Consider a finite-state stochastic process governed by an unknown objective
probability distribution. Observing the system, a forecaster assigns subje
ctive probabilities to future states. The resulting subjective forecast mer
ges to the objective distribution if, with time, the forecasted probabiliti
es converge to the correct (but unknown) probabilities. The forecast is cal
ibrated if observed long-run empirical distributions coincide with the fore
casted probabilities. This paper links unobserved reliability of forecasts
to their observed empirical performance by demonstrating full equivalence b
etween notions of merging and of calibration, and discusses implications of
this equivalence for the literature of forecasting and learning. Journal o
f Economic Literature Classification Numbers: C5, C11, C73, D83. (C) 1999 A
cademic Press.