Stock Prediction Project

Plan

• Baselines

  • LSTM
  • GRU
  • ResNLS

• Hyperparameter tuning

• Increase layers

• Increase features

• if ton: change dataset

  • Stock Mixer

• New ideas

  • Apply Transformers

• Implement ideas

  • TBD

• Experiment

  • Try different features and analysis

• Result analysis

  • Stock Mixer
  • ...baselines
  • ...new ideas

To-Do List

Progress 1 Simple Baseline

• Setup experiments (data, metrics, tools)
  • choose metrics
• Run baselines
  • write more metrics (3)
  • LSTM GRU (S&P500)
  • write model ResNLS (1)

Problem Setup

Dataset

	S&P500
Start Time	16-01-04
End Time	22-05-25
Train Days	1006
Val Days	253
Test Days	352

Metric

• Information Coeffcient (IC)
• Rank Information Coeffcient (RIC)
• Precision@N (prec@N)
• Sharpe Ratio (SR)

Method

Simple Baseline

$ X \in \mathbb{R}^{T \times F} \to p \in \mathbb{R} \to r \in \mathbb{R} $

$ T = 16, F = 5 $

$ r^{t} = \frac{p^{t} - p^{t-1}}{p^{t-1}}$

$ L = L_\text{MSE} + \alpha \sum_{i=1}^{N} \sum_{j=1}^{N} \max\left(0, -(\hat{r}{{t}{i}} - \hat{r}{{{t}{j}}})({r}{{t}{i}} - {r}{{{t}{j}}})\right) $

Result

Model	Metric 1	Metric 2	Metric 3

Reference

• StockMixer
• ResNLS
• MASTER

Acknowledgement

• https://github.com/Waterkin/stock-top-papers - for providing the list of top papers in stock prediction

Future Work

• MoE Timeseries