Confessions of a Fragilista: Talebian Redundancies and Insurance

I’ve been on a Taleb streak this year (here, here and here). Nassim Nicholas Taleb, that is, the options trader-turned-mathematician-turned public intellectual (and I even managed to get myself on his infamous blocklist after arguing back at him). Many years ago, I read Fooled by Randomness but for some reason it didn’t resonate with me and I wasn’t seeing the brilliance.

Last spring, upon reading former poker champion Annie Duke’s Thinking in Bets and physicist Leonard Mlodinow’s The Drunkard’s Walk, I plunged into Taleb land again, voraciously consuming Fooled, The Black Swan and Skin in the Game, followed by Antifragile just a few months ago.

Taleb is a strange creature; vastly productive and incredibly successful, everything he touches does not quite become gold, but surely stirs up controversy. What he’s managed to do in his popular writing (collected in the Incerto series) is to tie almost every aspect of human life into his One Big Idea (think Isaiah Berlin’s hedgehog): the role of randomness, risk and uncertainty in everyday life.

One theme that comes up again and again is the idea of redundancies: having several different and overlapping systems – back-ups to back-ups – that minimize the chance of fatally bad outcomes. The failures of one of those systems will not result in the extremely bad event you’re trying to avoid.

Focusing primarily on survivability – “absorbing barriers” – through the handed-down wisdom of the Ancients and the Classic, the take-away lesson for Taleb in almost all areas of life is overlapping redundancies. Reality is complicated, and the distribution from which events are drawn is not a well-behaved Gaussian normal distribution, but one of thick tails. How thick nobody knows, but wisdom in the presence of absorbing barriers suggest that taking extreme caution is a prudent long-term strategy.

Of course, in the short run, redundancy amounts to “wasted” resources. In chapter 4 of Fooled, Taleb relates a story from his option trading days where a client angrily calling him up about tail-risk insurance he had sold them. The catastrophic event from which the insurance protected had not taken place, and so the client felt cheated. This behavior, Taleb maintains quite correctly, is idiotic. After all, if an insurance company’s clients consist of only soon-to-be claimants, the company won’t exist for long (or it prices insurance at prohibitively high rates, undermining the business model).

Same thing applies for one of his verbose rants about airline “efficiency,” a rather absurd episode of illustrating “asymmetry” – the idea that downside risks are larger than upside gains. Consider a plane departing JFK for London, a trip scheduled to take 7h trip. Some things can happen to make the trip quicker (speedy departure, weather conditions, landing slot available etc), but only marginally; it would, for instance, not be possible to arrive in London after only an hour. In contrast, the asymmetry arises as there are many things that can delay the trip from mere minutes to infinity – again, weather events, mechanical failures, tech or communication problems.

So, when airlines striving to make their services more efficient by minimizing turnaround time – Southwest’s legendary claim to fame – they hit Taleb’s antifragile asymmetry; getting rid of redundant time on the ground, makes the process of on-loading and off-loading passengers fragile. Any little mistake can cause serious delays, delays that accumulate and domino their way through crowded airport networks.

Embracing redundancies would mean having more time in-between flights, with extra planes and extra mechanics and spare parts available at many airports. Clearly, airlines’ already brittle business model would crumble in a heartbeat.

The flipside efficiency is Taleb’s redundancy. Without optimization, we constantly use more than we need, effectively operating as a tax on all activity. Taleb would of course quibble with that, pointing out that the probability distribution of what “we need” must include Black Swan events that standard optimization arguments overlook.

That’s fine if one places as high a value on risks that Taleb does, and indeed they’re voluntarily paid for. If customers wanted to pay triple the money for airfares in order to avoid this or that delay, there is a market for that – it just seems few people value that price over the damage from (low-probability) delays.

Another example is earthquake-proving buildings that Nate Silver discussed in his The Signal and the Noise regarding the Gutenberg-Ritcher law (the reliably inverse relationship between frequency and magnitude of earthquakes). Constructing buildings that can withstand a high-magnitude earthquake, say a one-in-three-hundred-year event is something rich Californians or Japanese can afford – much-less so a poor country like the Philippines. Yes, Taleb correctly argues, the poor country pays its earthquake expenses in heightened risk of devastating damage.

Large redundancies, back-ups to back-ups, are great if you a) can afford them, and b) are risk-averse enough. Judging by his writing, Taleb is – ironically – far out along the right-tail of risk aversion; for most other people, we have more urgent needs to look after. That means occasionally “blowing up” and suffer hours and hours of airline delays or collapsing buildings after an earthquake.

Taleb rarely considers the trade-offs, and the different subjective value scales (or discount rates!) that differ between people. While Taleb may cherish his redundancies, most of us would rather eliminate them for asymmetrically small gains.

Insurance is a relative assessment of price and risks. Keeping a reserve of redundancies are subjective choices, not an objective necessities.

Do risk preferences account for 1.4 percentage points of the gender pay gap?

A few days ago, this study of gender pay differences for Uber drivers came out. The key finding, that women earned 7% less than men, was stunning because Uber uses a gender-blind algorithm. The figure below was the most interesting one from the study as it summarized the differences in pay quite well.


To explain this, the authors highlight a few explanations borne out by the data: men drive faster allowing them to have more clients; men have spent more time working for Uber and have more experience that may be unobserved; choices of where and when to drive matters. It is this latter point that I find fascinating because it speaks to an issue that I keep underlining regarding pay gaps when I teach.

For reasons that may be sociological or biological (I am agnostic on that), men tend to occupy jobs that have high rates of occupational mortality (see notably this British study on the topic) in the forms of accidents (think construction, firemen) or diseases (think miners and trashmen). They also tend to take the jobs in further removed areas in order to gain access to a distance premium (which is a form of risk in the sense that it affects  family life etc.). The premiums to taking risky jobs are well documented (see notably the work of Kip Viscusi who measured the wage premium accruing to workers who were employed in bars where smoking was permitted). If these premiums are non-negligible but tend to be preferred by men (who are willing to incur the risk to be injured or fall sick), then risk preferences matter to the gender wage gap.

However, there are hard to properly measure in order to assess the share of the wage gap truly explained by discrimination. Here with the case of Uber, we can get an idea of the amplitude of the differences. Male Uber drivers prefer riskier hours (more risks of having an inebriated and potentially aggressive client), riskier places (high traffic with more risks of accidents) and riskier behavior (driving faster to get more clients per hour).  The return to taking these risks is greater earnings. According to the study, 20% of the gap stems from this series of choices or roughly 1.4 percentage points.

I think that this is significantly large to warrant further consideration in the future in the debate. More often than not, the emphasis is on education, experience, marital status, and industry codes (NAICS code) to explain wage differences. The use of industry codes has never convinced me. There is wide variance within industries regarding work accidents and diseases. The NAICS codes industries by wide sectors and then by sub-sectors of activities (see for example the six-digits codes to agriculture, forestry, fishing and hunting here). This does not allow to take account of the risks associated with a job. There are a few study that try to account for this problem, but there are … well … few in numbers. And rarely are they considered in public discussions.

Here, the Uber case shows the necessity to bring back this subtopic in order to properly explain the wage gap.