mlp_stock

Stock price prediction using ensemble MLP in PyTorch.

Predict the index changes by the fluctuation of index and volume in the last 5 days.
Train data is the daily CISSM (Compositional Index of Shenzhen Stock Market) from 2005/01 to 2015/06, the test data is from 2015/07 to 2017/05.

Requirements

• Pytorch
• Numpy
• Pandas
• Matplotlib

Usage

1. Run train_net.py to train a group of MLPs with sz_train.csv, saved in /MLPs.
2. Run test_net.py to predict stock market trend (in sz_test.csv) using ensemble MLP.

Result

The train error rate (black) and test error rate (red) of a single MLP, changing with epoches.

The red line is asset sequence if we buy/sell CISSM-ETF according to our ensemble MLP, comparing with CISSM (black).

Tips

• Train samples are limited, using drop-out and early-stop to prevent overfitting.
• Simulated trading using this strategy, see https://xueqiu.com/P/ZH931230 .