Universal algorithms for well calibrated forecasting

Foster and Vohra (1997) have discovered an unexpected result: it is possible to predict any sequence of outcomes obtaining online, without using any hypotheses about the source generating data. The evaluation of the coherence between forecasts and the data is based on the tests of calibration. In this talk, we present an idea for constructing such randomized algorithms. Some applications of these algorithms in finance, game theory, and and for the recovery of functional dependencies will be presented.