Monday, May 29, 2006

Human agency in history: Niall Ferguson and Counterfactuals, Nicholas Taleb and Monte Carlo Simulations

If human agency is the traditional mainstay of narrative history, is human agency guided by the rigid deterministic collectivist rules that regulate a society or is there an element of contingent strategic action that jumps the bounds of these rules at critical points sometimes with elite seeking alliances and cultural exchanges with other societies?

Niall Ferguson’s notion of counterfactuals and Nicholas Taleb’s Monte-Carlo simulations are two ways of looking at the strategic behaviour of elites, in short human agency.

Perhaps poking fun at historians a bit, Taleb names the chapter he devotes to hindsight bias: "Skills in predicting past history. He defines hindsight bias as the “overestimation of what one knew at the time of the event due to subsequent information.” He uses apt analogies to convey what hindsight bias means: “Things are always obvious after the fact”, “Imagine taking a test knowing the answer”, “It is hard to imagine that people who witnessed history did not know at the time how important the moment was”, or “While we know that history flows forward, it is difficult to realize that we envision it backwards” (Taleb, Fooled By Randomness, 2nd ed., p. 56).

Taleb stresses the importance of choosing the right perspective: “A mistake is not something to be determined after the fact, but in light of information until that point.” In other words, base your historical interpretation on information that was available to a given historical actor, officer or soldier, up to a given point in time, a call for narrative history written from the perspective of participants?

Taleb combines his critique of after-the-fact history with a plug for Monte Carlo simulations, a mathematical technique from applied probability. “Far more a way of thinking than a computational tool,” Monte Carlo mathematics is a way of creating artificial histories and toying with the role of uncertainty in determining these histories.

A Monte Carlo generator creates thousands of random sample paths, a “succession of virtual historical events.” These historical paths are not just random walks or Brownian random motion. Given the constraints set in the Monte Carlo generator, some outcomes will have a higher probability or frequency than others. Choosing historical constraints (like Braudel does in his Mediterranean): “One sets conditions believed to resemble the ones that prevail in reality, and launches a collection of simulations around possible events.” You can imagine a huge collection of virtual histories analyzed after they are generated: “One can generate thousands, perhaps millions of random sample paths, and look at the prevalent characteristics of some of their features” (Taleb, 2005, 46). The difference from the normal after-the-fact history mentioned above is that we have thousands of possible histories (or worlds) not just the one history that was actually realized.

How can Monte Carlo techniques be applied to warfare? These mathematical methods are a lot more straightforward to apply to financial markets because detailed time series of prices for the assets traded on these markets are available. The very first historical application of these techniques was probably to warfare. The Rand Corporation think tank of the 1950’s applied Monte Carlo mathematics to war games simulating nuclear war during the cold war era.

It’s doubtful though that war games can adequately capture the continually changing panorama and mix of causal factors that enter real life decision-making. For history reconstructing the constraints that actors faced is an additional problem. The reconstruction of the battle of Waterloo that took place shortly after the battle was the first ambitious attempt to exhaustively assemble such historical constraints.

It is hard to pull the uncertain decisions out of the historical, the chance component of Clausewitz’s three causal factors in warfare (violence, chance, political objectives) because events have already been determined, so they appear deterministic.

In the end though, historical reconstruction is a matter of perspective. Choosing the appropriate perspective for writing history means eschewing an omniscient god-like perspective and adopting the narrative perspective that actual historical actors like officers and soldiers and the political leaders that guided them from above faced.

Of course Taleb is talking about traders on financial markets, not warfare, but there is a mapping from one domain to the other that makes a cross-pollination feasible. Niall Ferguson with his notion of virtual history has pointed the way.