Just how big a leap? And why? Why is it so difficult to go beyond a forecast of the weather, an inherently chaotic system, to using that forecast to protect life and property? What makes that seemingly incremental step so monumental?
It’s the shift from a purely physical system to an essentially human one. Humans are individually and corporately far more complex.
Intractably so, according to Richard Bookstaber, an academic economist, who’s also turned his hand to finance and hedge fund management in the world’s markets. He’s just published a new book, entitled The End of Theory: Financial Crises, the Failure of Economics, and the Sweep of Human Interaction. His book touches on weather (and even space weather!) a bit, but his main focus is the question: why can’t economics predict financial crises (e.g., 2007-2008)? He concludes that economic models are flawed in four basic respects.
- They can’t handle emergent phenomena – the way that the aggregate of human interactions can evolve in ways unrelated to individual intentions (his favorite examples are stampedes and traffic bunch-ups).
- The models fail to account for the ways history comes into play. The atmosphere’s future behavior will be determined solely by its initial state – not the prior circumstances that brought it there. By contrast, for human beings, history (experience) – not some aggregated cultural history, but unique individual histories – make a critical difference in behavior. Traditional economic models fail to capture this non-ergodic feature.
- Social behavior is not just uncertain – it’s radically uncertain, defying model characterization by the usual statistical approaches.
- The models confront what he calls “computational irreducibility”; the future is so complex, and the effect of human interactions so unfathomable, that people cannot possibly create models to anticipate the outcome.
Hmm. At this point, you and I might reasonably have two questions. First, what do attempts to model financial crises have to do with making weather forecasts suitable for impact-based decision support? Well, it’s not a perfect fit, but human and institutional decisions based on projections about financial markets and coming booms or busts would appear to be in some sense similar to decisions based on information about weather opportunity and risk. These same challenges face us. When we go beyond characterization of the atmosphere per se to messaging that attempts to help people make decisions (“turn around, don’t drown,” “break the grip of the rip,” “in case of earthquake, climb to higher ground,” “tomorrow you won’t have to irrigate,” and “for the next three hours, wind and solar energy will supply xx% of the electrical demand,” and more), we’re in essence trying to forecast not only the weather, but also the human response to a few words of warning delivered as a text or orally, or an image conveying the same content. Hardly surprising we find this difficult.
Second, doesn’t modeling of weather share some of the same four complexities described above? Yes, weather modelers might make such a case that to some extent we deal already with #’s 1,3, and 4, allowing for some differences of opinion we might all have with respect to details. But #2 is more problematic. You could argue that any given state of the atmosphere says much about where it was at the previous moment, but it’s possible to proceed without delving into that past. Moreover, it might be harder to make a case about “individual differences” in experience, or even what that means in the atmospheric instance.
But Mr. Bookstaber brings up a fifth issue, a pièce de résistance: George Soros’ idea of “reflexivity.” Think of this as the Heisenberg uncertainty principle on steroids. According to Heisenberg, the mere act of observing a physical particle’s position increases the observer’s uncertainty about the particle’s velocity; in turn, measuring its velocity increases uncertainty with respect to its position. Soros notes that economic actors – individuals, banks, corporations, et al. – avidly read economic predictions or central-bank signals of intent, and then actively, intentionally change their behavior in ways that suit their purposes but in the process render those forecasts less accurate, and policy shifts less useful. He uses an example from real estate: the belief that housing prices always rise makes buyers more willing to pay higher prices, and financing more readily available; hence house prices rise.
We see similar instances of this in meteorology. The finding that evacuation orders in the face of an oncoming hurricane find resistance in the threatened area; as many as half of those ordered to leave may choose to stay. At the same time, many who were asked to stay home instead hit the road, needlessly and counterproductively clogging escape routes for others. Another: viewed from the standpoint of a single homeowner facing a tornado strike on his/her home in fifteen minutes, flight might seem to make sense. But that fails to account for the actions of all other neighbors; if everyone tries to leave, the resulting traffic snarl increases the vulnerability for all.
With respect to economic forecasts of financial crises, Mr. Bookstaber offers an alternative to the economic models currently in favor. He calls this “agent-based economics,” in which the modeler doesn’t assume that all the actors in the financial markets are the same. Instead Mr. Bookstaber advocates building models that describe the major categories of players in financial markets (banks, central banks, hedge funds, big investors, et al.) and the rudiments of their goals and ways and means of engagement, and then running the models much as traffic modelers attempt to capture the features of traffic bunching. He doesn’t attempt to model precisely how the next financial crisis will unfold, but instead runs his models numerous times to build up what he calls “narratives” that capture the range of ways events might unfold. (To the meteorologist-reader, this is reminiscent of what ensemble forecasts do for us.)
The book is short. It is well written. The examples are clear, easily understood, and the arguments compelling and mind-expanding. A good use of a couple of hours for anyone working at the nexus of meteorology, social science, and weather services.
Summing up? The shift to impact-based decision support from forecasts of physical weather conditions per se is a big leap.
Like going to “lightning” from “lightning bug.”
In the next post, a look at the policy implications.