It’s a short line in a short presentation … but it carries a lot of implications:
First, what can we do differently in the future? Self-reflection is the key. It is now clear that some of the assumptions we made with respect to rating U.S. RMBS backed by subprime mortgages were insufficient to stand up to what actually happened. In addition, some have questioned whether our detailed analytical processes led us to wait too long to react to data that suggested a deviation from the expected trends. So we are focusing on getting in place the data, analytics, and processes to enhance our ability to anticipate future trends and process information even more quickly. [emphasis added – JH]
This is a little bit scary. I’ve done a lot of quantitative modelling – my entire professional career has been spent doing quantitative modelling – and I can tell you two things:
- Quantitative models do not do well when there is a trend change. This is because there is a lot more noise than signal in the market-place; a quant system will pick up the first one, two, three standard deviations as an exception that will revert before it changes the figure it takes as a base.
- Ain’t nobody can predict a trend change with reproducible accuracy. At best, you can pick up on the stress on the system implied by your data and assign a probability to the idea that it’s a trend change … e.g., when housing prices decline by 2% in a quarter, there might be a 25% chance that it’s a trend change as opposed to a 75% chance that it’s just noise. Which is not to say that estimating the chances of a change in trend is not useful; but which does mean that assigning a lot of weight to the idea that house prices will continue to decline by 2%/quarter over the medium term is quite aggressive
I will have to see how S&P fleshes out this idea – and how much disclosure they make of their future projections as part of the credit rating process.
With respect to projecting trend changes, lets look at one of the more respective organizations in the business – the National Bureau of Economic Research. How well do they do in determining trend changes? As they say:
On November 26, 2001, the committee determined that the peak of economic activity had occurred in March of that year. For a discussion of the committee’s reasoning and the underlying evidence, see http://www.nber.org/cycles/november2001. The March 2001 peak marked the end of the expansion that began in March 1991, an expansion that lasted exactly 10 years and was the longest in the NBER’s chronology. On July 16, 2003, the committee determined that a trough in economic activity occurred in November 2001. The committee’s announcement of the trough is at http://www.nber.org/cycles/july2003. The trough marks the end of the recession that began in March 2001.
So it took the NBER over a year and a half to look at all the data and determine where the bottom was. And S&P – under pressure by thousands of bozos who could have predicted the credit crunch ever-so-much-better, except that nobody asked them to – is going to try and predict the future?
It’s a scary thought – I hope that the implementation of the plans outlined in the S&P presentation is very, very restrained.