AI Leadership Summit 2025 Highlights
Back to all news articles
28/06/2018
Dr Gruen said the general consensus is that before the financial crisis, forecasting wasn't very good at telling us what we want to know, which is where are the turning points? Where are the risks?
“We're sort of making some progress now,” he said.
“The sort of forecast that you need to make if you want to find out who can forecast and who can't forecast, (is that) there is an x per cent chance of this happening because that's the only way you can then compare a forecast with what happened.
“The forecast…should be in some kind of rnge and it should be in a probabilistic form.”
However, he said it was important to distinguish this type of prediction competition from a tipping competition.
“A tipping competition is one of the reasons why we get so poor forecasting of recessions because recessions are rare and given our state of ignorance it's never in your interests if you want to get a good reputation for predicting what happened to predict a recession,” he said.
“That's actually one of the incentive mechanisms that's behind some of those numbers, there are plenty of other things, so you need people to be able to say I think that the chances of a recession have gone from per cent to per cent.
“Governments should host these kinds of prediction competitions; the private sector could do it.
“Anyone can host prediction competitions it's not that expensive now.
“That's something I think would be very worthwhile, how can we surface much more real-time data.”
In addition, he said there is no reason why the models that governments use to make forecasts shouldn’t be released publicly.
“There should be other possibilities for other people to use the same model and do different things,” he said.
“That would enrich the discussion and it would enrich the model because just as is the case with open source software, various people would build modules onto those models and improve them.
“I don't know any political party that has that policy but it's a good policy.”
“We're sort of making some progress now,” he said.
“The sort of forecast that you need to make if you want to find out who can forecast and who can't forecast, (is that) there is an x per cent chance of this happening because that's the only way you can then compare a forecast with what happened.
“The forecast…should be in some kind of rnge and it should be in a probabilistic form.”
However, he said it was important to distinguish this type of prediction competition from a tipping competition.
“A tipping competition is one of the reasons why we get so poor forecasting of recessions because recessions are rare and given our state of ignorance it's never in your interests if you want to get a good reputation for predicting what happened to predict a recession,” he said.
“That's actually one of the incentive mechanisms that's behind some of those numbers, there are plenty of other things, so you need people to be able to say I think that the chances of a recession have gone from per cent to per cent.
“Governments should host these kinds of prediction competitions; the private sector could do it.
“Anyone can host prediction competitions it's not that expensive now.
“That's something I think would be very worthwhile, how can we surface much more real-time data.”
In addition, he said there is no reason why the models that governments use to make forecasts shouldn’t be released publicly.
“There should be other possibilities for other people to use the same model and do different things,” he said.
“That would enrich the discussion and it would enrich the model because just as is the case with open source software, various people would build modules onto those models and improve them.
“I don't know any political party that has that policy but it's a good policy.”