| Quantitative trading refers to the use of mathematical,computer science and statistical methods,based on massive financial data and algorithm models,to build trading strategies and carry out automated trading operations.Through the support of information technology,quantitative trading can build trading strategies with the help of traditional financial technology,and build strategies and obtain certain excess returns by analyzing the performance of different market states and analyzing past data.The application of quantitative trading can be traced back to the 60 s of the 20 th century,with the rapid development of computer technology and the improvement of data collection technology,quantitative trading began to be popular in the 21 st century.Now,in developed financial market countries,quantitative trading has become the main body of market transactions,and in China’s domestic market,quantitative trading has also received more and more attention from institutions.Before using AI technologies such as machine learning to make quantitative investments,most strategies relied on econometric models such as time series to build.However,due to the non-stationary characteristics of financial markets,most traditional econometric models are not applicable,and their investment effectiveness is significantly low.The development of machine learning has provided a new perspective for the construction of quantitative trading strategies,and due to its high adaptability to non-stationary environments,quantitative trading strategies based on machine learning have begun to emerge.With the continuous development of information technology and artificial intelligence,how to rationally use various models,how to fully explore the performance of machine learning models,and how to form a more stable and accurate strategy when using machine learning and other algorithms to conduct transactions has become the most concerned issue for investors and financial institutions.From the application of machine learning in the field of quantitative trading,the ability of a single model to obtain excess returns in the quantitative field is generally limited,at this time,the stacking model,as an integrated algorithm,can effectively improve the overall return in the quantitative trading process,and is more and more favored by investors.How to improve the performance of the Stacking fusion model and build quantitative trading strategies based on the Stacking fusion model has received more and more attention from investment and academia.This study is based on the Stacking ensemble algorithm in the field of machine learning,which predicts stock prices in multiple industries and constructs trading strategies based on the predicted results.Subsequently,through theoretical analysis,a stock pool was constructed based on the MSCI China A50 index,which includes indicators such as operational,profitability,growth,and technical factors,to ensure that the information contained in stock price data can be fully explored.Secondly,factors that have a significant impact on improving prediction accuracy are selected as model input variables based on tree model screening and IV information values.In order to improve the prediction effect,this article optimizes the Stacking model by adding a basic learner layer on top of the two-layer Stacking model.Based on temporal cross validation and Bayesian optimization,the optimal parameters of each model layer are searched for to achieve the optimal prediction effect.On the basis of the above,this article constructs a short-term trading strategy and uses asset portfolios composed of leading stocks from different industries for daily price data prediction and backtesting.The backtesting results indicate that the model constructed in this article exhibits superior performance in both prediction and trading performance.The research in this paper shows that:(1)When only using BP neural network,random forest and other models to predict and trade a single stock price,it is impossible to completely and accurately predict the fluctuation trend of the price,and the excess return rate is relatively low and the rollback control is not ideal.In terms of prediction performance,the relevant index values such as precision and recall are low.(3)From the perspective of the constructed trading strategy,the optimized Stacking model constructed the intraday trading strategy with an average compound annualized return of 29.43% and a maximum pullback rate of 32.04% during the ten-year test,all of which performed well,indicating the feasibility of the strategy.(4)From the comparison between single variety and multi variety combinations,the multi variety combination can reduce the risk of the strategy and improve its profitability.The research in this article further explores the application of machine learning in the field of quantitative trading strategies,providing new references for various investors in market trading,and thus promoting the development of financial markets. |
