Continuing the theme of financial types talking to each other about predictions and predictability, this ‘Tea with the Economist’ interview of Stephen Roach, Chairman, Morgan Stanley Asia by Economist New York Bureau Chief Mathew Birk, carries interesting lessons about the limits of prediction.
Birk commends Roach for being one of the few to have predicted the Credit Crunch problems, to which Roach demurs in saying he was “too early”. He then furthers his modesty in saying that the “breakage” in the financial system was “in excess of anything I envisioned.”
Self-deprecation in assessing one’s predictive abilities will endear anyone to me. Even Roach, who later in the interview burns this hard-won credibility by laying the blame for the credit crunch at the door of regulators, forgetting how hard financial institutions lobbied regulators for greater freedoms in the 1990s.
But I digress. The predictive issues the interview raises are as follows. Issue one: it’s not enough (as any stock short-seller will confirm) to get the direction of a future change right. One must get the timing right too. Issue two: it’s not enough to anticipate a change. One must be able to judge it’s impact. Getting either timing or impact wrong is effectively to have missed the future.
.
Probability
On the latter topic — the problem of impact — Nassim Taleb is unrelenting, and he is right. Analysts routinely mix up probability and impact. They think that because an event has a low probability (‘it would be a 10-sigma event!’) it can be marginalized in the predictive number crunching. Of course, it can’t. The low-probability of a wildcard or black swan event is irrelevant because when it happens it will change the game, and that’s why, in every predictive situation of reasonable complexity and uncertainty, using statistical extrapolations (regressions and so on) to predict, is to dangerously paper over the cracks. It is precisely the cracks that businesses and policy makers need to worry about.
Determining the direction of change is hard enough. Assessing timing or extent of impact — a ‘total future impact index’ — is wickedly difficult. It’s a task not to be underestimated, and to simply extrapolate current trends (= assuming the trend’s timeline and impact stay the same as in the past) is the royal road to underestimating it.
This is the reason foresight for complex, uncertain, changing situations can only be grasped by NOT predicting (quantitatively or otherwise) but by exploring the limit-conditions of the plausible (What would happen if the timing of the change accelerated, or was significantly delayed? What if the impact was 10x or one tenth of what we expect? And so on.)
.
