Recent Developments at DG Competition: 2013–2014

The Directorate General for Competition at the European Commission enforces competition law in the areas of antitrust, merger control, and state aids. In 2013–2014 important antitrust decisions (Samsung and Motorola) were in the area of standard essential patents. In merger control the European Commission carried out an ex-post evaluation of past mergers in the INEOS/Solvay case in order to obtain valuable insights for the merger under review. In state aid, as a part of the State Aid Modernisation programme, guidelines were prepared to help EU member states to design and carry out ex post assessments of state aid schemes.


Introduction
The Directorate General for Competition (henceforth "DG COMP") at the European Commission enforces competition law in the areas of antitrust (anti-competitive agreements-including cartels-and abuses of dominant position), merger control, and state aids.
The workload of DG COMP can be characterized as follows: Over the past few years the European Commission has issued on average 10-12 antitrust decisions per year (the average duration of such cases between registration and decision is 4-5 years); 1 270-300 merger decisions per year, with an intervention rate (that is, the number of prohibitions, approvals subject to remedy, and withdrawal in the second phase 2 ) of about 5 %; 3 and around 400 state aid decisions per year. 4 The Chief Economist Team (which counts about 30 members) is involved in a subset of these cases. Typically it participates in antitrust cases that lead to intervention (although with a minor involvement in cartels, where cases largely hinge upon documentary evidence) and the most important merger cases (including all second phase cases), as well as the economic aspects of state aid cases.
The following three sections report on important and interesting work that was done last year at DG COMP: Section 2 deals with two recent antitrust decisions (Samsung and Motorola) in the area of Standard Essential Patents (SEPs), which have recently attracted a lot of attention both in academic environments and in the media (for example the frequent news of litigation among holders of SEPs-involving, for instance, Apple, Google, Samsung-and associated court judgments around the world).
Section 3 deals with INEOS/Solvay: 5 a merger case that involved the two top sellers in a chemical product market, in which DG COMP took advantage of the existence of previous recent mergers involving INEOS to carry out an ex-post evaluation of those past mergers in order to obtain valuable insights for the merger under review.
Section 4 also deals with ex-post assessments, but with a different perspective and application. As a part of the State Aid Modernisation programme, EU Member States (MSs) should evaluate their aid schemes themselves. DG COMP has prepared guidelines in order to help MSs to design and carry out ex post assessments of their schemes, and we shall briefly report on the main points touched upon in the guidelines.

Standard Essential Patents: Samsung and Motorola
In 2014, the European Commission issued two important decisions on SEPs. First, the Commission accepted commitments of Samsung and ended an antitrust investigation, 1 In 2013, there were 12: four cartel decisions and eight antitrust decisions, of which four were prohibition decisions and four were 'commitment' decisions: cases where the firm(s) that were involved offered remedies for their conduct and stopped the alleged anti-competitive conduct, and were not fined. 2 The European Commission's merger procedure has two phases: the first phase is the initial, standard investigation, which is followed by the second phase, in-depth investigation if the European Commission has competition concerns. Subject to a set of eligibility criteria a simplified procedure may apply to cases that do not raise competition concerns in the first phase. 3 For instance, in 2013 there were 269 decisions, of which 252 mergers were cleared in the first phase (166 under the simplified procedure), 13 after the notifying parties submitted remedies that addressed the identified competition concerns, and two (UPS/TNT and Ryanair/Air Lingus) were prohibited. 4 In 2013 there were 470 decisions with respect to state aid. DG COMP investigates state aid cases after receiving notifications by Member States, complaints (for instance by firms that claim that competition has been distorted by subsidies that are received by a rival), or on its own initiative (for instance, when the press reports about some state aid initiative that appears to be potentially distortive of competition). 5 Case No COMP/M.6905. according to which Samsung will not seek injunctions in Europe on the basis of its SEPs for smartphones and tablets against licensees who subscribe to a specified licensing framework. Second, the Commission adopted a decision that found that Motorola Mobility's seeking and enforcement of an injunction against Apple on the basis of a mobile telecommunications SEP constitutes an abuse of a dominant position and prohibited this conduct. In earlier years, the Commission had already investigated other cases in which SEPs were involved, such as Rambus. 6 The two recent decisions aim at ensuring effective access to a standard for all market players and to prevent "hold-up" by a single SEP holder. SEPs are patents that are essential to implement a specific industry standard. It is not possible to manufacture products that comply with a certain standard without accessing these patents. If SEP holders can enjoin products 7 that infringe on their SEPs from the market, implementers of this standard would lose the entire profit that is associated with these products and may thus accept onerous licensing terms to secure access to the relevant SEPs. 8 This may give the companies that own SEPs significant market power. Without further remedy, SEP holders could thus extract higher royalties than the value of the SEPs before they became part of the standard.
For this reason, standard setting organizations (SSOs) generally require their members to commit to license SEPs on fair, reasonable, and non-discriminatory (so-called "FRAND") terms when their patents are included in their standards. This commitment is designed to ensure effective access to a standard for all market players and to prevent "hold-up" by a single SEP holder.
In the present cases, Samsung sought injunctions, and Motorola sought and enforced an injunction on the basis of SEPs that were included in mobile telecommunications standards despite having made a FRAND commitment to the SSO: the European Telecommunications Standards Institute (ETSI). The Commission pointed out in its decisions that seeking an injunction that is based on SEPs may constitute an abuse of a dominant position if a SEP holder has given a voluntary commitment to license its SEPs on FRAND terms and where the SEP holder would likely obtain a fair remuneration without the use of injunctions. Since injunctions generally involve a prohibition of the product that is infringing the patent being sold, seeking (and enforcing) SEP-based injunctions could risk excluding products from the market although the SEP holder has previously committed to licensing its SEPs.

Samsung
Samsung owns SEPs that are related to various mobile telecommunications standards and has committed to license these SEPs on FRAND terms. In April 2011, Samsung started to seek injunctions against Apple on the basis of its SEPs. The Samsung SEPs in question related to the ETSI 3G UMTS standard, which is a key industry standard for mobile and wireless communications. In December 2012, the Commission informed Samsung of its preliminary view that the seeking of injunctions against Apple based on Samsung's SEPs in several EU Member States may in the circumstances of the case constitute an abuse of a dominant position. On 3 February 2014, Samsung submitted the final version of the commitments that addressed the Commission's competition concerns.
Two aspects of the assessment are particularly interesting from an economic point of view: First, the Commission assessed the potential anti-competitive effects of the use of injunctions on the basis of SEPs. It came to the view that by using injunctions on the basis of UMTS SEPs Samsung could potentially exclude Apple's UMTS-compliant mobile devices from the market. In order to avoid this exclusion, Apple may be induced to accept disadvantageous licensing terms.
Second, the Commission assessed whether denying injunctions to Samsung in the circumstances of the case would render it unlikely that Samsung would eventually obtain FRAND royalties. Importantly, an SEP holder always has the ability to seek damages for unauthorised use of its SEPs through damages actions before national courts. There may, however, be scenarios where despite the possibility to litigate for damages the SEP holder may not be in a position to obtain FRAND royalties without the use of injunctions. 9 In particular, damages are normally awarded for past infringement so that a longer period of negotiations (which may result from not accepting the opening offer) does not necessarily deprive the SEP holder of some of the deserved FRAND royalties. Moreover, the Commission is of the view that litigating for damages together with a judicial determination of the forward-looking royalties normally is an effective means for the SEP holder to obtain a FRAND royalty. 10 Therefore, in the circumstances of the Samsung case, the Commission came to the preliminary view that even without the use of injunctions, the risk of obtaining less than the FRAND remuneration was limited, and therefore Samsung should not have sought to enjoin Apple's products.
To address the Commission's concerns, Samsung has for a period of five years committed not to seek any injunctions in the European Economic Area (EEA) on the basis of its SEPs for smartphones and tablets against any company that agrees to a particular framework for licensing the relevant SEPs. The licensing framework provides for a negotiation period of up to 12 months; and if no agreement is reached, a third-party determination of FRAND terms will be made by a court or by an arbitrator.
The commitments therefore provide a "safe harbour" for all potential licensees of the relevant Samsung SEPs. Indeed, potential licensees that sign up to the licensing framework will be protected against SEP-based injunctions by Samsung.
From an economic perspective, this framework balances efficiently the interests of SEP-holders and implementers. Once an implementer has agreed to the framework, it is rational both for Samsung and the implementer to anticipate that in case of disagreement about the terms of the license, these will be determined by a neutral third party. On the assumption that the court or arbitrator will on average set FRAND royalties, it is likely that the parties would agree even before the third-party FRAND determination is triggered, which saves litigation/arbitration costs. Clearly, this procedure rules out the risk of hold-up since Samsung has committed not to resort to injunctions once an implementer has accepted this framework. Also the risk of "hold-out"-that is, Samsung having to accept a royalty that is below FRAND-is limited, since the negotiation period before entering third-party FRAND determination is limited and fixed upfront and the FRAND royalties of all SEPs that belong to a standard will be adjudicated together, which further speeds the process. 11 If an implementer chooses not to accept this framework, Samsung is not further bound by its commitments. In that case Samsung may opt to seek injunctions and to the extent that this would be granted by the courts, an implementer may thus face a risk of its products' being enjoined from the market after not having accepted the framework. Therefore, an implementer that is genuinely interested in licensing the relevant SEPs will likely agree to that framework.
More generally, other SEP holders may consider committing to the application of a similar framework. Not accepting a framework of third-party FRAND determination may constitute a signal that an implementer has no genuine interest in licensing certain SEPs and depending on the circumstances may justify that the SEP holder subsequently seeks injunctions against this implementer. Conversely, implementers may signal good faith by offering to the SEP holder that both parties enter into such a framework.

Motorola
Motorola sought and enforced an injunction against Apple in Germany on the basis of a SEP that relates to the ETSI's GPRS standard and that Motorola had committed to license on FRAND terms. Apple had agreed to take a licence and be bound by a determination of the FRAND royalties by the relevant German court. In its recent decision, the Commission found that Motorola infringed Article 102 TFEU by seeking and enforcing against Apple an injunction on the basis of one SEP that induced Apple inter alia to accept certain onerous licensing terms although Apple had agreed before to third-party FRAND determination. Motorola was ordered to eliminate any anticompetitive effect resulting from that infringement.
The Commission assessed the potential anti-competitive effects of the threat and the enforcement of injunctions on the basis of the relevant SEPs. The Commission concluded that Apple likely accepted certain disadvantageous licensing terms in order to ensure that its products were not enjoined from the market. Inter alia, Apple accepted that if it were to challenge any of the licensed SEPs, Motorola would be entitled to terminate the license in relation to all licensed SEPs ("termination clause"). Moreover, Apple agreed to take a license for Motorola's SEPs for certain of Apple's products for which it would have possibly not needed to take a license.
The Commission found that accepting the termination clause and further provisions would likely reduce Apple's future profit. First, this is evident by the revealed preferences of Apple that it had not included these provisions in its previous offers. Indeed, Apple accepted these clauses only after Motorola enforced an injunction in Germany. Second, the Commission analyzed the likely effect on Apple's future profits of these clauses. The impact of the termination clause on Apple's expected profits depends in part on how the existence of such a clause is reflected in the FRAND rates that are determined by a competent court. With regard to the agreement to take a license for certain products for which Apple may not have needed to take a license, Apple gave up the possibility not to pay any royalties for these products. Therefore, accepting that these products are covered by the license clearly increased Apple's expected royalty payments and therefore reduced its expected profits.
Moreover, the termination clause would reduce Apple's incentives to invalidate the licensed patents. The Commission found that Apple would anticipate that upon initiating invalidity actions, Motorola could terminate the licensing agreement and afterwards try to enjoin Apple's products from the market. The risk of losing the profits associated with the relevant products would render invalidity actions for Apple less profitable and Apple may therefore be more reluctant to initiate such actions, even if it obtained evidence that some of the licensed SEPs may be invalid. Given that licensees such as Apple normally have particularly strong incentives to try to invalidate the licensed patents, as this normally reduces the amount of royalties to be paid, such a clause increases the likelihood that weak patents are not challenged and possibly invalidated. Therefore, other implementers will have to pay royalties for patents that may have been invalidated in the absence of such a clause. 12 Finally, certain products of Apple were banned from Apple's online sale platform in the German market for a short period during which an injunction was enforced.
In its investigation, the Commission also assessed whether Motorola could have appropriated a FRAND remuneration for its SEPs without resorting to the seeking and the enforcement of an injunction. In this context, Apple in October 2011 made a binding offer that gave Motorola the right to set the royalties according to its equitable discretion and according to FRAND principles as regards the royalty rates and the method of calculation of the final amount of royalties. The offer also allowed for a full judicial review of the amount of FRAND royalties, whereby Motorola and Apple could submit their own evaluations, calculations and reasoning for consideration to the court.
The Commission concluded that at the latest after Motorola received this offer it could have received FRAND royalties without the further use of injunctions. The enforcement of an injunction would thus not have been necessary to obtain a FRAND remuneration.

Conclusion and Policy Implications
Not surprisingly, the reasoning in the Motorola case is similar to the justification for why the Commission accepted the commitments in the Samsung case. The common principle is that once a prospective licensee has committed to accept a license at terms that are determined by a competent third party, injunctions on the basis of SEPs should not be used any more.
This principle ensures on the one hand that potential licensees can avoid the risk of their products being enjoined from the market when not accepting terms that they would have rejected in the absence of the threat of an injunction. This effectively limits the risk of patent hold-up on the basis of SEPs. On the other hand, this principle does not imply a full ban of the use of injunctions on the basis of SEPs. In particular, this principle restricts predominantly the use of injunctions in situations where the risk of hold-out is very limited-that is, in situations where a SEP-holder can expect to achieve a FRAND remuneration for its SEPs without the use of injunctions.

Ex-post Evaluation in a Merger Case: INEOS/Solvay
Usually ex-post studies are used to evaluate the soundness of past merger decisions. In contrast, in the INEOS/Solvay merger case the Commission used ex-post evaluation as part of the assessment of the proposed transaction. The case involved the creation of a full-function joint venture of INEOS' and Solvay's chlorovinyls businesses in the European Economic Area (EEA). The centre of interest in the transaction was commodity suspension polyvinyl chloride (S-PVC) production in North-Western Europe (NWE) 13 that is used by different types of plastic compounders to make pipes, window frames, fittings, films, and sheets. In NWE INEOS was the clear leader, and Solvay was the second largest supplier.
INEOS already had a history of acquisitions in the S-PVC market: In 2008 it bought Kerling, which had production assets in the UK and Scandinavia; and in 2011 it bought Tessenderlo, which had factories in the Benelux and France. Through these acquisitions INEOS grew from a mid-sized company with 10-20 % 14 share of the S-PVC merchant market in NWE to the market leader with 30-40 % market share. At the time of the notification, INEOS faced four mid-sized competitors (Solvay, KEM ONE, Shin-Etsu, and Vinnolit) and four smaller, typically non-NWE competitors (Anwil, Borsodchem, Aiscondel, and Vestolit). Therefore, even before the proposed transaction INEOS held a very strong position in the market, especially taking into account that KEM ONE, the third largest player, faced serious financial difficulties after the economic crisis.
The past transactions of INEOS offered the possibility of an ex-post assessment of these mergers and of answering the question whether prices of S-PVC increased as a result of these mergers in the relevant geographic market. Under these circumstances, an ex-post assessment can provide information on three important aspects of the market: First, one can establish whether a company held market power already before the proposed merger. If indeed this was the case, merging with the second largest competitor would then potentially cause even larger price effects post-merger.
Second, if the past price increases that were due to the mergers differ across geographic regions then this can provide direct evidence of a Small but Significant and Non-transitory Increase in Price (SSNIP) in the past, hence giving us important information for the relevant market definition.
Finally, price increases that were due to past mergers would indicate that some of the assumptions that were made as part of the past clearance decisions may have to be revisited. In the INEOS/Kerling and INEOS/Tessenderlo clearance decisions four assumptions were made in order to reach a finding that the transactions would not lead to a lessening of competition: spare capacity held by rivals was an effective constraint on price increases; customers could easily switch across producers and therefore could mitigate incentives to raise prices; EEA competitors imposed an effective competitive constraint on the merging parties; and imports or the threat of imports were also seen as reducing the risk of price increases.

The S-PVC Market
S-PVC can be treated as a fairly simple homogenous product. 15 However, as is common in the chemical industry, S-PVC is co-produced with another product (in this case, caustic soda). This joint production is central to the profitability of the industry, because S-PVC prices are highly cyclical. S-PVC and caustic soda have different demand cycles, which provide a natural hedge against demand fluctuations. The stylised vertical production process is the following: First chlorine and caustic soda is produced in a fixed ratio; then chlorine and ethylene is transformed into an intermediate product: ethylene dichloride (EDC). EDC is cracked to vinyl chloride monomer (VCM); and finally VCM is polymerised to S-PVC.
EDC is tradable and if there existed a liquid and competitive market for EDC, S-PVC producers would not need to own the full vertical production process. However, in Europe most producers are fully vertically integrated (either through ownership or through long term contracts), and there is no liquid market for EDC. The absence of a liquid EDC market in Europe implies that vertical integration is the only way to use caustic soda sales as a hedge for S-PVC sales. Therefore vertical integration is key to the profitability of an S-PVC producer in Europe.
S-PVC production is characterised by capacity constraints and spare capacity. Competition in this market can therefore be characterised using a framework of capacityconstrained price competition, as captured by a Bertrand-Edgeworth (BE) model. The BE model was first used by the Commission in the Outokompu/Inoxum 16 decision, and it was also considered in the context of the INEOS/Solvay merger effects in conjunction with remedies and the parties' efficiency claims. The BE model captures the notion that capacity constraints can imply that firms face a residual demand that competitors cannot satisfy and therefore firms can charge prices that lie above costs. In this set-up, a merger can increase market power, because competitors of the merged entity will control less spare capacity. Therefore competitors might be increasingly unable to satisfy the merged entity's residual demand. While the BE model gives a reasonable account of the role of capacity in the S-PVC market, it also has the property that it predicts price ranges instead of one exact price for a market. Therefore merger simulations based on this model predict changes in the price range. Moreover, the model provides a simplified characterisation of how competition operates in the S-PVC market, and it cannot fully take into account important aspects of this case: of geographic product differentiation and vertical integration. For this reason the results of the model were only one element of the Commission's assessment.
Location of S-PVC plants is a key determinant of competition, although S-PVC is transported as far as 1,500 km within the EEA and EEA producers export a significant amount of their production outside the EEA. S-PVC producers sell to sophisticated industrial customers in a just-in-time manner. While this supply management method increases the return on capital by reducing working capital for the buyers, it also makes production more sensitive to supply-chain disruptions. The risk of these disruptions increases with the distance from producers to buyers, which generates a key non-price barrier. Transportation costs also form significant price barriers, especially when coupled with low margins. These two barriers to trade imply that, in principle, producers will have some additional market power over customers near their plants. Therefore, such markets are characterised by geographic product differentiation.
Geographic market definition in these markets is not straightforward, because these barriers are not prohibitive: One may observe S-PVC transported to almost everywhere within the EEA, but transported quantities generally decline with the distance. 17 Therefore, it is not clear at first sight where one should draw "borders" between geographic markets. However, past regional price movements can provide key evidence in this type of cases. Using a regression 18 that compared price movements between NWE and Rest of Europe (RoE), 19 the Commission established that prices in NWE increased 16 Case No COMP/M.6471-OUTOKUMPU/INOXUM. compared to RoE between 2007 and 2012, thus indicating a lack of arbitrage between the two regions. Moreover, in spite of this price divergence, RoE producers sold less S-PVC to NWE although their excess capacity increased. The inability to sell to a market with increasing relative prices indicated clearly that geographic product differentiation is strong enough to define NWE as a separate market within the EEA.

Identifying the Price Effect of Past Mergers
Although the evidence of increasing relative prices in NWE between 2007 and 2012 was relatively straightforward to establish and the price shifts arguably took place in the years following the previous mergers that involved INEOS, the Commission still had to prove the existence of a causal link between regional price differences and those mergers. The construction of this causal link was a complex exercise. Causality was a critical element in order to conclude that INEOS already held a certain degree of market power before the proposed merger. In order to establish this causal link, the Commission relied on an econometric analysis of price effects as well as additional evidence on volume changes and evidence from internal documents.
To perform its econometric analysis, the Commission had access to detailed transaction data of the Parties that recorded their invoices, values, and volumes, for the years 2007-2012. This dataset also included information on the location of customers, delivery mode, S-PVC grade, plant of production, and detailed cost items by plants at a monthly frequency.
Difference-in-differences methods were used to estimate the price effects. These methods compare transactions that were affected by the past mergers with transactions that were not affected by them. The affected transactions formed the treatment group, while non-affected ones formed the control group. The difference-indifferences method estimates merger effects by calculating how the average price difference of the treatment and control groups change after the merger: It looks at changes in relative prices. The control group is assumed to represent how the treatment group would have behaved had the merger not happened. Comparison to this benchmark leads to valid causal inference if the control group follows the same trend as the treatment group and the characteristics of the two groups are similar. The difference-in-differences methodology is used extensively in ex-post merger evaluation. 20 Two difference-in-differences identification strategies were employed to assess the merger price effects: The first compared the prices charged to INEOS' customers in NWE and RoE and calculated how much this price difference increased after the two mergers. This strategy controls for unobserved, EEA-wide shocks before and after the mergers by using sales to RoE customers as a control group. The second strategy looked at the price premium charged by INEOS relative to Solvay, compared them between NWE and ROE, and calculated how much the regional difference in this price premium increased after the mergers. This identification strategy controls for both EEA-wide, region-specific and firm-specific unobserved shocks before and after the mergers. As such it serves as an alternative to using firm-specific and regional covariates.
The second strategy is likely to underestimate the true merger effect significantly, because the mergers' overall effect on prices in the affected S-PVC market is differenced away. In particular, any upward impact on Solvay's prices due to the exercise of market power by INEOS post-merger would be removed from the measured merger effects. In this sense this method relies on INEOS' excess market power compared to Solvay, which is likely to be the result of geographic product differentiation (that is, customers that are close to the assets affected by each transaction suffer from a bigger loss of competition and a resulting price increase than customers that are farther away from these assets). 21 Therefore, this strategy is conservative in terms of the size of the price effect: It gives a lower bound that can be well below the actual effect. The advantage is that it produces particularly strong evidence for the existence of the price effect of the merger.
While the econometric analysis focused on identifying possible price effects of the two past mergers, there were other sources of economic evidence that helped to establish causality.
All in all, the Commission considered that the econometric evidence gathered, together with the evidence on the evolution of regional prices and the qualitative evidence from the market investigation was solid enough to conclude that INEOS held a certain degree of market power within NWE, already before the proposed merger with Solvay.

Conclusion and Commitments Offered
The pre-merger market power finding had three implications. First, it indicated that the proposed merger with Solvay was likely to lead to significant price effects. Second, it implied that the observed price differences between NWE and RoE can be interpreted as a conservative SSNIP for the NWE geographic market.
Finally, it showed that the assumptions in the past with respect to customerswitching, spare capacity, and competitors, which allowed the past mergers to proceed, did not fully hold up in practice.
One question not addressed so far concerns the relevance of ex-post evidence to estimate directly the magnitude of the price effect of the proposed merger. Admittedly, ex-post evidence is probably not suited to provide a direct estimate of future merger effects. Nonetheless, the significant price effects in the past suggest strongly that with the elimination of an additional competitor the future merger effects are likely to be actually even more pronounced than the past effects.
In conclusion, the Commission decided that the transaction, as originally notified, would have consolidated the degree of market power held by INEOS in the market for commodity S-PVC and lead to a significant impediment to effective competition. The theory of harm was unilateral effects: INEOS would have been able and was most likely to have the incentives to, for example, increase prices. This theory was based on the following arguments. First, the transaction removed INEOS' strongest competitor, Solvay, and increased the already high concentration level in NWE. Second, the merged entity would have faced insufficient competitive constraint from the remaining muchsmaller firms in NWE. Moreover, imports from third countries were limited. Finally, due to capacity constraints and geographic product differentiation, customers were unlikely to exercise sufficient buyer power to constrain the merged entity.
To address these concerns, the companies offered to divest three of INEOS' S-PVC plants in NWE (in Wilhelmshaven, Mazingarbe, and Beek Geleen). Since vertical integration was found to be crucial for viability in Europe, the S-PVC plants were offered together with the upstream chlorine and EDC production assets in Tessenderlo and Runcorn. These commitments effectively remove the overlaps between the Parties' S-PVC activities in NWE. Moreover, the parties have committed not to close the proposed transaction before concluding a binding agreement for the sale of the divested business to a suitable purchaser that must be approved by the Commission.

Ex-post Evaluation in State-Aid Schemes
During the last years, the Commission has promoted a State Aid Modernisation (SAM) programme that is aimed at directing public funds towards projects that could address genuine market failures and/or achieve equity objectives, as well as at refocusing the Commission's enforcement in the area. Faced with a vast number of state-aid cases (both aid to individual firms and aid schemes), DG COMP plans to focus on evaluating aid that is more important and a priori more likely to have a distortionary impact on the single European market; it will also offer more flexibility to Member States (MSs) in the implementation of less important aid, through the introduction of a new block exemption regulation that establishes less stringent criteria for state aid to be exempted.
As part of the SAM policy, there will also be a shift from an ex ante perspective (with aid schemes being approved ex-ante by the European Commission after an analysis of their expected costs and benefits) to an ex-post evaluation perspective, with MSs having to assess the effects of their own schemes in a rigorous way. The goal of this shift is to promote a culture of self-assessment of state-aid and therefore more efficient, cheaper and less distortive aid. To assist the evaluation work of the MSs, DG COMP has issued a guidance paper that illustrates the different quantitative methods that MSs could resort to in order to do a proper ex-post assessment of aid measures. 22

Evaluation Plan and the Crucial Early Role of the Evaluator
Most of the problems that are often faced by evaluators are related to either data limitation or the absence of a convincing identification strategy. In order to avoid the problem ex post, MSs should plan their evaluation ex ante, i.e. during of the design of the aid measures. Therefore, the guidance paper emphasises the importance of the evaluation plan, which should describe precisely the rationale for state intervention, its main benefits, and the possible distortions that it creates. The plan should then describe specific convincing identification strategies and ensure that the necessary data will be available in the relevant timeframe. With the notable exception of randomized controlled treatments (RCTs) that are discussed below and in which detailed data is normally gathered, it will normally be preferable to rely on existing data sources and administrative data, such as firms' balance sheets, merged employee-employer datasets, or large national surveys (such as innovation surveys). The evaluation plan should also describe the coverage and limitations of data sources, and make sure that the evaluation strategy is realistic given the existing data.
The evaluation plan should also describe the evaluator. An evaluator's main qualities are skills and independence, which should make academics natural candidates for this role. However, in certain MSs, statistical institutes, central banks or courts of auditors also enjoy the necessary skills and functional independence, as well as a thorough knowledge of data and institutional setups.
Early intervention of the evaluator is crucial. Having the evaluator drafting parts or the entire evaluation plan is the best guarantee for a sound and smooth evaluation process. Being involved early enough, the evaluator would be able to ensure that the design of the aid measure is suitable for evaluation and could propose alternative policies to be tested.

Methodology
A valid estimation is able to delineate precisely the causal impact of the policy itself. This causal impact is the difference between the outcome with the aid and the outcome in the absence of the aid. However, one does not, in general, observe what the outcome would be without the aid for the firms that receive the aid. This counterfactual therefore has to be constructed. This is normally done by finding the most comparable firm(s) or control group. The quality of this control group is crucial for the validity of the evaluation. The following sections briefly discuss the different methodologies to do so.

The Gold Standard: Randomized Control Trials ("RCTs")
The most favourable case for evaluation is when there is no systematic difference between beneficiaries and non-beneficiaries apart from the aid. In this case, the differences in the outcomes can be attributed to the policy. This is the case when beneficiaries are selected randomly.
RCTs are much easier to implement than most policy makers think and have been used in a growing number of applications. 23 As the evaluation of the state aid schemes has to be planned well in advance of the practical implementation of the scheme, early involvement of the experts at the stage of the design of the policy is crucial.
An example of RCT in practice is given by Bakhshi et al. (2013), who use a randomised control trial experiment to assess the effect of an innovative business support scheme. The pilot study began in Manchester in 2009. It was designed such that vouchers, or 'Creative Credits', would be randomly allocated to small and medium-sized businesses that applied to invest in creative projects. The research found that the firms that were awarded Creative Credits enjoyed a short-term increase in their innovation and sales growth in the six months that followed the completion of their creative projects. However, the positive effects were not sustained, and after 12 months there was no longer a statistically significant difference between the groups that received the credits and those that did not.
Moreover, running randomized experiments in the area of state aid schemes does not require the random selection of aid beneficiaries. It would for instance be possible to introduce elements of randomness in the eligibility, in the incentives to participate, or even in the level of advertisement of the programme. For a very innovative policy, it might be advisable to evaluate a pilot programme first. This pilot would likely be of a smaller size. Beneficiaries could then be chosen randomly. Randomization in the ramp-up of a scheme-for instance, by making available state aid to a randomly determined subgroup of the beneficiaries firms in the first year and giving the state aid gradually to the remaining beneficiaries over the following years (or alternatively, to advertise the scheme to a progressively larger audience)-could also be envisaged. 24 For a new policy, a period of ramp-up is in many cases an administrative necessity. Deciding randomly who can benefit first from a policy is often not less fair than other selection rules. It can also and can lead to very efficient evaluations.
The broad ideas presented above are mainly feasible for the implementation of totally new schemes (or a large variation in an existing one). Nevertheless, it is probably rather difficult to randomize eligibility for the continuation of a large and well-known existing scheme. However, even in this case, randomization could also be used at the margin. In particular, it may be still possible to select beneficiaries randomly for potentially more efficient, more targeted, and/or less distortive variants of the scheme.

Selection Effects
If they have not been chosen randomly, firms that receive aid are by definition in a different situation. Then, naively comparing beneficiaries with non-beneficiaries is likely to reflect pre-existing differences that led to the selection of beneficiaries in addition to the effect of the policy itself. Some of the differences between aid beneficiaries and non-aid beneficiaries are observable. It is then necessary to take these observable differences into account. This can be done by different means: for instance, through linear regression or matching.
However, some of these differences could also be unobserved. For example, a MS submitted a study that was aimed at assessing the employment effect of state support of investment projects in less developed areas. It matched aid beneficiaries with non-aid beneficiaries in the same region and sector and of a similar size and age. The problem is that these same criteria were used to determine eligibility for aid, which implied that the non-recipients of aid in this sample were in principle potentially eligible for aid as well. Most probably, therefore, they did not receive aid because they had no investment project that could have been subsidised in the first place.
Consequently, the comparison between the two types of firms highlights the effect of an investment project on employment and has possibly little to do with the effect of the aid itself.
The selection effect is generally a serious challenge, and the credibility of an evaluation largely relies on a very careful analysis of the design and context of the aid and on a proper research design to eliminate selection effects.

Quasi-Experimental Methods
Even though random experiments are without any doubt the gold standard, other methods have been developed to evaluate properly the effects of a policy. They all use exogenous variations of the environment, which create situations very close to experiments: natural or quasi-experiments. The guidance paper presents the most common of these methodologies: differences-in-differences, regression discontinuity design, and instrumental variables. 25 These three methodologies derive their validities from different assumptions. The best choice is normally driven by the context of the policy and the availability of data.
For example, Martini and Bondonio (2012) have examined an investment grant available throughout Italy (Law 488) using exogenous variation in eligibility. They compared firms that had their aid application approved (i.e., the aid-beneficiaries) with comparable firms that had their aid application rejected. This happened when the available budget had reached its limit. This group of "rejected applicants" had passed the first quality check. It means that they had a credible investment project and just did not receive money because of budget rationing. Therefore, this comparison is better that just comparing applicants with non-applicants: the selection bias is likely to be minimal. Then, the difference in performance between the successful applicants and the (closely) rejected applicants provided a reliable estimate of the effect of the aid.
The Guidance Paper insists that it is not the use of a specific econometric technique that allows the identification of the effects of a policy; it is the exogeneity of the control group and, ultimately, the quality of the counterfactual. With the proper research design, even simple matching on observables might provide interesting insight (Givord et al. 2013), but this is likely to be an exception rather than the rule.

Conclusion
By shifting the focus of state aid evaluation towards ex post assessment of aid measures by the Member States themselves, DG COMP has promoted a change of mentality. To the extent that ex post evaluation will be done properly and rigorously, it will allow the gathering of an important amount of knowledge on the effects of aid and create the conditions for making state intervention more targeted, cheaper, and less distortive.