The Elo model does not reset each year and so does not suffer from the same sort of early season data deficiency issues as season efficiency margins and RPI. The downside of this is that the model tends to lag the Adj EM and RPI models which quickly reflect changes in coaching or team personnel. The Elo model is used extensively at fivethirtyeight for their NFL and NBA predictive models. This Elo model has the following parameters: Starting Elo=1500, K=20, Home Court Advantage = 100; At the beginning of each new season, starting Elo is 0.75 previous season Elo + 0.25 Mean Elo (1505).