The procedure by which a prediction is made is called a "model." My job is to build models. I'd like to share a bit of what I know in the context of the Soros vs Buffett thread.
Over many generations, university statistics departments, economics departments and business schools have instilled the belief called "frequentism" in their students. Frequentism is the belief that the probability a real object will be found in a given state of nature is the limiting relative frequency of this state. The "limiting relative frequency" of a state is the proportion of occurrences of this state in the limit of an infinite number of observations.
According to people who have studied the matter, most practicing scientists subscribe to this belief; as frequentism was drilled into them by their professors, this is no surprise. Frequentism, however, is a trap, for it neglects the phenonenon which communications engineers call "noise." Because of this neglect, when a model is built under frequentism, it exhibits the phenomenon which I call "regression to the base-rates." This is that, when the model is tested, the observed relative frequencies lie closer to the base rates than the predicted probabilities. The prediction contains less information than the frequentist believes it to contain and may contain no information at all.
Building a model that reliably predicts outcomes in financial markets is difficult and sometimes impossible. In lieu of a reliable model, one strategy is to gamble on regression to the base-rates. This, roughly speaking, is the strategy called "value investing." Because of the pervasiveness of frequentism in peoples' thinking, value investing has been a fruitful strategy.
Investors exhibit belief in frequentism when they confuse signal with noise. For example, there is a 1 quarter drop in earnings and they take this to be a signal to sell. The value investor takes this "signal" to be noise and (if the price is right) buys.
Judged by their actions, Buffett gambles on the reliability of regression to the base-rates. Soros gambles on the reliability of predictive models. That Soros has been so successful suggests he has somehow evaded the trap of frequentism, thus being able to distinguish signal from noise.