Saturday, January 24, 2009

For real-time forecasting

The surprise November industrial production, current account and December capacity utilization figures, all released in a span of four days, led many market economists to sharply revise their growth and current account forecasts. In fact, in the week starting with January 12, when capacity utilization and current account were released, I was getting on average two to three "we revise our forecasts" emails per day.

I am not saying that forecasters should stick to forecasts forever; in fact, updating as new information becomes available is the logical way to go. But what makes the story interesting from my perspective is that I, having lived with a more gloomy scenario that I had repeatedly mentioned in my Hurriyet Daily News columns since mid-December, saw this as the economists suddenly waking up from Wonderland and facing the grim realities of Turkey.

In fact, if one digs deeper, the huge revisions beg one to question the models used in these forecasts. The IP and utilization figures were really dismal, but were they really dismal enough to lead forecasters to revise their 2009 growth forecasts from 1.5% to -1.0%? Whatever model/framework you use, two monthly figures can not justify such large revisions.

Having been there done that, I can personally testify that more often than not, there is not much beef behind such forecasting models. The issue is usually one of resources, as market economists do not have the luxury of academics to spend days or even weeks on a forecasting exercise. And even if they did and came up with the perfect model, past performance does not guarantee future performance, and the model is likely to be useless soon.

Of course, there are other factors at play as well. For one thing, in the world of forecasting, there is definitely incentive for herding, as I explained in a previous blog. Besides, it really does not make sense to continuously revise forecasts with every new data available. But I still feel that market economists would definitely benefit from having a real-time forecasting framework (like the one in a recent ECB paper) sit around for sporadic use at least.

No comments: