Pair Trading for Cointegrating Currencies

Statistical arbitrage uses various financial statistics to find pricing inefficiencies in mean-reverting trading pairs. This project explores the statistical arbitrage of the Canadian and Australian dollars.
A common type of statistical arbitrage is pair-trading. If two assets are cointegrated, then their price will converge to the mean price in the long term
Pair-trading recognizes deviations in the price of a stock from the mean and either shorts or longs the stock. Once the price reverts to the mean, profit can be made.
The trading signal is a statistical measure that indicates the moments of mean deviation and reversion
Trading model for the next day is predicted using LSTMs and position of the portfolio is determined from hedge ratios

Packages

keras
tensorflow
statsmodels
NumPy

Files

check_johansen_loop.py: checks cointegration with johansen test in loop
check_mean_reversion.py: checks mean reversion of two time series using Augmented Dickey Fuller test
AUDCAD_unequal_mod.py: main file
model.py: Trains LSTM Model and Predicts the trading signal
evaluation.py: evaluation using portfolio metrics and rmspe
seq_norm.py: window normalisation and data prep for LSTM model

Positions

Positions of securities are calculated by using the hedge ratios (hedge ratios determine capital allocation). Positions of two securities on training and test set:

When two cointegrating securities diverge due to market conditions, they tend to converge in few time periods

Usage

Clone the repo

git clone https://github.com/shimonanarang/pair-trading.git

Install the requirements

pip install -r requirements.txt

Navigate and run the following command

python AUDCAD_uneual_mod.py

References

Joao Frois Caldeira and Guilherme Moura. \Selection of a portfolio of Pairs Based on Cointegration: A Statistical Arbitrage Strategy". In: Bras. Financas 11.1 (Mar. 2013)
Lu, Anran, et al. Cluster-Based Statistical Arbitrage Strategy. Stanford University, (June, 2018)