Research on dynamic adjustment and risk management of quantitative portfolio optimization and high-frequency trading strategies based on machine learning

In this paper, the dynamic adjustment and risk management of QFT and HFT are researched. First of all, this paper presents a combination of reinforcement learning and deep learning. Through the introduction of Transformer model, we can make dynamic adjustment of the portfolio. Experiments indicate that this algorithm can raise the return of the portfolio by 20% and the Sharpe ratio by 15%. Secondly, we design a high frequency trading strategy based on GAN, and combine reinforcement learning to adjust the strategy dynamically. In the experiment, the successful rate of transaction is 30% higher, and the average profit is 25% higher. From the point of view of risk management, risk assessment is carried out with VaR and CVaR, and the risk is predicted by machine learning models, with a 10% reduction in VaR and a 12% reduction in CVaR. This study provides a new way of thinking for quantitative investment and high frequency trading. It is significant both in theory and practice.