• Constructed factors from the exchange's quotation information of the convertible bond market, selected 150
factors based on the correlation between factors and return rates in Python
• Trained, built LightGBM machine learning and LSTM deep learning models based on cleaned and
standardized features in Python with TensorFlow framework to predict return rates, achieving information
coefficients at 12%
• Performed cross validation and pnf back testing and fine-tuned hyperparameters to further improve model’s
information coefficients to 15%