Abstract: Aiming at the normal Gaussian distributional noise, greater breadth noise and oddity point noise of product sales series and combing a designed robust loss function with wavelet kernel function, we propose a new wavelet ν-support vector machine, named as robust wavelet ν-support vector machine (RWν-SVM). The RWν-SVM, which has a stronger robustness and simpler dual optimization problem than standard wavelet-support vector machine (Wν-SVM), can inhibit some types of noise and disturbing oddity point noise of product sales series effectively. Finally, the RWν-SVM is applied to the forecasts of car sales, and the results show that the forecasting model based on the proposed RWν-SVM is effective and feasible.