def get_har_forecast(dependent, lags, dist):
import sys
har_test = HARX(y=dependent, lags=lags, distribution=dist, rescale=True)
index = dependent.index
end_loc = np.where(index >= '2017-01-03')[0].min(
) #returns where index is greater than a date.
forecasts = {}
for i in range(
890): #878 values between the start and end of the OS dataset.
sys.stdout.write('.')
sys.stdout.flush()
res = har_test.fit(first_obs=i,
last_obs=i + end_loc,
disp='off',
options={
'maxiter': 10000000,
'eps': 0.01,
'ftol': 0.0001
})
#https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.fmin_slsqp.html
temp = res.forecast(horizon=1).mean
fcast = temp.iloc[i + end_loc - 1]
forecasts[fcast.name] = fcast
har_forecast = (pd.DataFrame(forecasts).T)
return (har_forecast)
def test_forecast_exogenous_regressors(self):
y = 10 * self.rng.randn(1000, 1)
x = self.rng.randn(1000, 2)
am = HARX(y=y, x=x, lags=[1, 2])
res = am.fit()
fcasts = res.forecast(horizon=1, start=1)
const, har01, har02, ex0, ex1, _ = res.params
y_01 = y[1:-1]
y_02 = (y[1:-1] + y[0:-2]) / 2
x0 = x[2:, :1]
x1 = x[2:, 1:2]
direct = const + har01 * y_01 + har02 * y_02 + ex0 * x0 + ex1 * x1
direct = np.vstack(([[np.nan]], direct, [[np.nan]]))
direct = pd.DataFrame(direct, columns=["h.1"])
assert_allclose(np.asarray(direct), fcasts.mean)
fcasts2 = res.forecast(horizon=2, start=1)
assert fcasts2.mean.shape == (1000, 2)
assert fcasts2.mean.isnull().all()["h.2"]
assert_frame_equal(fcasts.mean, fcasts2.mean[["h.1"]])
def test_forecast_exogenous_regressors(self):
y = self.rng.randn(1000, 1)
x = self.rng.randn(1000, 2)
am = HARX(y=y, x=x, lags=[1, 2])
res = am.fit()
fcasts = res.forecast(horizon=1, start=1)
const, har01, har02, ex0, ex1, _ = res.params
y_01 = y[1:-1]
y_02 = (y[1:-1] + y[0:-2]) / 2
x0 = x[2:, :1]
x1 = x[2:, 1:2]
direct = const + har01 * y_01 + har02 * y_02 + ex0 * x0 + ex1 * x1
direct = np.vstack(([[np.nan]], direct, [[np.nan]]))
direct = pd.DataFrame(direct, columns=['h.1'])
assert_allclose(direct.values, fcasts.mean)
fcasts2 = res.forecast(horizon=2, start=1)
assert fcasts2.mean.shape == (1000, 2)
assert fcasts2.mean.isnull().all()['h.2']
assert_frame_equal(fcasts.mean, fcasts2.mean[['h.1']])
HARCH,
HARX,
ConstantMean,
ConstantVariance,
EWMAVariance,
MIDASHyperbolic,
RiskMetrics2006,
ZeroMean,
arch_model,
)
from arch.univariate.mean import _ar_forecast, _ar_to_impulse
SP500 = 100 * sp500.load()["Adj Close"].pct_change().dropna()
MEAN_MODELS = [
HARX(SP500, lags=[1, 5]),
ARX(SP500, lags=2),
ConstantMean(SP500),
ZeroMean(SP500),
]
VOLATILITIES = [
ConstantVariance(),
GARCH(),
FIGARCH(),
EWMAVariance(lam=0.94),
MIDASHyperbolic(),
HARCH(lags=[1, 5, 22]),
RiskMetrics2006(),
APARCH(),
EGARCH(),
def model(self):
return HARX(**self.__params, lags=self.__lags)
def __bic_with(self, lag_list):
model = HARX(**self.__params, lags=lag_list)
result = model.fit(update_freq=0, disp='off')
return round(result.bic, 4)