This article defends economic forecasting against the critique of radical uncertainty, which is also called Knightian uncertainty. I grant such uncertainty exists but I argue it simply requires minor compensation from standard analytical methods.
Here are some criticisms against standard equilibrium analysis:
- Radical uncertainty
- Radical subjectivism
Here are my defenses:
- Categorical Certainty + Confidence Adjustment.
- Radical subjectivism is rejected as logically incoherent and empirically invalid.
- Irrationality as error exists, but rejection of the market in favor of public policy is a non-sequitor.
This article focuses on defense #1. Others have been discussed elsewhere on the site. Categorical certainty holds that it is possible to construct a set of categories such that no future event will occur outside of the predefined set. In other words, it is the claim that uncertainty in kind, or quality, does not exist. This is similar to but stronger than my previous claim, contra Mises, that case probability is feasible.
Two techniques to construct a collectively exhaustive set of categories include:
- In addition to the named categories, add an “other” category.
- Given category A, construct category not A.
Uncertainty in quantity is admitted as a stylized fact, but there are simple mathematical techniques which can be used to compensate. See this paper, in particular section 3. Tl;dr is below:
- Suppose you have a statistical point estimate which is subject to uncertainty. (Note: all point forecasts are subject to such uncertainty.)
- Given that it is subject to genuine uncertainty, the true value could be more or less than the supposed value. So the expected effect of uncertainty is 0.
- While the point effect change is 0, uncertainty does have the effect of reducing our confidence.
- By how much? By some amount between 0 and 100%. In other words, we should be between 0 and 100% confident in our statistical calculation which omits Knightian considerations.
- Given that X is some value between 0 and 1, the optimal guess is .5. Such a calculation maximizes expected value, minimizes risk, and so on.
- The simple result is that we simply multiply the confidence we have in our statistical estimate by .5 in order to include Knightian risk.