CSDN-用Python底层编写进行计量经济分析（三）

两两$x_i$和$x_j$交互项
$$e_i^2={\alpha }_0+{\alpha }_1{X}_{5i}+{\alpha }_2{X}_{6i}+\cdots +{\alpha }_8{X}_{13i}^{\prime }+{\alpha }_9{X}_{5i}^2+\cdots +{\alpha }_{16}{X}_{13i}^{\prime 2}+{\alpha }_{17}{X}_{1i}{X}_{2i}+\cdots +{\alpha }_{44}{X}_{12}{X}_{13}^{\prime }+{\epsilon }_i$$

$$H_0:{\alpha }_i=0,i=1,2,\cdots ,44;H_1:{\alpha }_i不全为0$$
统计量
$$WT(p)=n{R}^2\sim {\chi }^2(p)$$

我的代码写错了，这阵子忙完再改↓

import statsmodels.api as sm   # 线性回归
def white_test(x, e):
    '''
    x: 因变量
    e: 残差 e = y - y_predict
    '''
    xx = copy.deepcopy(x)
    for n, i in enumerate(x.columns[:-1]):
        for j in x.columns[n :-1]:
            xx['%s*%s'%(i,j)] = x[i]*x[j]
    print(xx.columns)
    # OLS-Regreession
    x1 = copy.deepcopy(xx)
    x1.insert(0, 'constant', 1)   # 添加常数项

    model = sm.OLS(e, x1) 
    result_wt = model.fit()
    
    # 计算
    wt = len(xx)*result_wt.rsquared   # result_wt.rsquared 可决系数
    # 1- stats.chi2.cdf(wt, x.shape[0])  p值
    chi = stats.chi2.ppf(1-0.5, df=xx.shape[1])
    print('卡方在0.95分位数的临界值是%.2f'%chi)
    if wt < chi:
        print("计算出来的wt统计量是%.2f, 接受H0，认为不存在异方差"%wt)
    elif wt > stats.chi.ppf(1-0.5, df=xx.shape[1]):
        print("计算出来的wt统计量是%.2f, 拒绝H0， 接受H1，认为存在异方差"%wt)
    return stats.chi.cdf(wt, df=xx.shape[1])

if __name__ == '__main__':
	x = pd.DataFrame(data=np.arange(20).reshape(10,2), columns=['x1','x2'])
	e = ...  # e为上一次回归计算出来的残差
	white_test(x, e)