rsyed0/mquery

Algotrading toolkit using customizable strategies, genetic algorithms, and RNN-based strategies

mquery

Setup:

python3 -m pip install pyalgotrade

python3 -m pip install keras

python3 -m pip install yfinance

• Requires Python 3 installation, not tested with Python 2
• Alternatively can use pip3

Existing features:

• Wrapper for fetching yfinance ticker history (yfinance_csv.py)
• Implementation of several trading indicators (strategies.py)
• Implementation of genetic algorithm to find optimal weighting of these indicators (pat_papertrade.py)
  • Snapshots of this algorithm's optimal model for certain tickers and timeframes (highlights/, recents/)
• Implementation of recurrent neural network (rnn_algotrade.py)
  • Works best with sectors that display repeated patterns (e.g. energy)

Planned features:

• Usage of quanttrader API to use live data/perform live trading
• Implementation of higher leverage trading strategies like calls/puts
• Hierarchical forecasting for stocks grouped by sector
• Web/front-end interface for stock analysis using Flask or Django
• Segmented RNN approach (volatility and bullishness), different normalizations

Usage:

python3 yfinance_csv.py [symbol] [period=1y] [interval=1d]

python3 yfinance_csv.py * [symbols...] [period] [interval]

• Fetches stock price of symbol on period at given interval using yfinance API
• Cleans data of possible issues (high < close/open, low > close/open)
• Can also fetch multiple symbols using * token

python3 single_strat.py [symbol] [period=1y] [interval=1d]

• Weighted-average of several indicators, based on weights/indicators set in code
• Generates line plot to show time-series exposure and performance of algorithm on period
• Useful for testing performance of single classifier, especially when creating custom

python3 pat_papertrade.py [symbol] [period=1y] [interval=1d]

• Weighted-average of several indicators, tuned using genetic reinforcement learning algorithm
• Generates line plot to show time-series exposure and performance of algorithm on period
• Can be customized by adding strategies to strategies.py
• Can use simple weighted-indicator model or a neural network model

python3 rnn_algotrade.py [symbol] [train_start] [train_end] [test_start] [test_end]

• LSTM RNN model trained to use indicator outputs as input
• Generates line plot to show time-series exposure and performance of algorithm on period
• Aiming to predict weighted future performance of stock on next 25 trading periods

python3 screener.py [period] [interval]

• Screen all stocks in local directory with given period and interval
• Uses genetic algorithm to determine bullishness at end of period
• Generates two scatter plots (exposure vs buy/sell, performance vs bullishness)
• Allows user to examine line plot and weights for each stock/classifier performance
• Saves each classifier in a file called screener_[mmddyy].txt

python3 multiple_series.py [symbols...] [period=1y] [interval=1d]

• Weighted-average of several indicators over several symbols, tuned using genetic algorithm
• Will accumulate shares of multiple symbols within same portfolio to maximize returns
• Generates line plot to show time-series exposure and performance of algorithm on period
• Can be customized by adding strategies to strategies.py