During a Long Now seminar, Nassim Taleb notes that low probability events are the most dangerous, in that they are the most unpredictable. The error term for an event with few observations is higher (the tails of any probability distribution); smaller n → larger ε. I began to ponder. My current occupation seems to be predicated on the notion that it’s possible to infer the future. But it’s sort of a joke: I have no idea what’s going to happen tomorrow, much less years from now.

The Austrian objection to this sort of inference: it’s impossible to know the future; the simple act of knowing will change any future state given a change in the past stock of knowledge. Meanwhile the entire modern politico-economic conceit is based upon estimating: next year’s revenue, national health care costs, pollution levels, future liabilities for failed institutions… Yet if we take these types objection seriously, then it’s fundamentally impossible to predict the future. But it’s also necessary. Why do people keep trying? Start with the following:

• It’s impossible for anyone to accurately predict the future. Anytime someone is correct, it is because of random luck. Those individuals who develop reputations as intelligent forecasters are simply the few who happen to be lucky over and over again.
• Simultaneously, it seems to be possible to choose almost any measurable indicator, and for any sufficiently short time period, the values will probably not going to be substantially different. For example Namibian infant mortality in 1997 isn’t substantially less awful than 1998. 2010 probably won’t differ substantially from 2009.

So: short periods of time seem provide a degree of confidence; the near past is like the present. This makes sense – the passage of time is necessary for change to occur. Neurons fire and motivate action. Actions are sometimes coordinated with other individuals. Only so many coordinations can take place within a relatively short period of time, thus tomorrow’s stock of matter and knowledge will probably resemble today’s stock of matter and knowledge. Again, most would respond that it’s a question of scale: predictions are mostly good one year out, but not so much beyond a threshold of… 10 years; depending on what you’re looking at.

Quantifiable is key: greater confidence in shorter time periods stems from the constraint that in general, fewer actions occur within shorter time periods. But when pressed, it becomes apparent that actions  are not homogenous in time (other dimensions, as it were, are required) and measurement as such is just convenient shorthand. Crude examples: there’s some variability in the time spent to brush one’s teeth on a daily basis; the Volatility Index went through a lot more variability in 2008 than in most years previous. There’s no deterministic relationship. To complicate the line of thought further, there’s good reason to believe that time (or the passage thereof) is simply an illusion, and measurements thereof are simply shorthand. From Yuval Dolev’s Time and Realism: Metaphysical and Antimetaphysical Perspectives, a summary of the logical argument for the “tenseless” (illusory) view of time:

… [time] is a mode of thought, speech, and experience, and not an element of reality itself.

1. Past, present, and future are … “incompatible determinants.” That is, these temporal categories are mutually exclusive: if an event e is past, then it is not present and it is not future; if it is present then it is not past and it is not future; if it is future then it is not past and it is not present.
2. Each event has all three determinants, that is, each event is past, present, and future. Indeed, it would be absurd to maintain of anything that it is only past, or only present, or only future.

The conclusion that follows is that since nothing in reality can have incompatible determinants, the distinction between past, present, and future is not a part of reality…

The immediate rejoinder is that [the event] is now past, but it is not now future or present. Rather it was future and was present… saying of [the event] that it is future at a time t1, and present at t2, and past at t3, is to no avail, for each of these times, t1, t2, t3, is itself past, present, and future. So now we’ve replaced the one original contradiction with three new ones…  And this can go on forever. The infinite regress, however, works in favor of [the tenseless view], for in every new stage, new contradictions are lurking. The regress could have undermined this argument only if at some point it would not have led to a contradiction-free moment or event. But it does not. [pp. 17-18]

There are also metaphysical and empirical arguments (relativity), and points to be made against each claim. Regardless, these arguments should at least cause one to think twice about assumptions as they relate to the passage of time as the only dimension in which action occurs, especially as these arguments depend on accurate delineations of comparable time scales: arguably the basis of all economic and financial modeling. To his credit, Dolev addresses the notion that the logical and metaphysical arguments are based upon language and modes of thought that may or may not be reflective of reality; the notions that time is “tensed” or “tenseless” are both constructs of human reasoning. Put another way:

“There are only two ways in which we can account for the necessary agreement of experience with the contents of its objects: either experience makes these concepts possible or these concepts make experience possible.” – Immanuel Kant, Critique of Pure Reason

So, [assuming passage of time is an illusion] language and notions of time have developed only as heuristics to allow people to interface effectively with the world and coordinate effectively with each other. This abstraction allows us to live, breed, and pass these crazy (but useful) ideas down to our offspring. Theories that work towards this end are propagated and codified; those that don’t work are forgotten. Differential equations, for example, allow people to come up with solutions to lots of interesting phenomena, but are incapable of solving, say, the n-body problem (for n ≥ 3, anyway). Are these theories actually descriptive, or just happen to be what works for certain applications, and so we construct our view of reality around tractable equations that we can grapple with while potentially ignoring a whole host of questions like the n-body problem?

All is not lost: physical sciences certainly have a clear advantage in this arena. There’s probably a strong bias towards reality description since we can at least observe and record reality / control experiments. For social “sciences,” pretty much everything should be thrown out the window. The few genuine ‘theories’ that survive are maybe reflective of some underlying reality, but used instead for the utility they provide in coordinating action. To put it in a very crude way: allowing someone to predict next year’s revenue is very useful if you’re in the business of,  say, exploiting excess labor, and so the Capital Asset Pricing Model gets included in textbooks and annual reports, since it usually works at coming up with results that allow proper coordination of action.

This digression doesn’t do much to advance the original query; but seems to be a fairly significant issue upon which I can find very little literature. The conclusion? I don’t know if there is one. Potentially it would indicate that people are better off trusting tried and true methodologies, however idiosyncratic they might seem. Such a notion isn’t that different from what I understand about legal theories. But it’s very unclear. These sorts of things keep me up at night.