textual metanoia

As a professor of statistics, with several popular books, and a roster of technical articles, Nassim Nicholas Taleb ought to know how to write in a passionate but measured way. Taleb's essay about contemporary risk management in Edge reads like an attack piece that, at best, races over alternative views and counter-research to make a point, or, at worst, races over these points to obfuscate errors and falsehoods. A quick analysis of Tableb's rhetoric bears out this point: he constantly uses language of certainty, bigness ("blow up"), and smugness and academic posturing ("the American Statistical Association had a special panel session on the "black swan" concept"). What you won't find are hedging words or precise language. Taleb even goes so far as to use the rhetoric that he criticizes others for using because of its misleading effects (when not understood in an appropriate analytical frame), e.g., "here I got hold of more than 20 million pieces of data". In addition to academic posturing, Taleb borrows terms from others, such as Mandelbrot's "monsters" without citing him (although this is surely excusable in an essay intended for a general audience). Taleb's "black swans", however, are (as far as I can tell), not much different than the point Mandelbrot was making decades ago. I'll admit, no one listened to Mandelbrot back then (and they should), so a rehashing of these old ideas isn't a terrible thing, but let's not celebrate the originality of the "black swan". Taleb's suggestion that academia doesn't properly document its failures is also unoriginal, with projects like Open Notebook Science starting to get traction and countering this trend.

Although Taleb was a "quant" financier himself for many years, he seems to be out of touch with contemporary practices, or glosses over them to portray only those that work with his theory. Taleb criticizes statistical methods for failing to model very rare occurrences, because there is not enough empirical history to inform the model (if we have never seen it occur, we can't model it). While it's true that we can't model some unknown directly, we can still develop models for prediction by getting the rest of the parts correct (though a dubious reality with contemporary techniques), and then "stress test" certain scenarios to measure the effect. Of course, Taleb mentions stress testing, but he seems to mistake it for "sampling", which is when you use historical data as input for the model. Stress testing typically uses made up scenarios as an input, such as "What if counterparty X defaults on this instrument?". There is no historical evidence of such an event, but if the model is correct in general, changing a single parameter has a chance of accurately modeling reality (but also requires knowing if certain effects are multi-casual, correlated, or stand in intractable relationships). It's difficult to model, for sure, but not because we don't have empirical evidence. We aren't looking for Truth in our prediction models, we are looking for likely and pragmatic outcomes.

A current popular "revolution" (as some say) in economics is the introduction of behaviourism. Taleb's theoretical framework doesn't seem sophisticated enough to recognize the behavioural qualities of a financial system (since he uses mathematical techniques that have a family resemblance to those he is criticizing). Mackenzie's analyses have been very influential as of late in the sociology of economics; Mackenzie recognizes the artificiality of the financial system and argues that these analytical techniques may actually be performative, thus creating the reality they require. Problems still can arise for a performative system, however, since it requires everyone to play along in the game. What we have been seeing in recent months is a lack of confidence in the system, or, put another way, people being less willing to play the game. Iliquidity is sure to occur.

Finally, Taleb's analysis is ultimately two-fold: he criticizes the application of the mathematical underpinnings of risk management, and then criticizes the people who use this math. Taleb doesn't seem to be too bothered by the math itself, since he continually offers kind words for statisticians, but he has some criticisms of the application of the math used for modeling. Taleb's main point, however, is that the risk managers and executives who use these systems misunderstand the mathematics behind them and thus make bad decisions. Stepping back though, we realize that this isn't much of damming criticism; people are frequently stupid, misguided, or undereducated, this is hardly unique to risk management. Should the risk managers better understand the systems? Sure, but there are also technical fixes for these sorts of problems. As software develops, more and more of the decisions can be offloaded to the computer. In fact, years ago when LTCM was offering its "insurance" the entire process was mechanical and without the need to human input (turns out that was a bad idea though, but that was due to unexpected occurrences, not bad decisions).