Fintech Stock Index Forecasting Study Based On LSTM Combinatorial Network Transfer Model

Since the late last century,the financial industry has started to develop healthily under the double impulse of reform and innovation.In the new century,especially in the last decade or so,the so-called "ABCD" cutting-edge high technology has swept the world from weakness to development,giving rise to the development of Fintech,bringing power and vitality to the real economy and people’s lives.If there is a market,there is investment,and the Fintech windfall has attracted many market investors.Fintech stock indices and stock index fund derivatives prepared by trading institutions enter the market,and risk warning and related market forecast become important.The current research on Fintech stock indices or similar small-scale thematic stock indices is rare,and their special features may make the use of traditional model algorithms limited.Drawing on existing research on stock market data prediction,this paper constructs a deep learning LSTM combined network structure and its transfer model for fitting and predictions,introducing a CNN structural layer to combine with the LSTM.The CNN structure consists of a convolutional-pooling unit,and the LSTM structure consists of a combination of two LSTM units and the fully connected layers.For the nonlinear and complex features of Fintech stock index data,the CNN structure serves to extract features more accurately and efficiently and share parameters to reduce calculation amount.Meanwhile the special threshold setting in the LSTM on this basis enables the model to selectively learn long-time memory from sequential data,and the features learned by the combined network model are more accurate based on the feature transfer of the CSI 300 data.In addition,the single machine learning model algorithm SVR and the integrated machine learning model algorithm XGBoost were also selected separately as controls for prediction comparison.In this paper,six technical indicators of the Fintech index,which are called" opening,closing,highest,lowest,dollar volume and trading volume",are selected as features to fit and predict closing prices.Firstly,the overall statistical descriptive analysis of Fintech stock index data is conducted.The smoothness ADF test finds that the closing price series is a non-smooth series,and the skewness value,kurtosis value and kernel density estimation show that the Fintech index data series is basically a single-peaked distribution slightly skewed to right.In the model construction,for SVR and XGBoost,grid search and Optuna framework are used for tuning according to the different ranges of free parameters to be selected,respectively.A trial application of CNN layer is performed in the LSTM combinatorial network model design based on Adam optimizer,and the similarity between the CSI 300 stock index and the data in this paper is analyzed before the transfer.By calculating the absolute and relative errors of the predictions,the final short-term and medium-long-term results show that the LSTM combinatorial network transfer model>LSTM combinatorial network model>XGBoost>SVR.The LSTM combinatorial network transfer model fits best and significantly outperforms the two traditional machine learning models,SVR and XGBoost,while the transfer effect compared with the original LSTM combinatorial network model is also significant.In this paper,we integrate the deep network model with transfer learning idea,and propose that the combinatorial network transfer model combining CNN and LSTM is superior in theory and practical application,and complete the applicability comparison with other methods.To a certain extent,it complements the research dimensions and comparative ideas of thematic stock index forecasting,provides possible guiding directions and certain reference values for market participants of Fintech thematic stock indices,and also provides industry development information for management agencies and departments of Fintech industry to help improve China’s stock index market and the development of Fintech.