NeuronQuant is a set of tools and an engine meant to let you easily and intuitively build your own automated quantitative trading system. It is designed to let financial, developer and scientist dudes (together sounds great) explore, test and deploy market technical hacks.
While the project is still at an early age, you can already write, or use, buy/sell signal algorithms, and portfolio allocation strategies. Then just plug it in the system and watch it from your dev-console or the web app run on backtest or live-trading mode.
In addition the project propose facilities to build a distributed system and 21st century application (big data, fat computations, d3.js and other html5 stuff), tools to mix languages like Python, node.js and R and a financial library. You will find some goodies like machine learning forecast, markowitz portfolio optimization, genetic optimization, sentiment analysis from twitter, ...
- Debian based distribution: apt-get install git libzmq-dev node npm R python2.7
- git clone https://github.com/Gusabi/ppQuanTrade.git
- ./setup.py install (or run: "pip install -r scripts/requirements.txt")
- R packages, run: "R < ./scripts/install_r_packages.R --no-save"
- Node.js modules, run from [root]/server "npm install"
- Edit [root]/config/local.sh to suit your machine
- Run a mysql database and edit [root]/config/mysql.cfg
- Create the tables with "./neuronquant/data/database.py -c", then populate it with neuronquant/data/database.py -a symbols.csv (see database/QSDATA for a huge list)
- Done !
To run a backtest manually, configure algos.cfg and manager.cfg file, and then run
./backtest.py --tickers google,apple --algorithm DualMA --manager OptimalFrontier --start 2005-01-10 --end 2010-07-03
Or in realtime mode:
./backtest.py --tickers random,6 --algorithm StdBased --manager Equity --delta 2min --live
More examples available [root]/scripts/run_backtest.sh
As mentionned you can easily write your own algorithms. Here is the equity manager example, which allocates the same weight to all of your assets:
from neuronquant.ai.portfolio import PortfolioManager
class Equity(PortfolioManager):
""""
dispatch equal weigths
"""
def optimize(self, date, to_buy, to_sell, parameters):
allocations = dict()
if to_buy:
fraction = round(1.0 / float(len(to_buy)), 2)
for s in to_buy:
allocations[s] = fraction
for s in to_sell:
allocations[s] = - self.portfolio.positions[s].amount
return allocations, 0, 1Strategies triggering buy or sell signals are used within the great zipline backtester engine and therefore use quite the same scheme, plus the manager, and some config parameters. Here a classic momentum strategie:
class Momentum(TradingAlgorithm):
'''
https://www.quantopian.com/posts/this-is-amazing
'''
def initialize(self, properties):
self.max_notional = 100000.1
self.min_notional = -100000.0
self.add_transform(MovingAverage, 'mavg', ['price'], window_length=properties.get('window_length', 3))
def handle_data(self, data):
''' ---------------------------------------------------------- Init --'''
self.manager.update(self.portfolio, self.datetime.to_pydatetime())
signals = dict()
notional = 0
''' ---------------------------------------------------------- Scan --'''
for ticker in data:
sma = data[ticker].mavg.price
price = data[ticker].price
cash = self.portfolio.cash
notional = self.portfolio.positions[ticker].amount * price
capital_used = self.portfolio.capital_used
# notional stuff are portfolio strategies, implement a new one, combinaison => parameters !
if sma > price and notional > -0.2 * (capital_used + cash):
signals[ticker] = - price
elif sma < price and notional < 0.2 * (capital_used + cash):
signals[ticker] = price
''' ---------------------------------------------------------- Orders --'''
if signals:
order_book = self.manager.trade_signals_handler(signals)
for ticker in order_book:
self.order(ticker, order_book[ticker])
if self.debug:
self.logger.info('{}: Ordering {} {} stocks'.format(self.datetime, ticker, order_book[ticker]))
self.logger.info('{}: {} / {}'.format(self.datetime, sma, price))Rememeber that managers and algorithms should be configured in their own *.cfg files or through the webapp.
You can setup your trading environment by running from the root directory:: ./scripts/run_labo.py and access to the shiny frontend page. From there configure the backtest, run it and explore detailed results.
Projects and websites below are awesome works that i heavily use, learn from and want to gratefully thank:
-
r-bloggers http://www.r-bloggers.com/
-
zipline http://github.com/quantopian/zipline and quantopian http://wwww.quantopian.com
-
coursera http://www.coursera.org/
-
udeacity http://www.udacity.com/
-
Babypips http://www.babypips.com/