def testPolicy(self,
symbol="JPM",
sd=dt.datetime(2008, 1, 1),
ed=dt.datetime(2009, 12, 31),
sv=100000):
self.sv = sv
#Read in the stock price
if type(symbol) is str:
symbols = list()
symbols.append(symbol)
else:
symbols = symbol
prices_df = get_data(symbols,
pd.date_range(sd, ed),
addSPY=True,
colname='Adj Close')
#prices_df['cash'] = 1.0
#prices_df = prices_df.drop(['SPY'], axis = 1)
prices_df = prices_df.fillna(method='ffill')
prices_df = prices_df.fillna(method='bfill')
lookback_days = 30
#date_range = prices_df.index[(lookback_days-1):]
#Create an empty dataframe df_trades
df_trades = pd.DataFrame(data=0,
index=prices_df.index,
columns=symbols)
#print df_trades.head(20)
sma, p_sma = p_sma_ratio(prices_df, lookback=lookback_days)
K, D = stochastic_K_D(prices_df, lookback=lookback_days)
BB, upperband, lowerband = bband(prices_df, lookback=lookback_days)
for i in range(1, df_trades.shape[0] - (lookback_days - 1)):
#The long entry
if BB[symbol][i - 1] <= -1.01 and p_sma[symbol][i] > p_sma[symbol][
i - 1] and D[symbol][i - 1] < 20:
if self.netHolding == -1000:
df_trades.iloc[i + lookback_days - 1] = 2000
elif self.netHolding == 0:
df_trades.iloc[i + lookback_days - 1] = 1000
#The short entry
elif BB[symbol][i -
1] >= 1.01 and p_sma[symbol][i] < p_sma[symbol][
i - 1] and D[symbol][i - 1] > 80:
#
if self.netHolding == 1000:
df_trades.iloc[i + lookback_days - 1] = -2000
elif self.netHolding == 0:
df_trades.iloc[i + lookback_days - 1] = -1000
#The exit position
elif (BB[symbol][i - 1] >= 0.95 and BB[symbol][i - 1] < 1.0) or (
BB[symbol][i - 1] > -1.0 and BB[symbol][i - 1] <= -0.95):
df_trades.iloc[i + lookback_days - 1] = -self.netHolding
self.netHolding += df_trades.iloc[i + lookback_days - 1].values
return df_trades
def __init__(self, sym='SPY', sd=dt.datetime(2008,1, 1), \
ed=dt.datetime(2009,12,31),sv=100000,verbose=False):
self.sym = sym
self.verbose = verbose
self.shares = 0
self.position = 0
self.sv = sv
self.cash = sv
self.pp = 0
#period
n = 28
chart = False
#take in, normalize data
df = ut.get_data([sym], pd.date_range(sd - dt.timedelta(100), ed))[sym]
normed = df / df.ix[0]
#get features
sma = ind.SMA(normed, n)
smap = ind.standardize(normed / sma)
bbp = ind.standardize(ind.bband(normed, n, chart)[0])
stdev = ind.standardize(normed.rolling(window=n, min_periods=n).std())
mom = ind.standardize(ind.momentum(normed, n, chart))
momvol = ind.standardize(mom / stdev)
#daily returns & combined features
rets = pd.DataFrame(df)
daily_rets = rets[sym].copy()
daily_rets[1:] = (rets[sym].ix[1:] / rets[sym].ix[:-1].values) - 1
daily_rets.ix[0] = 0
#print df
#df = pd.DataFrame(df).assign(normed = normed).assign(smap=smap).\
# assign(stdev=stdev).assign(bbp=bbp).assign(mom=mom)[sd:]
df = pd.DataFrame(df).assign(normed = normed).assign(smap=smap).\
assign(bbp=bbp).assign(mom=mom)[sd:]
#print df
daily_rets.ix[0] = 0
df = df.assign(dr=daily_rets)
#print df
self.df = df
self.market = df.iterrows()
self.curr = self.market.next()
self.action = self.Action()
#self.features = ['smap', 'stdev', 'bbp', 'mom']
self.features = ['smap', 'bbp', 'mom']
def testPolicy(symbol = "AAPL", sd=dt.datetime(2010,1,1), \
ed = dt.datetime(2011,12,31), sv=100000):
chart = False
n = 14
#symbol = [symbol]
prices = get_data(symbol, pd.date_range(sd, ed))
#print prices
#ffill and drop SPY
prices.fillna(method='ffill', inplace=True)
prices.fillna(method='bfill', inplace=True)
prices = prices[symbol]
#Generate indicators
sma = ind.SMA(prices, n)
smap = ind.standardize(prices / sma)
bbp = ind.standardize(ind.bband(prices, n, chart)[0])
stdev = ind.standardize(prices.rolling(window=n, min_periods=n).std())
mom = ind.standardize(ind.momentum(prices, n, chart))
momvol = ind.standardize(mom / stdev)
orders = prices.copy()
orders.ix[:, :] = np.nan
smap_cross = pd.DataFrame(0, index=smap.index, columns=smap.columns)
smap_cross[smap >= 1] = 1
smap_cross[1:] = smap_cross.diff()
smap_cross.ix[0] = 0
orders[(smap < 0.95) & (bbp < 0) & (stdev < 0.1) & (mom < 0)] = 1000
orders[(smap > 1.05) & (bbp > 1) & (stdev > 0.3) & (mom > 0)] = -1000
orders[(smap_cross != 0)] = 0
orders.ffill(inplace=True)
orders.fillna(0, inplace=True)
orders[1:] = orders.diff()
#orders.ix[0] = 1000
#orders.ix[-1] = -1000
#orders = orders.loc[(orders!=0).any(axis=1)]
#orders.ix[0] = 0
#orders.ix[-1] = orders.ix[-2]*-1
#print orders
return orders
# split to Bid/Ask
dfdict = func.splitbidask(dfprices)
#-----------------------------------------------------------------------------
# indicators testing
#-----------------------------------------------------------------------------
dfbid = dfdict['bid']
dfbid.head()
dfbid.dtypes
df1 = ind.trix( dfbid, [10,15])
df1 = ind.sma( dfbid, window = [3,5,9])
df1 = ind.sma( dfbid, 'volume', [5])
df1 = ind.bband( dfbid, 21)
df1.head(30)
df1 = ( dfbid.pipe( ind.bband, 21)
.pipe( ind.sma, [3,5,9])
.pipe( ind.trix, [7,11])
#-----------------------------------------------------------------------------
# optimization for indicators
def addEvidence(self, symbol = "IBM", \
sd=dt.datetime(2008,1,1), \
ed=dt.datetime(2009,1,1), \
sv = 100000):
# add your code to do learning here
self.sv = sv
#Read in the stock price
if type(symbol) is str:
symbols = list()
symbols.append(symbol)
else:
symbols = symbol
prices_df = ut.get_data(symbols,
pd.date_range(sd, ed),
addSPY=True,
colname='Adj Close')
#prices_df['cash'] = 1.0
#prices_df = prices_df.drop(['SPY'], axis = 1)
prices_df = prices_df.fillna(method='ffill')
prices_df = prices_df.fillna(method='bfill')
prices_df = prices_df.drop(['SPY'], axis=1)
lookback_days = 15
#Create an empty dataframe df_trades
df_trades = pd.DataFrame(data=0,
index=prices_df.index,
columns=symbols)
#print df_trades.head(20)
sma, p_sma = p_sma_ratio(prices_df, lookback=lookback_days)
K, D = stochastic_K_D(prices_df, lookback=lookback_days)
BB, upperband, lowerband = bband(prices_df, lookback=lookback_days)
#discretize the indicators
p_sma_bins = pd.to_numeric(
pd.cut(p_sma[symbol], 10, labels=range(0, 10)))
K_bins = pd.to_numeric(pd.cut(K[symbol], 8, labels=range(0, 8)))
BB_bins = pd.to_numeric(pd.cut(BB[symbol], 10, labels=range(0, 10)))
states = 100 * K_bins + 10 * p_sma_bins + BB_bins
#print type(states[0])
#Instantiate a Q-learner
converge = False
count = 0
#portvals_prev = [sv, sv, sv, sv, sv, sv, sv, sv, sv, sv]
portvals_prev = [sv, sv, sv, sv, sv]
#df_tmp = df_trades
while (not converge and count < 500):
action = self.learner.querysetstate(states[0])
if (action == 0): #long
self.netHolding = 1000
elif (action == 1): #Nothing
self.netHolding = 0
else: #short
self.netHolding = -1000
#print df_trades.loc['2008-01-02':'2008-01-31']
# print states.index
# print states.shape[0]
# print prices_df.loc[states.index[0]]
for i in range(1, states.shape[0]):
day = states.index[i]
day_before = states.index[i - 1]
if (action == 0):
impact_sign = 1
elif (action == 1):
impact_sign = 0
else:
impact_sign = -1
#r = (prices_df[symbol].loc[day] - prices_df[symbol].loc[day_before] * \
# (1+impact_sign * self.impact)) * self.netHolding / self.sv
r = self.netHolding * (prices_df[symbol].loc[day] /
prices_df[symbol].loc[day_before] - 1 -
self.impact * impact_sign)
# if self.netHolding != 0:
# r = prices_df[symbol].loc[day]/prices_df[symbol].loc[day_before] - 1 - self.impact
# else:
# r = 0
#r = ((prices_df[symbol].loc[day] - prices_df[symbol].loc[day_before]) * self.netHolding \
# - impact_sign * self.impact * df_trades.loc[day_before] * prices_df[symbol].loc[day_before])/self.sv
action = self.learner.query(states[day], r)
if (action == 0): #long
if self.netHolding == -1000:
df_trades.loc[day] = 2000
elif self.netHolding == 0:
df_trades.loc[day] = 1000
elif (action == 2): #Short
if self.netHolding == 1000:
df_trades.loc[day] = -2000
elif self.netHolding == 0:
df_trades.loc[day] = -1000
self.netHolding += df_trades.loc[day].values
# print prices_df.loc[day]
# print df_trades.loc[day]
# print self.netHolding
# print ""
portvals, cr, stdev, mean = compute_portvals(df_trades,
start_val=sv,
commission=0,
impact=self.impact)
#diff_portvals = abs(portvals[-1]- portvals_prev)
diff_portvals = abs(portvals[-1] - np.mean(portvals_prev))
#print portvals_prev
if (count >= 10 and diff_portvals < 50.0):
converge = True
else:
count += 1
portvals_prev.pop(0)
portvals_prev.append(portvals[-1])
print count
def testPolicy(self, symbol = "JPM", \
sd=dt.datetime(2009,1,1), \
ed=dt.datetime(2010,1,1), \
sv = 10000):
#Read in the stock price
if type(symbol) is str:
symbols = list()
symbols.append(symbol)
else:
symbols = symbol
prices_df = ut.get_data(symbols,
pd.date_range(sd, ed),
addSPY=True,
colname='Adj Close')
#prices_df['cash'] = 1.0
#prices_df = prices_df.drop(['SPY'], axis = 1)
prices_df = prices_df.fillna(method='ffill')
prices_df = prices_df.fillna(method='bfill')
prices_df = prices_df.drop(['SPY'], axis=1)
lookback_days = 15
#Create an empty dataframe df_trades
df_trades = pd.DataFrame(data=0,
index=prices_df.index,
columns=symbols)
#print df_trades.head(20)
sma, p_sma = p_sma_ratio(prices_df, lookback=lookback_days)
K, D = stochastic_K_D(prices_df, lookback=lookback_days)
BB, upperband, lowerband = bband(prices_df, lookback=lookback_days)
#discretize the indicators
p_sma_bins = pd.to_numeric(
pd.cut(p_sma[symbol], 10, labels=range(0, 10)))
K_bins = pd.to_numeric(pd.cut(K[symbol], 8, labels=range(0, 8)))
BB_bins = pd.to_numeric(pd.cut(BB[symbol], 10, labels=range(0, 10)))
states = 100 * K_bins + 10 * p_sma_bins + BB_bins
netHolding = 0
for i in range(1, states.shape[0]):
day = states.index[i]
#day_before = states.index[i-1]
#r = (prices_df[symbol].loc[day] - prices_df[symbol].loc[day_before])* netHolding / sv
action = self.learner.querysetstate(states[day])
if (action == 0): #long
if netHolding == -1000:
df_trades.loc[day] = 2000
elif netHolding == 0:
df_trades.loc[day] = 1000
elif (action == 2): #Short
if netHolding == 1000:
df_trades.loc[day] = -2000
elif netHolding == 0:
df_trades.loc[day] = -1000
netHolding += df_trades.loc[day].values
#portvals, cr, stdev, mean = compute_portvals(df_trades, start_val = sv, commission=0, impact=self.impact)
# print states[day]
# here we build a fake set of trades
# your code should return the same sort of data
# dates = pd.date_range(sd, ed)
# prices_all = ut.get_data([symbol], dates) # automatically adds SPY
# trades = prices_all[[symbol,]] # only portfolio symbols
# trades_SPY = prices_all['SPY'] # only SPY, for comparison later
# trades.values[:,:] = 0 # set them all to nothing
# trades.values[0,:] = 1000 # add a BUY at the start
# trades.values[40,:] = -1000 # add a SELL
# trades.values[41,:] = 1000 # add a BUY
# trades.values[60,:] = -2000 # go short from long
# trades.values[61,:] = 2000 # go long from short
# trades.values[-1,:] = -1000 #exit on the last day
# if self.verbose: print type(trades) # it better be a DataFrame!
# if self.verbose: print trades
# if self.verbose: print prices_all
return df_trades
def __init__(self, sym='SPY', sd=dt.datetime(2008,1, 1), \
ed=dt.datetime(2009,12,31),sv=100000,verbose=False, experiment=False, impact=0.0):
self.sym = sym
self.verbose = verbose
self.shares = 0
self.position = 2
self.sv = sv
self.cash = sv
self.pp = 0
self.impact = impact
self.experiment = experiment
#n = 3 #period
chart = False
#take in, normalize data
df = ut.get_data([sym], pd.date_range(sd - dt.timedelta(100), ed))[sym]
#print df
#print sd, ed
normed = df / df.ix[0]
#print normed
#daily returns & combined features
rets = pd.DataFrame(df)
daily_rets = rets[sym].copy()
daily_rets[1:] = (rets[sym].ix[1:] / rets[sym].ix[:-1].values) - 1
daily_rets.ix[0] = 0
#Simple moving average
sma5 = normed.rolling(5).mean()
sma15 = normed.rolling(15).mean()
sma20 = normed.rolling(20).mean()
sma25 = normed.rolling(25).mean()
sma30 = normed.rolling(30).mean()
sma40 = normed.rolling(40).mean()
#print sma5
# Volatility
vol5 = normed.rolling(5).std()
vol10 = normed.rolling(10).std()
vol20 = normed.rolling(20).std()
vol30 = normed.rolling(30).std()
# Bollinger bands
sma_bb = normed.rolling(5).mean()
sma_bb_std = normed.rolling(5).std()
bb = (normed -
(sma_bb - 2 * sma_bb_std)) / ((sma_bb + 2 * sma_bb_std) -
(sma_bb - 2 * sma_bb_std))
# Moving average convergence/divergence
#ema12 = pd.ewma(np.asarray(normed), span=12)
#ema26 = pd.ewma(np.asarray(normed), span=26)
#macd = ema12 - ema26
# Momentum
momentum2 = normed / normed.shift(2) - 1
momentum5 = normed / normed.shift(5) - 1
momentum10 = normed / normed.shift(10) - 1
# Combine into new dataframe
df = pd.DataFrame(df).assign(sma5=normed / sma5 - 1).assign(\
bb=bb).assign(momentum2=momentum2).assign(normed=normed).\
assign(vol10=vol10).assign(vol20=vol20).assign(vol30=vol30).assign(vol10=vol10)\
.assign(vol5=vol5).assign(\
momentum5=momentum5).assign(momentum10=momentum10)[sd:]
df = df.assign(dr=daily_rets)
#print df
# Determine optimal features for states
corr_df = df.corr().abs()
corr = corr_df['dr'][:]
corr.ix[0] = 0
corr['normed'] = 0
icorr = np.asarray(corr)
scorr = icorr.argsort(
)[-4:][::-1] # select top 3 features and daily returns
scorr = scorr[1:] # remove daily returns from possible features
optimal_ftrs = []
for i in scorr:
optimal_ftrs.append(corr_df.columns[i])
if experiment:
#print "Experiment 1 ACTIVATED"
n = 14
sma = ind.SMA(normed, n)
smap = ind.standardize(normed / sma)
bbp = ind.standardize(ind.bband(normed, n, chart)[0])
stdev = ind.standardize(
normed.rolling(window=n, min_periods=n).std())
mom = ind.standardize(ind.momentum(normed, n, chart))
momvol = ind.standardize(mom / stdev)
df = pd.DataFrame(df).assign(normed=normed).assign(
smap=smap).assign(stdev=stdev).assign(bbp=bbp).assign(
mom=mom)[sd:]
optimal_ftrs = ['smap', 'stdev', 'bbp', 'mom']
#df = pd.DataFrame(df).assign(normed = normed).assign(smap=smap).assign(bbp=bbp).assign(mom=mom)[sd:]
#df = pd.DataFrame(df).assign(normed = normed)[sd:]
self.df = df
self.market = df.iterrows()
self.curr = self.market.next()
self.action = self.Action()
self.features = optimal_ftrs