Predicting the Stock Market Volatility Based on Deep Learning and Boosting Tree Methods

Accurate forecasting of volatility is critical for options trading, as option prices are directly related to the volatility of the underlying product. However, industry-leading pricing algorithms never stop evolving. In order to create better trading channels and prices for options, cash stocks, bonds and foreign exchange on numerous exchanges around the world, we need to take volatility forecasting models to the next level. For that purpose, based on the data provided by Optiver, we convert the fine-grained data in the stock trading panel into the feature data. Then we use the K-Means algorithm to cluster the features, and introduce the KNN algorithm to calculate the features at similar times. After the feature engineering is completed, we build a combined machine learning method comprising of LightGBM and neural networks, where RMSPE is the loss function and evaluation metric. Our integrated model reached 0.2176 on the test set and 0.2193 on the forecast dataset for the next three months, which can get a silver medal prize in the Optiver Realized Volatility Prediction Kaggle competition. © 2022 Association for Computing Machinery.