Black Swans, Financial Crises and more

I’ve spent the last couple of days in Sydney at a conference organized by the Paul Woolley Centre for the Study of Capital Market Dysfunctionality. It’s striking that this is the only research group of which I’m aware that takes dysfunctionality, rather than the Efficient Markets Hypothesis as a starting point.

Various people have asked me about the paper and slides, so I’m putting them up for download.

Black Swans and Financial Regulation (presentation

Unawareness and financial innovation presentation and paper)

19 thoughts on “Black Swans, Financial Crises and more

  1. Useful summary of past issues John.

    But the $64 questions are:

    – How would we cope today with a much more fragile system?
    – Is another collapse less likely today or just as likely?
    – Are we dealing now with a white swan – we know what can happen now but we arent changing the settings appropriately?

    Here is alternative progressive economics’ chance to become a real science by predicting when and how the current white swan will unravel….bearing in mind the last economist who tried this was sent to Coventry ….. or was it Kingston where he seems actually quite happy for speaking truth to power and hasnt done badly either.

    But seriously if there is any offerings from you or your colleagues I think many of us spectators would be appreciative – an alternative to not making predictions would be to identify why you cant say where the present trends are heading or be sure of the timing – a gap analysis as it were.

  2. I basically agree with the argument and conclusions of the paper which I scanned rapidly. I understood all the English but not all the Maths. The English argument and prescriptive conclusions appear to agree with the empirical reality that unfolded round about the GFC.

    The statement, “In games with a secure strategy, reject strategies that are subject to unfavorable surprises.”, particularly resonated for me. I think that is a good precautionary strategy or principle.

    In games apparently without a secure strategy there may be no secure strategy or a secure strategy as yet undiscovered. If the number of potential strategies is large and the number of potential unfavourable surprises or pitfalls is also large, it would likely take an exponentially (or combinatorially) enormous number of tests to possibly discover a secure strategy. Just saying. Don’t know if this thought could be helpful.

  3. I was also pleased to see you mention the Precautionary Principle detailed in your paper and its origin in environmental management theory.

    Its was intriguing to see this familiar issue being considered in a very different way, and with different terminology and problem framing to that I’m familiar with. I confess to somewhat glazing over the game theory to analysis of decisions and options you outline but I think I got the gist enough to see this is a very very different approach to environmental / risk management that you find in the ISO 31000 tool box and national guidelines.

    Correct me if I’m wrong but the paper seems to start with and focus on the concept of the ‘Decision Maker’ with Nature itself being anthropomorphised. This is very different to the science/engineering approach which focuses on system components in an ‘objective’ external fashion which are modelled as operating at arms length and following various functions, ideally describable by algorithms. The latter leads to tools such as fault-tree, failure mode and event analysis.

    ISO 31000 includes such things as human reliability analysis but I have not seen human decision making placed so centrally – though it can be incorporated using such tools as Bayes Nets. Conversely the occurrence of hazardous events is nothing new at all (in contrast the Black Swan still seems novel to many economists) but simply a given i.e. at some point “s@#& happens” and management planning starts from there.

  4. Ikonoclast,
    “I understood all the English but not all the Maths.”

    Just Yesterday I came across a whole book called The Algebra of Conscience. I cannot understand all the maths in the book – but I am dubious that conscience works in such a regular fashion that algebra is the best way of understanding it. I think if it were so regular and algebraic then the tradition of Conscience Algebra would go back to Classical times like normal algebra and be taught in schools.

    But it is quite interesting in that the book starts with a sentence in English then translates it into algebra. It says the following is a moral truism (which is quite an overstatement but nonetheless I guess if the author was to quibble about what a moral truism was for too long he would never get around to his algebra):

    “The world can always incline someone to reject temptation, but in the case when the world itself inclines someone toward temptation, that person can resist only under the condition that the very idea that of his succumbing to temptation is terrifying to him.”

    The above sentence told in algebra is: f(1,b) =1, f(O,b) =b.

    The working was done by the author introducing “the variables a and b defined on the Boolean set {0,1}, where 0 means ‘to yield to temptation’, and 1 means ‘to resist temptation’. Let variable a represent the external world; a=O means that the world inclines a person to give in to temptation, and a=1 means that the world inclines one to resist. Let variable b represent a person’s image ofthe self; b=O means that a person sees himself falling for temptation, and b=1 means that he sees himself resisting. Now we introduce the Boolean function A=f(a,b) which describes a person’s behavior in reality; A=O means that he falls for temptation, A=l means that he resists. The following equations correspond to the semantic content of the text given above:
    f(1,b) =1, f(O,b) =b.”

    But I still think this is dubious. The author goes on to do algebra on parts of Crime and Punishment, and on Hamlet which is quite a feat indeed – The maths goes up and down on the page and makes faces with dashes for a reason I am quite uncertain about but because of this it makes the work look like 1960s/70s style poetry.

    This is an example of Hamlet as told by algebra:

    Let us designate Hamlet by H and Claudius by C. In the first ethical system, Hamlet would then be
    H=aaa+b+ b
    IH I=3/4 =max
    (1) has correct images of himself and of Claudius; (2) doubts the correctness ofhis images.
    In the second ethical system, Hamlet would be
    H=a IHI =3/4=med.

    And so on. Although I cannot understand the algebra – I think probably Hamlet is better told in words rather than in algebra. I think perhaps this is the same with economics – if Hamlet is better told in words probably economics is too since economics is even more complicated than Hamlet being it is unbounded by just being a few Acts written on paper like Hamlet is – where the economics discussion is strictly of quantities then maths would be appropriate for that .

  5. @ZM

    Alternatively perhaps the author is related to the originator of this gem

    MCNOLEG, O. 1996. The integration of GIS, remote sensing, expert systems and adaptive co-kriging for environmental habitat modeling of the Highland Haggis using object-oriented, fuzzy-logic and neural-network techniques. Computers & Geosciences, 22, 585-588.

    Yes its a real article – and cited several times – I think by people who were testing the observation powers of referees and a little skeptical of modelling.

  6. Newtownian,


    The book is written by Vladimir A. Lefebvre and called Algebra of Conscience : Revised Edition with a Second Part with a new Foreword by Anatol Rapoport

    Volume 30 in the Theory and Decision Library, Series A: Philosophy and Methodology of the Social Sciences (General Editors: W. Leinfellner (Vienna) and G. Eberlein (Munich))

    I found it as an electronic book – so it is maybe sold as part of a digital library package I would assume unless this is a popular area for research (?)

  7. Returning to the sentence: “In games with a secure strategy, reject strategies that are subject to unfavorable surprises.”

    John, I tend to think that computer game theory might provide useful tools and methods for economic investigation and analysis. When I say computer game theory I mean the theory of automating artificially intelligent play of algorithmic games using computer programs. This of course involves creating artificially intelligent and rational “agents”. Such games range from noughts and crosses (simple) through draughts, chess, go and on to RTS games such as Starcraft, Cossacks and Supreme Commander (in order of complexity).

    All of these games are, in theory, completely solvable by algorithmic methods (by the perfect strategies or sequences of moves of two or more opponents playing perfectly). In practice only the simplest games are solvable. Noughts and crosses is algoritmically solvable by a tree search. That is to say perfect non-losing moves can be computed for both opponents through all variations. The game always end in a draw between two perfect opponents. It would be possible to lay out or print out the entire move tree with all winning/losing and all drawing variations.

    Chess, for example, also is entirely algorithmically solvable in theory but in practice the permuations and combinations are too great to allow computation and analysis of the move tree to find the perfect game between two perfect opponents. Each side’s move is called a ply. Chess has a typical branching factor (of move possibilities) of about 32 moves. One suspects that the perfect game of chess would run to at least 200 ply (100 moves) therefore the number of positions to be calculated and evaluated in a full tree search would run to at least 200 to the power of 32.

    Games more complex than chess (with an 8×8 array) like Go (with a 19×19 array) become even more complex. Then an RTS game with a virtual 3D array of millions of pixel points (any unit being capable of being centered on all or many of these pixel points) and with frame rates of say 50 fps become stupendously complex.

    The key point here is that algorithmic complexity (even at the “mere” chess level) leads to the requirement that heuristics be applied to assist algorithmic search methods. This is done to save computational time (along with the alphabeta algorithm which can truncate searches which turn worse than something already found).

    In summary, the techniques used along with the alphabeta search algorithm are to;

    (a) employ well known heuristics for good moves;
    (b) search the heuristically likely good moves first;
    (c) limit the average search depth by time available (say 3 mins per move);
    (d) use the alphbeta algorithm itself to truncate a search that turns worse than an already searched alternative.*

    * It stops completely evaluating a move when at least one possibility has been found that proves the move to be worse than a previously examined move.

    A key general point to be made here is that extreme algorithmic complexity in a fully determined and deterministic system still leads to the necessity to use heuristics. A very complex, indeed wickedly complex, system (like an economy) which is known to be indeterminate in some way or ways or strongly suspected to be indeterminate in some way or ways needs heuristics… perhaps even more so.

    Thus, whether the system is deterministic and very complicated or likely non-deterministic (in part) and wickedly complicated makes no difference to method we should choose to create or model intelligent rational agents making “their” best decisions possible given best information possible. The method should be, just as it is for a game;

    (a) employ well known heuristics for good moves;
    (b) search the heuristically likely good moves first;
    (c) limit the average search depth by time available;
    (d) use the alphbeta algorithm itself to truncate a search that turns worse than an already searched alternative.

    But how would one translate a game model to an economic model? An economic model of this sort would have different players, actors or intelligent rational agents, all auotmated. Some of these players are like each other and have similar goals (wealth accumulation). Some are somewhat dissimilar but still have the same goal (wealth accumulation). Then there is one special player we can call the regulator. Thus we have a regulator and various classes of competing accumulators.

    The regulator has a special goal. This goal is to increase the total wealth of the system quickly and smoothly AND to avoid large perturbations (booms and busts). To sum up, the accumulators have only the goal of greatest self-accumulation. The regulator has the goal of overall system accumulation AND the avoidance of large swings or perturbations.

    The regulator imposes the financial rule set. In computer game theory this would be called the Legal Move Generator. I see no need to change this terminology. The regulator defines the legal moves for the accumulators (as in legal transactions and legal financial instruments). In the simplest model, the accumulators are only permitted legal moves and in this model legal move parameters could be changed for different test runs of the system. I envision a computerised model tree-searching outcomes ie. behaviours of the system over time under different parameters.

    At the other end of the model, raw wealth enters the system as raw materials of nature. Initially, one can suppose this supply to be an infinite stock. Yes, I of all people know the objections to this but we must start with a simplified model. However, the flow of raw material will be governed by the amount of capital that is or can be applied to its extraction. We could model just a few basic resources; maybe just three namely energy, water and minerals. We could deem that every basic product requires energy and every product also requires water and/or minerals in different proportions as seems appropriate for modelling purposes. We could deem that several products and services are produced namely food, manufactured products, general services and financial services (because financial services are “special”).

    At the level of accumulators (those trying to accumulate wealth), you have primary, secondary, tertiary producers. There might be a case for having quaternary producers and making these producers the FIRE sector (or at least the Finance and maybe Insurance sector if not the Real Estate sector). This model wants to pay special attention to Finance and its regulation so I suspect it needs its one modelling category. The regulator could be called perhaps the quinary sector. Yes, I had to look up the term.

    Given the way some economists model sectors and sectoral interactions, I am sure such economists would know how to structure the rules of the model to govern real flows and capital flows according to current basic business models and accounting. I think it is important to model real flows of the real (simplified) goods and services as well as capital flows. The regulators remit to avoid booms and busts includes not just remit to avoid financial booms and busts but also the remit to avoid severe gluts and shortages of real goods and real services.

    It would be a matter a giving each intelligent agent up the levels (at least several at each level) a set of goals and methods (algorithms and heuristics) for achieving these goals. But the ultimate goal of each competitive wealth accumaltor is self-weath withour concern for total system wealth. The goal of the regulator is to maximise total system wealth and stability along the lines outlined above.

  8. Ikonoclast,

    The problem with that is you would have to go to a great deal of effort and then only get a very ordinary result based on generalisations.

    The biggest computer we have in Australia allowed medical researchers to fully model the behaviour of some sort of flu cell , I think it was. Anyway – they could not get a good model of this simple behaviour until they got this great big computer.

    But this little cell is much less complex than an economy.

    The computer which is just the next step up from our current great big computer will require a whole nuclear reactor to power it. It is unlikely Australuans so much desire a big computer one step up from a cell modelling computer if it needs a whole nuclear reactor. A computer big enough to model the whole economy more accurately might need several nuclear reactors – this would be too dangerous I think – so then we are stuck with far less accurate models.

  9. @ZM

    I did not say we would model the whole economy. I outlined a very simple model to test several sets of competitive intelligent or rational agents with just a few members in each set. These agents would be primary accumulators (primary producers), secondary accumulators (manufacturers), tertiary accumulators (services except capital financial funding, insurance and risk hedging services) and quaternary accumulators (capital financial funding, insurance and risk hedging services). They are competitive accumulators in that they try to accumulate and maximise their own wealth without regard to the overall wealth of the system. In addition there would be a quinary sector of one regulator whose goal is to maximise and increase total wealth of the system and to do this in a stable (non boom / non-bust) kind of way.

    We can’t say before attempting developments and iterations of this model whether it would teach us something or not; whether we would make useful discoveries or not.

    But it seems to me that this basic model captures something fundamental about the agents. The accumulators are totally or primarily focused on the goal on individual wealth accumulation. In theory, they would accept that the total overall health of the system does help or hinder them but in practice they do not concern themselves practically with ensuring that overall health (at least not until you allow provisions for common interest philanthropy into the model). On the other hand, the regulator is a special kind of agent with a different goal.

    In one could get a working model of this (and I admit the devil is in the detail) then interesting tests could be conducted along the lines of;

    (a) programming the accumulators to seek the greatest achiveable absolute wealth; or
    (b) programming the accumulators to seek the greatest achiveable relative wealth compared to the average achieved by all accumulators.

    Then watch what happens to the overall system. Of course results are determined by starting parameters and the effectiveness of the goal seeking heuristics and algorithms. But different agents in the one level and at each level could be programmed with different goal-seeking methods (heuristics and algorithms) or different weightings of certain factors and we could see how this affected competitive behaviour and the overall system.

    We could even introduce human agents as some of the accumulators in some of the tests. Human players (in a complex model) would play very differently from automated agents for sure and would even likely find both new strategies and loopholes. The legal rules of the regulator would exist as written laws in a manual and also as the programmed “laws” of allowed transactions and actions in the program. But human players might, indeed would, find ways (without actual hacking) of breaking the programmed laws by exploiting programming errors, bugs, glitches, unintended features etc. This latter can and does occur in such complex programs because designers and programmers program AI bearing in mind how the believe their programmed laws work and not bearing in mind (because they don’t know them) the mistakes, bugs and glitches they have inadvertantly included in the programmed laws.

    Getting humans to seek such “exploits” can and does help debug the program. (It’s a form of system testing after all.) But it can do much more than that in this kind of program. It can elucidate what can happen in the model when rational agents are not literal and law abiding ie. when they are imaginative and devious. If someone finds a good loophole and exploits the heck out of it the massive detriment of other players, AI or human, then it could be instructive to see what happens to the overall model. Is overall wealth creation aided or hindered? Is overall stability affected? What happens in the model when other players, human or AI, “catch on” and oppose or imitate the exploitative behaviour in some way. The AI could contain an over-watch element and watch what other players do and possibly even then seek to imitate or oppose. I am talking of a sophisticated but far from impossible program given current AI theory and practice.

  10. Looks good stuff.
    The formal scheme consider two type of cognitive error: coalescence (failure to distinguish distinct outcomes) and restriction (failure to consider an outcome), taken together as unawareness. Unawareness has consequences,fine. But error has many more forms, see Bacon and Kahneman. Please read them. Bacon is more fun, but Kahneman has proof.

    One common form of error, in nature as in human society, results from deception. The Bank of England didn’t think that Fred Goodwin was a competent and honest banker because of unawareness, but because he fooled them into thinking so. The possibility I am being deceived can be brought under the “restriction” heading. Is this enough? Deception isn’t neutral; it’s done for a reason, against the interest of the deceived. The possibility of fraud puts the precautionary principle on steroids.

  11. I want to add an addendum to my previous long comments. It refers to the statement:-

    “The regulator defines the legal moves for the accumulators (as in legal transactions and legal financial instruments).”

    More precisely, we would need to create an open-ended automated regulatory agent. An economic modelling program needs to be more open-ended than a mathematical array game program like a chess program. The open-ended automated regulatory agent would follow the general principle that “All is permissible except that which is proscribed” which roughly follows the spirit and letter of our legal and regulatory systems.

    However, it gets a bit more complicated than the above. There are also procedures which are prescribed; obtaining regulatory permissions before proceeding with contracts and processes, paying regulatory fees, paying taxes in legal tender, not refusing payment in legal tender (though also accepting it in other forms) and so on. Then there are institutional and customary ways of doing business which are effectively somewhat prescribed.

    It ought not be too difficult to devise a Hierarchical Rule-Based regulating agent and attendant modelled insitututional environment system. A proposed transaction or contract between two rational free agents is tested as follows (I won’t add the Yes/No dichotomies here as the outcomes are rather obvious as in permit/do not permit);

    1. Does the proposed transaction or contract violate a prohibition?
    2. Does the proposed transaction or contract meet prescribed conditions and fees for execution?
    3. Is the proposed transaction or contract executable in practice both in the modelled “insitutional milue” and in the actual transaction system of the model?

    At the level of point 3, the regulator or overall program must have at least 2 distinct modules relating to the modelled “insitutional milue” and the actual transaction system of the model. For neatness the regulating agent can model the insitutional milue rules it creates by “law and regulation” and also model any attendent cultural milue modelling (if any). The checking for actual transaction executabity is more of an overall executive control function of the entire program rather than properly part of the regulator function.

  12. Unfortunately deception doesn’t appear to be significant unless it leads to a loss in performance – some hedge funds trawl the market looking for unrealistic or overenthusiastic evaluations to short. In general shorters are reviled by the market, which implies that the market eg Wall St is built on deception. There are those that want shorters restricted or banned saying that they cause Company failure via market manipulation. Others argue that Company failure is reflected, not caused by, falling market prices. The evidence is that shorters increase market efficiency by facilitating discovery and those that call for shorters to be restricted and/or regulated may not be acting in the best interest of investors.

  13. @Newtownian

    The lead author’s name is interesting: O. McNoleg: assuming the O. stands for Oleg, this is a character mentioned in Blackadder, The Cavalier Years, as in:

    Baldrick: Sir, please don’t kill me, I have a cunning plan to save the king.

    Blackadder: Forgive me if I don’t do a cartwheel of joy, your family’s record in the department of cunning planning is about impressive as Stumpy Oleg McNoLeg’s personal best in the Market Harbor marathon! [sighs] All right, what’s the plan?

    Perhaps this is a hint as to the paper’s seriousness…but stranger things have happened.

  14. @Newtownian
    The funny thing is, with the exception of the haggis, those are (or were – I’ve been out of the field for some years) quite reputable modelling techniques for spatial data.

    Unfortunately my work’s security software has blocked access to the online journal (you’d think they’d trust a university library, but no), but I’ll take a look at it online tonight and get back to you.

  15. @rog
    Outright deception (cooking the books) is difficult in relation to financial markets, but not impossible. My example – Fred Goodwin – was about deceiving regulators. The US subprime mortgage scandal involved large-scale deception both ways: customers and investors. (The securities simply did not have corresponding records of servicing, mortgage by mortgage). The Greek government deceived Eurostat about its true levels of debt, coached by Goldman Sachs. That’s three examples, pretty significant ones I’d say, to set against an a priori argument.

  16. @James Wimberley

    Amd w hat about manipulating the LIBOR? “A senior banker from a UK bank has admitted conspiring to defraud over manipulating the Libor lending rate. The banker, who can not be named for legal reasons, is the first person in the UK to plead guilty to the offence. Two men have already pleaded guilty in the US to fraud offences linked to the rigging of Libor, for years the benchmark by which trillions of pounds of financial contracts are based.” – BBC , 7 Oct, 2014.

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 )

Connecting to %s