INNER CODE UNIT · Python
ll
dmarienko/QuantTradingResearch · models/kalman_filter_adjust.py:20
ll = -0.5*(N * np.log(2*np.pi) + np.log(ev) + err**2/ev).sum()
if ll > ll_max:
ll_max = ll
best = p
best
b, err, evars = kalman_regression_estimator(X[constituents].values, Y.values, **best, intercept=False)
cut_off = slice(15, None)
e_s = pd.Series(err, index=Y.index)[cut_off]
v_s = pd.Series(np.sqrt(evars), index=Y.index)[cut_off]
b_a = pd.DataFrame(b.T, index=Y.index, columns=constituents)[cut_off]
b_a_w = b_a.div(b_a.sum(axis=1), axis=0)
fig(16, 5)
k_prediction = pd.DataFrame(X[constituents].values * b.T, index=Y.index)[cut_off].sum(axis=1)
plt.plot(k_prediction)
plt.plot(Y, 'g'); plt.title('Kalman estimator tracking Russell1000');