Every month, on the first Friday, the US Bureau of Labour Statistics announces the Non-Farm Payrolls number. This is widely seen as an important macroeconomic datum for the US economy, and large surprises in it causes swings in nearly every financial market on the planet. It is considered to be difficult to forecast accurately, despite it having seasonal variations, and so is fair game in nearly every financial institution for a sweepstakes.
We are no different at my workplace. The sweepstakes is simple: whoever chooses a number that's closest to the actual as reported by the Bureau wins the pot. When I began participating a couple of years ago, the entry fee was £1 with the winner collecting it all. Last month, the boss, claiming that in these difficult times of credit crunch we should strive to make someone very happy, urged us to raise the stakes to £5. Amazingly, people agreed. He then proceeded to win handily. It turns out he ran some statistical regression or the other, and his guess came surprisingly close to the actual number.
Now, over the past several months, payroll numbers have been plummeting in the US. It's not too surprising, then, that fitting a trend line should be a reasonably accurate forecast. Previously, I had generally chosen randomly - as far away from the average opinion as possible, so that if the result were an outlier, that is, far from the average, I'd claim the pot. Never worked, dash it. This month, I briefly considered getting fancy with my forecast. I tried this model (ARIMA) and that (linear regression). Both of them gave me numbers that were not too different from the consensus of 79 economists polled by Reuters. What was the point of using these models, then? The last time I had chosen the Reuters consensus, I had been so far away from the announced number that it was embarrassing even to admit I had participated in the sweepstakes. But perhaps now my luck was turning? After all, only two days earlier, I had won £10 on the National Lottery. (Woo-hoo. 10x returns! Take that, financial practitioners!) So I chose a number (-675,000) close but not too close to the consensus (-648,000). The rest of the punters were all over the place; some predicting a drop in employment of a million even.
At this point, I can modestly admit that I won. About bloody time, too, I say. The number came in at -651,000. Kudos and backslaps (and suspicions of unfair advantage by means of statistical tests) came my way. I counted my gains. Fifty quid, folks. Fifty freakin' quid.
[In unrelated news, the wife's just announced that she is going shopping.]
0 comments:
Post a Comment