Stock rise and fall prediction method based on multi-task self-supervised learning

Abstract

The invention provides a stock rise and fall prediction method based on multi-task self-supervised learning, which can automatically learn valuable features from original financial signals and predict the rise and fall of stocks by taking the features as signals. Specifically, the invention realizes a stock technical data sequence encoder based on Transform and an attention mechanism, designs a plurality of self-supervision auxiliary tasks to train the encoder, and encodes stock sequence data by using the trained encoder. Finally, based on the long-short-term memory neural network and the feedforward neural network, sequence representation is learned, and the rise and fall of the stock is predicted. A large number of experiments carried out on a real stock data set show that the features learned by the method are effective signals for predicting the rise and fall of the stock, and the method has leading accuracy in the aspect of predicting the rise and fall of the stock.

Description

technical field [0001] The invention relates to the technical fields of financial quantification and artificial intelligence, in particular to a method for predicting stock rise and fall based on multi-task self-supervised learning. Background technique [0002] Stock trend forecasting is an important task in the financial industry, which has a huge impact on individual investors and the national economy. Using predictive models to predict stock trends has always been people's expectation. During the long research, researchers found that generating representative features is the core of predictive models. However, in practical applications, the characteristics and patterns of stock price fluctuations are highly unstable and diverse. Therefore learning meaningful stock features is challenging for market trend prediction. [0003] Traditional forecasting methods usually use the daily data of stocks as time series signals, and then use classic machine learning methods, such a...

AI Technical Summary

Problems solved by technology

[0005] 2) Artificial features cannot make full use of informative stock signals, which inevitably leads to suboptimal predictions

However, for sequences of financial data, there are no direct labels that can be used to construct auxiliary tasks for self-supervised learning models

Furthermore, it is challenging to incorporate multiple auxiliary tasks in a unified approach to mine complex financial signals

Images