Modeling fraud in the financial system
Today we have a guest post by Dan Tedder. Actually it’s a letter he sent me after listening to my EconTalk podcast with Russ Roberts which he kindly agreed to let me post. Dan’s bio is below the letter.
I think this letter is profound (although I don’t completely agree about the Markov stuff), because it points out something that I see as a commonly held blindspot by people who think about regulation and modeling. Namely, that any systemic risk model of the financial system that doesn’t take account of lying isn’t worth the memory it takes up on a computer.
That brings us to the following question: can we incorporate lies into models? Can we anticipate and model fraud itself, in addition to the underlying system? Or do we give up on models and rely on skeptical people to ferret out lies? Or possibly some hybrid?
——
Hi Cathy,
I really liked your interview, and I think you are right on in pointing to a lack of ethics. I would say further that what we need is rigorous honesty in all aspects of the financial system. I agree with your objections to conflicts of interest. Allowing such conflicts to exist demonstrates a lack of rigorous honesty on the part of the participants. In my opinion a lot of bankers and folks on Wall Street should be headed to jail. The inability of the SEC to file charges and prosecute them further demonstrates the lack of honesty and character in the financial system and the government. So why am I telling you things you already know?
My father was a successful businessman. Years ago I was invited to invest in an ice cream franchise by another faculty member. I spent several days developing models using Excel. Finally, I decided to talk to my father. I called him and he immediately asked me to tell him about the present owners and their accounting. I told him the husband was in jail and accounting was five years behind. Further, his wife was probably taking money out of the till.
He stopped me right there, and pointed out that I needed to look no further. The present owners were not honest and therefore the opportunity was too risky. No telling what liabilities they had incurred and passed on to the franchise. I felt like an idiot. My modeling was a total waste of time because it assumed the present owners were honest. In fact, they were dishonest and no defensible model could be constructed based upon their accounting or lack thereof.
I think the complexity of our present financial problems will largely disappear if we try to focus more on the obvious. First, it is obvious that bankers, accountants, modelers, and other participants must be rigorously honest. Second, George Box, a statistician at the University of Wisconsin, studied the stock market and found through time series analysis that stock market prices are Markov processes. So in modeling stock prices we need only worry about today and tomorrow. The best indicator of tomorrow’s price is today’s price. The best indicator of what will happen tomorrow is where we are today, and probably our models of the larger process should also be Markovian. Third, apply the KISS method, “Keep it simple, stupid.” Instead of worrying about the mathematical model, worry about the honesty of the participants. The financial system cannot tolerate dishonesty. Making sure the bankers are honest will go a long way toward balancing the books.
Regards, Dan
——
Daniel William Tedder is Associate Professor Emeritus, School of Chemical and Biomolecular Engineering, and Adjunct Professor, School of Mechanical Engineering, both at the Georgia Institute of Technology. He attended Kenyon College and received a Bachelor’s in Chemical Engineering at the Georgia Institute of Technology. He obtained MS and PhD degrees in Chemical Engineering at the University of Wisconsin, Madison. He was a staff engineer in the Chemical Technology Division of the Oak Ridge National Laboratory before joining the faculty at Georgia Tech. He served as an independent technical reviewer at the Nuclear Regulatory Commission after retiring from Georgia Tech. He has numerous publications, has edited 11 books, and has authored one book, Preliminary Chemical Process Design and Economics, which is available from Amazon. He is an expert in chemical separations and in actinide partitioning, an advanced method for radioactive waste management.
mathbabe 
Reminds me of Warren Buffet’s investment philosophy.
Sent from my iPad
LikeLike
Buffett’s old investment philosophy maybe. The Buffett philosophy now is to invest in too big to fail (Wells Fargo) or government subsidized (Goldman Sachs) companies and rely on insurance float growth to propel Berkshire’s growth.
LikeLike
Haha, how quickly his BAC ‘gamble’ is forgotten…on top of his $5bb at 6% annual interest with another 5% due at repayment (he leap-frogged stockholders and even preferred shareholders in case things went further south) his 700mm $7 warrants (with 10 yr. execution exp. date instead of the 3 yr exp. goldman gave him in in 2008) are now already up 60-70%…
Genius indeed…
LikeLike
Modelling fraud seems like a silly idea to me. The solution to bad or dishonest models hurting the economy is not more models. The solution is to make it so that the people using the bad models are never in a position to be able to hurt the wider economy in the first place. You do this by attacking the root cause of the problem, which is the moral hazard caused by banks being ‘too big to fail’. If the banks are never TBTF then they can use all the shitty models they want, because they will be punished by bankruptcy.
Also on stock prices being Markovian, to quote Mr McEnroe: You. Cannot. Be. Serious. Are we going to be claiming that markets are efficient next?
LikeLike
Ummm… there was lots of fraud before there was TBTF.
LikeLike
Yes, but TBTF gives extra incentive to create fraudulent models.
My point is, why try to detect bad models in a system which encourages bad models? The effort would be better spent first trying to fix the incentives.
LikeLike
Great! Let’s fix the incentives that we can! But there will always be bad incentives that we can’t fix, by the very nature of banking and leverage.
LikeLike
The NYSE Equity markets are efficient. At least in the econometrician’s “this is not falsified sense” which is the only non-handwavy sense that matters. The time frames for efficiency matter, but anecdotally none of the patterns discovered in the 1970s work anymore, same with the 1980s and same with the 1990s and so on.
The stock market is not a Markov process for the same reason as the above. However, Box is an eminent statistician and so I’m genuinely curious about what he found.
LikeLike
No, inefficiencies which were identified in the past do still exist.
For an example see [1]. This paper was published in 1986. The inefficiencies around stock index inclusions persists to this day. Banks and prop trading firms trade on this.
If you are interested in this area I would recommend reading Andrei Shleifer’s “Inefficient Markets”.
1. http://www.economics.harvard.edu/faculty/shleifer/files/demandcurves.pdf
LikeLike
I have read Inefficient Markets and I’m a researcher in finance. I have actually done research in the paper you brought up and I believe Shleifer gives a non-EMH violating, albeit contrived reason for the price increase in the conclusion of the paper involving certification of quality or some such behavior. You could also contrive “liquidity” reasons for price increases in the included firms IF you knew something about the market microstructure before and after S&P 500 inclusion.
Market efficiency typically requires a plausible risk-return structure (Fama) OR the persistence of an exploitable arbitrage net of transactions costs (Malkiel). The “handwavy” sense I was referring to is certainly prevalent.
How many times have you heard people assert that large downward or upward price moves aren’t possible in an efficient market OR that if market moves don’t appear to have any human discernible reason – then markets are inefficient.
LikeLike
How does the inefficiency in Shleifer’s paper not violate the EMH? (semi-strong).
I can’t tell if it is a rhetorical question or not, but I’ve never heard anyone claim that large up or down moves aren’t possible in an efficient market. If significant news comes out then a large move will occur, surely this is completely in line with the EMH?
Also I’m curious as to why you single out the NYSE as being an efficient venue?
LikeLike
At the top of my wish list is being able to express agreement or disagreement (ideally, with anonymity being optional) with a single mouse click. We have what appear to be a couple heavyweights duking it out over the EMH and our only options are to sit quietly by or jump into the ring? If this intermediate form of contribution existed, I would vote for raising Blah’s fist skyward at this point.
LikeLike
The Physics of Finance blog has written at length pushing wacko anti-EMH arguments like Leon mentions. Just check it out, I’m not up for rehashing. Pops up lots of places.
LikeLike
Thanks Bindicap. That shit is everywhere you look and there is no need to direct people to Google.com
I single out NYSE because I (and every other researcher/trade/investor looking anywhere evar) have seen persistent/exploitable violations of EMH over longer timeframes in lower-liquidity markets and developing countries with bad information disclosure. I could say also NASDAQ and many dark pools are basically efficient..I single out equities because there is some decent evidence that some options were mispriced before the 87 crash and this kind of lines up with most people’s intuition about finance.
So on the Shanghai exchange in China, for instance, even naive momentum-Markov chain based strategies can yield supernormal profit. Traders have a little joke about this: in America arbitrageurs take advantage of momentum and make it disappear make momentum disappear. In China, momentum makes arbitrageurs who try to take advantage of it disappear. There’s a good reason for this: government insiders use momentum as a signal to reap supernormal profits and since trades are recorded on the Shanghai exchange and international investors cannot make these investments – outsiders who would make the momentum disappear risk gov’t interference. =
On the other hand, in the US, despite there being a wealth of papers on technical analysis and momentum investing – virtually NONE of the momentum strategies have been profitable net of transactions costs for the average hedge fund (that isn’t plugged into the exchange) and even less so for the average investor. Of course, you can go through paper by paper and actually build in realistic transactions costs to see which phenomenon that have been published persist. I’ve actually done this years ago in my capacity as a research assistant for a Professor. The result? Unless you were next to the exchange and were an institution – nothing you read in a paper would yield supernormal profit. Even so, usually using a non-CAPM risk adjustment could kill the rest.
I haven’t read the paper and don’t have time right now, but Shleifer does say something about certifying quality right? That alone can be built into an “efficiency” framework. I’m not sure if Shleifer points this out, but being added to the S&P makes a stock a lot more tradeable and this added liquidity could also explain a lot about the price jump.
EMH is certainly false over certain time frames and for certain asset classes (you rarely hear EMH arguments about people’s wages even though you could). But it is *close* to truth in a deep way. If you’re the type who needs to read a leftist economist defending something before you believe it, my views are very much in line with:
http://noahpinionblog.blogspot.com/2013/02/in-defense-of-emh.html
LikeLike
well my stepbrother has made $70mn using momentum trading. I guess he chose the markets where it works?
LikeLike
OR it wasn’t actually momentum trading or he just got lucky. as a certain number of investors are going to do.
LikeLike
Whatever the explanation I do not believe it was luck. He used computer models to identify situations in futures contracts where further price movements might be expected, says he always had exposure to a minimum of two hundred different contracts to spread the risk and take advantage of the trading edge his models gave him. And he said he was a momentum trader. I do not know what the underlyings were he traded but I doubt he was picky – anything where he could identify a tendency for trends to persist.
He was no amateur, did do it full-time.
LikeLike
Also market efficiency is well defined, there is no hand waving. See the definitions of weak, semi-strong and strong efficiency.
LikeLike
True. It seems to me that people have given up on the problem of TBTF though. I find it strange that OWS don’t focus on this as their main issue, particularly as people criticise them for lacking concrete goals.
LikeLike
We do! At least in Alt Banking we do.
LikeLike
“So in modeling stock prices we need only worry about today and tomorrow”
What is the obsession with modeling stock prices ? If people can do it well enough to make money, great. If they lose money doing it they won’t be doing it very long. Its frequently academics telling other people how things should be done but never actually putting their money where their mouth is.
As for stock prices being markov processes, i’ll only point out that large stock gains are more likely in a period of large stock gains, and large stock drops more likely in a period of large stock drops, so this alone is pretty strong anecdotal evidence to the contrary. But if you think you can make money with that theory, then build a model around it and please make yourself very rich. I will be very happy for you (and even defend your right to spend the majority of your well earned profits as you see fit, rather than having a moral duty to give it mostly back to the goverment).
LikeLike
All that’s required of phenomena to attain variablehood is to be observed (or inferred) and deemed important. Dishonesty is both extremely important and observable, so it’s variable-worthy.
BFD: dishonesty variables have their place, but they’re nowhere near the panacea Prof. Tedder paints them to be.
LikeLike
The modeling of fraud can enter at the level of the very foundation : the assumptions. For instance, all five (sometimes six, depending on the analysis) of the explicit assumptions of the Black-Scholes(-Merton) model are, in ANY sense of the word rigor, false or incomplete or self-contradictory.
Any pretension of an efficient market are immediately destroyed by secrecy, which is the sine qua non of derivational instruments : efficiency obviously depends on rapid and full disclosure of relevant information – which secrecy inhibits at every level.
It took centuries for alchemy to become the science-based field of chemistry (still hindered by inconsistent international notation, however). The blanket-application of Markov processes must be carefully applied even here (e.g. the memory effect of vorticity in reaction rates). The nonsensical bias that Markoviana applies to history-burdened human transactions is an unsupportable extension of Gaussian Frequentism that not only has never been proven (much less shown to be Popper-falsifiable) and flies in the face of applicable fundamental principles Bayesian probability.
LikeLike
I work in a quant shop and when we build stock selection models we always incorporate a measure of fraud, in terms of the reliability of revenue/expense recognition, governance practices, litigation risk and so on. It works well when used properly and has allowed us to side-step disasters like Sino-Forest.
On a broader basis however, it’s not so much whether it’s a fraud or not but how much will that fraud hurt you. It is no different from any other event that causes an enormous destruction of value very quickly (maybe a bit because of subsequent drops in confidence etc. but not terribly so). From a portfolio perspective, this can be alleviated.
The real problem arises only when your “portfolio” cannot be managed, for whatever reason. HSBC is the perfect example – you can almost trivially kick out HSBC stock/debt/etc. out of your portfolio but you can’t realistically kick HSBC the bank out of your economy.
LikeLike
badmax : “… but you can’t realistically kick HSBC the bank out of your economy.” The hell you can’t : you revoke its charter to serve the public commons and kill it!
LikeLike
You could, but they employ approx. 43,000 people in the US alone. It’s not a very good solution, you are shifting the consequences on those undeserving.
LikeLike
Yes, honesty and fair dealing are very important in finance and are the reason for quite a lot of rules and culture. Failures are very damaging on both the small and large scale, as comes to light periodically. It’s not nearly enough though.
Even though the ice cream investment is outside finance and banking, I think there is an honest way to proceed and draws some lessons. Disclosure of the situation is critical, and valuation can’t be naive. You shouldn’t even consider buying an interest in such a small business unless you have hands on experience managing some similar retail business and are prepared for the complications hinted at. Modeling will be done with some more grounding and accounting for the uncertainty of the situation (big error bars).
I’m not too hung up on the Markov issue. For most purposes with stocks, it’s the right framework, but there are definitely cases where history matters (eg, mortgage prepay risk, though sometimes you can pull the relevant history into a hidden state). For his lesson that character matters, history is very important (too complex to pull into a hidden state). I think he’s rightly anti-anti-EMH here, and that’s the main point.
And KISS is definitely another good lesson. Try to tear down complexity and make sure what remains has a good reason.
LikeLike
Cathy and Dan:
Awesome—thank you!
Cathy, we connected on LinkedIn last week (and I’ve already turned on another MathBabe on to you; said she giddily after I shared it with her on Friday, “this is my new favorite blog”).
Dan, I’m a fee-only fiduciary on the business of investment advice in your former stomping ground of Madison, Wisconsin.
The dilemma so well cited in this post is one close to my heart, and job (I fought for many painful years to ensure that the two eventually correlated). I see the infantile and obscene reality of “Wall Street” up front and center every day and worse yet what it does to people who get hoodwinked (nay robbed, even when not by outright corruption) by it. It’s infuriating. And it’s stupid as we continue allowing it to be propped up every single day by our corrupt institutions, and corrupt individuals who comprise them.
But in our age of information technology it doesn’t have to be this way anymore.
My theory is this—for those in US society seeking roles of elected office and fiduciary duty (in other words any and all seeking public office/trust and as such claiming to act in “our” best interest, or an interest other than their own selfish self-interest) a character transparency measure is required. I call it the “Four Dimensional Quotient” or “4DQ”— inner, outer, deeper and greater as initially outlined by C.G. Jung—and the model and method for it is entirely possible today thanks to very smart people like such as yourselves. Don’t think you need to man up (yes, it’s mostly men behaving badly here as Cathy has so clearly and numerously pointed out) to such as thing? Fine, then go lie, cheat and steal in the private sector where the free market can have its way with you, idiot. But no longer is the public realm freely accessible for your ego-inflating sophomoric BS.
My learning is that good people—people of character integrity—do good things. It does not matter if such people are rich/poor, young/old, fat/skinny, black/white, Democrat/Republican, Liberal/Conservative, yadda/yadda—people acting from integrity (which the 4DQ intends to measure before and not after society gets fleeced by our Madoff’s, Spitzer’s and Nixon’s) do what’s right. And how much more obvious could it possibly be that we have anything but in far too many of our political and fiduciary “leaders” today?
I call this new social science Faustomics, after Goethe’s lifework and blueprint for shadow retrieval for the Western masculine psyche. I’ve got a lame blog for it http://www.faustomics.com that I way too little time to devote to (one son, two adopted daughters—all teenagers, a business, 19 year marriage, a non-profit to save our Amish raw milk farmer from bogus criminal charges… you know, little stuff keeps getting in the way). Also, BTW for the full economic platform of which Faustomics is only the last of four prongs, see http://www.lebowskiparty.com. In any event, I’m hoping that by “meeting” here we can collaborate in what ever way on this *important* (!!!) topic.
So inspired by you both—thanks again.
Brendan
PS: Dan, the Markov process seems entirely sensible as I understand it from your description. The probability theory behind it, or some close variation of it, is one of the tenets on which we base our investment advice (the Fama/French Three Factor model is another). IFA.com is our website on it if interested. Indeed, the price is right—or less wrong than is any other individual’s (can you say “hedge fund manager”?) opinion/forecast of it.
LikeLike
So, some time ago, Dan considered investing in a non-financial business that was (presumably) owned by a fellow academic. He detected fraud, and elected not to invest.
This sounds like a terrific basis for…. banning academics from business? Outlawing ice-cream franchises? Preventing investors from doing due-diligence? Stopping husbands and wives from working together?
Obviously, fraud in banking is a problem. It’s also a problem when it takes place in pharmaceutical companies, industrial procurement, public infrastructure contracts, and ice-cream franchises. There are conflicts – and therefore incentives for fraud – in literally every industry under the sun, so it would probably make sense to focus on ethics in society as a whole, wouldn’t it?
LikeLike
Yeah, the transaction here is all outside finance and banking, no doubt under exemptions for limited private security offerings. Probably a lot of people would be amazed at what would be needed for a formal registered securities offering that would lay bare all the background and risks in a case like this.
You can still draw conclusions based on the experience, though, and I believe the points about honesty are reasonable ones. And the point about naive modeling going wrong is important.
There’s probably a lot of crosswise talk on this thread. Seems like there is an underlying tone that finance is less honest and needs change here. But the real lesson is a transaction with some sketchy elements would need extensive documentation and disclosure up the wazoo as a formal finance transaction, and in finance you need to be scrupulously honest because when you lose your reputation, you are done.
LikeLike
Cathy, you are right to be skeptical about whether stock market prices are Markovian. Mandelbrot showed in the fifties that they follow a Cauchy/Mandelbrot distribution. Such distributions have neither a mean nor a calculable variance. This renders, for instance, the Black-Scholes-Merton equations untenable and unrepresentative of reality. As a number of modelers in this area assume a normal distribution, this makes their risk assessments completely unrealistic.Along with their salaries and bonuses, their view of reality seems to be contained within the bubbles they themselves create.
It would be salutary to see a theory that incorporates control fraud as part of its intrinsic structure.
LikeLike
Well, a Cauchy process is Markov. Fat tails is a separate issue.
LikeLike
“the Black-Scholes-Merton equations untenable and unrepresentative of reality. As a number of modelers in this area assume a normal distribution, this makes their risk assessments completely unrealistic.”
uh…no and lol. Most prop traders haven’t used gaussians for anything deep in options pricing since the 90s…
LikeLike
Not from what I have read. What distributions do they assume are operative?
LikeLike
In the laziest cases, they could just bootstrap a distribution from the historical time series of financial returns…
More sophisticated players use more sophisticated techniques.
LikeLike
I perhaps should have pointed out that fat tails is the most important factor in assessing such risks. I was thinking of the Gauss-Markov process.I stated it badly. The point I wanted to make was that if you assume normality, or a Gaussian distribution as it is often called, as is done with Black-Scholes-Merton, then your risk assessments will be way out.
Control fraud is where the top management initiates and controls the fraud and inevitably involves accounting irregularities. Maybe a theory incorporating control fraud can begin with such accounting irregularities. It seems to be a decent place to start.
LikeLike
Mathbabe: it might be useful to write a post about how Black-Scholes(-Merton) is/isn’t used in quantitative finance. There are a LOT of sins in finance but “assuming Gaussian returns in BS” is not one of them! Much like arguments about the efficient markets hypothesis I feel like some well intentioned mathematicians come across as hopelessly naive by missing the point as to how the model is applied in practice.
Nobody now or at any time in the past 20 years has “believed BS” in the sense that they plugin some values (interest rate(s), underlying price, option strike/type, etc) and compare the formula output to the prevailing market price (i.e. believe all the assumptions and follow the math to its logical conclusion).
Instead, it’s used to reduce a high-dimensional space down to a lower dimensional space so that two sides of a transaction can better manage the communication complexity (reducing a whole manifold down to a scalar function). For example the two parties can agree to fix certain values (interest rates and the underlying price and/or the strike price or both) and plug them into the equation and then consider something like dPrice/dVlt about that point. It’s mostly used to translate between prices and volatilities because everybody understands that (go-forward, unknowable, not even directly observeable) volatility is the true quantity of interest but a trade of the option has to occur at a particular price and that price annoyingly depends on many other small details. It’s really just a convention and the underlying assumptions do not have to be valid for it to serve this useful purpose.
LikeLike
Nice post but off the mark replies. Funny how most miss the prof’s key point about finance: it’s the people and their integrity that matter, if you want a proper investment, not the model. And none mentioned Gresham’s dynamic.
Guess the answer depends on who you ask.
LikeLike
The title may have misled but that’s admittedly a lame excuse for missing the author’s point. As penance, a link to Gresham’s law: http://en.wikipedia.org/wiki/Gresham's_law
LikeLike
Wow, at Georgia Tech, I knew many people who knew Dr. Tedder and knew of him myself, but didn’t take a class. I’m wondering if the ice cream franchise is Jake’s, which I always thought was great but many locations closed down. I had wondered why and the fraud makes sense.
LikeLike