Causal Impact分析报错ValueError：指定日期不在数据索引中

我是一名初学者，尝试对股票数据进行简单的Causal Impact分析，但每次尝试绘制影响图时都会触发如下ValueError：

ValueError: 20201019 not present in input data index.

尽管打印DataFrame索引时能明确看到该日期，系统仍提示无法找到该日期。我已尝试了网上及文档中的所有方法但均无效，不知下一步该如何处理，恳请提供帮助。

完整代码 #

import yfinance as yf
from causalimpact import CausalImpact
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from statsmodels.tsa.stattools import adfuller

training_start = '2020-09-01'
training_end = '2020-10-19'
treatment_start = '2020-10-20'
treatment_end = '2020-10-23'
end_stock = '2020-10-24'

y = ['BTC-USD']
y = yf.download(tickers = y,
                start = training_start,
                end = end_stock,
                interval = '1d')
y = y['Adj Close'].rename('y')

stocks = ['ZAL.DE', 'SQ', 'CRSP', 'JD', 'DE', 'KTOS', 'GOOG', 'TPB']
x = yf.download(tickers = stocks,
                start = training_start,
                end = end_stock,
                interval = '1d')
x = x.iloc[:,:len(stocks)]

x.columns = x.columns.droplevel()

x.index = x.index.tz_localize(None)

df = pd.concat([y,x], axis=1).dropna()

df_training = df[df.index <= training_end]

test = adfuller(x = df_training.y)[1]

if test < 0.05:
    print('The time series is stationary')
else:
    print('The time series is not stationary')

differencing = df_training.pct_change().dropna()

test = adfuller(x = differencing.y)[1]

if test < 0.05:
    print('The time series is stationary')
else:
    print('The time series is not stationary')


plt.figure(figsize = (8,6))
sns.set(font_scale = 1.2)
sns.heatmap(differencing.corr(),
            annot = True,
            fmt = '.1g',
            cmap = 'YlOrBr',
            center =  True,
            linewidth = 1,
            linecolor = 'black')
#plt.show()

df_final = df.drop(columns = ['ZAL.DE'])


df_final = df.set_index(pd.date_range(start='2020-09-01', periods=len(df_final.index)))
pre_period = [pd.to_datetime(training_start), pd.to_datetime(training_end)]
post_period = [pd.to_datetime(treatment_start), pd.to_datetime(treatment_end)]

impact = CausalImpact(data=df_final, pre_period=pre_period, post_period=post_period)
impact.plot()

报错信息 #

Traceback (most recent call last):
  File "C:\Users\MGerdes\PycharmProjects\EconometricsCausalInference\Google Causal Impact.py", line 71, in <module>
    impact = CausalImpact(data=df_final, pre_period=pre_period, post_period=post_period)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\MGerdes\PycharmProjects\EconometricsCausalInference\server\Lib\site-packages\causalimpact\main.py", line 206, in __init__
    processed_input = cidata.process_input_data(data, pre_period, post_period,
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\MGerdes\PycharmProjects\EconometricsCausalInference\server\Lib\site-packages\causalimpact\data.py", line 120, in process_input_data
    pre_data, post_data = process_pre_post_data(fmt_data, pre_period, post_period)
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\MGerdes\PycharmProjects\EconometricsCausalInference\server\Lib\site-packages\causalimpact\data.py", line 266, in process_pre_post_data
    checked_pre_period = process_period(pre_period, data)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\MGerdes\PycharmProjects\EconometricsCausalInference\server\Lib\site-packages\causalimpact\data.py", line 389, in process_period
    raise ValueError("{point} not present in input data index.".format(
ValueError: 20201019 not present in input data index.

问题根源 #

你的代码存在两个关键错误，导致索引不匹配：

1. df_final赋值被覆盖：你先删除了ZAL.DE列得到df_final，但紧接着又用原始的df（包含ZAL.DE列）重新赋值覆盖了df_final，之前的列删除操作完全无效。
2. 错误重置索引：你用pd.date_range生成连续日期替换了原始的交易日索引。股票数据仅包含交易日（排除周末、节假日），而连续日期包含非交易日，导致你指定的training_end（2020-10-19）不在新的连续索引中。

修改步骤 #

1. 保留正确的df_final赋值，不要覆盖：
   删除错误的索引重置代码，保留原始的交易日索引（yfinance下载的索引已经是正确的datetime格式，且已移除时区信息）。
2. 确保pre_period和post_period的日期存在于df_final的索引中。

修改后的代码片段 #

替换原代码中df_final相关的部分为：

# 正确删除ZAL.DE列，保留原始索引
df_final = df.drop(columns=['ZAL.DE'])

# 直接使用原始交易日索引，不需要重置
pre_period = [pd.to_datetime(training_start), pd.to_datetime(training_end)]
post_period = [pd.to_datetime(treatment_start), pd.to_datetime(treatment_end)]

# 可选：验证日期是否在索引中
print(f"training_end在索引中: {pd.to_datetime(training_end) in df_final.index}")
print(f"treatment_start在索引中: {pd.to_datetime(treatment_start) in df_final.index}")

impact = CausalImpact(data=df_final, pre_period=pre_period, post_period=post_period)
impact.plot()

额外提示 #

• 股票数据存在缺失值时，dropna()会删除整行，导致索引不连续，这是正常现象，CausalImpact支持非连续的datetime索引。
• 运行前可以打印df_final.index确认所有需要的日期都存在。

内容的提问来源于stack exchange，提问作者mhgerd01