First Trading Algorithm¶
Pairs Trading¶
Pairs trading is a strategy that uses two stocks that are highly correlated. We can then use the difference in price between the two stocks as signal if one moves out of correlation with the other. It is an older strategy that is used classically as a guide to beginning algorithmic trading. There is a fantastic full guide and write up on Investopedia you can find here! I highly recommend reading the article in full before continuing, it is entertaining and informative!
Let's create our first basic trading algorithm! This is an exercise in using quantopian, NOT a realistic representation of what a good algorithm is! Never use something as simple as this in the real world! This is an extremely simplified version of Pairs Trading, we won't be considering factors such as cointegration!
Part 1: Research¶
You can perform research process in your local environment or via the Quantopian platform. Codes included below are for running under local environment ONLY. These codes are NOT to be used in Quantopian platform! Please refer to the online notebook "Pairs Trading Research" in Quantopian platform
Imports¶
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
import quandl
Get Data from Quandl¶
Use get_pricing() in Quantopian platform.
United Airlines and American Airlines
start = '07-01-2015'
end = '07-01-2017'
united = quandl.get('WIKI/UAL',start_date=start,end_date=end)
american = quandl.get('WIKI/AAL',start_date=start,end_date=end)
united.head()
american.head()
american['Adj. Close'].plot(label='American Airlines',figsize=(12,8))
united['Adj. Close'].plot(label='United Airlines')
plt.legend()
It looks like their behaviors are highly correlated to each other!
Spread and Correlation¶
np.corrcoef(american['Adj. Close'],united['Adj. Close'])
We're going to assume that because they're highly correlated, any significant difference in the spread of their prices may be a trading opportunity.
Use .axhline() to draw a horizontal line
spread = american['Adj. Close'] - united['Adj. Close']
spread.plot(label='Spread',figsize=(12,8))
plt.axhline(spread.mean(),c='r')
plt.legend()
Normalizing with a z-score¶
def zscore(stocks):
return (stocks - stocks.mean()) / np.std(stocks)
zscore(spread).plot(figsize=(14,8))
plt.axhline(zscore(spread).mean(), color='black')
plt.axhline(1.0, c='r', ls='--')
plt.axhline(-1.0, c='g', ls='--')
plt.legend(['Spread z-score', 'Mean', '+1', '-1']);
If there is ever a dip either below -1 or some other arbitrary value below the mean or there is ever a peak above +1 or some other arbitrary value (e.g. 1.5), We expect there will be a revision to the mean and eventually the spread will get back down. The reason we think that is because in the past, these two stocks are highly correlated, which means the actual spread of value when we normalize it shouldn't be that big of a difference.
Rolling Z-Score¶
Our spread is currently American-United. Let's decide how to calculate this on a rolling basis for our use in Quantopian
#1 day moving average of the price spread
spread_mavg1 = spread.rolling(1).mean()
# 30 day moving average of the price spread
spread_mavg30 = spread.rolling(30).mean()
# Take a rolling 30 day standard deviation
std_30 = spread.rolling(30).std()
# Compute the z score for each day
zscore_30_1 = (spread_mavg1 - spread_mavg30)/std_30
zscore_30_1.plot(figsize=(12,8),label='Rolling 30 day Z score')
plt.axhline(0, color='black')
plt.axhline(1.0, color='red', linestyle='--');
Part 2: Implementation of Strategy¶
WARNING: YOU SHOULD NOT ACTUALLY TRADE WITH THIS!
The part applies to Quantopian platform ONLY!
There are two ways to get to the development environment:
- Click on
Qat the upper left corner and then clickStart Coding - Click on
Researchand then clickAlgorithms
import numpy as np
def initialize(context):
"""
Called once at the start of the algorithm.
"""
# Every day we check the pair status
schedule_function(check_pairs, date_rules.every_day(), time_rules.market_close(minutes=60))
# Our Two Airlines
context.aa = sid(45971) #aal
context.ual = sid(28051) #ual
# Flags to tell us if we're currently in a trade
context.long_on_spread = False
context.shorting_spread = False
def check_pairs(context, data):
# For convenience
aa = context.aa
ual = context.ual
# Get pricing history
prices = data.history([aa, ual], "price", 30, '1d')
# Need to use .iloc[-1:] to get dataframe instead of series
short_prices = prices.iloc[-1:]
# Get the long 30 day mavg
mavg_30 = np.mean(prices[aa] - prices[ual])
# Get the std of the 30 day long window
std_30 = np.std(prices[aa] - prices[ual])
# Get the shorter span 1 day mavg
mavg_1 = np.mean(short_prices[aa] - short_prices[ual])
# Compute z-score
if std_30 > 0:
zscore = (mavg_1 - mavg_30)/std_30
# Our two entry cases
if zscore > 0.5 and not context.shorting_spread:
# spread = aa - ual
order_target_percent(aa, -0.5) # short top
order_target_percent(ual, 0.5) # long bottom
context.shorting_spread = True
context.long_on_spread = False
elif zscore < -0.5 and not context.long_on_spread:
# spread = aa - ual
order_target_percent(aa, 0.5) # long top
order_target_percent(ual, -0.5) # short bottom
context.shorting_spread = False
context.long_on_spread = True
# Our exit case
elif abs(zscore) < 0.1:
order_target_percent(aa, 0)
order_target_percent(ual, 0)
context.shorting_spread = False
context.long_on_spread = False
record('zscore', zscore)