Machine Learning Methods Used In The Convertible Bond Valuation

Firstly, using the single factor model based on stock price, this paper makes an empirical analysis on the 13 on-the-run convertible bonds and found that the theoretical value calculated by this model is generally higher than the market price, which is consistent with the conclusions of many previous studies. This shows that this pricing model, as well as the market mechanism and the assumption behind it, is not entirely suitable for China’s convertible bond market. Using this single factor model to simulate the price of convertible bonds has obvious errors.Then,based on the inevitable defects of this model, this paper creatively transforms ideas and jumps out of the model of the formula to carry out the valuation of convertible bonds from the perspective of machine learning. For the complex factors affect convertible bond price and cann’t do any assumptions of, almost all aspects of the factors are chosen to analyze by machine learning methods.Compared with several machine learning methods, random forest method is the best method to simulate the price of convertible bonds. The training set’s NMSE of 5-fold cross validation is as low as 0.009, and test set’s NMSE is 0.046. Further valuation of random forest model is significantly better than the single factor model based on stock price for 13 on-the-run convertible bonds above.The effect of fitting and forecasting may be better than any one based on the formula model. Therefore, in the actual valuation of convertible bonds, we can use the random forest regression method to get the corresponding model, and predict the new data to obtain the valuation of convertible bonds.