Tuesday, April 19, 2011

Challenges with predicting the future

You may have heard that at last week's Masters Golf Tournament in Augusta, Georgia, young (not quite 22 years of age) Rory McIlroy was leading by four strokes after the third round, with just Sunday to play. The golf pundits expressed great confidence that Rory would achieve victory given his stellar play the first three days, and that no one was likely to be able to catch him. Sadly, this wasn't to happen as he finished Sunday with a rather poor 80. And from out of nowhere, South African Charl Schwartzel won the tournament. My point? With just one day predictions were far from accurate, even when given by the most knowledgeable of sports people.

When a doctor tells a woman that her "due date" is October 31, how likely is it that he'll be correct? Well below 50% it seems: Wikipedia, in an x-rated post, reports that it's below five percent. Mike Brown in How I Killed Pluto and Why It Had It Coming explained how, when his wife was pregnant with their first child, he wondered what percentage of babies were born before versus after the date; how many were born within a day, a week, etc. of the date. He attempted to develop statistics on this. No doubt such stats would be helpful in providing some degree of confidence as to when a baby will, in fact, arrive.

I spoke last Tuesday at a DST Global event in Singapore, where I delivered a talk titled "Performance Measurement: One Size Doesn't Fit All." Later I sat on a panel with Nick Wade of Northfield and an old friend, Trevor Persuad of Russell. The topic of ex ante risk came up, and I commented how difficult it is to make any accurate predictions, thus the need to couch such statistics with a declaration that qualifies what the number(s) represents and the assumptions underlying it.

It's necessary in many cases to offer predictions; it just important to understand what their basis is and potential for being correct.

p.s., As an aside, Trevor mentioned that you can't manage risk with Value at Risk, and I thought this was an insightful comment. Recall that VaR is based on a portfolio's history: how does one adjust their history? This is perhaps another topic that is worthy of some discussion.

1 comment:

  1. Two different but both very good perspectives

    An excellent paper by Carol Alexander on Multi-Asset Market Risk Analysis: UK Gilts and the Banking Crisis argues that "Extreme losses during the crisis could have been predicted using (historical) data from the 1970’s."



    Aaron Brown's excellent article on Risk versus Portfolio Management in which he argues

    "Every year, a few thousand people discover that VaR is not a measure of risk. They always seem to think they are the first people to notice that. You don’t get good portfolios by maximizing expected return subject to a VaR constraint"


    For what its worth I'm somewhere in-between but that's because I am a portfolio manager first and risk manager second.


Note: Only a member of this blog may post a comment.