Abstract: Considering the high volatility and skewness of real sales series, the latest forecasting method in statistics, exponential weighted quantile regression, was applied to get higher forecast accuracy. The cost model for supply chain system based on CPFR was built, which includes the retailer's cost, the manufacturer's cost and the supply chain total cost. This model forecasts directly the quantile of the sales series, which not only avoids the forecast mistakes based on hypothesis of research at present, but also makes the forecast results approach the real results of the demand model. The numerical analysis illustrated that quantile regression forecast is better than traditional methods in the demand forecast by giving real examples.