Behavioural economics and compliance with the DMA

The Digital Markets Act (DMA) regulates the power of large digital platforms in Europe with the primary objectives of enhancing fairness and contestability in digital markets. It became applicable in March 2024, prescribing a range of DOs and DON’Ts that large platforms must follow. In this article, Zita Vasas and Abul Fazal [1] review the enforcement and compliance process so far, specifically assessing the role behavioural insights have played.

First, for context, the authors outline why behavioural economics has become relevant to competition policy, particularly in the digital sector. Then, they briefly explain how the emergence of ex ante regulations affects the Commission’s enforcement process relating to its concerns.

They make three observations of the DMA’s compliance process during its initial years in practice. So far, it has been:

• largely effective at removing and preventing major behavioural concerns from the past;
• relatively swift at identifying, investigating, and addressing potentially anti-competitive conduct; but
• there are still concerns regarding compliance and there remain a number of unresolved issues.

The authors conclude by highlighting the following challenges to the process going forward:

• improved choice architecture may not materially change outcomes in many services;
• the need for empirical testing; and
• the need to adapt processes to new (AI-enabled) services.

The growing relevance of behavioural economics to competition policy, particularly in the digital sector

For some time, competition authorities have been concerned that providers of digital services can influence users’ decisions by altering the design of their choice architecture — the environment that determines which options are presented to users and how — in ways that may harm competition.

The behavioural economics of choice architecture

These concerns are grounded in the extensive behavioural economics literature, which catalogues numerous factors that influence decision-making in ways that (seemingly) depart from the rational behaviour classical economics assumes.

In particular, consumers’ decisions may vary depending on various features of the choice architecture.[2] For instance:

• the number of options presented: the design of the choice architecture can steer users towards a particular option either by providing too few alternatives (effectively restricting choice) or too many (choice overload);
• the information provided: familiar options can gain advantage if the information provided about alternatives is too little (shrouding) or too much (cognitive overload);
• the contextual signals presented: the architecture can steer users towards particular options by increasing their salience or prominence relative to others’ — for instance, by presenting the preferred option in vivid colours and others in grey, or by using social cues and prompts such as reviews, popularity metrics, or “scare screens” warning users about potential risks of third-party providers;
• the relative effort of the selection process: the architecture determines how laborious it is to select each option. Some options can be made easier to choose (e.g., through defaults or pre-installations), while others can be made more difficult — for instance, by complexification, which makes the administrative steps unnecessarily onerous, leading to a higher likelihood of users abandoning the process before completion;
• the timing of the selection process: user choice may differ depending on whether an architecture requires a decision when the user is more likely to devote attention to it (such as during initial setup of a device), or when the user is focused on something else (such as when opening a search engine to look something up). In the latter case, users are more likely to stick with the default option than consider alternatives so that they can swiftly complete the action they were doing.[3]

Why choice architecture matters in digital services in particular

Authorities have been particularly concerned that digital platforms can optimise their choice architecture to give an advantage to their own complementary services (self-preferencing) at the expense of rivals — making these markets less contestable. The focus on digital platforms broadly stems from two main reasons.

First, the impact of choice architecture on users is likely to be especially significant in digital markets given the following characteristics that make users susceptible to the influence of choice architecture in digital settings.

• Numerous decisions. Users face many small choices, leaving them prone to choice overload and decision fatigue, which amplify the power of defaults and the influence of contextual signals.
• Low stakes for users. For each specific decision, the stakes for the user are relatively low — the marginal benefit of choosing the best option is limited, as is the potential for long-term regret from choosing a less optimal one.
• High capabilities and incentives for provider. For digital service providers, both the ability and the incentive to influence user choices are high. Small shifts in many users’ behaviour can generate large aggregate rewards, and access to user data enables providers to test and optimise their choice architecture continuously.[4]

Second, many digital services combine (a) a platform that generates significant network effects — creating a degree of user lock-in or dependence — with (b) an ecosystem of interoperable products and services offered by the platform provider and by third parties. This structure creates opportunities for self-preferencing, whereby platforms can leverage users’ dependence on the core service to distort competition in complementary markets.

It is these features — network effects, user dependence and ecosystem integration — that make self-preferencing and choice architecture particular concerns in digital markets. In contrast, authorities are generally less concerned about, for example, a supermarket that designs its store layout to favour its own-brand beans over (better quality) third-party ones. The influence of the choice architecture may be just as powerful as in a digital platform, but consumers’ dependence on it is not: they can easily switch to another supermarket that makes it easier to find their preferred brand.

Anti-competitive choice architecture is easier to identify than pro-competitive architecture

Importantly, the nature of these concerns is that – while it is often possible to recognise what “bad” or anti-competitive choice architecture looks like – it is hard to define what “good” or pro-competitive design should look like.

There are several reasons for this asymmetry.

• There is no neutral or bias-free choice architecture. Every interface or presentation involves design choices — the order of options, their framing, their visual emphasis — all of which influence behaviour. A fully neutral design, that does not influence users, is not a coherent alternative; every design affects users in some way, even if unintentionally.
• A fully informed choice is not practical – if it is even ideal at all. In theory, users could make fully informed choices – where they understand the pros and cons of each option available to them – but this is unrealistic. Consumers do not have the time, information, or cognitive capacity to process all available alternatives in a perfectly rational way. Nor would they want a choice architecture that tried to promote fully informed choices.
• The biases in choice architecture can both harm and benefit consumers. Many design features that influence behaviour are not necessarily harmful — they can help users make faster or better decisions by reducing cognitive effort. In choices with low-stakes, consumers often prefer convenient short-cuts, such as recommendations, popularity indicators, or defaults, because these lower the mental cost of decision-making. However, these can introduce trade-offs that are difficult to assess. For instance, warnings about third parties’ services may be “scare screens” that deter users from choosing competitors, but also genuinely protect users from risks. Similarly, popularity metrics can help users avoid wasting time assessing bad options, but also make incumbents with a historical advantage difficult to challenge.
• The effect of a given structure can vary in different situations. There will rarely be a design that is pro-competitive in every circumstance. For instance, a specific number of options may be sufficient in one context, too few in another, and lead to choice overload in a third.

These features can make effective regulation challenging. Typically, a regulator can identify specific elements of a choice architecture that clearly distorts user choice and give an unreasonable advantage to the platform provider. Examples include deceptive prompts, excessive restriction of alternatives, or asymmetric burdens imposed on users depending on whose service they aim to use.

However, it is far harder to specify what a pro-competitive design would look like. A particular choice architecture could be adjusted in many ways, some of which may superficially appear to better promote user choices, but still skew their decisions in practice. Moreover, there will always be changes that could channel more users to competitors, but that is not the goal of competition policy – authorities aim to protect the competitive process, not to arbitrarily improve competitors’ outcomes.

As we will explore, this asymmetry affects how authorities address their concerns about digital choice architecture – both in terms of how they specify their objectives, and how they prioritise assessment and intervention.

The emergence of ex ante regulation of digital markets

In the past, authorities have attempted to address their concerns about choice architecture in digital services through ex post investigations. That is challenging to do effectively, as competition cases react to specific instances of harm after they occur.

In contrast, the advent of ex ante regulation in digital markets, exemplified by the DMA, may help address the limitations of an ex post regime. In addition to removing existing practices that likely have been anti-competitive, an ex ante regime seeks to prevent harm before it arises, by imposing ongoing obligations and compliance assessments on large digital platforms (gatekeepers) to ensure fairness and contestability in the affected markets.

The challenge with addressing behavioural issues through ex post investigations of digital services

For at least the past decade, insights from behavioural economics have featured in European competition authorities’ ex post investigations of digital services. The importance of choice architecture has featured in cases affecting most of the GAFAM firms.[5]

• Alphabet (Google). In Google Shopping (2017), the European Commission found that Google abused its dominance in general search by giving its own comparison-shopping service prominent placement at the top of search results, while demoting rival services. Since users tend to engage more with the first results shown, this design diverted traffic towards Google’s own service, and away from competitors.[6] In Google Android (2018), the Commission found that Google abused its dominance by requiring device manufacturers to pre-install Google Search and Chrome. In 2022, the General Court largely upheld this finding, noting that status quo bias makes users reluctant to download alternatives and that pre-installation has the same practical effect as a default, based on behavioural evidence on user switching.[7]
• Amazon. In 2021, the Italian Competition Authority (Autorità Garante della Concorrenza e del Mercato, or AGCM) found that Amazon abused its dominance in online marketplace intermediation by tying its logistics service, Fulfilment by Amazon (FBA), to a set of exclusive seller benefits, most notably the Prime label. The investigation showed that offers with the Prime label were much more visible to consumers, steering buyers towards FBA sellers.[8]  Likewise, in its Buy Box investigation, the Commission preliminarily found that Amazon’s rules for selecting the Buy Box winner and granting Prime eligibility favoured Amazon’s own offers and FBA users, steering consumers towards those default options. To address these concerns, Amazon committed to treat all sellers equally in Buy Box rankings and to display a second competing offer, making it easier for consumers to see and choose alternative deals.[9]
• Apple. In 2024, the Commission fined Apple over €1.8 billion for abusing its dominance in the distribution of music streaming apps through the App Store. The Commission found that Apple’s anti-steering provisions prevented developers from informing iOS users about alternative or cheaper subscription options outside the App Store. The Commission noted that these restrictions negatively affected users – who could not make informed and effective decisions on where and how to purchase subscriptions – and that many may have paid higher prices as a result. It also noted non-monetary harm in the form of a degraded user experience: users faced a cumbersome search to find external offers, or never subscribed at all.[10]
• Microsoft. In 2004, the Commission found that Microsoft had abused its dominance by tying Windows Media Player with the Windows operating system – a finding later largely upheld by the General Court. The Court noted that this practice altered competition by ensuring the widespread presence of Windows Media Player on personal computers, creating a disincentive for users to install third-party players and for manufacturers to pre-install competing media software.[11] Similarly, in 2009 Microsoft agreed to address the Commission’s concerns about tying Internet Explorer to Windows by displaying a browser choice screen to Windows users. The choice screen was rolled out in March 2010 and was deemed successful by the Commission, with around 84 million browser downloads by November 2010.[12]

Broadly speaking, this form of intervention has proved ill-suited to resolve the competition concerns that arise in digital markets. The main criticisms are the following.[13]

• The assessment process is too long. The process to identify concerns, launch an investigation, assess the issue and reach a decision is long. The criticism is that by the time a conclusion is reached it is often rendered obsolete given the fast-paced nature of digital markets.
• The scope is too narrow. Focussed and backward-looking investigations make it difficult for authorities to either provide guidance that would help providers avoid their concerns, or create meaningful deterrence effects that would reduce the incentive to develop anti-competitive architecture in the first place.
• The source of market power is not necessarily anti-competitive behaviour. If a digital platform achieves significant market power stemming from intrinsic characteristics of digital markets (such as economies of scale and network effects) but without any anti-competitive actions, ex post competition law is clearly not the right tool to enhance competition as absent anti-competitive behaviour, an abuse of dominance cannot be established.
• Remedies are ill-suited to address concerns. The interventions required to develop, implement, and monitor the performance of adequate remedies are better suited to an ongoing regulatory regime rather than to one-off investigations.

The shift to ex ante regulation of digital platforms

Several jurisdictions are considering adopting, or have already adopted, ex ante regulation of digital platforms to address competition concerns more quickly and effectively. Here, we focus on the implementation of the European Union’s DMA. But, the UK is taking a similar approach in its new digital markets regime, where the Competition and Markets Authority (CMA) recently designated Google with Strategic Market Status (SMS) in general search and search advertising services, [14] and proposed designating Apple and Google with SMS in mobile ecosystems.[15] While the European Union and the UK have moved first with dedicated regimes now in force, other jurisdictions – such as Australia, Brazil, India, Japan and South Korea – have also considered legislative proposals, or are yet to fully implement comparable frameworks.[16]

The DMA draws on behavioural insights from earlier competition cases, but aims to overcome the limits of ex post enforcement. This is evident both in the substantive obligations imposed on designated firms and in the process put in place to ensure compliance.

The requirements of the DMA

Behavioural economics underpins some of the DMA’s overarching requirements and several of the operational duties imposed on designated digital gatekeepers.

• Overarching requirements.[17] The DMA aims to enhance fairness and contestability in digital markets. Article 8 sets out that gatekeepers must implement provisions in a way that is effective in achieving these overall objectives. In addition, Article 13 prohibits circumvention, including via the use of behavioural techniques. Thus, the gatekeepers are expected to comply with not only with the word, but also the spirit of the rules.
• Specific requirements. In practice, much of the DMA includes formal rules that guide what the Commission regulates and how – at least in the initial stages of its implementation.[18] Certainly, the articles of the DMA set out requirements and prohibitions that relate to specific conduct on particular platform services that gatekeepers must comply with. These provisions directly target practices previously scrutinised from a behavioural perspective in ex post investigations.

For instance, Article 6(3) specifically addresses the use of pre-installations and defaults by gatekeepers – common focal points of ex post assessments prior to the DMA. The article requires that end users must be able to easily uninstall apps (with narrow exceptions for essential apps), easily change defaults on a gatekeeper’s operating system, web browser, and virtual assistant, and must be prompted at first use of these services to choose from the main available providers (a choice screen).

Similarly, Article 6(5) addresses gatekeepers’ ability to manipulate the ranking of search results to give preferential treatment to their own services. Wherever a gatekeeper ranks results (e.g., in search results, app-store listings, or marketplace options), the DMA requires that the gatekeeper must not favour its own products/services over similar third-party offerings, and must apply transparent, fair, non-discriminatory ranking conditions.

The compliance and assessment process in the DMA

Unlike ex post enforcement, the DMA’s compliance regime introduces ongoing and iterative assessment. It is designed to assess, monitor, and refine how gatekeepers’ design actually affects users’ choices and competition in practice.

The core features are (see Figure 1).

• Designations. Effective enforcement began with the first gatekeeper designations on 6 September 2023.
• Initial compliance reports. Designated gatekeepers prepare compliance reports, demonstrating how their services meet requirements. The first round of reports were published on 6-7 March and 1 November 2024.
• Initial scrutiny and dialogue. The Commission reviews the reports and holds workshops to discuss the gatekeepers’ approach, providing the opportunity to consider evidence on the impact that the chosen design, and alternatives, would have on business users and end users. The first round of compliance workshops took place on 18-26 March and 25 November 2024.
• Targeted scrutiny and specification. Article 8 of the Act covers specification investigations, which allow the Commission to further specify how obligations apply in practice. In this context, the Commission has opened two specification proceedings: one on Apple’s interoperability obligations under Article 6(7); and another on Apple’s iOS connectivity features for connected devices. The Commission adopted decisions for both in March 2025.[19]
• Specific non-compliance proceedings. Where concerns arise, the Commission can open focused investigations into potentially non-compliant conduct. To date, it has opened six non-compliance cases. The first five (against Apple, Meta and Google) were launched within a month of receiving the initial compliance reports. The sixth was launched in June 2024 (against Apple).

The Commission’s non-compliance investigations can result in: (a) no adverse findings, either by providing evidence that allays the Commission’s concerns, or by adjusting the architecture to address concerns; or (b) adverse findings and fines.

Figure 1: Timeline of key events in the DMA compliance process

Importantly, the compliance process is an annual cycle, allowing the Commission to monitor compliance and performance on a continual and iterative basis, updating its understanding and assessment. The second round of compliance reports were published in March 2025.

After the first two years, the Commission launched a consultation on its first review of the DMA and issued a call for evidence on the effectiveness of the DMA.[20][21]

In principle, therefore, the DMA was designed to tackle both the specific concerns identified in the Commission’s ex post assessment of the choice architecture in digital services, and also serve the broader principle that firms should not design their core services in ways that hinder competition, even as they evolve.

In the next sections, we assess how well the compliance process has worked in practice and the issues that may emerge.

How well has the compliance process worked in practice?

In the first compliance cycle, the Commission designated seven companies as ‘gatekeepers’ and opened six non-compliance investigations.

As shown in Table 1: (a) one investigation has been closed with no adverse findings, following Apple committing to changes to address concerns with its user choice obligations; [22] and (b) two investigations have resulted in infringement decisions and significant fines: Apple’s App Store steering and Meta’s “pay or consent” model.[23]

The remaining three cases, covering Google’s search and Play Store practices, and Apple’s contractual terms for alternative app distribution, are still ongoing.

Table 1: Summary of DMA non-compliance investigations to date

Reviewing the DMA compliance process so far, we highlight three observations. The ex ante approach has been:

• largely effective at removing and preventing major behavioural concerns from the past, particularly on the end user side, and with regards to choice screens;
• relatively swift at identifying, investigating, and addressing potentially anti-competitive conduct, certainly in comparison with the ex post regime that preceded it; but
• there are still concerns regarding compliance and there remain unresolved issues, particularly for business users of designated platforms.

Removing and preventing major behavioural concerns of past investigations

A clear success of the first round of compliance assessments is that many of the concerns the Commission had had with firms’ choice architecture have been removed and have not re-emerged.

First, the ex ante regime appears to encourage parties to self-censor, avoiding the worst and most blatant behavioural techniques – such as flagrant self-preferencing – to a greater degree and with a broader scope than they likely would have avoided under a purely ex post regime.

This is particularly the case with browser and search engine choice screens as Apple’s and Google’s choice architecture largely avoids past techniques: their services are not more salient than competitors; nor are they presented as literal or de facto defaults (in the choice screens); neither does the design include any apparent complexification for rivals, such as various additional screens; and messages do not deter users from selecting third party services. Additionally, there is no indication of information overload – for instance, Google includes a short description, and Apple provides short titles – and social cues – such as number of reviews – are not included in the choice screens to steer users towards established providers.

This observation, to some extent, should be expected and represents the low hanging fruit for gatekeepers seeking to comply with the requirements. However, that should not diminish the extent to which the initial compliance cycle demonstrates the advantage of an ex ante regime over ex post investigations.

Second, the gatekeepers have implemented a wide range of changes to their services in order to comply with the DMA. For instance, Apple has introduced a browser choice screen, created a centralised menu for managing default apps, and now allows users to delete pre-installed apps that were previously not possible to remove – including the App Store, Messages, Camera, Photos, and Safari.[24] Google has introduced choice screens for both browsers and search engines, and has also implemented significant changes to the design of its search results to avoid preferential treatment of its own services.[25] Similarly, Microsoft has also introduced new ways for users to manage default applications[26] and it no longer promotes other Microsoft products in Windows.[27]

A relatively swift process

Another broad success is the pace and effectiveness of the assessment process.

First, it is fair to say that the Commission has been remarkably fast and active in enforcing the DMA. The annual compliance process is up and running. The Commission issued preliminary findings in two non-compliance investigations within a few months of opening them, and issued decisions in roughly a year (see Table 1 above). Compare that with the timelines of ex post investigations – both by the Commission and others – and the comparative speed of the DMA regime is clear. Consider, for instance, the Google Shopping case, in which the Commission issued the infringement decision six and half years after it opened the investigation in November 2010.[28]

Second, the iterative nature of the process allows the Commission and the gatekeepers to find solutions in a relatively efficient and effective way. In particular, gatekeepers have the opportunity to discuss implementation with the Commission and update their design where required. For instance, following the Commission’s findings that its earlier measures still favoured Google services such as Shopping, Hotels and Flights, Google proposed further changes in June 2025.[29] Similarly, the Commission closed its investigation assessing whether Apple had provided end-users with sufficient ability to un-install any software applications and to easily set the default web browser, and other settings on iOS, after Apple had announced it would implement new measures to comply with the DMA.[30] These updates included design changes, such as easier browser selection through the choice screen and changes to the positioning of the browser icons on the device’s home screen and dock.[31]

Unresolved compliance issues

Despite the achievements listed above, there remain concerns around the use of behavioural techniques in gatekeepers’ compliance.

First, this is reflected in the non-compliance decisions against Apple and Meta.

• In relation to Apple’s App Store, the Commission found that (i) Apple restricted third-party developers’ ability to provide end-users with links to external websites when promoting their offers within app, [32] which made it unduly difficult for end-users to locate the offer, [33] (ii) Apple’s disclosure sheet contained a warning for unsubstantiated risks to user’s privacy and security, [34] which could deter end-users from following through, and that (iii) Apple did not allow third-parties to include additional data on the URL, forcing end-users to re-enter information, [35] which increases frictions.[36]
• In relation to Meta’s “pay or consent” model, the Commission found Meta’s model was non-compliant, as it forces end users to choose between either sharing their personal data for use in targeted advertising, or paying a monthly subscription fee to access Meta’s social media services.[37] In its decision, the Commission notes that one of the reasons why users may be hesitant to switch to the fee-based model is that end users of Meta’s social networking services have grown accustomed over the years to using them for free.[38] Here, the Commission may be concerned about the impact of status quo bias, that is, our tendency to stick with what we are used to. The Commission also notes that an equivalent alternative to its current free-to-use service should (among other things) be presented in a neutral manner with choice flows that allow end users to freely choose that alternative.[39]

Second, some of the ongoing non-compliance investigations are clearly concerned about behavioural issues. For instance, the Commission’s preliminary investigation found that Apple made it overly burdensome and confusing for end users to install apps when using such alternative app distribution channels.[40]

Third, business users remain concerned about their ability to compete. For instance, DuckDuckGo claimed in November 2024 that switching search engines and browsers on Android devices did not become easy (complexification) [41] and we are unaware of further changes since then by Google in this respect.

Finally, there are some clear gaps in the application of the DMA’s principles. No choice screens have been introduced for virtual assistants as no gatekeeper has been designated with respect to these services. This, in particular, indicates one of the structural challenges in the DMA compliance process. Its major achievements so far relate to mature products and services, where users’ preferences have already been shaped by years of use. But once formed, those habits and preferences can be hard to change, even if the constraints of the original choice architecture that moulded them is loosened. In contrast, users’ preferences for relatively new products and services, such as virtual assistants, are less likely to have ossified, and so the principles that underpin the DMA’s ex ante regulation could have greatest impact here. But the rules that determine its application do not (yet) apply.

Considerations for the compliance process going forward

The DMA’s compliance process will need to evolve; its success will depend on how effectively it adapts. We highlight three relevant considerations:

• improved choice architecture may not materially change outcomes in many services;
• the need for empirical testing; and
• the need to adapt processes to new (AI-enabled) services.

First, mature markets may be resilient to changes in choice architecture design. The DMA cannot be expected to reverse the impact of past practices.[42]  Where users’ preferences have formed under the choice architecture authorities have been concerned about, it is unlikely the DMA can now undo those habits and preferences. For instance, Pape-Rossi (2025) finds that removing Google’s “one-click advantage” did not shift traffic towards alternative map providers but instead increased user friction.[43]

While there are meaningful relative gains for smaller rivals – with Mozilla reporting that Firefox daily active users doubled in Germany and France on iOS devices over the first year after the launch of browser choice screens, [44] and Opera and Vivaldi also reporting substantial growth in their user traffic [45] – these are unlikely to materially erode the leading providers’ overall market shares (at least in the short run and without any disruptive innovation).

The interventions that could shift deeply embedded preferences likely go well beyond what DMA compliance would contemplate. For instance, Allcott et al. (2025) found that a significant share of users paid to try Bing rated it more positively and continued using it afterwards (albeit some due to inertia and not because they realised they preferred Bing). Users who were simply prompted to make an active choice between search engines – without any financial incentives – largely stayed with Google.[46]

Second, empirical testing must inform compliance assessments and prioritisation going forward. For the reasons set out above, it will be difficult to categorically say which particular design is most effective in enhancing contestability in practice. For instance, the choice screens Apple and Google introduced show up to 12 alternatives.[47] That, in principle, should make the markets more contestable than they were without choice screens. However, in practice, will users engage with assessing 12 alternatives? Even if they do, could user engagement be further improved by showing fewer alternatives? If so, is that a priority, or even necessary, to comply with the DMA?

These are empirical questions that can only be answered via testing. We note that Google first introduced stratified randomisation showing the top five alternatives in random order before the others but subsequently changed the design.[48] The role that empirical testing played in the assessment of that specific change is important – as, from first principles, we expect stratified randomisation to better engage users, as it reduces the risk of choice overload.[49]

Empirical testing has played a role in the compliance process already, and the Commission has criticised parties for not providing results of A/B testing or other empirical evidence.[50] However – for understandable reasons – there is very little available in the public domain on the scope of any such testing, the alternatives considered, or their results. Given the inherent empirical nature of establishing the impact of behavioural biases, these issues will increasingly matter in practice, as proper compliance assessments and prioritisation can only be carried out once design choices have been tested and evaluated.

Third, a key question is how the DMA will adapt to new services and disruption, and in particular AI-functions. The DMA’s initial designation and compliance process is informed by lessons from the past.[51] However, the formalism that has been developed to assist the regulation of familiar services provided by familiar firms will become less important, as new services emerge, with new choice architecture, in new contexts. How the DMA identifies, assesses and addresses these new challenges will be the true litmus test of ex ante regulation.

---

References