Pair Trading of GLD and GDX

In [1]:

    

import numpy as np


    

In [2]:

    

import pandas as pd


    

In [3]:

    

import matplotlib.pyplot as plt


    

In [4]:

    

import statsmodels.api as sm


    

In [5]:

    

df1=pd.read_excel('GLD.xls')


    

In [6]:

    

df2=pd.read_excel('GDX.xls')


    

In [7]:

    

df=pd.merge(df1, df2, on='Date', suffixes=('_GLD', '_GDX'))


    

In [8]:

    

df.set_index('Date', inplace=True)


    

In [9]:

    

df.sort_index(inplace=True)


    

In [10]:

    

trainset=np.arange(0, 252)


    

In [11]:

    

testset=np.arange(trainset.shape[0], df.shape[0])


    

In [12]:

    

model=sm.OLS(df.loc[:, 'Adj Close_GLD'].iloc[trainset], df.loc[:, 'Adj Close_GDX'].iloc[trainset])


    

In [13]:

    

results=model.fit()


    

In [14]:

    

hedgeRatio=results.params


    

In [15]:

    

hedgeRatio


    

    
Out[15]:





Adj Close_GDX    1.631009
dtype: float64

In [16]:

    

spread=df.loc[:, 'Adj Close_GLD']-hedgeRatio[0]*df.loc[:, 'Adj Close_GDX']


    

In [17]:

    

plt.plot(spread.iloc[trainset])


    

    
Out[17]:





[<matplotlib.lines.Line2D at 0x23ce082ec18>]






    

In [18]:

    

plt.plot(spread.iloc[testset])


    

    
Out[18]:





[<matplotlib.lines.Line2D at 0x23ce06a7240>]






    

In [19]:

    

spreadMean=np.mean(spread.iloc[trainset])


    

In [20]:

    

spreadMean


    

    
Out[20]:





0.05219623850035999

In [21]:

    

spreadStd=np.std(spread.iloc[trainset])


    

In [22]:

    

spreadStd


    

    
Out[22]:





1.944860873496509

In [23]:

    

df['zscore']=(spread-spreadMean)/spreadStd


    

In [24]:

    

df['positions_GLD']=np.nan


    

In [25]:

    

df['positions_GDX']=np.nan


    

In [26]:

    

df.loc[df.zscore>=2, ('positions_GLD', 'positions_GDX')]=[-1, 1] # Buy spread


    

In [27]:

    

df.loc[df.zscore<=-2, ('positions_GLD', 'positions_GDX')]=[1, -1] # Short spread


    

In [28]:

    

df.loc[abs(df.zscore)<=1, ('positions_GLD', 'positions_GDX')]=0 # Exit spread


    

In [29]:

    

df.fillna(method='ffill', inplace=True) # ensure existing positions are carried forward unless there is an exit signal


    

In [30]:

    

positions=df.loc[:, ('positions_GLD', 'positions_GDX')]


    

In [31]:

    

dailyret=df.loc[:, ('Adj Close_GLD', 'Adj Close_GDX')].pct_change()


    

In [32]:

    

pnl=(np.array(positions.shift())*np.array(dailyret)).sum(axis=1)


    

In [33]:

    

sharpeTrainset=np.sqrt(252)*np.mean(pnl[trainset[1:]])/np.std(pnl[trainset[1:]])


    

In [34]:

    

sharpeTrainset


    

    
Out[34]:





2.3324949367127323

In [35]:

    

sharpeTestset=np.sqrt(252)*np.mean(pnl[testset])/np.std(pnl[testset])


    

In [36]:

    

sharpeTestset


    

    
Out[36]:





1.5139137120213677

In [37]:

    

plt.plot(np.cumsum(pnl[testset]))


    

    
Out[37]:





[<matplotlib.lines.Line2D at 0x23ce0b844a8>]






    

In [38]:

    

positions.to_pickle('example3_6_positions')