, Volume 32, Issue 1, pp 151-159

The reproducible properties of correct forecasts

Rent the article at a discount

Rent now

* Final gross prices may vary according to local VAT.

Get Access

Abstract.

Each period, one outcome out of finitely many possibilities is observed. Each period, a forecaster announces some probability for the future outcomes based on the available data. An outsider wants to know if the forecaster has some knowledge of the data generating process. Let a test be an arbitrary function from sequences of forecasts and outcomes to {0,1}. When the test returns a 0 the test is said to reject the forecasts based on the outcome sequence. When the test resturns a 1 the test is said to not reject the forecasts based on the outcome sequence. Consider any test that does not reject the truth, i.e. it does not reject when the announced forecasts are the conditional probabilities of the data generating process. Based on Fan’s (1953) Minimax theorem, I show that it is possible to produce forecasts that will not be rejected on any sequence of outcomes.

Journal of Economic Literature Classification Number:D83 Learning
I thankfully acknowledge financial support from the National Science Foundation grant SES 0109650. I also thank a referee for useful comments. All errors are mine.