Theory and praxis of DAOs

Evident in the emergence of several new projects, decentralized autonomous organizations (DAOs) are currently experiencing new popularity. Even though early implementations such as the infamous TheDAO¹ already surfaced a long time ago, the conceptualized theory surrounding DAOs is still in a very nascent stage.

This report examines the current status quo, proposes a working taxonomy of DAOs, and looks at relevant industry efforts.

1. Decentralized autonomous organizations

Soon after Bitcoin rose to initial popularity, DAOs quickly became a new dream for many early developers and adopters. The idea of having a digital bearer instrument like Bitcoin opened up countless opportunities. Everything seemed possible all of a sudden.

DAOs promised to revolutionize how we structure socio-economic relationships. No one really knew how to do this and it was – and still is - a very explorative process.

So what exactly do DAOs promise?

The ability to structure large-scale, merit-based communities with minimal entrance barriers, factor in externalities, and – much in the Bitcoin spirit – break free from relying on traditional institutions. While Ethereum has been playing a dominant role in enabling DAOs, smart contract risk related to its programming language, Solidity, was ironically also a significant impediment for the maturing of DAOs.

In the following, this report explores the current status of DAOs. However, before doing that, we have to define what precisely a DAO is.

In the absence of a commonly agreed-upon definition, Binance Research proposes the following working definition for DAOs:

“An organizational form that coordinates the efforts and resources of members via an a priori binding, formalized, and transparent set of rules that are agreed upon in a multilateral fashion.”

This definition abstracts away from the underlying technology and is strongly driven by functionality. It will be developed in a step by step approach by examining the three blatantly obvious components of any DAO: decentralization, autonomy, and organizations.

1.1 Decentralisation

Usually, one of the core elements of any definition of a DAO is the notion of “decentralization”. This often talked about, yet poorly understood concept is avoided as it warrants a more detailed analysis on its own.

Within the crypto-verse, the term “decentralization” either refers to the physical distribution of hardware (think: validating nodes & full nodes) or the diffusion of political influence. In this case, the latter is relevant. Instead of having a classical hierarchical power distribution, the common approach of any DAO is to foster coordination on a peer-to-peer basis. This element can be described as a multilateral decision-making process.

“[…] agreed upon in a multilateral fashion.”

1.2 Autonomy

The second major component of any DAO is “autonomy”, which can be best described by the ability to govern itself².

This ability to be self-governed arises from using smart contracts to stipulate software encoded “rules of the game” for any given DAO. Accordingly, a DAO is governed by an a priori binding, formalized, and transparent set of rules. The respective smart contracts may execute organizational functions in an automated manner, irrespective of the involvement or absence of any other party.³

“[…] via an a priori binding, formalized and transparent set of rules […].”

The autonomy of any DAO could thus be generalized as “the degree to which smart contracts govern activity”.

When adopting a new institutional economics perspective, it is possible to describe all activities by value streams. This description is especially apt for a transactional ecosystem that is not based on (interpersonal) trust. The ability of a DAO to initiate value streams - i.e., to act and therefore be autonomous - thus hinges on the ability to allocate and manage resources (i.e., funds). The autonomy of any DAO is consequently inherently linked to the DAO’s direct control over resources.

“[…] the efforts and resources of […].”

1.3 Organisation

By focusing on what is governed by multilateral agreements and how a DAO exerts control over resources, these first two components are foremost defining the scope of any DAO.

Besides the scope, there must be an eventual consensus on how to act within this “multilaterally agreed upon autonomy”. Essentially the question is: how can a DAO decide over what components it has control?

In order to answer this question, some deliberation and a subsequent decision-making process must be present.

The deliberation must formalize some agreed-upon discussion process, which could be either discretionary or rule-based. Usually, discussed upon proposals are subsequently voted on. By definition, this voting process has to be conducted on-chain, as it would otherwise not be possible to have certainty about rule adherence.

“An organizational form that coordinates […].”

Generally, the only way to bootstrap this organizational form - the DAO - is by using decentralized ledger technology that is capable of executing smart contracts. Contractual arrangements also protect the involved parties against contract breaches by providing means of retaliation that go as far as to state coercion. Unlike smart contracts, they can, however, not restrict the scope of eligible action a priori by testing requests against business validation rules and thus render (smart) contract breaches impossible.

2. The DAO: a many-faced organization

2.1 Forms of DAOs

With a better understanding of what a DAO is we can take a look at actual implementations thereof. It has already been well described⁴ that this organizational entity can take on manifold appearances.

With a better understanding of what a DAO is, we can take a look at actual implementations thereof. It is already well described that this organizational entity can take on manifold appearances.

They can, for example, be:

• A blockchain: Dash, for example, funnels a part of their block rewards into an investment pool that can be allocated by members. Arguably, even Bitcoin should be considered a (primitive) DAO, as it heavily leverages incentives to execute a specific function, i.e., agree on the state of the UTXO set.
• An ecosystem: some DAOs such as Aragon, for example, have created an entire ecosystem - including apps, a digital judiciary system, and more.
• A protocol: as of now, the most successful DAO is certainly MakerDAO. MakerDAO and its DAI is a core part of the current DeFi stack and is primarily governed by MKR holders, the DAO members.
• A mutual insurance: NexusMutual allows protecting against smart contract risk by ensuring against a smart contract failure.
• Natural resources: social and economic activists have already long asked the question of why the only non-humans having a legal entity should be organizations. New Zealand, for example, already granted legal personhood rights to a river. Coherently, Terra0 is attempting to create a DAO to allow forests to be self-governed by auctioning off licenses to log trees, accumulating capital to purchase, and eventually own itself. Finally, such “forest DAO” might use its capital and not only buy itself, but surrounding forests. While this may sound esoteric, such structure is unique and allows for recognizing often neglected commons. Giving commons an ability to participate in a marketplace on a level-playing field, could be a first step to limit externalities of business operations. Presumably, however, this hinges on respecting and enforcing the rights of such commons.

2.2 Facilitators

The emergence of DAOs is greatly facilitated by multiple “operating systems” (OS) for DAOs that make it possible to get set-up quickly and use “plug-and-play” like solutions. The most popular, currently emerging frameworks are Aragon, DAOstack, and Colony. It should, however, be noted that these projects are still in an early stage of their development, and many use cases thereof are simply experimentations. Aragon is the currently most mature OS. These operating systems are complemented with - oftentimes - minimalistic funds allocation frameworks such as Moloch, MetaCartel, and TheDAO.

The following charts show the most relevant DAO facilitators as defined by the number of projects that used the respective smart contracts and the USD denominated resources that are locked within them.

Chart 1 - USD denominated amount of locked assets (as of November 15th 2019)

Chart 1 tellingly shows that, by the amount of (USD denominated) assets locked within a DAO’s smart contract, TheDAO significantly outperformed any other DAO. Notably, for TheDAO, the asset (i.e., Ether) valuation from 2016 is used. Using the current valuation leads to the net worth of TheDAO’s to rise by a factor of 10.

Chart 2 - Number of summoned DAOs/successful fund requests

Chart 2 highlights the significant number of DAOs that already exist. Instead of, however, doing a case by case analysis of the “summoned” DAOs, or the initiating DAO OS’es, this analysis will take a macro perspective.

3. Taxonomy

With diminishing entrance barriers, it is increasingly likely that more and more projects will use one of the existing DAO frameworks to set up their own DAO. And that is precisely why there is a growing need for a framework that helps with systematically clustering DAOs.

Subsequently, Binance Research proposes a tiered taxonomy to classify DAOs that builds on the previously identified elements of DAOs: the scope of multilateral agreements, resource management, and the discussion and voting process⁶.

Generally speaking, every taxonomy is a classification exercise that aims to account for all differentiating features and potential implementation thereof. At the same time, however, a taxonomy should not allow overlaps. Thus, taxonomy elements are to be mutually exclusive but collectively exhaustive.

Most importantly, taxonomies are to only account for the differentiating - and, therefore, most relevant - elements. Taxonomies should not be complete structures allowing for every potential implementation. For this reason, the following working taxonomy of DAOs consists of only four main ../components. Design choices in these elements are likely to have repercussions on several other aspects of DAOs that will be briefly mentioned subsequently but are not considered crucially important.

The working taxonomy of Binance Research itself is tiered and structured as follows:

• Tier 1: Core element
• Tier 2: Mutually exclusive design choice
• Tier 3: Non-exclusive implementation variant

Chart 3 - Overview of the DAO taxonomy elements

3.1 What defines the “DNA” of decentralized autonomous organizations?

It is possible to make a rough distinction between DAO taxonomy elements. The “D” and “A” of DAOs is foremost defined by the scope of multilateral agreements and resource management. Another way to frame these components would be to look at the function of any DAO. Is the DAO’s goal some “funds allocation” (e.g., MolochDAO) or “system design” (e.g., governance rights associated with MakerDAO’s MKR token)?

This taxonomy does, however, not aim to allocate DAOs into such predefined categories. It is merely an agnostic framework that will make it possible to identify DAO clusters later on.

3.1.1 The scope of multilateral agreements

The “scope of multilateral agreement” defines to what extent all decisions are agreed upon multilaterally. The primary distinction within this taxonomy element is whether the scope of multilateral agreements is unrestricted or restricted.

If it is restricted, it is necessary to make a further distinction based on what is exempted from a multilateral agreement. Generally, exemptions can either be incorporated within natural language or software.

This is a relevant distinction, as the chosen medium defines how changes to the content thereof are enforced. Software upgrades are compelling, whereas natural language texts lack hard enforcement and are somewhat soft guidelines. Changes to natural language texts, such as a constitution defining specific core values, could be deemed to be less critical than software upgrades. Nonetheless, natural language texts may still be required to express a nuanced idea or account for and link to existing legal contracts. Software upgrades, on the other hand, are not only binding, but usually relate to core operational elements, such as the deployed smart contracts, or proposed changes to a native blockchain protocol if a DAO would be running on its blockchain.

Further reasons for this structure are twofold:

• Notably, an opt-in is not considered a multilateral agreement in this context. The notion of multilateral agreements was introduced to reflect less hierarchic power dynamics and decisions amongst peers. Agreeing to a preconceived set of rules does not allow for a heterarchical⁷, collaborative decision-making process. Similarly, opt-outs are also excluded from the scope but indirectly covered in the chapter on voting.

• Nestled within this concept is a general distinction between horizontal and vertical mechanisms⁸. Horizontal mechanisms reflect forking and constitute a non-coercive exit strategy that bypasses any governance problem. It risks, however, the existing network effects and may fracture in-group cohesion. For this reason, horizontal governance mechanisms are largely ignored for a more detailed assessment of vertical governance mechanisms that directly incorporate solutions on stackable smart contracts.

3.1.2 Resource management

As sketched out earlier, resource management matters, as it profoundly impacts the autonomy of any DAO.

The exclusive usage of on-chain resources allows a DAO to directly exert control and initiate action via a smart contract that may or may not be upgradeable. Depending on the exact set-up of the respective smart-contract, a trade-off between the level of multilateral agreement and practicability (i.e., upgradeability) can be ensured, while counterparty risk could be avoided.

When using off-chain assets (such as fiat currencies), a natural person or legal entity (in jargon “legal personhood”) must be assigned control over the respective assets. This legally identifiable point of contact has legal control over off-chain assets and thus introduces counterparty risk. It would be possible to abuse the legal authority over the DAO’s assets and disregard decisions of the DAO (i.e., do a good old exit scam)⁹.

3.2 Moving from the DNA to the prefrontal cortex

Meaningful distinctions between DAOs target not only the scope or function but also how the organization of a DAO is operationalized. Generally, the organizational aspect of DAOs is defined by two different elements: discussion and voting.

Based on the initially developed definition, the rules of a DAO must be “a priori binding, formalized and transparent”. To see how DAO, a “DAO in the wild” really is, one can make one meaningful distinction for the taxonomy element “discussion” - is the deliberation rule-based or discretionary?

3.2.1 Discussions

Rule-based discussions can be executed on-chain via pre-voting. The core idea is that a smart contract is used for implementations such as token curated registries or prediction markets. As these implementations are based on smart contracts, they all are binding, formalized, and transparent.

Alternatively, a rule-based discussion could also be executed off-chain by linking a smart contract event to a legal contract.

If the discussion is not rule-based, it must be discretionary. By definition, discretionary discussions must be based on an off-chain medium. These could either be legal contracts that are not initiated or otherwise directly linked to smart contracts of the DAO, or - more commonly - various fora or messenger services.

Linked discretionary discussions are usually involving an identity-based off-chain discussion that will be referred to in an on-chain transaction, once a rough consensus has been reached. By doing this, an informal and open discussion can be integrated into a rules-based system.

Most DAO frameworks (for example, Aragon or Metacartel) have made use of (mostly pseudonymous) fora to enable a natural language discussion of members.

3.2.2 Voting structures

The differentiating feature for voting structures of DAOs is whether voting is based on a liquid stake or merit. In the opinion of Binance Research, voting and membership in a DAO are intrinsically linked, as a member must be able to signal its view.

The concept of liquid stake describes a voting process that is executed with a transferable, stake-based token. The implementation can follow a “dedicated voting” structure, further defining whether the stakeable token must be native or can be foreign.

This is based on the assumption that DAOs always use tokenized voting rights. Consequently, the real difference here arises from the fact of what is represented by these voting rights – i.e., if there is dedicated voting. Are voting rights “native token”, issued by the DAO to be a bearer right to vote, or are they “foreign token” and thus merely representing financial commitment?

At this point, we should acknowledge – but quickly move on subsequently - that liquidity can have a significant impact on the ability of a user to enter or exit a DAO with such “native token”. If there is no demand or supply in primary or secondary markets, such a system is inherently flawed, opening up interesting long-tail events.

Liquid stake voting can not only be dedicated but also weighted. This differentiation defines to what extent one token represents one vote. Alternatives to such a simple set-up could be Vitalik’s Quadratic Voting, where voting costs are increasing quadratically, or other functions where n token represent m votes.

Moving on, voting can also be based on merit instead of a stake. Naturally, this would still involve token. However, these tokens would not be transferable freely, but merely display the salience of any owner. The idea is to have tokens that cannot be moved for any purpose besides voting and are thus restricted to internal usage. As the name suggests, such tokens would usually be issued based on the merit and activity of an individual.

The relevant distinction here is on what level the identification takes place. Is the “know-your-member” process based on account addresses? If so, it would open the opportunity of individuals being represented more than once. On the other hand, is it based on government ID’s or social media networks (e.g., Twitter handle, IG ID, etc.)? Similarly to stake based tokens, non-transferable merit tokens could still be weighted.

3.3 What else is there to DAOs?

The respective design choices in the above elements are likely to influence numerous other areas, such as:

• Compounding smart contract risk: right in the tradition of TheDAO, DAOs are still at risk of smart contract failure. As more complex set-ups require multiple, usually interacting smart contracts, smart contract risk compounds. Nonetheless, formal verification of smart contract code or insurances might be able to remedy or manage this risk.
• Added value functionalities: to what extent does the chosen structure allow a subsequent enhancement with elements such as delegated or secret voting? It is also possible to leverage an ID level (applicable to liquid stakes and merit stakes) for member onboarding - new members could need to be championed, i.e., recommended by an existing member. This list could easily be expended.
• Social cohesion: one of the many factors contributing to the activity and membership base of a DAO is the ease and procedure of leaving a DAO. Curiously, insights for social cohesion can be drawn from a study on the symbolic effect of a wedding. Coherently, the symbolism of engaging in a partnership for life has led married couples to be more stable than couples in a registered partnership – the analogy would suggest that similar effects could be achieved in defining and framing a DAO membership accordingly.
• Dispute resolution capacities: are likely to be influenced by design choices for the discussion and voting processes.
• Degree of autonomy: is most likely additionally influenced by the broader environment of the DAO. Exemplary questions will be if the DAO is operating atop a proprietary or “foreign” blockchain or what the costs of conducting a censorship attack would be.

4. DAOs and the law

Interestingly, the connection between DAOs and the legislative environment is currently in flux. The very first “connection” was, in fact, an attempt to deny such connection and simply disregard the legal environment. TheDAO, for example, wanted to exist independent of any legislation. However, such a strategy merely results in giving up control over how the DAO will is treated under the law.

“If you don’t formalize a legal structure for a human-created entity, courts will impose one for you. And a general partnership, unless properly formalized or a deliberately created structure, is a very bad thing.”

Fundamentally, DAO’s must have a legal entity. A legal entity also allows DAOs to subsequently integrate legal and smart contracts. As of now, the following two strategies can be identified:

• DAOs can form a legal entity that will enable them to opt-out of any legal jurisdiction. This will, however, have repercussions on the DAO’s ability to manage off-chain funds.
• DAOs can attempt to fully integrate and leverage one of the recently created legal entity structures to obtain legal personhood.

The first strategy relates to a DAO using something called “Qualified Code Deference” (QDC) to opt out. The QDC is primarily based on work from Gabriel Shapiro and can be described by members agreeing to a DAO charter and, by doing that, forfeiting their right to legally dispute the results and operations of smart contracts. However, as Shapiro’s QDC includes setting up an “unincorporated association of individuals”, it would not be possible for DAOs to have legal personhood. Subsequently, such DAOs would not be able to enter legally binding agreements and would, therefore, not be able to directly manage off-chain resources.

The second strategy is less radical and merely aims to fully integrate DAOs within the legislative environment by obtaining legal personhood. Legal personhood can be attained in various ways and not only allows DAOs to exercise control over off-chain assets but, for example, also allows them to enter into contractual agreements and offer liability protections to participants. It is, therefore, a precondition for linking legal and smart contracts.

The Maltese ITAS act, for example, allows DAOs to be certified as an “Innovative Technology Arrangement” and subsequently opt for legal personhood. Similarly, Virginia passed a bill that enables DAOs to “elect to be a blockchain-based limited liability company” (BBVA). Notably, this has already been used by, for example, dOrg.

Once a DAO is incorporated and has legal personhood, it may wish to integrate legal contracts (in jargon “wet law”) and smart contracts. Various attempts have been made to achieve this goal:

• An initial starting point was the attempt to - instead of writing smart contract code - merely write a written text that would express a legal intent.
• Alternatively, “code” could be a written text that merely describes a contractual arrangement. Instead, however, being legally binding these first efforts were rather initial attempts resembling quick hacks and not binding solutions.
• A more mature approach is to transform natural language agreements into machine-readable contract objects. OpenLaw is doing this with its markup language.

While incorporating smart contracts into wet law is still in a rather rudimentary stage, it is, nonetheless, a more pragmatic approach that reflects opinions stated in contemporary legal theory¹⁰.

5. What is in “Il Principe” for DAOs?

Machiavelli wrote The original Principe with the intent to teach a young noble how to navigate a very particular institutional environment. So what would Machiavelli recommend DAOs? What is the landscape they have to navigate, and how could this landscape be relating to the current institutional environment?

As of now, there are 193 Nation-States in the United Nations and more than 275 jurisdictions worldwide. In the European Union alone, there are more than 40,000 legal acts, 15,000 court verdicts, and 62,000 international standards¹¹.

This daunting landscape has historically been fundamental to limit uncertainty, with the trustworthiness of data being entirely dependent on the accountability of the provider. If there is a legal system that is capable of enforcing its laws against public and thus regulated entities (i.e., data providers), it is possible to hold them accountable for wrong data.

As a result, transactions in traditional databases are final-by-law.

However, they are only final if they can be enforced and thus require some level of trust in the institutional environment. Especially for international operations that involve many various jurisdictions, the related risk is amplified. Managing this risk can be costly (e.g., via insurances and third-party services).

In DLT-based processes and information, on the other hand, data credibility is assured not by a third party (Davidson, Sinclair, De Filippi and Potts, 2016), but by economic incentives. Why is this relevant to DAOs?

Because it shows the inherent clash that will occur whenever DAOs interact with the physical world, either directly (e.g., using a bank account) or indirectly (e.g., using a token that represents a physical asset). One of the most progressive pieces of legislation regarding tokenization comes from Lichtenstein, where tokens are defined as “containers” for (all kinds of) physical assets and are thus protected by the established legal system.

While this is undoubtedly a significant advancement, it reinstates a need for a higher-level legal system. To the extent that ties with the established institutional environment are to be avoided in the first place, such a structure would thus not be sufficient. A more pragmatic and less idealistic scenario is one, where DAOs are not radically replacing, but rather complementing existing forms of organization. As such, they would also need to be integrated into the broader institutional environment and would therefore significantly benefit from this and other related efforts¹².

6. Conclusion

Have DAOs already grown into the big dreams of enthusiasts of the first hour?

Maybe not, but the “The DAO” trauma has been overcome; several DAO frameworks are pioneering a way into a future, and an increasing amount of supporting elements (e.g., formal verification of smart contracts) is rapidly advancing. Meanwhile, the established legislative environment is increasingly perceptive of tokenization and “Innovative Technology Arrangements”.

However, no report about DAOs could be complete without mentioning the incentive system of DAOs that must be deployed to coordinate members. Incentives are firmly rooted in game theory and are an integral part of any DAO. A careful calibration thereof is required to successfully avoid problems such as the tragedy of the commons, byzantine behavior, principal-agent theory, or prisoner dilemmas.

Incentives do not necessarily have to be positive, but can also relate to penalties for actions that “are to be avoided”. Moreover, it is possible that vested interests are not succinct, as they could be increasingly undermined by the ability to use various derivatives. A large token holder could, for example, have an economic benefit from purposefully “breaking” a DAO and shorting an associated token thereof.

However, incentive design for DAOs is still very much a green field. It is to be seen whether experimentation, maturing DAO OS’es, and self-regulation are sufficient or whether a general “Blockchain Governance Code” must be imposed. Incentive driven governance is dependent on many different variables and relates to many design elements of the proposed taxonomy. More game-theoretic research must be undertaken that exceeds the scope of this report. Subsequently, incentives have not – yet – found their way into the proposed working taxonomy.

The fact that this area is still very much in need of rigorous analysis is also evident in the voter turnouts of existing DAOs. When looking at participation rates from DAOs across the board, this becomes obvious. Due to their well-designed UI, Aragon Governance Proposals (AGPs) can easily be assessed as an example. On average, AGPs had a participation rate of 6% of all ANT token (the Aragon governance token) that came from less than 50 unique addresses.

These figures clearly show that DAOs still have many challenges to overcome to reach the oftentimes proclaimed goal of being “hyperscalable” and effectively coordinate a large number of DAO members.

This report was inspired by a Devcon V workshop organized by Gnosis and greatly benefited by comments from Philippe Honigman.

7. Bibliography

• Cambridge Dictionary (2019). “Autonomous”. Accessible online at: https://dictionary.cambridge.org/dictionary/english/autonomous
• COALA (2018). “Governance Of Blockchain Systems”. Available online at: https://coala.global/wp-content/uploads/2019/02/BRI-COALA-Governance-of-Blockchains.pdf
• Crumley (1995). “Heterarchy and the analysis of complex societies.” Available online at: https://www.researchgate.net/publication/227664129_Heterarchy_and_the_Analysis_of_Complex_Societiess
• Elrifai (2019). “6 Basic legal principles of DAO’. Available online at: https://www.youtube.com/watch?v=6Vi39duSpeY
• Honigman (2019) “What is a DAO”. Available online at: https://medium.com/hackernoon/what-is-a-dao-c7e84aa1bd69
• Hacker (2017). Corporate Governance for Complex Cryptocurrencies? A Framework for Stability and Decision Making in Blockchain-Based Organizations. Available online at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2998830
• Leblebici (2019). Vertical and Horizontal Coordination Mechanisms. Available online at: https://ko.coursera.org/lecture/designing-organization/4-1-4-coordination-and-teamwork-B1wxx?authMode=signup
• Markežič and Blaj (2019). “Legally DAO”. Available online at: https://docs.google.com/presentation/d/108Q0ACj7MNU5PGhgwzlVzT7tZ-m4VFeCDxsT_r1klhA/edit#slide=id.g56c3efd1e2_1_0
• Pasquale (2018). A Rule of Persons, Not Machines: The Limits of Legal Automation. Available online at: https://digitalcommons.law.umaryland.edu/cgi/viewcontent.cgi?article=2616&context=fac_pubs
• Popper (2016). “A Hacking of More Than $50 Million Dashes Hopes in the World of Virtual Currency”. Available online at: https://www.nytimes.com/2016/06/18/business/dealbook/hacker-may-have-removed-more-than-50-million-from-experimental-cybercurrency-project.html
• Shapiro (2019). “Drafting wet contracts in a smart contract world”. Available online at: https://www.youtube.com/watch?v=hzbMPLxiht4
• Verstraete (2017). “The Stakes of Smart Contracts.” Available online at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3178393
• Walch (2019). "Deconstructing ‘Decentralization’: Exploring the Core Claim of Crypto Systems”. Available online at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3326244