4 ms·
These metrics are totally meaningless. If some metric claims that there is a 50% chance of recession in the next 12 months, I'd expect it to be right roughly ha
by scyclow 7y ago
These metrics are totally meaningless. If some metric claims that there is a 50% chance of recession in the next 12 months, I'd expect it to be right roughly half the time it makes that confident a claim. Similarly, I'd expect a recession one third of the time that the metric says 33%.
The year leading into the Great Recession was mostly around 36%. One year before the 2001 recession it was 24%. And then we have spikes around > 40% in 1999, 2002, 2003, 2006, 2016, and 2018, none of which "paid off".
So how are we supposed to interpret these numbers? That a 40% chance of a recession has a 15% chance of actually predicting a recession?
- throwawaymath 7y ago> If some metric claims that there is a 50% chance of recession in the next 12 months, I'd expect it to be right roughly half the time it makes that confident a claim. Similarly, I'd expect a recession one third of the time that the metric says 33%. What if the forecast is Bayesian rather than frequentist?
- scyclow 7y agoIt's fine if the underlying model is Bayensian, but if you're going to present a time series graph then those past numbers don't mean a whole lot. Why not adjust the past data to be frequentist at that point?