Python 熊猫中每行的滚动计算

我尝试对数据帧的每一行执行以下测试和计算：

将过去10年的每日std作为长期波动率（L_vol）

将上个月的每日性病记录作为vol_hat查找

如果上个月的每日std在上五分位或下五分位，则将指示器（状态或st）值变为1，否则为0

将上述项目放入此公式中，以调整回报率： 缩放比例=拉伸最小值（1-st）*

我可以在每行上执行所有这些操作的唯一方法是通过以下循环

    # creat a function to see if we are in extreme state volatility
    def state(series):
       result=0
       last_vol = series[-21:].std()
       rolling_vol = series.rolling(21).std()
       if (last_vol <= rolling_vol.quantile(0.2)) or (last_vol >= rolling_vol.quantile(0.8)):
            result = 1
       return result

    # create 4 columns based on data from each day's observation
    df_spy['state'] = 0
    df_spy['L_vol'] = 0
    df_spy['sc_ret'] = 0
    df_spy['vol_hat'] =0
    for i in range(2500,len(spy)):
        ret = df_spy['spy'].iloc[i]
        serie = df_spy['spy'].iloc[:i]
        vol_hat = serie[-21:].std()
        df_spy['state'].iloc[i] = state(serie)
        df_spy['vol_hat'].iloc[i] = vol_hat
        df_spy['L_vol'].iloc[i] = df_spy['spy'].iloc[:i][-2500:].std()
        st = df_spy['state'].iloc[i]
        lv = df_spy['L_vol'].iloc[i]
        df_spy['sc_ret'].iloc[i] = (st*ret*min(lv/(vol_hat),2)) + (1-st)*ret

df_spy

Date    
1980-01-03  -0.005106
1980-01-04  0.012355
1980-01-07  0.002723
1980-01-08  0.020036
1980-01-09  0.000918
... ...
2020-10-26  -0.018590
2020-10-27  -0.003026
2020-10-28  -0.035288
2020-10-29  0.011947
2020-10-30  -0.012130
10297 rows × 1 columns