ar_utility.py 文件源码

def least_square_lagged_regression(u_array):
    """
    u_array, q, T+1, p
    """
    q,T,p = u_array.shape
    T -= 1
    # t0, t1 term is t1 regressed on t0
    lagged_coef_mat = np.zeros([T,T],dtype = np.object)
    for t0 in range(T):
        for t1 in range(t0,T):
            tmp_coef = np.zeros([p,p])
            for i in range(p):
                # least square regression u_t+h[i]  u_t
                tmp_y = u_array[:,t1+1,i]
                tmp_x = u_array[:,t0,:]
                # (X'X)^{-1} X' Y
                tmp_coef[i,:] = np.linalg.inv(tmp_x.T.dot(tmp_x)).dot(tmp_x.T.dot(tmp_y))

            lagged_coef_mat[t0,t1] = tmp_coef

    return lagged_coef_mat