Wednesday, 5 March 2014

What do we want? Inductive models! When do we want them? When we have an evidence base!

After a researchers' discussion, I usually adjourn to the pub with others from the group.  Following the last meeting however, I rushed off to a talk by the Executive Director of Financial Stability at the Bank of England.

I sat next to a Chinese PhD student who introduced himself and asked my name.  "Andrew" I said. He looked at the ticket he was holding.  "Ah!" he replied, "The same as the speaker.  An important name."

We had come to hear Andrew Haldane in conversation.  The academic conversing with him was introduced as Professor Andrew Gamble.  "No" I observed, "just common".

Andrew Haldane may have a commonplace name, but for someone embedded at the core of the establishment, his views are not commonplace.  He has attracted attention in the past for speaking in favour of the Occupy Movement.  In the course of his conversation with Professor Gamble, he criticized economists' over-reliance on deductive models and expressed the view that greater efforts should be made to develop more pragmatic inductive models.  

Amongst the questions asked by members of the audience was one from an economics student at Sheffield University.  She bemoaned the fact that she was learning the type of model that Haldane had criticized.  What, she asked, should she and her fellow students do about it?  "Protest!" was his succinct response.

Haldane’s objection to many of the standard economic models was that they were devised to evaluate risk rather than to address uncertainties.  The difference, explained Prof Gamble, was the difference between known unknowns and unknown unknowns.  Risk analysis identifies a spectrum of outcomes and assigns probabilities to each, then uses those probabilities to help prioritise actions. So for example, if an unbiased coin is tossed, probabilities can be assigned to whether it will, on coming to earth, be seen to display heads or tails.  If it falls down a crack in the floor, those calculations become invalid.  

Although the probability of a coin disappearing in this way is small, there are numerous other improbable things that could happen to the coin, each of which could affect knowledge of whether it displays heads or tails (or lands on its edge!).  There is a high probability therefore, that at least one improbable event will occur, with an impact that has not been considered in the risk analysis.

Because deductive models tend to focus on known unknowns, these get over-emphasised.  By contrast, pragmatic models based on rules of thumb, implicitly adjust for the accumulated improbables.

No comments:

Post a Comment