Cybernetic regulation in the age of algorithmic finance

March 26, 2014

A lawyer, an economist and a Python programer walk into a bar. This is no joke. It is the future of financial regulation, at least according to a provocative proposal by Andrei Kirilenko, Professor of the Practice of Finance at the MIT Sloan School of Management.

At the recent meeting on the future of financial standards, jointly organized by SWIFT and the London School of Economics and Political Science, Kirilenko spoke of one of the fundamental problems of modern financial regulation: the translation of legal requirements into the computer code that drives much of the activity in today’s markets. Specifically, when can we say that a code is compliant?

For Kirilenko and his colleagues, this question is critical to how the financial services industry behaves both in the short and long run: a firm that over-interprets regulation may invest more resources in compliance than are necessary; a firm that interprets the rules too weakly may, conversely, risk becoming the object of fines.

To address this, Kirilenko advocates what is best described as ‘coding’ regulation. In addition to (or perhaps, in substitution of) standard cost-benefit analyses, Kirilenko proposes adding a section to regulatory documents that specifies how novel rules operate in terms of Boolean logic. If. Then. Zeroes. Ones. Rules are, after all, instructions that “allow some things, prohibit others, and are neutral over a range of possible values.” They are, hence, logical gates within a normative circuit board. New financial regulations can thus be represented as specific configurations of circuits linking Boolean gates to form a mesh of decisions, possibilities and potential outcomes. And in aid of implementation, such circuits should be made as explicit as possible (they are, after all,more practical than cost benefit analysis), thereby resolving the problem of the under-determination of compliance in code.

Capturing rules, however, requires translating ‘legalese’ into some form of high-level logic, and for this Kirilenko proposes putting lawyers in direct conversation with coders, programmers and systems developers. The outcome of this conversation should be a logical representation of the rules made by lawyers and regulators in a form that can be applied to specific real-world technical systems. A cybernetic reconstruction of the natural language of the legal domain translated into a flowchart out of which code can be developed. (Note that, in a sense, this cybernetic transformation evokes Philip Mirowski’s notion of markomata in the regulatory domain).

Kirilenko’s idea is certainly tantalizing, not the least for sociologists of knowledge, markets and organizations. From an infrastructural perspective, it stresses the importance of inversion, of raising and making visible the plumbing and tracks of finance in order to regulate and change the market (and, interestingly, it calls for an explicit link between legal infrastructures, market engineers, and standard-setters). In epistemic terms, the proposal invites recombining knowledge within finance, creating spaces similar to Galison’s trading zones where communities bearing different languages (e.g. lawyers and technologists) can exchange, modify and build upon specific designs on rules and regulations. Similarly, from an organizational perspective, the idea of coding natural language into Boolean gates highlights the importance of interactional expertise in the regulation of finance: regulators and lawyers need to be able to converse code, just as much as coders need to be able to understand the basics of law and regulatory regimes. If carried to its consequences, this bold proposal necessitates re-conceptualizing how knowledge is created, distributed and organized in financial institutions.

A number of issues remain unsolved about Kirilenko’s proposal (which he aptly calls ‘Financial Regulation 2.0’). The Wittgensteinian critique of rule following, for instance, questions the possibilities of translation and highlights, perhaps, the limits of transforming natural language statements into Boolean relations. Law and regulation are as much about the formal content of texts as they are about their interpretation, indexicality and judgment: indeed, an important part of the work of regulators is justifying rule-breaking (as seen, for instance, in the production of letters of no-action, a fact highlighted by another conference participant). Of course, sociological questions also arise around the process of transforming flowcharts into computer codes: coding is perhaps well described through Andrew Pickering’s metaphor of the mangle, with programmers negotiating their practices and implementations with the affordances and material resistance of the systems on which they work. Will programmers working on different systems, languages and beholding different hacking cultures create the same code out of the same flowchart?

Reservations notwithstanding, Kirilenko’s call for a sort of ‘algorithmic regulation’ remains a fascinating experiment (one that, as far as I could see, is being reproduced across the world): after all, Kirilenko has a unique combination of experience and expertise that places him in a privileged position to talk about finance, technology and regulation. Not only is he an accomplished academic, but he also served as Chief Economist for the Commodities Futures Trading Commission between 2008 and 2012—a period that saw, among other things, the Flash Crash of 2010. Perhaps fire is best fought with fire, and coding regulation is a better mechanism for dealing with a complex market ecology of humans, algorithms and machines than the more traditional forms of rule making. Then again, perhaps it is not. In any case, Kirilenko reminds us that these are certainly interesting (cybernetic) times for the social studies of finance.

About these ads

12 Responses to “Cybernetic regulation in the age of algorithmic finance”

  1. excellent piece, Juan Pablo! And a good point re Wittgensteinian critique of rule following. A case in point (a bit of shameless self-promotion here) is my research on cash-settled currency derivatives and how they were made bets and then re-made as legitimate transactions depending on legal interpretations and creative rule following (JCE 2014).

  2. Juan Pablo Pardo-Guerra Says:

    Absolutely, Sveta! (and great article, by the way: SSF and socio-legal studies still have much to talk about, and yours is a terrific contribution; I have it on my sociology of markets reading list!). Practitioners sometimes forget how interpretative law and regulation can be, perhaps because of the recent historical growth in legal work within financial markets (back in the 1970s, the rulebook of the LSE was 20 pages long; today, FCA regulations, guidances and so forth run in the hundreds of pages). The interesting thing is that, perhaps only anecdotally, I have the impression that the earlier generation of people in finance are more aware of this than current generations; It seems that people who were on the floors before automation are more attuned to this pragmatic finitism of law and regulation than those who just joined the ranks of the highly rationalized financial industry.

  3. marthapoon Says:

    This is an excellent post JP, one of the best ever on Socializing Finance!

    Does Kirilenki’s ‘Financial Regulation 2.0′ apply to all algorithms, or are his proposals specific to trading in stock markets?

  4. Juan Pablo Pardo-Guerra Says:

    Thanks Martha!

    I think Kirilenko is mostly concerned with financial regulation (which is his professional background), but it does pose the question of whether this is something applicable to other densely algorithmic spaces.

  5. danielbeunza Says:

    Hello JP, excellent post. Kirilenko’s proposal sounds eminently sensible, but based on my own research at the NYSE this might be a very misguided proposal. What I noticed at the NYSE is that modifying technology to comply with regulatory mandates requires a deep understanding of the social structure of an organization. In the case of the NYSE, not even the managers of the Exchange had it. It took the Exchange a near-death experience, two rounds of automation and new hires to get automation right. Based on this, I wonder whether lawmakers are the ones best capable of translating into code the practices (and corresponding social roles) that define an organization.

  6. Juan Pablo Pardo-Guerra Says:

    Thanks Daniel! I completely agree: regulators are not necessarily the best at translating the practices of an organization into code, not the least because they often lack the type of experiential and organizational knowledge that goes with ‘understanding’ a marketplace. Who knows what type of hybrids and chimeras will emerge from the experiments that Kirilenko is running. Indeed, it would be fascinating to see the process of translation in practice, and to better gauge how the different imaginaries, politics and approaches of lawyers and technologists are contested in the act of making regulatory language logical. That would be awesome fieldwork!

    In the context of Kirilenko’s proposal, your comment does raise the question of the genealogies and generational dynamics of regulatory knowledge, though, which is an aspect that I think deserves more attention within the field. Market organizations, like their technologies, change. yet the human agents (the biological tails of organizations, as Yuval once put it) populating these organizations can be quite resilient. The market contains all kinds of experience and expertise, fragmented bits of (very opinionated) knowledge about how best to organize the world. For a techie, the nature of organizational problems are notably different than for an ex-floor trader. Who is best suited to regulate and outline the codes of an organization in particular and the market in general? Certainly, some form of iterative learning, however painful, can build consensus. But it is a question that necessitates more empirical answers.

  7. Jed Harris Says:

    Fascinating topic. I’d like citations for Daniel Beunza’s work. It echos many other experiences I’ve seen — usually not documented very clearly — in which organizations damage or destroy themselves through brutal technical interventions, because they do not understand or value their own implicit social practices.

    The managers are often the last people to ask since they do not participate in most of the “routine” practices — especially when those practices are working well — and their self-esteem is enhanced by believing those practices aren’t as important as their “leadership”.

    Looking at this from the software side, any given regulation is only a very thin layer on a deep assemblage of practices, de facto ontology (embedded in legal and conventional language), institutional structure, and so on. Much of that has to be reflected in code before anything like coded regulations can work.

    This can be done, and I think it would have very positive social effects, but the investment would be large and long term. More than that, I think it would require a degree of transparency that is anathema to current financial institutions. They would have to depict their innermost operations in concrete, honest detail.

    One interesting effect would be that we would be able to accurately model the operation of financial institutions. We’d still have to add some models of human decision making to the institutional models, but those would be very constrained. This would be a very big advance over existing (economic) models that pretty much ignore this whole aspect of how the system actually works. Especially under stress, these models would be much better guides.

    I think Wittgenstein’s critique was valuable in being an early harbinger of Moravec’s Paradox (see Wikipedia). We are now on our way to resolving that but only by paying the full price of recognizing the enormous capabilities we bring to even the simplest human activities.

  8. danielbeunza Says:

    Hello JP, well put. Generations are indeed key to the market. Prior to the Flash Crash, I never heard any of the young Wall Street technologists raise concerns about algorithms, whereas the “old guard” on the floor of the NYSE was anticipating problems like it since the first efforts to design single-stock circuit breakers (LRPs) in 2007.

  9. danielbeunza Says:

    Hello Jed — delighted that the problem of “damaging automation” resonated with your own experience. And as you point out, key to the problem of automation is getting the input from the people who have the tacit knowledge. Here’s the link to my work, joint with Yuval Millo:

  10. Juan Pablo Pardo-Guerra Says:

    Thank you very much, Jed (and sorry for the late reply).

    I think that regulation is ultimately a way of configuring social relations, and in this sense there is an interesting epistemological discussion about what we can know about such relations. If we can map out all relations, a type of well-orchestrated, logical regulatory structure should work. If we can’t, there will always be epistemic surprises.

    Now, some would argue that, at the end of the day, relations (and thus regulation) can always be made explicit and therefore codified. Others, however, would tend to argue that there is an irreducible tacit dimension to social relations and that regulation is thus inherently incomplete. I think the evidence would seem to suggest that the second is a better description of the scope of regulation. Some financial innovations, for instance, emerge precisely out of the unavoidable cracks created by regulation; they are epistemic surprises, of sorts.

    Perhaps an interesting exercise would be to see how translations from social relations to code work in other settings. I can think, for instance, of the work by colleagues at Edinburgh (e.g. Robin Williams and Neil Pollock) on SAP.

  11. Tim Wheaton Says:

    Interesting points. With the huge amount of computer controlled trading it should happen. We have all seen the results of ignorance and lack of understanding of financial instruments amongst bank executives and regulators

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


Get every new post delivered to your Inbox.

Join 515 other followers

%d bloggers like this: