def gen_data(symbols, dates, holding_days, window, Ybuy, Ysell):
"""
generate normalized indicator values (dataX),
and the classification based on holding_days return (dataY)
"""
# Get prices (adj close and tipical prices)
prices = get_prices(symbols, dates)[0] # adj close prices
tp_prices = get_prices(symbols, dates)[2] # tipical prices
# get all normalized indicator values as dataX
dataX = get_all_indicators(prices, tp_prices, window)[1]
# ret: holding_days return (e.g. 21 days return)
ret = prices.copy()
ret[holding_days:] = (prices[holding_days:] /
prices[:-holding_days].values) - 1
ret.ix[:holding_days] = 0
### Classification
# Generate dataY to classification values LONG(1), SHORT(-1) or HOLD(0)
# based on holding_days return.
dataY = ret.copy()
dataY.ix[:, :] = np.NaN
dataY[(ret > Ybuy)] = 1 # if ret > Ybuy, Long
dataY[(ret <= Ybuy) & (ret < Ysell)] = -1 # else if ret < Ysell, Short
dataY.fillna(0, inplace=True) # else hold, do nothing
# print ret
# print dataY
# convert data from dataframe into np array (to fit for RTLearner I created)
dataX_m = dataX.as_matrix() # convert to np array
dataY_m = dataY.as_matrix().T[0] # convert to 1d np array
return dataX_m, dataY_m
def build_order_list(dataY, symbols, dates, holding_days, output_filename):
"""
construct order_list from dataY, also consider holding_days
dataY: 1d np array, Y data (classification) for each day with Long, short or do nothing result
"""
# get prices for index template
prices = get_prices(symbols, dates)[0]
dataY[0] = 1 # benchmark entry
# build order_list
order_list = [] # a list of both entry and exit orders
count = holding_days
for sym in symbols:
for i in range(dataY.shape[0]):
if count > 0 and count < holding_days:
count -= 1
continue
if dataY[i] == 0:
continue
if dataY[i] != 0:
count = holding_days - 1
exit_i = i + holding_days
# write into order_list
if dataY[i] > 0:
order_list.append(
[prices.index[i].date(), sym, 'BUY', 200])
if exit_i < dataY.shape[0]:
order_list.append(
[prices.index[exit_i].date(), sym, 'SELL', 200])
elif dataY[i] < 0:
order_list.append(
[prices.index[i].date(), sym, 'SELL', 200])
if exit_i < dataY.shape[0]:
order_list.append(
[prices.index[exit_i].date(), sym, 'BUY', 200])
"""
# print order_list
for order in order_list:
print " ".join(str(x) for x in order)
"""
# write order_list to csv file
with open(output_filename, "wb") as f:
writer = csv.writer(f)
writer.writerow(['Date', 'Symbol', 'Order', 'Shares'])
writer.writerows(order_list)
f.close()
def test_run():
# Input data
dates_in_sample = pd.date_range('2008-01-01', '2009-12-31')
symbols = ['AAPL']
prices = get_prices(symbols, dates_in_sample)[0]
order_list = []
sym = 'AAPL'
for day, next_day in zip(prices.index[:-1], prices.index[1:]):
if prices.ix[day, sym] < prices.ix[next_day, sym]:
order_list.append([day.date(), sym, 'BUY', 200])
order_list.append([next_day.date(), sym, 'SELL', 200])
elif prices.ix[day, sym] > prices.ix[next_day, sym]:
order_list.append([day.date(), sym, 'SELL', 200])
order_list.append([next_day.date(), sym, 'BUY', 200])
# write orders to csv file
with open("orders-bestpossible.csv", "wb") as f:
writer = csv.writer(f)
writer.writerow(['Date', 'Symbol', 'Order', 'Shares'])
writer.writerows(order_list)
f.close()
def main(information):
a_list_of_prices = indicators.get_prices(information)
print(a_list_of_prices)
strategies(a_list_of_prices)
def compute_portvalue(symbols, dates, orders_file, start_val=100000):
# Read in order.csv data
orders = pd.read_csv(orders_file,
index_col='Date',
parse_dates=True,
na_values=['nan'])
# sort orders by date
orders = orders.sort_index()
# Get prices of a given symbols and date range
prices = get_prices(symbols, dates)[0]
# Step 1: 'prices' dataframe
prices['CASH'] = 1.0 # add Cash -- all = 1.0
# print prices.ix[:10,:]
# Step 3: 'holdings' dataframe
# initialize 'holdings' from 'prices', fill columns by 0
# then add start value to the first colunm of cash
holdings = pd.DataFrame(0, index=prices.index, columns=prices.columns)
holdings.ix[0, -1] = start_val # should be holdings.ix[:,-1] = start_val
# read orders from the 'orders' dataframe
# calculate leverage, update holdings, etc.
# if one order void leverage, then cancel all trades on that day
for row in orders.itertuples():
# pass orders in non trading days
if row.Index not in prices.index:
continue
# read in shares
if row.Order == 'BUY':
shares = row.Shares
else:
shares = -row.Shares
# calculate leverage_before the order placed
hld_before = holdings.ix[row.Index, :].copy()
val_before = prices.ix[row.Index, :] * hld_before
leverage_before = sum(abs(val_before[:-1])) / sum(val_before)
# calculate leverage after the order placed
hld = holdings.ix[row.Index, :].copy()
hld[row.Symbol] += shares
hld['CASH'] -= prices.ix[row.Index, row.Symbol] * shares
val = prices.ix[row.Index, :] * hld
leverage = sum(abs(val[:-1])) / sum(val)
# print "lev-bef: ", leverage_before, "lev: ", leverage
if True: # leverage < 1.5 or leverage < leverage_before:
# make order happen
holdings.ix[row.Index, row.Symbol] += shares
holdings.ix[row.Index,
'CASH'] -= prices.ix[row.Index, row.Symbol] * shares
holdings.ix[row.Index:, :] = holdings.ix[row.Index, :].values
# Step 4: 'values' dataframe
values = prices * holdings
# Step 5: 'portvals'
portvals = values.sum(axis=1)
return portvals
def build_orders(symbols, dates, holding_days, window, output_filename):
# Get prices (adj close and tipical prices)
prices = get_prices(symbols, dates)[0]
tp_prices = get_prices(symbols, dates)[2]
# get all indicator values (not normalized)
indicators = get_all_indicators(prices, tp_prices, window)[0]
# bollinger band precent
bbp = indicators['BBP']
# price/sma
price_sma_ratio = indicators['PRICE_SMA']
# momentum
momentum = indicators['MOMENTUM']
# cci
cci = indicators['CCI']
### Use the four indicators to make rule based decision
# Holdings starts as a NaN array of the same shape/index as price.
orders = prices.copy()
orders.ix[:, :] = np.NaN
# Apply our entry order conditions all at once. This represents our TARGET SHARES
# at this moment in time, not actual orders.
# orders didn't consider holding_days at this stage
orders.ix[0, :] = 200 # benchmark entry
orders[(momentum > 0)
& ((price_sma_ratio < 0.95) | (bbp < 0) | (cci < -50))] = 200
orders[(momentum < 0)
& ((price_sma_ratio > 1.05) | (bbp > 1) | (cci > 150))] = -200
orders.ffill(inplace=True)
orders.fillna(0, inplace=True)
# construct order_list from orders dataframe, also consider holding_days
order_list = [] # a list of both entry and exit orders
count = holding_days
for sym in symbols:
for i in range(orders.shape[0]):
if count > 0 and count < holding_days:
count -= 1
continue
if orders.ix[i, sym] == 0:
continue
if orders.ix[i, sym] != 0:
count = holding_days - 1
exit_i = i + holding_days
# write into order_list
if orders.ix[i, sym] > 0:
order_list.append(
[orders.index[i].date(), sym, 'BUY', 200])
if exit_i < orders.shape[0]:
order_list.append(
[orders.index[exit_i].date(), sym, 'SELL', 200])
elif orders.ix[i, sym] < 0:
order_list.append(
[orders.index[i].date(), sym, 'SELL', 200])
if exit_i < orders.shape[0]:
order_list.append(
[orders.index[exit_i].date(), sym, 'BUY', 200])
# print order_list
for order in order_list:
print " ".join(str(x) for x in order)
# write order_list to csv file
with open(output_filename, "wb") as f:
writer = csv.writer(f)
writer.writerow(['Date', 'Symbol', 'Order', 'Shares'])
writer.writerows(order_list)
f.close()
def main(information):
a_list_of_prices = indicators.get_prices(information)
print(a_list_of_prices)
strategies(a_list_of_prices)