Cochrane on Point vs. Density Forecasting

I recently blogged on Manski's call for uncertainty estimates for economic statistics.  Of course we should also acknowledge the uncertainty in economic forecasts (with or without acknowledgment of data uncertainty, and with is better than without).

Some of us have been pushing applied interval and density forecasting for years, and of course Bayesians have been pushing for centuries. The quantitative finance and risk management communities have been largely receptive, whereas macroeconomics has been slower, not withstanding the Bank of England's justly-famous "fan charts."

From a recent post in John Cochrane's Grumpy Economist:

... conditioning decisions on a forecast, cranked out to two decimal places, is a bad idea. Economic policy should embrace uncertainty! ... This is really a big deal. ... All forecasts ... should have error bars. ... Knowing what you don't know is real knowledge.

Here, here!

Surely statistical "error bars" as conventionally calculated are themselves often too tight, as they rely on a host of assumptions and in any event fail to capture unknown and unknowable sources of forecast uncertainty. But they're certainly a step in the right direction.