Prototyping Trading Strategies with Python¶
Getting Started with Python for Algorithmic Trading¶
© Ran Aroussi
@aroussi | aroussi.com | github.com/ranaroussi
February, 2017
About me¶
History¶
- Programming since forever
- Operating in Online / AdTech space since 1997
- Trading since 2013
- Switched to automated trading in 2014
Co-Founder of Native Alpha¶
Trading mostly Futures and Options using automated and semi-automated strategies, with a 100% technical / quantitative approach
Developer of ezIBpy¶
A Pythonic wrapper for the IbPy library, that ease the communication with Interactive Brokers for market data and order execution.
Available via github.com/ranaroussi/ezibpy
Developer of QTPyLib¶
A simple, event-driven, algorithmic trading system written in Python, that supports backtesting and live trading using Interactive Brokers for market data and order execution (QTPyLib stands for: Quantitative Trading Python Library).
Available via github.com/ranaroussi/qtpylib
Agenda¶
- Why Pyhton for Algorithmic Trading?
- The Python Toolbox: Platforms and Libraries for Trading
- Data Wrangling and Working with Market Data
- Using Technical Indicators with Python
- Coming Up with Trading Ideas
- Writing and Testing your first Trading Strategy
- Strategy Development Workflow
* This is the first webinar of a multi webinar series
Why Python?¶
Python's Benifits¶
- A very high-level language with clean and readable syntax
- Fast learning curve
- Object-oriented, multi-paradigm
- A rich standard library of modules
- One of the most used languages in IoT, data science, finance, and web development
- A strong open-source community
"Hello World" in Python¶
In [ ]:
print("hello world")
Compared to C#...¶
In [ ]:
public class Hello
{
public static void Main()
{
System.Console.WriteLine("Hello, World!");
}
}
Why Python for Trading?¶
Python for Algorithmic Trading¶
- Python's syntax is attractive for science, and particularly finance
- Fast learning curve
- Solid Machine Learning libraries
- Quick protopyting, fast enough execution
- Many Algorithmic Trading Libraries and Tools (Zipline, PyAlgoTrade, PyBacktest, QTPyLib)
- Pandas!
The Python Toolbox¶
Recommended Libraries¶
- TA-Lib - technical indicators
- scikit-learn - machine learning algorithms
- Tensorflow - deep learning / neural network
- SciPy - scientific functions
- statsmodels - statistical functions
- SQLAlchemy - relational database abstraction library
- TsTables - tick data storage and retrieval
- ezIBpy - Interactive Brokers bridge
- OandaPy - Oanda bridge
Python 2 or Python 3?¶
Python 3!¶
Data Wrangling and Working with Market Data¶
Pandas - the times series and tabular data wrangler¶
- Pandas is one of the best libraries (in any language, IMO) for data exploration and manipulation
- Main objects are
DataFrameandSeries - Has many available built-in operations
In [3]:
import pandas as pd
import numpy as np
from pandas_datareader import data
spy = data.get_data_yahoo("SPY", start="2000-01-01")
spy.head()
Out[3]:
In [4]:
ratio = spy["Close"] / spy["Adj Close"]
spy["close"] = spy["Adj Close"]
spy["open"] = spy["Open"] / ratio
spy["high"] = spy["High"] / ratio
spy["low"] = spy["Low"] / ratio
spy["volume"] = spy["Volume"]
spy = spy[['open','high','low','close','volume']]
spy.head()
Out[4]:
In [5]:
spy.to_csv('~/Desktop/sp500_ohlc.csv')
In [6]:
spy.shape
Out[6]:
In [7]:
# calculate returns
spy['return'] = spy['close'].pct_change()
spy['return'].describe()
Out[7]:
In [8]:
# slicing
spy[5:10][["close", "return"]]
Out[8]:
In [9]:
spy[ spy['return'] > 0.005 ]['return'].describe()
Out[9]:
In [10]:
spy['close'].plot()
Out[10]:
In [11]:
spy['return'].plot.hist(bins=100, edgecolor='white')
Out[11]:
In [12]:
# resampling
spy['return'].resample("1A").sum() * 100
Out[12]:
Using Technical Indicators with Python¶
In [35]:
# pandas' rolling mean
spy['ma1'] = spy['close'].rolling(window=50).mean()
spy['ma2'] = spy['close'].rolling(window=200).mean()
In [13]:
spy[['ma1', 'ma2', 'close']].plot(linewidth=1)
Out[13]:
In [14]:
# rolling standard deviation
spy['return'].rolling(window=20).std().plot()
Out[14]:
In [15]:
# Calculating log returns and volatility
spy['logret'] = np.log(spy['close'] / spy['close'].shift(1))
spy['volatility'] = spy['logret'].rolling(window=252).std() * np.sqrt(252)
spy[['close', 'volatility']].plot(subplots=True)
Out[15]:
In [17]:
# pure python bollinger bands
spy['sma'] = spy['close'].rolling(window=20).mean()
spy['std'] = spy['close'].rolling(window=20).std()
spy['upperbb'] = spy['sma'] + (spy['std'] * 2)
spy['lowerbb'] = spy['sma'] - (spy['std'] * 2)
plot_candlestick(spy[-100:])
plt.plot(spy[-100:][['upperbb', 'sma', 'lowerbb']], linewidth=1)
Out[17]:
In [36]:
# using TA-Lib..
import talib as ta
spy['rsi'] = ta.RSI(spy['close'].values, timeperiod=2)
spy[-100:][['close', 'rsi']].plot(subplots=True)
Out[36]:
In [19]:
# having some fun :)
data = spy[-100:]
fig = plt.figure()
gs = gridspec.GridSpec(2, 1, height_ratios=[4,1])
ax0 = plt.subplot(gs[0])
plt.plot(data['close'])
ax0.set_ylabel('close')
ax0.fill_between(data.index, data['close'].min(), data['close'], alpha=.25)
ax1 = plt.subplot(gs[1], sharex=ax0)
ax1.set_ylabel('RSI 2')
ax1.plot(data['rsi'], color="navy", linewidth=1)
ax1.axhline(90, color='r', linewidth=1.5)
ax1.axhline(10, color='g', linewidth=1.5)
Out[19]:
Coming Up with Trading Ideas¶
Trading Ideas Sources¶
- Books
- Podcasts
- Forums
- Workshops and Meetups
- SSRN (Social Science Research Network)
- Observations
- Bugs and miscalculations 😁
Strategy Development Workflow¶
My Workflow¶
- Quick proof of concept of a trading idea
- Use a vectoried backtest first for fast testing and dumping ideas
- Move to an event-based backtest for trading ideas that shows potential
- Scrutinize and Optimize
- Forward test
- Live Trading
Vectorized vs. Event Based Backtesting¶
Cons:¶
- Danger of look-ahead bias
- Works of pre-existing, pre-loaded data
- Code usually cannot be used "as-is" for live trading
Cons:¶
- Takes longer to write and backtest
- Not the best choise for prototyping IMO
Basic Strategy Example¶
- BUY ON CLOSE when SPY drops 0.5% (or more)
- SELL ON CLOSE of next day
Vectorized Prototyping¶
In [20]:
portfolio = pd.DataFrame(data={ 'spy': spy['return'] })
portfolio['strategy'] = portfolio[portfolio['spy'].shift(1) <= -0.005]['spy']
In [21]:
portfolio.fillna(0).cumsum().plot()
Out[21]:
Measure Performance with Sharpe Ratio¶
Annualized Sharpe Ratio¶
In [22]:
def sharpe(returns, periods=252, riskfree=0):
returns = returns.dropna()
return np.sqrt(periods) * (np.mean(returns-riskfree)) / np.std(returns)
Initial (and very basic) metrics¶
In [23]:
# benchmark sharpe
sharpe(portfolio['spy'])
Out[23]:
In [24]:
# strategy sharpe
sharpe(portfolio['strategy'])
Out[24]:
In [25]:
# time in market
len(portfolio['strategy'].dropna()) / len(portfolio)
Out[25]:
EOY Returns¶
In [38]:
eoy = portfolio.resample("A").sum()
eoy['diff'] = eoy['strategy']/eoy['spy']
print( np.round(eoy[['spy', 'strategy', 'diff']] * 100, 2) )
Enother Example¶
- GO LONG where 50-day SMA > 200-day SMA
- GO SHORT where 50-day SMA < 200-day SMA
In [41]:
ma_portfolio = spy[['close', 'return']].copy()
ma_portfolio.rename(columns={'return':'spy'}, inplace=True)
In [42]:
ma_portfolio['ma1'] = ma_portfolio['close'].rolling(window=50).mean()
ma_portfolio['ma2'] = ma_portfolio['close'].rolling(window=200).mean()
In [43]:
ma_portfolio[['ma1', 'ma2', 'close']].plot(linewidth=1)
Out[43]:
In [28]:
ma_portfolio['position'] = np.where(
ma_portfolio['ma1'].shift(1) > ma_portfolio['ma2'].shift(1), 1, -1)
ma_portfolio['strategy'] = ma_portfolio['position'] * ma_portfolio['spy']
In [29]:
ma_portfolio[['strategy', 'spy']].cumsum().plot()
Out[29]:
Initial metrics¶
In [30]:
# benchmark sharpe
sharpe(ma_portfolio['spy'])
Out[30]:
In [31]:
# strategy sharpe
sharpe(ma_portfolio['strategy'])
Out[31]:
In [32]:
# time in market
len(ma_portfolio['strategy'].dropna()) / len(ma_portfolio)
Out[32]:
In [40]:
eoy = ma_portfolio.resample("A").sum()
eoy['diff'] = eoy['strategy']/eoy['spy']
print( np.round(eoy[['spy', 'strategy', 'diff']] * 100, 2) )
To Be Continued...¶
This was the first webinar out of 4 webinars we're planning.
Up Next...¶
Prototyping Trading Strategies- Backtesting & Optimization
- Live Trading
- Using Machine Learning in Trading
Q&A Time¶
Prototyping Trading Strategies with Python¶
Thank you for attending!¶
© Ran Aroussi
@aroussi | aroussi.com | github.com/ranaroussi
February, 2017