Next, since any δ < δ***, when δ***= pα >1, taking into account the best response of Big, the saller fir will prefer outcoe (2) over. Hence, (2) is played in equilibriu. 122

Volue V/ Issue 2(12)/ Suer 2010 To suarize the above discussion it needs to be stressed that, even when penalties are high enough to block the cartel foration in static settings, (i.e. pα > 1) there could be adverse effects of leniency progras on the incentives to the firs to collude in a dynaic setting. There could be a threat of retaliation and of enhanced collusion in the industries with an interediate level of asyetries (i.e. in industries where π β(1-pα) <K< π (1-pα) ). This iplies that, in this kind of industries, a strong ephasis on the protection of leniency applicants needs to be introduced and particular attention should be paid to industries where exit costs are high. Next, we continue our analysis with the discussion of the case where pα<1. In this setting again two sub-cases, depending on the size of δ****(k,s,α) can arise. When pα<1 and (1-β)π <S)) we obtain that δ****>1, and, hence, the distribution of outcoes can be described as is done in the left panel of Figure 3. When pα<1 and (1-β)π >S)) we obtain that δ****<1, and, hence, the distribution of outcoes is given in the right panel of Figure 3. 12 δ pα < 1, δ**** > 1 high exit cost δ** δ pα < 1, δ**** < 1 low exit cost δ** (3) (3) δ* δ *** (3) (3) (2) (2) (1) (2) K K1 K2 K1 K2 K Figure 3. Equilibriu outcoes when pα<1. In the case where pα<1 the following regularities are satisfied for any paraeter values: δ***=pα<1, δ*>0, δ**>0, δ**>δ*. Note also that when pα<1, (1-β)π >S, and δ****<1, we have that δ****-δ***=s(pα - 1)/((1-β)π - S)<0. Hence, inequality δ****<δ*** holds for any paraeter values. This is also depicted in the right panel of Figure 3. Moreover, in both cases described above (naely, δ****<1 and δ****>1,with pα<1), we have that δ**-δ***>0, when K>pαπ (1-β) (denoted by K 1 in Figure 3) and δ*-δ***>0, when K>pαπ (denoted by K 2 in Figure 3). 13 Finally, based on the above analysis, we conclude that the following proposition holds. It relates four industry types to an environent with rather weak law enforceent (the product of the rate of capture and the coefficient of proportional fine is saller than one). This situation applies to ost European countries and to EU antitrust law as well. The US anti-cartel enforceent sees to be stricter. It only grants a single fine reduction to the first reporter. In addition it generally uses criinal law and includes possibility of iprisonent also for violations of antitrust law. For a ore detailed insight into these atters see the appendix. Having established that the environental condition of the above proposition holds we can ove on to the industry characteristics and copleent these stateents with policy iplications. Proposition 2: When traditional antitrust enforceent is weak (i.e. pα < 1), the following four effects of ipact of industry structure on incentives to self-report can be found: Proposition 2(1): In industries with little asyetry (K is high) and low discount rate the first best outcoe with self-reporting and copetitive pricing afterwards (equilibriu (2)) can be achieved. 12 We do not provide detailed proofs of these results here, since they follow the sae lines as in case where pα>1. Detailed proofs and clarifications are available fro authors upon request. 13 Proofs of these siple regularities are available fro authors upon request. 123

Journal of Applied Econoic Sciences This result shows that a leniency progra can have the desired effect, i.e. destabilizing cartel and inducing ore copetition. However, the ai of the leniency progra to establish copetition in cartelized industries is only achieved when firs within an industry are ore or less of the sae size and little ephasis is put on future profits. This result is counterintuitive to the reasoning where syetry creates a coon focal price and facilitates collusion. Clearly asyetry can be an iportant aspect for the stability of cartels. By far the largest nuber of industries can be described along the above lines. The firs within these industries have a strong incentive to start copeting healthily, deviate fro collusion or report to the AA upon introduction of the progra. A new question that can be raised is whether collusion in these types of industries is as harful to society as collusion is in industries characterized by a higher discount rate. Proposition 2(2): In industries characterized by a high discount rate (δ > δ**) retaliation is always the ost attractive strategy for any type of fir (regardless of asyetry). Outcoe (3) arises in equilibriu. Industries that do put an ephasis on future profits will have a reason to do so. Mostly these industries are coprised of rather larger firs than the previously entioned type of industry. 14 Natural onopolists are willing to ake great investents in networks since they know they will be able to profit fro these for any years. The introduction of a leniency progra has no effect what so ever in this type of industry since there is usually only a single fir active in the arket. In other words there is no collusion in the first place. The prevention and regulation of onopolists lies beyond the scope of this paper. Proposition 2(3): In industries with low exit costs (S < (1-β)π ), high asyetry (K<pαπ (1-β)), and a low discount rate there is a threat of retaliation on the self-reporting fir. Outcoe (1) with selfreporting and retaliation arises in equilibriu. The event of fierce copetition isn t always an illustration of healthy copetition (where healthy refers to pricing at or slightly above arginal cost). A war on prices or a period of repetitive advertising on prices ight be an indication of duping or predatory pricing. These practices reduce welfare, since investents are wasted through the destruction of capital. Moreover the consequence of this aggressive behaviour is the exit of firs fro the industry and healthy copetition is further away than it ever was. Proposition 2(3) shows that the introduction of a leniency progra in soe cases ay lead to aggressive pricing (retaliation can occur as a response to self-reporting) and a loss of welfare. Besides the welfare loss the antitrust authority will also have to spend resources in the future to regulate the newly created onopolists. This scenario is especially likely to occur in asyetrical industries with a relatively low sunk cost and a greater ephasis on future profits. When sunk costs are low, sall fir will not lose uch when predated upon, and will therefore be ore inclined to report. On the other hand, due to the low discount factor for the big fir the potential fine of collusion will outweigh the future gain fro cartel. This will cause retaliation. It is the proise to protect any party to self-report to the antitrust authority that helps to overcoe this scenario. This proise needs to be clear and credible, though resources need not be wasted. Therefore a proise to protect should be incorporated in the leniency guidelines. Proposition 2: In industries with high exit costs (S > (1-β)π ) and discount factor in the range δ* < δ < δ** collusive equilibriu is sustainable even after leniency progras are introduced. Moreover, in industries with high exit costs is sustainable for a bigger range of discount factors copared to low exit cost industries. The worst effect the introduction of a leniency progra can have is the strengthening the stability of cartels. The results of the analysis show this does occur, however. In industries characterized by high sunk costs this scenario is ore likely to occur. The reason the stability of the cartel is increased lies in the possibility for large firs to use the leniency progra as a eans to increase the trust they put in 14 Note also that discount rates do, in fact, change with tie for all industries. Thus, according to the odel, a leniency policy that has one effect when the interest rate in the econoy is law ight have a different effect when the interest rate is high. Another interpretation is when higher discount rate is associated with higher industry stability. The discount rate (δ) can be interpreted as probability that the fir will survive in the next period. So that higher δ iplies higher industry stability. In this sense, industries that are coprised of larger firs, which are stronger and have higher probability of survival, would be ore stable and, as a result, can be viewed as industries characterized by the higher discount rate. 124

Volue V/ Issue 2(12)/ Suer 2010 other firs not to report to the AA. The larger fir trusts the saller not to report to the AA. It can do so since the saller fir knows the punishent of the larger fir is severe. This scenario can never be prevented copletely. The chance, this scenario develops however, can be lowered by developing a coparable proise to protect the reporting fir as is described in the previous paragraph. Besides this type of policy approach the leniency prograe should always go accopanied by an effort of traditional antitrust law enforceent directly aied at the industries described above. When the threat of a fine due to the ore traditional prosecution increases, ore firs will choose to abandon the cartel. 4. Conclusions The analysis of this paper adds to the current econoic literature on leniency progras for cartels and illegal price-fixing activities, but the reasoning can also be applied to corporate whistle-blowing progras and witness protection progras. It reveals a nuber of adverse effects of the introduction of leniency progras in view of industry asyetry. The ain conclusion is that the introduction of a leniency progra, regardless of the size of the fine, ight facilitate the stability of cartels in certain industries. This is ainly due to a leniency progra s inability to reove the threat of punishent on a self-reporter (or whistle-blower) by its forer partners. After a fir is convicted it reains with sufficient resources to retaliate on the reporting party. It enables soe firs to use coercion as a eans to increase trust in the cartel. Though the reoval of trust is the ai of the progra, the introduction of the schee actually provides colluding firs with the eans to stabilize the cartel. Increasing the size of the fine and liiting the nuber of fine reductions to the first party to report isn t sufficient to (fully) overcoe the adverse effect of the introduction of the leniency progra. The analysis in this paper iplies that the progra s effectiveness largely depends on the environent and the type of industry to which it is being applied. Raising the rate of capture (through e.g. liiting the nuber of fine reductions) and the size of the penalty do help to diinish the adverse effect. The size of the fine can for instance be increased by putting a greater ephasis on aggravating circustances, such as coercion. It will however not be sufficient to tackle cartels in industries with an interediate level of asyetry. When an AA is unable to raise sufficient resources to increase the rate of capture through traditional law enforceent, in this type of industry it should direct its focus on the proise to protect self-reporters fro retaliation by forer collusive partners. Since the current type of policy approach is sufficiently effective in a great nuber of industries, diversification of the progra can give rise to a ore efficient use of resources. Custoization of the progra, where it coes to protection, size and nuber of the fine reduction, paralleled by a traditional effort of law enforceent aied at industries in which the adverse effect is likely to occur will help to ake the progra ore effective. The analysis of Section 3 iplies that, even when penalties are sufficiently high to block the foration of cartels, the leniency progra can still withhold firs fro self reporting. Because of the threat of retaliation after a cartel is uncovered, trust between cartel ebers will be stronger in those industries characterized by an interediate degree of asyetry and barriers to entry, due to the introduction of the schee. This iplies that in these kinds of industries a strong self-reporter s protection progra should be introduced besides the leniency progra. When penalties are lower (i.e. the product of the rate of capture and the coefficient of the proportional fine is lower than one), which is currently generally the case in ost European countries, the effectiveness of leniency progras largely depends on the environent the firs find theselves in and on the type of the industry. In this case the focus of the copetition authority should be on those industries characterized by a low to interediate degree of asyetry and an interediate to high discount rate. Since, in these types of industries, regardless of any barriers to entry, chances are that, the introduction of a ild leniency progra facilitates collusion. It serves to strengthen trust between colluders, rather than to create a breach of trust. Another effect of the introduction of a leniency progra is reversion to aggressive pricing in order to retaliate on the reporting partner. Though this ight at first look like healthy copetition it eventually reduces welfare. In an environent of high fines (product is greater than one) this is a ore likely scenario and it will occur in industries characterized by an interediate level of asyetry and low barriers to entry. However also in an environent of low fines aggressive pricing can be the effect of the introduction of the progra, especially when barriers to entry are low. Besides having to spend resources on regulating these new (sei) onopolists, the destruction of capital associated with the retaliation strategy is detriental to welfare. 125

Journal of Applied Econoic Sciences To suarize the above analysis, in industries characterized by high barriers to entry/exit (such as high exit costs) and degree of asyetry leniency progras ay be ineffective and give rise to increased cartel strength. In industries with low exit costs leniency progras ay be ore effective, but retaliation is ore likely to occur as a response to self-reporting. Policies aied at the reoval of this threat of punishent through aggressive behaviour need to be considered in order to reove these kinds of hard core cartels. A first eans is to eploy higher fines in order to reove a bigger part of the illegal gains. Putting ore ephasis on aggravating circustances, such as coercion, in the fining guidelines can also be an effective approach. Another regulatory easure is to introduce the proise to protect the reporting party after reporting in the leniency application. In general though a leniency progra along can not be fully effective in its ai to prevent and prosecute all cartels. A certain aount of effort will always need to be directed towards certain industries beside the leniency progra. The currently eployed odel calls for the decision to collude to be ade prior to the decision to introduce a leniency policy. This, to soe extent, liits the applicability of the odel only to analysis of pre-existing cartels. The natural extension would be to enrich this odel by eploying a sequential repeated gae fraework where in each period in the presence of leniency progras the following two stage gae is played. In the first stage firs decide whether to for cartel or not, the second stage is the revelation decision with the subsequent retaliation decision. Siilar structure is ipleented in Houba et al. (2009) with syetric firs and in the absence of retaliation. However, incorporating asyetric firs and possibility of retaliation in such an advanced fraework ay ake the odel intractable and difficult to solve. In this anuscript we forego excessive technical difficulties and adopt a siplified version of the odel in order to concentrate on how possibility of retaliation can influence the effectiveness of leniency progras for already fored cartels. References [1] Becker, G. 1968. Crie and Punishent: an Econoic Approach, Journal of Political Econoy, 76:169-217. [2] Buccirossi, P., and Spagnolo, G. 2006. Leniency Progras and Illegal Transactions, Journal of Public Econoics, 90(6-7):1281-1297. [3] Chaberlin, E. 1929. Duopoly: Value Where Sellers Are Few, Quarterly Journal of Econoics 3:63-100. [4] Chen, J., and Harrington, J. 2007. The ipact of the corporate leniency progra on cartel foration and the cartel price path, in The Political Econoy of Antitrust, ed. by V. Ghosal, and J. Stennek. Elsevier. [5] Chen, Z., and Rey, P. 2007. On the Design of Leniency Progras, IDEI Working Papers 452, Institut d conoie Industrielle (IDEI), Toulouse. [6] D.O.J. 1993. US Corporate Leniency Policy, http://www.usdoj.gov/ [7] D.O.J. 2003. US Sentencing Guidelines (Chapter 8: Sentencing of Organizations), http://www.ussc.gov/2003guid/chap8.ht. [8] EC 2002. Coission Adopts New Leniency Policy for Copanies which Give Inforation on Cartels, Brussels, http://europa.eu.int/co/copetition/antitrust/leniency/. [9] EC. 2006. Guidelines on the ethod of setting fines iposed pursuant to Article 23(2)(a) of Regulation No 1/2003, Official Journal of the European Union, (2006/C 210/02), Brussels. [10] Ellis, C., and Wilson, W. 2001. Cartels, Price-Fixing, and Corporate Leniency Policy: What Doesn t Kill Us Makes Us Stronger, ieo, University of Oregon, Oregon. [11] Feess, E., and Walzl, M. 2004. Self-reporting in Optial Law Enforceent when there are Criinal Teas, Econoica 71:333-348. [12] Harrington, J. 2008. Optial Corporate Leniency Progras, The Journal of Industrial Econoics, LVI(2):215-246. 126

Volue V/ Issue 2(12)/ Suer 2010 [13] Hinloopen, J. 2003. An Econoic Analysis of Leniency Progras in Antitrust Law, De Econoist, 151:415-432. [14] Houba, H., Motchenkova, E., and Wen, Q. 2009. The Effects of Leniency on Maxial Cartel Pricing, Tinbergen Institute Discussion Paper 09-081/1. [15] Kovacic, W. 2006. Bounties as induceents to identify cartels, in European Copetition Law Annual 2006, ed. by C. Ehlerann, and I. Atanasiu. Hart Publishing, Oxford/Portland, Oregon. [16] Motchenkova, E. 2004. Effects of Leniency Progras on Cartel Stability, CentER Discussion Papers Series 2004-98, Tilburg University, Tilburg. [17] Motchenkova, E., and Kort, P. 2006. Analysis of Current Penalty Schees for Violations of Antitrust Laws, Journal of Optiization Theory and Applications, 128(2). [18] Motta, M., and Polo, M. 2003. Leniency Progras and Cartel Prosecution, International Journal of Industrial Organization 21:347-379. [19] OECD. 2002. Fighting Hard Core Cartels: Har, Effective Sanctions and Leniency Prograes, Paris, http://www.oecd.org/dataoecd/41/44/1841891.pdf. [20] Rey, P. 2003. Towards a Theory of Copetition Policy, in Advances in Econoics and Econoetrics: Theory and Applications, ed. by M. Dewatripont, L. Hansen, and S. Turnovsky. Cabridge University Press. [21] Spagnolo, G. 2004. Optial Leniency Progras, Centre for Econoic Policy Research, Discussion paper series, 4840, http://www.cepr.org/pubs/new-dps/showdp.asp?dpno=4840, (revised 2008). [22] Spagnolo, G. 2008. Leniency and Whistleblowers in Antitrust, in Handbook of Antitrust Econoics, ed. by P. Buccirossi. MIT Press. [23] Stigler, G. 1964. A Theory of Oligopoly, The Journal of Political Econoy 72 (1):44-61. [24] Tirole, J. 1988. The Theory of Industrial Organization. MIT Press, Cabridge. 127

Journal of Applied Econoic Sciences APPENDIX Historical Overview and Structure of LPs Table 1. Fining Systes and Structure of Leniency Progras Country Size of Fine Liitation of Fine EU US UK Base level of fine is deterined by the gravity and duration Base level of fine is deterined by the gravity, illegal gains and daage to society Seriousness and relevant turnover for a basis 10% of the total annual turnover of the year before conviction No upper bound 10% of total UK turnover of the year before conviction No of Fine Reductions Max Fine Reduction first reporter Max Fine Reduction other reporters >1 100% Up to 75% 1 100% No fine reduction >1 100% 50% Sources: DoJ (2003), EC (2006), EC (2002), and OECD (2002). Table 2. Tiing of Introduction of Copetition Law and Leniency Progras General Copetition Law Leniency Progra Country First Initiative Organization in charge LP introduction LP latest version EU 1959 Coission 1996 2002 US 1890 DoJ 1978 1997 UK 1998 Office of Fair Trade 1998 2004 Netherlands 1998 NMa 2002 2006 Gerany 1958 Bundeskartellat 2002 2006 Sources: DoJ (2003), EC (2006), EC (2002), and OECD (2002). 128