BECG theblog

The emperor has no clothes, but the empress is seriously over-dressed.

This phrase came to me in the shower (a veritable temple of insight) as I pondered the question of why I am having so much trouble writing about the relationship between connectivity and the economic meltdown of the day. I realized I have so much to say (and I candidly admit that it is of substance, as embarrassing as I find this admission to be) that I don't know where to start. Or how to start. Somehow all my drafts have an 'approachability quotient' in the same (exceptionally low) range as do those imposing stone edifices on Mount Rushmore. 

Infauxmation

This is a beginning stab at addressing my earlier question about what we can learn from the current economic crisis. It will probably take several posts to lay out my hypotheses and opinions, so please indulge me, and feel free to chime in (I mean it. Having a conversation with yourself is not what the web is made for and besides, I've got too many of those going on in my head already).

To begin, I want to pick up on two of the items I mentioned in that last post:

1. Mistaking wishful thinking for reality, and
2. Ignorance of the assumptions and limitations that are built into the mathematical models upon which high-stakes financial bets were placed. 

  

I. Mistaking wishful thinking for reality

This is extremely common in all areas of life, but particularly so in the financial  arena. Often, it evolved into a group delusion, akin to the psychiatric disorder, folie á plusieurs (madness of many). Nobody bothered to pay heed the few who checked the facts and thought for themselves. That which did not fit with consensus reality went unheard and unseen, because with delusional systems, inconsistencies with one's world view literally 'don't compute.' It's odd how history repeats itself. This is something   Daniel Defoe (1660 - 1731) wrote about the South Sea Bubble of 1720:

    Some in clandestine companies combine;
    Erect new stocks to trade beyond the line;
    With air and empty names beguile the town,
    And raise new credits first, then cry 'em down;
    Divide the empty nothing into shares,
    And set the crowd together by the ears.

II. Ignorance of the assumptions and limitations that built into the mathematical   models upon which high-stakes financial bets were placed.

Apparently, many of the problems that caused financial systems to implode were flaws in the financial models used to predict  outcomes. These models include but are in no way limited to so-called 'neural networks' (Incidentally, 'neural networks'  do not replicate the brain's computational actions in any known way. They are simply pared down, highly abstracted, regression equations).* The term "neural networks" smacks of marketeering: taking a form of statistical analysis that is in the public domain and can be done by anyone with the appropriate background, giving it a catchy new name, and making it sound far more special than it actually is.

Regression equations predict the likelihood of an outcome or outcomes, based on the input of a series of input variables (in other words, data about various characteristics that are hypothesized to be good predictors of that outcome. The highest possible adjusted R² (this is also called the 'regression coefficient), indicating a 100% likelihood that the array of independent or input variables will produce a particular outcome, simply does not occur in nature. The reason is common sense: any outcome has multiple predictors, some of which are known, some of which are knowable, and some of which are neither. Moreover, every predictive model by definition comes along with an error term (this is a statistic that shows the average amount of error that one can expect) and a confidence interval (this shows the band within which 68.2%,  95.4%, 99.6, 99.8%, 99.9%, and on and on, of the true scores are likely to fall. Note that one can never reach 100%. It is an asymptotic curve, that is a curvilinear line that approaches nearer to the 'destination' (in this case, 100% of the scores falling within the confidence interval) but, though infinitely extended, would never meet it.

In fact, one of the central tenets of mathematical prediction is that no equation, no matter how perfect it is, can ever predict an individual outcome with 100% certainty. Moreover, there are several critical assumptions that must be met for the equation to be valid (regardless of whether it is statistically significant or not). An unique feature of the current situation is that, due to advances in computer and communications technology, it is now possible to run equations on truly massive data sets. This enables the mathematician to achieve higher and higher levels of statistical significance and power (meaning that the likelihood of getting a result that is way off by chance alone is greatly reduced). However, as the data sets get bigger, so does the likelihood of the highly improbable occurring, as Nassim Nicholas Taleb notes in his book, The Black Swan: the Impact of the Highly Improbable. Well, guess what? They did. In a sense, the financial community, in the throes of its folie á plusieurs, failed to take into account that their seemingly brilliant decisions were made on a foundation of infauxmation, that is something masquerading as highly credible information, but is distorted, inaccurate, presented without necessary caveats, or just plain wrong.

In Charles L.L.D. MacKay's 1820 book, Memoirs of Extraordinary Popular Delusions and the Madness of Crowds (worth reading based on the flamboyantly weird title alone, but also worth reading for its content), the author relays the following story:

An enthusiastic philosopher, of whose name we are not informed, had constructed a very satisfactory theory on some subject or other, and was not a little proud of it. "But the facts, my dear fellow," said his friend, "the facts do not agree with your theory."—"Don't they?" replied the philosopher, shrugging his shoulders, "then, tant pis pour les faits;"—so much the worse for the facts!

In short, financiers, government officials, and consumers in the throes of folie á plusieurs  and 'armed' with infauxmation -- both amplified by speed, volume, and computing power -- constitute a marriage made in Hades.

Remember, this is just a beginning. There is much more to say and to discuss. For example, per George Santayana ("Those who forget the past are condemned to repeat it.”), there is the question of forgetting and of how we manage to get ourselves into these binds over and over again without, it seems,  learning a thing. Another topic is the impact of elected (and selected) officials -- especially the creeping devastation that results when ideology dominates governance, crowding out the rule of law.

Oh yes, there is much more--so stay tuned!

  

* Statisticians, please forgive any simplifications I have made in the interest of increasing the comprehensibility of the concept's description.

Have you ever noticed...

• That people* institute preventative measures only after the fact (or that over time, they manage to convince themselves that 'now, it's different' and the dreadful thing will not happen again as, predictably, they dismantle or fail to maintain these measures -- thus ensuring that it will)?
• That far too often, the measures people institute are designed to prevent that which has already happened rather than that which looms ahead (and that the few who see it and say so are typically viewed as nut cases or worse until, posthumously, they -- or someone whose timing is better -- are proclaimed geniuses and heroes)?

These are just two of the quirks of the human psyche that I see expressing themselves in the current economic meltdown.

Others include: mistaking wishful thinking for reality, ignorance of the assumptions and limitations that are built into the mathematical models upon which high stakes financial were based, narcissism, and more. 

My aim is neither to be cynical nor to depress you (or myself). There are good lessons here. What are they? As they say in Jamaica, 'soon come!'

* In the interest of transparency, I will candidly admit that I am a person.

HAL is Not the Culprit

I've been largely silent, consumed by the news of the day, but I have decided to break my silence, realizing that it was the sheer volume of my thoughts that was preventing me from speaking.

So often, I hear or read that computers are the problem. Our relatively newfound ability to locate and process mass quantities of data somehow 'caused' the current financial crisis. What is odd and funny to me (not in the ha-ha sense) is that the machines are not really the problem. It's that those using them -- by omission, commission, or both -- are unknowingly wielding tools about which they know so little. (I refer to the mathematical formulae, the hardware and the software necessary to process mass quantities of data, which comprise a key element of the deck of cards collapsing around us).

Please forgive what may seem like an oversimplification, but statistical analyses basically boil down to two types of methods: 

• Those that rely on experimental data (basis=ANOVA, the analysis of variance), and 


• Those that rely on post-hoc analysis of trend or pattern-level data (most basic building block=the Pearson product moment correlation also known as the Pearson r. It is the basis of regression analysis, latent variable analysis--AKA factor analysis and all its permutations, path analysis and more).

Both of these, are  inferential statistics. That is, they are statistics that result from calculations on a sample; if the results are dramatic enough that they are unlikely to be due to chance alone, one can infer that the result will hold true for the population as a whole, within a range of variation known as the confidence interval and limited by the likelihood of type 1 (false positive) or type 2 (false negative) errors. Any inferential statistic, by definition, contains something called an error term, because one is predicting something that applies to an entire population (be it human, financial, or otherwise) from a sample. Predictive models simply cannot predict a single case of anything. Note: In the case of a census, no inference is necessary because the population parameters are known.

Moreover, there are certain assumptions built into all of these models which, if violated, render the outcome invalid. My favorite is called: homocedasticity. This is a basic assumption of regression analysis and means that the variation of x scores around the regression (y) line falls within certain limits and is not scattered all over the place. 

Several points about statistical analysis, inference, and prediction of outcomes:

1. When you start putting together multiple calculations using multiple sources of data upon which you have made inferences, you'd better remember to include the error term for each in your calculations. BTW, the error term is additive. If you put together too many equations and data sources, the error term gets bigger and bigger, and you might as well flip the script and admit that the error term is larger than the confidence interval, and if any prediction is to be made with confidence, it is that your result will be wrong.


2. If you also fail to take seriously the assumptions of each type of analysis you are using and make sure they are met, you are doubly doomed.


3. Often times, there is no empirical basis for the input data other than wishful thinking on the part of the data source. For instance, just because you believe that your privately held company is worth 20X gross revenue, even though you have never tested that theory, and no company like yours has ever sold for more than 2X revenue, doesn't make it so. My hunch is that this sort of wishful and -- in some cases delusional -- thinking is also a factor in the nation's current economic implosion.

The overarching point is that the machines are just doing the bidding of the people who run them. Any self-respecting statistician knows the above points to be true, but the statisticians have never been in charge. The people who run the show are the ones who hired the statisticians who used technology to perform calculations. 

Whether the statisticians bowed to the wishes of their employers, or had themselves forgotten that no matter how perfect the strength of an association, the type one or type two error scores, or etc., no inference can ever predict any one specific outcome--or whether they were clear about the limits of prediction and were simply ignored by their employers--is immaterial. The point is it's easy to blame a machine, even for doing what you told it to. The last time I checked, machines weren't able to defend themselves. 

Oh, by the way, there is no such thing as AI, unless one is referring to a certain extremely talented basketball player with a mind of his own. To think that a bunch of equations could ever mimic the complexity, the quirkiness, and the multidimensionality (not sure if this is a real word--if not, hope the meaning is clear) of the human mind -- which exists not just in the head but also in the finger, the small intestine, and etc.-- is surely delusional. 

Truly, it is a modern day version of Pygmalion but with a less happy outcome. At least the original Pygmalion fell in love with the statue of a woman. After praying to Venus to bring his beloved statue to life and having his wish granted, the couple bore a son and a daughter. I think we now are seeing just how unappealing the offspring of a person and an algorithm can be.