Stock index trend prediction method based on Laplace operator

Abstract

The invention discloses a stock index trend prediction method based on a Laplace operator. The stock index trend prediction method comprises the steps that 1, intra-day MACD is calculated; step 2, calculating RSI (Received Signal Interference) within a day; 3, the intra-day RCR is calculated; step 4, splicing into a high-frequency factor matrix according to the intra-day MACD, the intra-day RSI and the intra-day RCR sequences; 5, performing Laplace operator operation on the high-frequency factor matrix to obtain a Laplace matrix; step 6, straightening the Laplace matrix into a factor vector; and step 7, constructing a prediction model. Matrix transformation is carried out on stock index factors to form matrix representation of conventional indexes, a Laplace operator is used for carrying out second-order difference operation on a factor matrix, the difference characteristics of the factors are achieved while the correlation of the stock index factors is mined, the method is used for deeply mining the correlation and time sequence characteristics of the factors, and the performance of the overall prediction model is improved.

Description

technical field [0001] The invention relates to a stock index trend prediction method based on a Laplacian operator, and belongs to the technical field of computer technology and intelligent prediction. Background technique [0002] At present, the existing mainstream stock index trend prediction methods mainly include methods based on trend extrapolation, methods based on time series analysis, methods based on machine learning and methods based on deep learning: [0003] (1) Trend extrapolation method: This method performs trend fitting based on the sequence trend of a period of time before the forecast point. Commonly used trend extrapolation models include linear models, exponential models, etc. [0004] (2) Time series analysis method: This method is a classic statistical forecasting method, and the representative one is the ARIMA model. The general steps are to first test the model for stationarity, then determine the order of the model, estimate the parameters, and fi...

AI Technical Summary

Problems solved by technology

[0008] (1) Due to the small amount of information, the trend extrapolation method is effective for relatively stable trends in the near future, and the prediction accuracy is poor in most cases

[0009] (2) The forecasting method based on time series must require the series to satisfy stationarity, and in essence, this method can only capture linear relationships, but not nonlinear relationships

[0010] (3) The method based on machine learning is essentially to establish a nonlinear mapping relationship between high-frequency factors and trend ups and downs, which greatly improves the accuracy of predictions, but requires a large number of factor design, selection and testing to ensure certain In addition, the shallow model is weak in the nonlinear fitting ability of the sequence, etc.

[0011] (4) The prediction method based on the deep learning model has the biggest advantage over the method based on machine learning in the "memory" ability of the time series pattern and the ability of the deep network to extract features, but due to the complexity of the parameters and the interpretability of the model Poor performance and complex model maintenance

Images