In [13]:
import pandas as pd
import matplotlib.pyplot as plt
# Lesson 01-02 - Working with multiple stocks
print "Lesson 01-02 - Working with multiple stocks"
def getEmptyDataFrameWithRange(startDate, endDate):
dates = pd.date_range(startDate, endDate)
return pd.DataFrame(index=dates)
def loadData(symbols, column, startDate, endDate, dropNan=True):
dfResult = getEmptyDataFrameWithRange(startDate, endDate)
for symbol in symbols:
dfSymbol = pd.read_csv('data/{}.csv'.format(symbol), index_col='Date', parse_dates=True, usecols=['Date',column], na_values=['nan'])
dfSymbol = dfSymbol.rename(columns={column: symbol})
# join (left join) the 2 dataframes and remove rows with NaN
dfResult = dfResult.join(dfSymbol)
if dropNan == True:
dfResult = dfResult.dropna()
dfResult.sort_index(inplace=True) # sort by index columns
return dfResult
symbols = ['GLD', 'GOOG', 'IBM', 'SPY', 'XOM']
column = 'Adj Close'
df = loadData(symbols, column, '2010-01-01', '2012-12-31')
%pylab inline
pylab.rcParams['figure.figsize'] = (15, 10)
# values plot
axes = df.plot(title=column, fontsize=10)
axes.set_xlabel('Date')
axes.set_ylabel('Value')
plt.show()
# normalization
dfNorm = df / df.ix[0,:] # divide all rows by the first one
# normalized plot
axes = dfNorm.plot(title='{} (normalized)'.format(column), fontsize=10)
axes.set_xlabel('Date')
axes.set_ylabel('Value')
plt.show()
In [14]:
# Lesson 01-04
print "Lesson 01-04 - Statistical analysis of time series"
print "\n\t\tGlobal statistics"
print "\nMean:"
print df.mean()
print "\nMedian:"
print df.median()
print "\nStandard deviation:"
print df.std()
print "\nSum:"
print df.sum()
print "\n\n\t\tRolling statistics"
symbol = 'SPY'
dfRolling = loadData([symbol], 'Adj Close', '2012-01-01', '2012-12-31')
axes = dfRolling[symbol].plot(title='{} rolling mean'.format(symbol), label=symbol)
# compute rolling mean using a 20-day window
window = 20
dfRolling.sort_index(inplace=True) # sort by index columns
rollingMean = dfRolling[symbol].rolling(window=window, center=False).mean()
rollingStd = dfRolling[symbol].rolling(window=window, center=False).std()
rollingMean.plot(label='Rolling mean', ax=axes)
(rollingMean-rollingStd*2).plot(label='-2σ', ax=axes)
(rollingMean+rollingStd*2).plot(label='+2σ', ax=axes)
axes.set_xlabel('Date')
axes.set_ylabel('Price')
axes.legend(loc='upper left')
plt.show()
def dailyReturns(data):
result = data
result[1:] = (result[1:]/result[:-1].values) - 1 # compute daily returns
result.ix[0, :] = 0 # set the first row values to 0
return result
# calculate daily returns
dfReturns = dailyReturns(dfRolling.copy())
# plot daily returns
axes = dfReturns.plot(label='{} daily returns'.format(symbol))
axes.set_xlabel('Date')
axes.set_ylabel('Daily return')
plt.show()
# calculate cumulative returns
dfReturns = dfRolling.copy()
dfReturns = dfReturns.div(dfReturns.ix[0]) - 1 # compute daily returns
# plot cumulative returns
axes = dfReturns.plot(label='{} cumulative returns'.format(symbol))
axes.set_xlabel('Date')
axes.set_ylabel('Cumulative return')
plt.show()
In [15]:
# Lesson 01-05
print "Lesson 01-05 - Incomplete data"
symbols = ['SPY']
column = 'Adj Close'
startDate = '2012-07-01'
endDate = '2012-09-30'
dfMissing = loadData(symbols, column, startDate, endDate, dropNan=False)
dfMissing.plot(title='Missing values')
dfMissing.fillna(method='ffill').plot(title='Missing values (fill forward)')
dfMissing.fillna(method='bfill').plot(title='Missing values (fill backward)')
dfMissing.fillna(method='ffill').fillna(method='bfill').plot(title='Missing values (fill forward + backward)')
plt.show()
In [16]:
# Lesson 01-06
print "Lesson 01-06 - Histograms and scatter plots"
print "\n\t\tHistograms"
symbols = ['SPY', 'XOM']
column = 'Adj Close'
startDate = '2009-01-01'
endDate = '2012-12-31'
dfDaily = dailyReturns(loadData(symbols, column, startDate, endDate))
dfDaily.plot(title='Daily returns')
plt.show()
mean = dfDaily['SPY'].mean()
print 'SPY mean:', mean
std = dfDaily['SPY'].std()
print 'SPY standard deviation:', std
dfDaily['SPY'].hist(bins=20, label='SPY')
plt.legend(loc='upper right')
plt.axvline(mean, color='w', linestyle='dashed', linewidth=2, label='SPY mean')
plt.axvline( std, color='r', linestyle='dashed', linewidth=2, label='SPY std')
plt.axvline(-std, color='r', linestyle='dashed', linewidth=2, label='SPY -std')
plt.title('SPY daily returns histogram')
plt.show()
dfDaily.hist(bins=20)
plt.legend(loc='upper right')
plt.show()
dfDaily['SPY'].hist(bins=20, label='SPY')
dfDaily['XOM'].hist(bins=20, label='XOM')
plt.title('SPY + XOM daily returns histogram')
print "\n\t\tScatterplot"
symbols = ['SPY', 'XOM', 'GLD']
column = 'Adj Close'
startDate = '2009-01-01'
endDate = '2012-12-31'
dfScatter = dailyReturns(loadData(symbols, column, startDate, endDate))
betaXOM, alphaXOM = np.polyfit(dfScatter['SPY'], dfScatter['XOM'], 1)
betaGLD, alphaGLD = np.polyfit(dfScatter['SPY'], dfScatter['GLD'], 1)
print 'XOM: beta={}, alpha={}'.format(betaXOM, alphaXOM)
print 'GLD: beta={}, alpha={}'.format(betaGLD, alphaGLD)
print 'Correlation:'
print dfScatter.corr('pearson')
dfScatter.plot(title='SPY vs XOM (scatterplot)', kind='scatter', x='SPY', y='XOM')
plt.plot(dfScatter['SPY'], betaXOM*dfScatter['SPY'] + alphaXOM, '-', color='r')
plt.show()
dfScatter.plot(title='SPY vs GLD (scatterplot)', kind='scatter', x='SPY', y='GLD')
plt.plot(dfScatter['SPY'], betaGLD*dfScatter['SPY'] + alphaGLD, '-', color='r')
plt.show()
In [17]:
# Lesson 01-07
print "Lesson 01-07 - Sharpe ratio and other portfolio statistics"
print "\n\t\tDaily portfolio value"
symbols = ['SPY', 'XOM', 'GOOG', 'GLD']
startDate = '2009-01-01'
endDate = '2011-12-31'
column = 'Adj Close'
allocs = [ .4, .4, .1, .1]
startValue = 1000000
# load data
dfPort = loadData(symbols, column, startDate, endDate)
# normalization: prices[t] / prices[0]
dfPort = dfPort / dfPort.ix[0,:]
# allocations: normed * allocs
dfPort = dfPort * allocs
# position values: alloccd * startValue (values for each day)
dfPort = startValue * dfPort
# portfolio values per day = posVals.sum(axix=1)
portfolioValues = dfPort.sum(axis=1)
portfolioValues.plot()
plt.show()
In [24]:
import scipy.optimize as spo
def error(line, data):
value = np.sum((data[:,1] - (line[0] * data[:,0] + line[1])) ** 2)
return value
def fitLine(data, errorFunction):
# generate initial guess
line = np.float32([0, np.mean(data[:,1])]) # slope=0, intercept=mean(y values)
# call optimizer to minimize error function
result = spo.minimize(errorFunction, line, args=(data,), method='SLSQP')
return result.x
# Lesson 01-08
print "Lesson 01-08 - Optimizer: nuilding a parameterized model"
print "\n\t\tDaily portfolio value"
# original line
originalLine = np.float32([4, 2])
xOrig = np.linspace(0, 10, 21)
yOrig = originalLine[0] * xOrig + originalLine[1]
# print original line's coefficients and plot it
print "Original line: C0 = {}, C1 = {}".format(originalLine[0], originalLine[1])
plt.plot(xOrig, yOrig, 'b--', linewidth=2.0, label='Original line')
plt.legend(loc='upper left')
# add noise to the original line
noise = np.random.normal(0, 3.0, yOrig.shape)
data = np.asarray([xOrig, yOrig + noise]).T
plt.plot(data[:,0], data[:,1], 'go', label='Data points')
plt.legend(loc='upper left')
# try to fit a line to the data
fitLine = fitLine(data, error)
# print fitted line's coefficients and plot it
print "Fitted line: C0 = {}, C1 = {}".format(fitLine[0], fitLine[1])
plt.plot(data[:,0], fitLine[0]*data[:,0] + fitLine[1], 'r--', linewidth=2.0, label='Fitted line')
plt.legend(loc='upper left')
Out[24]:
In [ ]: