def test_attribute_non_profit_series(self):
"""Test attributing series quantities besides profit.
"""
# Alpha source
alpha_sources = [ReturnsForecast(
self.returns, name=i) for i in range(3)]
weights = np.array([0.1, 0.3, 0.6])
alpha_model = MultipleReturnsForecasts(alpha_sources, weights)
emp_Sigma = np.cov(self.returns.as_matrix().T)
risk_model = FullSigma(emp_Sigma, gamma=100.)
tcost_model = TcostModel(self.a, self.b, self.sigma, self.volume)
hcost_model = HcostModel(self.s, self.s * 0)
pol = SinglePeriodOpt(alpha_model, [risk_model, tcost_model,
hcost_model], [],
solver=cvx.ECOS)
tcost = TcostModel(self.volume, self.sigma, self.a, self.b)
hcost = HcostModel(self.s)
market_sim = simulator.MarketSimulator(self.returns,
costs=[tcost, hcost],
market_volumes=self.volume)
p_0 = pd.Series(index=self.universe, data=1E6)
noisy = market_sim.run_backtest(p_0, self.returns.index[1],
self.returns.index[10], pol)
# Select tcosts.
def selector(result):
return result.leverage
# linear fit attribution
attr = market_sim.attribute(noisy, pol, selector,
parallel=False, fit="linear")
base_line = noisy.leverage
for i in range(3):
self.assertItemsAlmostEqual(
attr[i] / weights[i] / sum(p_0), base_line / sum(p_0))
self.assertItemsAlmostEqual(attr['RMS error'], np.zeros(len(noisy.v)))
# least-squares fit attribution
attr = market_sim.attribute(noisy, pol, selector,
parallel=False, fit="least-squares")
for i in range(3):
self.assertItemsAlmostEqual(
attr[i] / weights[i] / sum(p_0), base_line / sum(p_0))
# Residual always 0.
alpha_sources = [ReturnsForecast(
self.returns * 0, name=i) for i in range(3)]
weights = np.array([0.1, 0.3, 0.6])
alpha_model = MultipleReturnsForecasts(alpha_sources, weights)
pol = copy.copy(pol)
pol.alpha_model = alpha_model
attr = market_sim.attribute(noisy, pol, selector,
parallel=False, fit="least-squares")
self.assertItemsAlmostEqual(attr['residual'], np.zeros(len(noisy.v)))
def setUp(self):
self.sigma = pd.read_csv(DIR + 'sigmas.csv',
index_col=0, parse_dates=[0])
self.returns = pd.read_csv(DIR + 'returns.csv',
index_col=0, parse_dates=[0])
self.volume = pd.read_csv(DIR + 'volumes.csv',
index_col=0, parse_dates=[0])
self.a, self.b, self.s = 0.0005, 1., 0.
self.portfolio = pd.Series(index=self.returns.columns, data=1E6)
self.tcost_term = TcostModel(self.a, self.b, self.sigma, self.volume)
self.hcost_term = HcostModel(self.s)
self.Simulator = MarketSimulator(self.returns,
costs=[self.tcost_term,
self.hcost_term],
market_volumes=self.volume)