In most sports, a game ends with a win, a draw or a loss. Before a given game, one can compile past data by picking similar games, whose results are known. If, in this sample, w% are wins, d% a draws and l% are losses, then the probability of a win in the future game is w/100 (or w%).
The question is: what to include in this sample? All historical confrontations between both teams?
The issue is similar to the one found when conducting a relative valuation of a stock. Which comparables should we use?