Cool:

The software models in question estimate the level of financial risk of a portfolio for a set period at a certain confidence level. As Benoit Mandelbrot, the fractal pioneer who is a longtime critic of mainstream financial theory, wrote in Scientific American in 1999, established modeling techniques presume falsely that radically large market shifts are unlikely and that all price changes are statistically independent; today’s fluctuations have nothing to do with tomorrow’s—and one bank’s portfolio is unrelated to the next’s. Here is where reality and rocket science diverge. Try Googling “financial meltdown,” “contagion” and “2008,” a search that reveals just how wrongheaded these assumptions were…

The causes of this fiasco are multifold—the Federal Reserve’s easy-money policy played a big role—but the rocket scientists and geeks also bear their share of the blame. After the crash, the quants and traders they serve need to accept the necessity for a total makeover. The government bailout has already left the U.S. Treasury and Federal Reserve with extraordinary powers. The regulators must ensure that the many lessons of this debacle are not forgotten by the institutions that trade these securities. One important take-home message: capital safety nets (now restored) should never be slashed again, even if a crisis is not looming.

For its part, the quant community needs to undertake a search for better models—perhaps seeking help from behavioral economics, which studies irrationality of investors’ decision making, and from virtual market tools that use “intelligent agents” to mimic more faithfully the ups and downs of the activities of buyers and sellers. These number wizards and their superiors need to study lessons that were never learned during previous market smashups involving intricate financial engineering: risk management models should serve only as aids not substitutes for the critical human factor. Like an airplane, financial models can never be allowed to fly solo.

Check out the full piece

>

Source:
After the Crash: How Software Models Doomed the Markets
Scientific American, November, 2008

http://www.sciam.com/article.cfm?id=after-the-crash&sc=DD_20081121

Category: Markets

Please use the comments to demonstrate your own ignorance, unfamiliarity with empirical data and lack of respect for scientific knowledge. Be sure to create straw men and argue against things I have neither said nor implied. If you could repeat previously discredited memes or steer the conversation into irrelevant, off topic discussions, it would be appreciated. Lastly, kindly forgo all civility in your discourse . . . you are, after all, anonymous.

34 Responses to “How Software Models Doomed the Markets”

  1. debreuil says:

    I would love to be able to audit some of the code out there. I would bet that a lot of it isn’t just ‘didn’t think of that scenario’, but actually willfully changing formulas until they say ‘buy more and you can’t go wrong’. That kind of thing is usually pretty easy to spot in code. Wishful thinking is the source of many bugs, so you get tuned for it : ).

  2. ironman says:

    Mandelbrot is absolutely correct in observing that the assumption made by many of these models that stock prices are statistically independent is incorrect. Stock prices obey a power law relationship – they are highly interdependent with dividends. Once you understand that, a lot of the stock market’s behavior falls right into place and most changes in stock prices may be recognized as being rationally-driven, which means that behaviorial economics would only have limited utility in explaining the market.

    Speaking of that “better model”:

    * The basic power law relationship between stock prices and dividends per share
    * Why models that assume independence are wrong, but useful
    * Using the two to explain an “unexplainable” event: the Black Monday Crash of 1987.

    And finally, the evidence that stock prices are rationally driven (even in the current investing climate!)

    There’s more, of course – the links at the bottom of the last post cited will show you quite a bit more, and there’s quite a bit that I haven’t posted as yet.

  3. paulyarbles says:

    The assumptions of human rationality and resultant models give the illusion of that economics can be a discipline that is quantifiable in the same way as physics or chemistry. This illusion allows the present-day economic orthodoxy to rest on a ‘solid scientific’ foundation. Needless to say that this orthodoxy is quite beneficial to those who already have a great deal of wealth. Especially those in big finance.

    It’s well known that economics can be quantifiable in some way but the maths that result have no predictive power. Yet the assumptions and modeling persists. Cui bono?

    Accepting that human beings are irrational destroys the legitimacy of the economics profession as it is currently practiced. An economics that does accept the truth of human irrationality and, thus, is more in line with the real world would lose its most important role as servant of wealth and power.

    It will take time for economists to come up with an economics that incorporates human irrationality and is still beneficial to the super wealthy. Rest assured they are hard at work on it as you read this. Once that project is finished expect to see how human irrationality leads to the status quo.

  4. Bruce in Tn says:

    “has already left the U.S. Treasury and Federal Reserve with extraordinary powers. ”

    Yes, and the trend of congress to abdicate their powers to the executive branch is trouble as the governing of our democracy goes forward. Congress has many times in my lifetime “gone along” with undeclared wars, now they are giving money to the executive branch that has essentially no rules. TARP understood for one thing, now goes for another…

    Seems to me the executive branch is getting far too much in the way of unfettered extraordinary powers. Why isn’t congress willing to fight for this?

  5. wally says:

    In all the analysis about who got it wrong, it is time to observe that some did get it right: events of the past few years make the debt-deflationists look pretty good. This is a complete collapse caused by credit rising to unsustainable levels; it can only be fixed after the losses are recognized and taken or after the currency is destroyed – no other options.
    So, for the future: what level of credit triggers these events? How is it created? What limits or controls should be enacted? Should the Fed or some other agency be watching credit levels rather than GNP or employment or – god forbid – the stock market?
    This is not a simple business-cycle recession. It is part of a complex system of bubble-and-bust that we have been in for an extended period. It is time for economists to pull their collective heads out and re-examine some of their fundamental theories about modern economies.

  6. rww says:

    Lots of things are hard to understand but everything, once understood, is pretty simple. Beware anyone who cannot explain in simple terms what they do. Complexity obscures; simplicity illuminates. It’s Ptolemy vs. Copernicus.

  7. ironman says:

    Not Ptolemy vs Copernicus, so much as Brahe vs Kepler. ;-)

  8. rww says:

    Too “obscure”, Ironman.

  9. ironman says:

    Ah, but the personalities are more compelling, rww! The vision of a bigger than life guy who wore a brass nose prosthesis because he lost part of his in a sword fight, and who could describe the reality he saw in exquisite detail versus the vision of a dedicated number cruncher who could make a newer reality jump out of the data the other recorded. Sure, we know who Hollywood would choose….

  10. jmborchers says:

    This is my trade. Software. There is no such thing as a good model which shows reality when non-scientific inputs are part of the model (IE behavior). Models are only good when they are based on facts.

  11. Jason G. says:

    I think it is not a failure of the quants… though their output (the models) did certainly fail.

    I think it was that the quants were steered in a specific direction to placate the agendas of their managers. If I am told to find out ways to lever up, I find ways to lever up. The quant goes into a meeting saying “as long as X and Y are true, then we’re safe”. The manager walks out of the room having heard “we’re safe levering up”.

    There are more than a few quants that had good models… look at Taleb or Wilmott. Their models couldn’t scale to make oversized profits at the big firms, so their models were ignored and managers went looking for others who would give them the answers they wanted.

  12. VoiceFromTheWilderness says:

    Stock prices don’t obey any law. That’s the beginning of going wrong with financial models. Stock prices are a projection from a trillion dimensional space onto a 1-D space. As long as all your model looks at is stock prices, it will always be surprised. These are elementary observations. Stock prices are not an independent variable, they are a dependent variable. If you don’t look at the actual independent variable you’ll never understand why things change. (I’ve repeated myself). A simple example: 9/11. Would any rational person expect a financial model (treating stock price as an independent statistical variable) to predict the price drop on 9/13 (I forget the exact day). No we wouldn’t, but we can predict it, easily. Because we know that the causative variable was an action taken for entirely non-market reasons, the causes of which (right or wrong, this isn’t a morality tale) had been building for many years. In order to predict that event in a model one would have had to incorporate trillions of variables at different scales and that model would have to have built into it an ‘understanding’ of things like cultural identity, emotional response, historical patterns, personal relationships (access to money and weapons on a global scale), details about transportation and how it works, vulnerability of structures to various kinds of attack — notice that these variables aren’t even of the same type, or on the same scale. The point that is that in fact *every* price movement is governed by precisely these same kinds of forces.

    Blaming quants is just as silly as blaming any body else — funny though how the ones getting paid to make decisions are the least subject either to blame or to consequence. The reality is that models are tools, and like any tool only as useful as the hand guiding them. Blaming model makers for not delivering the fantasy dreams of the Masters of the Universe ensures the perpetuation of the fantasy that one can be ‘in charge’ without knowing what one is doing, without knowing anythingabout the tools one is using or how they work. It is certainly true that the tool builders want to believe their tools can do anything — if their owners aren’t smart enough to know better, then their owners will perish with the tool builders in the disaster that follows. Unless of course the owners manage to transfer the cost onto someone else, and of course the first step in transferring cost is transferring… blame

  13. Dude says:

    I always assumed a lot of the market behavior was an ‘emergent’ behavior from individual investment decisions similar to the emergent behavior of a flock of birds or a herd of animals. I can’t believe the models assumed so much independence in prices over time.

  14. Gene says:

    The remark attributed to Disraeli would often apply with justice and force: “There are three kinds of lies: lies, damned lies, and statistics” – Mark Twain

  15. spooky says:

    what bullshit.

  16. RW says:

    I’m forced to come down somewhere between Ironman and Dude: Investing does possess a rational foundation in the perceived present value of dividend streams but there is also the robust and eminently repeatable observation that only a few of us monkeys actually invest that way; e.g., http://tinyurl.com/6ktzas (ht Brad Delong)

  17. gorobei says:

    I find this whole “blame the models” thing to be absurd. Does anyone really think that quants didn’t read Mandelbrot? Of course we did. He’s a second rate thinker but a good salesman. Do his ideas add a single penny to the bottom line? No.

    Most people seem confused as to what quants and their models do. 95% are not in the prediction business, they are in the arb biz: if you say you can price options, I’ll give you a model for binaries; if you say you can price knockins, I’ll give you a model for transformers.

    You want to do mortgages? No problem! What are we buying, and what can we sell? Type, type, type and here is model that gives us three cents per dollar. You ask about the risk? From a math point of view, very little, but note the sensitivity to credit and liquidity. That’s a job for the traders and the risk committee.

    Oh, you want to arb on-the-runs vs off-the-runs? Gosh, better get a Nobel-prize winner to do that, because this is a sales job not a quant job. Oh, the “model” was wrong? Sorry, the model was right, and 5 minutes of thought would have an investor why.

  18. AGG says:

    Hamlet’s Mill and fat tails; They never happen until they happen. Black Swans or Sigma 25 events are an effort by people to model reality. Sure, things happen cyclicaly but it is a fools errand to try to guarantee you won’t get stuck at some point. The bottom line is that if you swim, you’re going to get wet. It’s unavoidable. The man that convinces you he can beat the odds is just trying to make a little money.

  19. aftf says:

    Gorobei is right. Most models are pretty much the same and even the best traders make plenty of losing trades. Judgment and leadership separates the successful firms from the losers.

    It also needs to be pointed out that what models say is not static because the inputs are changing all the time. Traders don’t put on positions and then walk away to wait for the money to come in. They are constantly rebalancing their positions to take out profits and adjust for risk.

  20. Irene Flowers says:

    There’s to say that existing models were faulty mostly because of engineering limitations. The quants tended to make of a necessity a virtue by coming up with all sorts of justifications in principle. But there’s only so far one can go selling technical compromises as features, as opposed to recognize them as bugs.

    Traditional models needed to be built so that they would be analytically solvable otherwise calibration was not possible. Analytic solvability however limits very much the extent to which models can reproduce a realistic price dynamics.

    So models tend to be in a constant state of flux. Model parameters keep change. The solution would be to embed the flux into the model itself, making the dynamics on parameters endogenous. But to do this correctly one requires non-parametric models.

    That’s the future in my opinion. But again, one needs technology and computing power. Fortunately nowadays access to teraflop computing is becoming extraordinarily cheap with massively parallel architectures. I see quants moving on rapidly to new model classes to seize that power. This would have happened anyway, but the crisis will likely speed up the renewal process.

  21. ajjhp510 says:

    Agree with Gorobei. Models are fine. Quants know the assumptions that the models imply – and so do the good traders.

    Failure has to come from the Risk Mangement side of the firm. It is there job to aggregate the books across desks/firm and say we have too much of X or Y or if correlations go to 1 we are fucked!

    I traded a big FI book on the street for years. What I can tell you is that the Risk Management guys need some sort of independence or the trader just runs over them. Generally the guy that makes the firm money (ie trader) wins the fight. Ofc I have no idea how to fix this.

  22. “So models tend to be in a constant state of flux. Model parameters keep changing.(-ed.) The solution would be to embed the flux into the model itself, making the dynamics on parameters endogenous. But to do this correctly one requires non-parametric models.”

    LSS: “Nothing=Anything”

    Can ‘models’ be a useful guide? Of course. Can they provide ‘Answers’? ..Riight~

    re: Risk, from above: “From a math point of view, very little, but note the sensitivity to credit and liquidity.” This: “sensitivity to credit and liquidity.”, is a helluva Caveat. “Credit” y “Liquidity”, in view of the History of Markets, are the Exceptions, not the Rule..

    Past that, “gorobei”, “Irene Flowers “, and “ajjhp510″, if any of y’all would care to be a little more discursive, it’d be appreciated. Ths is a subject, central to many–whether they know it, or not–that few know anything about..

  23. willid3 says:

    not sure we can blame the software it self. software is based on what some one (who hopefully knows what they are talking about) tells the developer. and it can only do what those ‘experts’ have told it do, and based on what information those same ‘experts’ have given it. now if some one ignores some thing (because they believe it be unimportant) and it turns out to be very important, we tend to blame it. instead of those experts . now developers can make mistakes in their coding true but its usually easy enough to identify that, because it used the wrong numbers or some how corrupted them. but this sounds more like a failure by design. cause the failure was ignoring some thing altogether. and economics seems to tend to be more of a behavioral science not a rational one. otherwise there would never be days that nobody could explain why the market did some thing (instead they use it was a day investors are taking profits, which translate into, I don’t know what happened or why)

  24. gorobei says:

    Mark Hoffer,

    Hey, we’re happy to pontificate up to Barry’s tolerance level.

    What’s the subject, exactly? I can bore people for hours on:

    1. Bank compensation policies and ownership structure
    2. Gauss vs Cauchy and what it means for you
    3. IB credit lines and gamblers’ ruin
    4. Hedging exotics and the underlying effects.

  25. gorobei, fire away until BR comes by w/ the buckets..

    2. and 4. are prob. the 200-level ones, out of that bunch, the other two are more readily accessible.

    as a +, they’ll lend insight into the ‘brittleness’ of the pricing edifice..

  26. danm says:

    It’ s very simple, in portfolio management risk measures are based on volatility. More often than not, models use 3 years worth of data. While MPT clearly states expected risk and returns, portfolio managers use historical data.

    Most models would have told a portfolio manager who was at 0% in Nortel when the stock was at 120$ that he was being reckless.

    I rest my case.

  27. themis says:

    As debreuil points out, the software itself might be loaded with defects.
    From http://thedailywtf.com/Articles/The-Great-Excel-Spreadsheet.aspx

    —-

    In going back previous editions of the spreadsheet, somehow, they managed to send completely idiotic numbers to their customers for three full years (at least). Not a single customer, not a single manager ever noticed the inconsistency for what were supposed to be trivial multiplications; not a single one of them noticed that “The-most-important-figure-on-this-chart-we-base-all-our-decisions-on” was random garbage.

    In the end, he found out that whenever Helen needed to create a new row, she would simply copy and paste some random row and then adjust the values. At some point, however, she must have messed up and the spreadsheet ended up in a weird state, with formulas referring to cells in other rows, or sometimes even referring to nothing, creating a whole bunch of inconsistent values.

    Excited at the chance to clear his name, Maxim revealed his findings to the lead analyst. However, instead of relief, he only shrugged and responded “Hmph… well, we usually just use our gut for recommendations, anyway.”

  28. gorobei says:

    Mark,

    as you note, 1 and 3 are the easy ones.

    #1 basically killed Wall St. Once you switch from a partnertship to a LLC, the game is over: yes, increased access to capital markets lets you do bigger deals, but 1) the compensation time-horizon shrinks from 30+ years to 5 years or so, and 2) the people at risk become a diffuse pool.

    Without a pool of 1000 or so of partners and limited partners (the retired MDs,) it’s all too obvious what happens next: people lever the one big asset (firm reputation,) and make great returns for a decade. Which brings us to point#3:

    Few people were trying to be evil, they just optimized given the structure in place. GS, for example, lined up a $40 billion short-term credit facility: seemed prudent and conservative. Then they paid their top producers really well: again, very reasonable, theys guys made the money for the firm.

    But the firm completely missed where the real risk was: you can’t lever reputation! Traditionally, most partnerships fall into one of four categories: doctors, laywers, accountants, and investment banks: all charge high fees because the service is opaque and thus putting the senior guy’s money at risk is a good way to signal and ensure quality.

    The model only works if owners have real skin in the game: give them an option (e.g. LLC protection on the downside,) and they will lever to the max.

  29. constantnormal says:

    Ahem. Allow me to point out the elephant in the discussion forum.

    Is not TMA merely another way to model the future based on the dependent variables of price and volume?

    It may not be as precise as these more “sophisticated” automated trading algorithms, but the premise is the same, and it suffers from the same fallibility.

    Looking for a method of predicting movements in a hugely complex system of human interactions is fine, but don’t bet your future on it. You can back-test algorithms to the Big Bang, but there’s no guarantee that the future will conform. I suspect that one could wrap this into the context of The Halting Problem and logically show it to be impossible.

    Instead, design hedging strategies that always work — which is no small feat in itself.

  30. constantnormal says:

    @gorobei — how about a skeleton outline (at least) of #2. History (esp. mathematical) and personalities are always fun for me. A long time ago, in another life, I had a fair understanding of the work of Gauss and Cauchy. But that was a long time ago.

  31. gorobei says:

    constantnormal, hang on a little while, the explanation is proving to be a bit longer than I anticipated :(

  32. gorobei,

    thanks for the additional…upon re-reading my post @December 14th, 2008 at 9:10 pm, I see I was less than clear.

    I was meaning that 2. and 4. Should be explained, b/c 1. and 3. were less difficult, more accessible..

    tho, seeing this: December 14th, 2008 at 11:55 pm, thanks again~

  33. Yohei says:

    gorobei,

    I am looking forward to your future post as well.

    Recognized your nom de plume immediately and so picked mine, and rather appropriately, I think!

  34. gorobei says:

    Yohei, you have good taste in movies.