AB
AB
AiBoss

All Categories

project

StockMixer - A stock price prediction framework launched by Shanghai Jiao Tong University

StockMixer is a multilayer perceptron (MLP) architecture developed by Shanghai Jiao Tong University for stock price prediction, possessing both simplicity and powerful predictive capabilities. The architecture is based on three steps: indicator mixing, time mixing, and stock mixing...

What is StockMixer?

StockMixer, developed by Shanghai Jiao Tong University, is a multilayer perceptron (MLP) architecture for stock price prediction, boasting both simplicity and powerful predictive capabilities. The architecture processes and predicts stock data through three steps: indicator mixing, time mixing, and stock mixing, effectively capturing the complex correlations between stock indicators, time, and individual stocks. Time mixing utilizes multi-scale time segment information, while stock mixing leverages the influence of market conditions on individual stocks, achieving more accurate predictions. StockMixer demonstrates superior performance in multiple stock market benchmark tests, outperforming many advanced prediction methods, while reducing memory usage and computational costs.

StockMixer's main functions

  • Indicator MixWe use matrix multiplication and activation functions to simulate the interaction between internal indicators of each stock-time pair, and extract high-level latent features that are informative about future stock trends.
  • Time MixingBased on the information exchange of multi-scale time segments, it captures the time trends and patterns in stock price movements, and supports the model to extract features from different time scales.
  • Stock MixingLearn about stock status from the overall market to individual stocks and back to the overall market, simulate the complex correlations between stocks, and model stock correlations more robustly.
  • Predicting stock pricesBy combining the features obtained from the above mixture, the closing price of the next trading day can be predicted.

StockMixer's technical principles

  • Multilayer Perceptron (MLP) ArchitectureIt is based on MLP, which processes stock data with linear computational complexity and a simple architecture.
  • Information exchangeBased on the MLP structure, information is exchanged between different dimensions (indicators, time, and stocks), enhancing the model's expressiveness.
  • Residual connectivity and layer normalizationResidual connections are used to maintain a balance between input and mixed features, and layer normalization reduces the impact of data offset.
  • Multiscale time segmentsThe time series is segmented into segments of different sizes to capture features at different time scales.
  • The impact of stocks on the market and the market on stocksBased on two MLP structures, we learn the potential stock states and use these states to influence individual stocks, simulating the correlation between stocks.
  • Loss function designBy combining point regression and pairwise ranking perceived loss, the optimized model minimizes the mean squared error between the predicted and actual returns, maintaining the relative order of stocks with high expected returns.

StockMixer's project address

Application scenarios of StockMixer

  • Quantitative InvestmentStockMixer helps quantitative investment strategists predict stock price movements and develop trading strategies to achieve excess returns.
  • Risk ManagementFinancial institutions can better assess and manage market risks in their investment portfolios.
  • Algorithmic tradingIt provides rapid stock price predictions in high-frequency trading, helping algorithmic trading systems make immediate buy and sell decisions.
  • Portfolio optimizationInvestment managers optimize their portfolios by selecting stocks with better expected performance or avoiding potential risks.
  • Market analysis and researchAnalysts analyze market trends, make industry comparisons, and study the impact of specific events on stock prices.
Back