Apropos of nothing here is a great quote from Willem Buiter (& not just for the allusion to Winnie the Pooh):
"Never mind that anyone providing point forecasts of anything without also offering at least some information about of the rest of the probability distribution of future outcomes (variance, skewness, kurtosis, single-peakedness etc) is either a fool, a knave (or both) or caters to an audience consisting of bears of very little brain."
I wonder is there any chance that some of our local macroeconomic experts ( & the hacks who lap up these forecasts) like to take this on board?
http://blogs.ft.com/maverecon/2009/10/another-quarter-of-negative-gdp-growth-in-the-uk-situation-hopeless-but-not-serious/
5 comments:
Ok Kevin but be somewhat fair. How do you communicate kurtosis to the Six One News in the 30 second slot you have been given? Possible but certainly not trivial. I have never worked as a forecaster but I have been asked to talk about my work in national media. You have some chance on radio and in print media. But once you are on telly (rare for me thankfully) you have about 30-60 seconds to say your bit.
I have enough issues with the first moments of these forecasts. It's impossible to evaluate the validity of a model and its predictions without a discussion of the underlying assumptions and parameterisation. I understand why it’s not possible to debate this on the national airways, but I would like to hear more, perhaps at one of the conferences on economic recovery. Out of curiosity I tried to find concrete details of how HERMES works on the ESRI website, but couldn’t find anything. To be fair to them I presume it’s there somewhere. I’m just surprised that given the current climate, when it’s entirely possible that the nature of some of these macro variables have changed substantially, there isn’t more debate on the assumptions as opposed to the kind of vague discussions we seem to be having. Some sensitivity analysis too please.
I don't think any of us care about kurtosis although I look forward to saying "lepto-kurtic" the next [& probably last] time I am on the telly. But it is possible to say "We think that growth will be 6.5% plus or minus 2%" or "The ESRI says.." as the case may be. The conspiracy theorist in me says that the forecasters might prefer people not to know just how wide the intervals are. I don't know.
In fact when political opinion polls are quoted they routinely refer to sampling error. Why not in this context? Its simply about being honest.
So when the forecasters make their statements they should specify confidence intervals. Its then up to the media whether they want to mention them. The good one's will.
Broadcast media, in any event, is not representative in general as there is much more scope in print media for details like this.
mark - there are papers outlining the assumptions of HERMES. the medium term reviews always include sensitivity analysis.
Thanks Liam, I’ll have a look. I don’t doubt the quality of the forecasts, I only meant that as a non macro person I’d like to hear more specific discussion on this. And by sensitivity analysis I don’t doubt it’s there I was only getting at Kevin’s point that the way the media tend to report a single point estimate can be misleading
Post a Comment