def run(prices, ts, ratio_func, start_cash):
count = len(ts)
orders = np.zeros((count, 2), dtype=np.int)
positions = np.zeros((count, 2), dtype=np.int)
cash = np.zeros(count)
cash[0] = start_cash
cash_delta = np.zeros(count)
vxx_ratio = np.zeros(count)
vxz_ratio = np.zeros(count)
for i in range(1, count):
ratio = ratio_func(ts[i-1])
vxx_ratio[i] = ratio[0]
vxz_ratio[i] = ratio[1]
orders[i] = calc_adjustments(
positions[i-1],
(prices['VXX'][i-1], prices['VXZ'][i-1]),
ratio,
cash[i-1])
positions[i] = positions[i-1] + orders[i]
cash_delta[i] = calc_cash_delta(
orders[i],
(prices['VXX'][i-1], prices['VXZ'][i-1]),
positions[i-1])
cash[i] = cash[i-1] + cash_delta[i]
if DEBUG:
print('vxx ratio, vxz ratio, vxx px, vxz px, vxx orders, vxz orders, '
'vxx positions, vxz positions, cash, cash delta')
for i in range(0, count):
print('{}, {}, {}, {}, {}, {}, {}, {}, {}, {}'.format(
vxx_ratio[i], vxz_ratio[i],
prices['VXX'][i], prices['VXZ'][i],
orders[i][0], orders[i][1],
positions[i-1][0], positions[i-1][1],
cash[i], cash_delta[i]))
vxx_change = (prices['VXX'] - utils.lag(prices['VXX'])) * positions[:, 0]
vxz_change = (prices['VXZ'] - utils.lag(prices['VXZ'])) * positions[:, 1]
notional_value = (np.absolute(positions) * prices[['VXX', 'VXZ']]).sum(1) \
+ cash
returns = (vxx_change + vxz_change) / notional_value
# np.save('/Users/Conor/code/blog/python/kelly/returns.npy', returns)
np.savetxt('/Users/Conor/code/blog/python/returns/cash.txt', cash)
# np.savetxt('/Users/Conor/code/blog/python/returns/positions.txt', positions)
# np.savetxt('/Users/Conor/code/blog/python/returns/stockA.txt', prices['VXX'] / 5.)
# np.savetxt('/Users/Conor/code/blog/python/returns/stockB.txt', prices['VXZ'] / 5.)
return returns, cash
def population_std_dev(close_prices, lookback):
N = float(lookback)
prices = log(close_prices / utils.lag(close_prices))
results = np.zeros(np.size(prices))
results[:] = np.NAN
for i in range(lookback, len(prices)):
bounds = range(i - (lookback - 1), i + 1)
results[i] = sqrt(
((prices[bounds] - prices[bounds].sum() / N)**2).sum() / (N - 1))
return annualise(results)
def population_std_dev(close_prices, lookback):
N = float(lookback)
prices = log(close_prices / utils.lag(close_prices))
results = zeros(size(prices))
results[:] = NAN
for i in range(lookback, len(prices)):
bounds = range(i-(lookback-1), i+1)
results[i] = sqrt(
((prices[bounds] - prices[bounds].sum() / N)**2).sum() / (N - 1))
return annualise(results)
def get_random_intraday_returns(mu, sigma, day_count):
# prices = np.random.normal(mu, sigma, 391 * day_count)
prices = np.cumsum(np.random.normal(0, 1, 391 * day_count) * sigma) + mu
log_returns = (np.log(prices / utils.lag(prices, np.NaN)))
log_returns[0] = 0.
return np.cumsum(log_returns)
def test_lag_multiple_vec(self):
np_utils.assert_array_equal(
np.array([[0, 0], [1, 4], [2, 5]]),
utils.lag(np.array([[1, 4], [2, 5], [3, 6]])))
def test_lag_nan(self):
np_utils.assert_array_equal(
np.array([np.NAN, 1, 2, 3, 4]),
utils.lag(np.array([1., 2., 3., 4., 5.]), np.NAN))
def test_lag(self):
np_utils.assert_array_equal(np.array([0, 1, 2, 3, 4]),
utils.lag(np.array([1, 2, 3, 4, 5])))
def pandas_std_dev(close_prices, lookback):
prices = log(close_prices / utils.lag(close_prices))
return annualise(pd.rolling_std(prices, window=lookback))
def pandas_std_dev(close_prices, lookback):
prices = log(close_prices / utils.lag(close_prices))
return annualise(pd.rolling_std(prices, window=lookback))