Antifragile: Things That Gain from Disorder

Home > Other > Antifragile: Things That Gain from Disorder > Page 53
Antifragile: Things That Gain from Disorder Page 53

by Taleb, Nassim Nicholas


  BOOK II: Modernity and the Denial of Antifragility

  City-states: Great arguments in support of the movement toward semiautonomous cities. Benjamin Barber, Long Now Foundation Lecture (2012), Khanna (2010), Glaeser (2011). Mayors are better than presidents at dealing with trash collection—and less likely to drag us into war. Also Mansel (2012) for the Levant.

  Austro-Hungarian Empire: Fejtö (1989). Counterfactual history: Fejtö holds that the first war would have been avoided.

  Random search and oil exploration: Menard and Sharman (1976), controversy White et al. (1976), Singer et al. (1981).

  Randomizing politicians: Pluchino et al. (2011).

  Switzerland: Exposition in Fossedal and Berkeley (2005).

  Modern State: Scott (1998) provides a critique of the high modernistic state.

  Levantine economies: Mansel (2012) on city-states. Economic history, Pamuk (2006), Issawi (1966, 1988), von Heyd (1886). Insights in Edmond About (About, 1855).

  City-States in history: Stasavage (2012) is critical of the oligarchic city-state as an engine of long-term growth (though initially high growth rate). However, the paper is totally unconvincing econometrically owing to missing fat tails. The issue is fragility and risk management, not cosmetic growth. Aside from Weber and Pirenne, advocates of the model, Delong and Schleifer (1993). See Ogilvie (2011).

  Tonsillectomies: Bakwin (1945), cited by Bornstein and Emler (2001), discussion in Freidson (1970). Redone by Avanian and Berwick (1991).

  Orlov: Orlov (2011).

  Naive interventionism in development: Easterly (2006) reports a green lumber problem: “The fallacy is to assume that because I have studied and lived in a society that somehow wound up with prosperity and peace, I know enough to plan for other societies to have prosperity and peace. As my friend April once said, this is like thinking the racehorses can be put in charge of building the racetracks.”

  Also luck in development, Easterly et al. (1993), Easterly and Levine (2003), Easterly (2001).

  China famine: Meng et al. (2010).

  Washington’s death: Morens (1999); Wallenborn (1997).

  KORAN and Iatrogenics:

  Semmelweiss: Of the most unlikely references, see Louis-Ferdinand Céline’s doctoral thesis, reprinted in Gallimard (1999), courtesy Gloria Origgi.

  Fake stabilization: Some of the arguments in Chapter 7 were co-developed with Mark Blyth in Foreign Affairs, Taleb and Blyth (2011).

  Sweden: “Economic elites had more autonomy than in any successful democracy,” Steinmo (2011).

  Traffic and removal of signs: Vanderbilt (2008).

  History of China: Eberhard (reprint, 2006).

  Nudge: They call it the status quo bias and some people want to get the government to manipulate people into breaking out of it. Good idea, except when the “expert” nudging us is not an expert.

  Procrastination and the priority heuristic: Brandstetter and Gigerenzer (2006).

  France’s variety: Robb (2007). French riots as a national sport, Nicolas (2008). Nation-state in France, between 1680 and 1800, Bell (2001).

  Complexity: We are more interested here in the effect on fat tails than other attributes. See Kaufman (1995), Hilland (1995), Bar-Yam (2001), Miller and Page (2007), Sornette (2004).

  Complexity and fat tails: There is no need to load the math here (left to the technical companion); simple rigorous arguments can prove with minimal words how fat tails emerge from some attributes of complex systems. The important mathematical effect comes from lack of independence of random variables which prevents convergence to the Gaussian basin.

  Let us examine the effect from dynamic hedging and portfolio revisions.

  A—Why fat tails emerge from leverage and feedback loops, single agent simplified case.

  A1 [leverage]—If an agent with some leverage L buys securities in response to increase in his wealth (from the increase of the value of these securities held), and sells them in response to decrease in their value, in an attempt to maintain a certain level of leverage L (he is concave in exposure), and

  A2 [feedback effects]—If securities rise nonlinearly in value in response to purchasers and decline in value in response to sales, then, by the violation of the independence between the variations of securities, CLT (the central limit theorem) no longer holds (no convergence to the Gaussian basin). So fat tails are an immediate result of feedback and leverage, exacerbated by the concavity from the level of leverage L.

  A3—If feedback effects are concave to size (it costs more per unit to sell 10 than to sell 1), then negative skewness of the security and the wealth process will emerge. (Simply, like the “negative gamma” of portfolio insurance, the agent has an option in buying, but no option in selling, hence negative skewness. The forced selling is exactly like the hedging of a short option.)

  Note on path dependence exacerbating skewness: More specifically, if wealth increases first, this causes more risk and skew. Squeezes and forced selling on the way down: the market drops more (but less frequently) than it rises on the way up.

  B—Multiagents: if, furthermore, more than one agent is involved, then the effect is compounded by the dynamic adjustment (hedging) of one agent causing the adjustment of another, something commonly called “contagion.”

  C—One can generalize to anything, such as home prices rising in response to home purchases from excess liquidity, etc.

  The same general idea of forced execution plus concavity of costs leads to the superiority of systems with distributed randomness.

  Increase of risk upon being provided numbers: See the literature on anchoring (reviewed in The Black Swan). Also Mary Kate Stimmler’s doctoral thesis at Berkeley (2012), courtesy Phil Tetlock.

  Stimmler’s experiment is as follows. In the simple condition, subjects were told:

  For your reference, you have been provided with the following formula for calculating the total amount of money (T) the investment will make three months after the initial investment (I) given the rate of return (R):

  T=I*R

  In the complex condition, subjects were told:

  For your reference, you have been provided with the following formula for calculating the total amount of money An the investment will make three months after the initial investment An-1 given the rate of return r.

  Needless to mention that the simple condition and the complex one produced the same output. But those who had the complex condition took more risks.

  The delusion of probabilistic measurement: Something that is obvious to cabdrivers and grandmothers disappears inside university hallways. In his book The Measure of Reality (Crosby, 1997), the historian Alfred Crosby presented the following thesis: what distinguished Western Europe from the rest of the world is obsession with measurement, the transformation of the qualitative into the quantitative. (This is not strictly true, the ancients were also obsessed with measurements, but they did not have the Arabic numerals to do proper calculations.) His idea was that we learned to be precise about things—and that was the precursor of the scientific revolution. He cites the first mechanical clock (which quantized time), marine charts and perspective painting (which quantized space), and double-entry bookkeeping (which quantized financial accounts). The obsession with measurement started with the right places, and progressively invaded the wrong ones.

  Now our problem is that such measurement started to be applied to elements that have a high measurement error—in some case infinitely high. (Recall Fukushima in the previous section.) Errors from Mediocristan are inconsequential, those from Extremistan are acute. When measurement errors are prohibitively large, we should not be using the word “measure.” Clearly I can “measure” the table on which I am writing these lines. I can “measure” the temperature. But I cannot “measure” future risks. Nor can I “measure” probability—unlike this table it cannot lend itself to our investigation. This is at best a speculative estimation of something that can happen.

  Note that Hacking (2006) does not
for a single second consider fat tails! Same with Hald (1998, 2003), von Plato (1994), Salsburg (2001), and from one who should know better, Stigler (1990). A book that promoted bad risk models, Bernstein (1996). Daston (1988) links probabilistic measurement to the Enlightenment.

  The idea of probability as a quantitative not a qualitative construct has indeed been plaguing us. And the notion that science equals measurement free of error—it is, largely but not in everything—can lead us to all manner of fictions, delusions, and dreams.

  An excellent understanding of probability linked to skepticism: Franklin (2001). Few other philosophers go back to the real problem of probability.

  Fourth Quadrant: See the discussion in The Black Swan or paper Taleb (1999).

  Nuclear, new risk management: Private communication, Atlanta, INPO, Nov. 2011.

  Anecdotal knowledge and power of evidence: A reader, Karl Schluze, wrote: “An old teacher and colleague told me (between his sips of bourbon) ‘If you cut off the head of a dog and it barks, you don’t have to repeat the experiment.’ ” Easy to get examples: no lawyer would invoke an “N=1” argument in defense of a person, saying “he only killed once”; nobody considers a plane crash as “anecdotal.”

  I would go further and map disconfirmation as exactly where N=1 is sufficient.

  Sometimes researchers call a result “anecdotal” as a knee-jerk reaction when the result is exactly the reverse. Steven Pinker called John Gray’s pointing out the two world wars as counterevidence to his story of great moderation “anecdotal.” My experience is that social science people rarely know what they are talking about when they talk about “evidence.”

  BOOK III: A Nonpredictive View of the World

  Decision theorists teaching practitioners: To add more insults to us, decision scientists use the notion of “practical,” an inverse designation. See Hammond, Keeney, and Raiffa (1999) trying to teach us how to make decisions. For a book describing exactly how practitioners don’t act, but how academics think practitioners act: Schon (1983).

  The asymmetry between good and bad: Segnius homines bona quam mala sentiunt in Livy’s Annals (XXX, 21).

  Stoics and emotions: Contradicts common beliefs that Stoicism is about being a vegetable, Graver (2007).

  Economic growth was not so fast: Crafts (1985), Crafts and Harley (1992).

  Cheating with the rock star: Arnavist and Kirkpatrick (2005), Griffith et al. (2002), Townsend et al. (2010).

  Simenon: “Georges Simenon, profession: rentier,” Nicole de Jassy Le Soir illustré 9 janvier 1958, N° 1333, pp. 8–9, 12.

  Dalio: Bridgewater-Associates-Ray-Dalio-Principles.

  BOOK IV: Optionality, Technology, and the Intelligence of Antifragility

  The Teleological

  Aristotle and his influence: Rashed (2007), both an Arabist and a Hellenist.

  The nobility of failure: Morris (1975).

  Optionality

  Bricolage: Jacob (1977a, 1977b), Esnault (2001).

  Rich getting richer: On the total wealth for HNWI (High Net Worth Individuals) increasing, see Merrill Lynch data in “World’s wealthiest people now richer than before the credit crunch,” Jill Treanor, The Guardian, June 2012. The next graph shows why it has nothing to do with growth and total wealth formation.

  FIGURE 39. Luxury goods and optionality. On the vertical the probability, on the horizontal the integral of wealth. Antifragility city: the effect of change in inequality on the pool of very rich increases nonlinearly in the tails: the money of the superrich reacts to inequality rather than total wealth in the world. Their share of wealth multiplies by close to 50 times in response to a change of 25% in dispersion of wealth. A small change of 0.01 in the GINI coefficient (0 when perfect inequality, 1.00 when one person has all) equivalent to 8% rise in real Gross Domestic Product—the effect is stark regardless of the probability distribution.

  Camel in Arabia: Lindsay (2005).

  Obliquity: Kay (2010).

  Real options literature: Trigeorgis (1993), review in Dixit and Pindyck (1994), Trigeorgis (1996), Luehrman (1998), McGrath (1999)—the focus is on reversible and irreversible investments.

  Translational gap: Wooton (2007); Arikha (2008b); modern Contopoulos-Ioannidis et al. (2003, 2008), commentary Bosco and Watts (2007).

  Criticism of Wootton: Brosco and Watts (2007).

  Epiphenomena and Granger-causality: See Granger (1999) for a review.

  Lecturing birds how to fly: There are antecedents in Erasmus, “teaching fish how to swim.” Adages, 2519, III, VI, 19. “Piscem nature doces I’χθὺν νήχεσθαι διδάσκεις, id est piscem nature doces. Perinde est ac si dicas : Doctum doces. Confine illi, quod alibi retulimus : Δελφἶνα νήχεσθαι διδάσκεις, id est Delphinum natare doces.” The expression was first coined in Haug and Taleb (2010), posted in 2006, leading to a book, Triana (2009). We weren’t aware of the Erasmus imagery, which we would have selected instead.

  Education and its effect on growth and wealth: Pritchett (2001), Wolf (2002), Chang (2011).

  Schumpeter’s ideas on destruction for advancement: Schumpeter (1942). Criticism by Harvard economists about lack of technical approach in McCraw (2007).

  Amateurs: Bryson (2010), Kealey (1996).

  Scientific misattribution of the works of Bachelier, Thorpe, and others: Haug and Taleb (2010). Discussion in Triana (2009, 2011).

  Jet engine: Scranton (2006, 2007, 2009), Gibbert and Scranton (2009).

  Busting the episteme theory of cybernetics: Mindell, 2002. I thank David Edgerton for introducing me to his works.

  Cathedrals and theoretical and axiomatic geometry: Beaujoan (1973, 1991), Portet (2002). Ball (2008) for the history of the construction of Chartres cathedral.

  Epistemic base and conflation: The epistemic base is sort of the x, not f(x). A great way to see the difference between x and f(x) in technology, offered by Michael Polanyi: one can patent f(x), a technique, but not x, scientific knowledge. In Mokyr (2005).

  Epistemic Base: Mokyr (1999, 2002, 2005, 2009). The biggest problem with Mokyr: not getting ωC. Further, this notion of the East missing trial and error (also see argument about China): see Tetlock in Tetlock et al. (2009). Mokyr and Meisenzahl have a different spin, with microinventions feeding macroinventions. Still intellectually weak.

  Techne-Episteme in economics: Marglin (1996), but the tradition did not go very far.

  Needham’s works on China: Winchester (2008).

  Tenure: Kealey (1996): “Adam Smith attributed the English professors’ decay to their guaranteed salaries and tenured jobs. (As compared to Scottish Universities.)”

  Fideism: Popkin (2003).

  Linear Model: Edgerton (1996a, 1996b, 2004). Edgerton showed that it was a backward-fit idea, that is, fit to the past. Edgerton also writes: “This profoundly academic-research-oriented model of twentieth-century science is all the more surprising in view of the long tradition of stressing the non-academic origins of modern science [emphasis mine], particularly the craft traditions, and the insistence of much history of science, strengthened in the last 20 years, on the significance of industrial contexts for science, from dyeing to brewing to engine making.”

  Convexity bias: It was discovered early in commodity and financial futures; Burghardt and Hoskins (1994), Taleb (1997), Burghardt and Liu (2002), Burghardt and Panos (2001), Kirikos and Novak (1997), Pieterbarg and Renedo (2004). Many people blew up on misunderstanding the effect.

  Example of detection and mapping of convexity bias (ωA), from author’s doctoral thesis: The method is to find what needs dynamic hedging and dynamic revisions. Among the members of the class of instruments considered that are not options stricto-sensu but require dynamic hedging can be rapidly mentioned a broad class of convex instruments: (1) Low coupon long dated bonds. Assume a discrete time framework. Take B(r,T,C) the bond maturing period T, paying a coupon C where rt = ∫rs ds. We have the convexity д2B/дr2 increasing with T and decreasing with C. (2) Contracts where
the financing is extremely correlated with the price of the Future. (3) Baskets with a geometric feature in its computation. (4) A largely neglected class of assets is the “quanto-defined” contracts (in which the payoff is not in the native currency of the contract), such as the Japanese NIKEI Future where the payoff is in U.S. currency. In short, while a Japanese yen denominated NIKEI contract is linear, a U.S. dollars denominated one is nonlinear and requires dynamic hedging.

  Take at initial time t0, the final condition V(S,T) = ST where T is the expiration date. More simply, the security just described is a plain forward, assumed to be linear. There appears to be no Ito term there yet. However should there be an intermediate payoff such that, having an accounting period i/T, the variation margin is paid in cash disbursement, some complexity would arise. Assume ∆(ti) the changes in the value of the portfolio during period (ti,ti-1), ∆(ti)= (V(S,ti)-V(S, ti-1)). If the variation is to be paid at period ti, then the operator would have to borrow at the forward rate between periods ti and T, here r(ti,T). This financing is necessary to make V(S,T) and ST comparable in present value. In expectation, we will have to discount the variation using forward cash flow method for the accounting period between ti-1 and ti. Seen from period T, the value of the variation becomes Et [exp[-r(ti,T)(T-ti)] ∆(ti)], where Et is the expectation operator at time t (under, say, the risk-neutral probability measure). Therefore we are delivering at period T, in expectation, as seen from period t0, the expected value of a stream of future variation Et0[Σ exp[-r(ti,T)(T-ti)] ∆(ti)]. However we need to discount to the present using the term rate r(T). The previous equation becomes V(S,T)|t=t0= V[S,t0]+ exp[r(T)] Eto [Σ exp[-r(ti,T)(T-ti)] ∆(ti)], which will be different from ST when any of the interest rate forwards is stochastic. Result (a polite way to say “theorem”): When the variances of the forward discount rate r(ti,T) and the underlying security STare strictly positive and the correlation between the two is lower than 1, V(S,T)|t=t0 ≠ ST. Proof: by examining the properties of the expectation operator. Therefore: F(S, t0) = F(S,t0+∆t), while a nonlinear instrument will merely satisfy: E[V(S,t0)]=E[V(S,t0+∆t)].

 

‹ Prev