股票数据的日历效应：月份效应

最新推荐文章于 2024-11-14 13:48:50 发布

finance褪黑素

最新推荐文章于 2024-11-14 13:48:50 发布

阅读量1.1k

点赞数 26

文章标签： python 数据挖掘大数据金融机器学习

本文链接：https://blog.csdn.net/celiaweiwei/article/details/136321990

版权

日历效应（二）：月份效应

文章目录

日历效应（二）：月份效应

完整代码和数据可关注gzh’finance褪黑素’回复关键词【2002】免费+无套路获取！

1. 获取数据

import baostock as bs
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from datetime import datetime, date

# 登陆系统
import baostock as bs
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from datetime import datetime, date
lg = bs.login()
code = 'sz.399006' 
start = '2018-01-01'
end = '2023-1-01'
# 获取指数基金指数历史数据
# 沪深300指数
hs300_price = bs.query_history_k_data_plus(code, "date,code,open,high,low,close,preclose,pctChg",
              start_date=start, end_date=end, frequency="d")
# 整合为DataFrame格式
data_list = []
while (hs300_price.error_code == '0') & hs300_price.next():
    data_list.append(hs300_price.get_row_data())
hs300 = pd.DataFrame(data_list, columns=hs300_price.fields)
hs300.to_csv('hs30.csv')
hs300.to_excel('hs30.xlsx')

hs300=pd.read_csv('hs30.csv',index_col='date')

del hs300['code']
hs300.head()

2.描述性统计分析

# from pandas_datareader import data
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
# 导入数据
hs300 =pd.read_csv("hs300.csv")
del hs300['code']
# 标准化数据格式
hs300["dates"]=pd.to_datetime(hs300["date"])
# 获取月份的数据
hs300["month"] = hs300["dates"].dt.month_name()
# 月份字符串转换成数值
days ={'January':1,'February':2,'March':3,'April':4,'May':5,
       'June':6,'July':7,'August':8,'September':9,'October':10,
       'November':11,'December':12}
hs300["month"] = hs300["month"].apply(lambda x: days[x])
# 计算对数收益率
hs300_close=hs300['close']
hs300_lgreturn=np.log(hs300_close)-np.log(hs300_close.shift(1))
# 计算收益率
hs300_return=hs300_close/hs300_close.shift(1)
# 分组描述
df1=pd.DataFrame(hs300_lgreturn.groupby(hs300["month"]).describe()) 
df2=pd.DataFrame(hs300_lgreturn.describe())
df3=pd.DataFrame(np.array(df2).reshape(1,8))
df3.columns=['count','mean','std','min','25%','50%','75%','max']
df3.index=['All']
hh=pd.concat([df1,df3],axis=0)

# 偏度、峰度、标准差、J-B统计量
import numpy as np
import scipy.stats as stats

def self_JBtest(y):
    # 样本规模n
    n = y.size
    y_ = y - y.mean()
    """
    M2:二阶中心钜
    skew 偏度 = 三阶中心矩 与 M2^1.5的比
    krut 峰值 = 四阶中心钜 与 M2^2 的比
    """
    M2 = np.mean(y_**2)
    skew =  np.mean(y_**3)/M2**1.5
    krut = np.mean(y_**4)/M2**2

    """
    计算JB统计量，以及建立假设检验
    """
    JB = n*(skew**2/6 + (krut-3 )**2/24)
    pvalue = 1 - stats.chi2.cdf(JB,df=2)
    data={'偏度':skew,'峰值':krut,'JB检验':JB,'P值':pvalue}
    data1=pd.DataFrame(data,index=[1])
    return data1

a1=pd.DataFrame()
a2=self_JBtest(hs300_lgreturn)
a2.index=['All']
for i in range(1,13):
    data=hs300[hs300['month']==i]
    w_close=data["close"]
    ret = np.log(w_close) - np.log(w_close.shift(1))  # 这里用对数收益率，因为对数收益率的统计特性比较好
    a=self_JBtest(ret)
    a1=pd.concat([a,a1],axis=0)
a1.index=[1,2,3,4,5,6,7,8,9,10,11,12]
result=pd.concat([a1,a2])
result

	偏度	峰值	JB检验	P值
1	1.639012	36.181934	4169.198151	0.000000e+00
2	1.679215	33.155897	3299.021498	0.000000e+00
3	1.083555	26.000192	1467.689222	0.000000e+00
4	4.808498	40.821891	5266.979476	0.000000e+00
5	6.327881	50.633985	8906.937585	0.000000e+00
6	4.094596	34.286906	3878.669472	0.000000e+00
7	0.746986	24.804115	1612.073478	0.000000e+00
8	1.155379	22.347403	1186.442555	0.000000e+00
9	3.695878	29.039659	2503.398098	0.000000e+00
10	5.440282	40.069552	5597.018945	0.000000e+00
11	5.579888	43.347194	6279.564813	0.000000e+00
12	4.809296	33.516392	3967.098890	0.000000e+00
All	-0.289843	4.182104	72.875276	1.110223e-16

ds=pd.concat([result,hh],axis=1)
del ds['25%']
del ds['50%']
del ds['75%']
ds1=pd.DataFrame()
for i in range(13):
    ds0=pd.DataFrame(np.array(ds[i:i+1]).reshape(9,1))
    ds1=pd.concat([ds1,ds0],axis=1)
ds1.columns=['January','February','March','April','May',
       'June','July','August','September','October',
       'November','December','All']
ds1.index=ds.columns
ds1

	January	February	March	April	May	June	July	August	September	October	November	December	All
偏度	1.639012	1.679215	1.083555	4.808498	6.327881	4.094596	0.746986	1.155379	3.695878	5.440282	5.579888	4.809296	-2.898430e-01
峰值	36.181934	33.155897	26.000192	40.821891	50.633985	34.286906	24.804115	22.347403	29.039659	40.069552	43.347194	33.516392	4.182104e+00
JB检验	4169.198151	3299.021498	1467.689222	5266.979476	8906.937585	3878.669472	1612.073478	1186.442555	2503.398098	5597.018945	6279.564813	3967.098890	7.287528e+01
P值	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	1.110223e-16
count	92.000000	86.000000	90.000000	82.000000	75.000000	81.000000	89.000000	88.000000	83.000000	66.000000	86.000000	90.000000	1.008000e+03
mean	0.000918	0.001794	-0.001471	0.000430	0.000300	0.004687	0.001271	-0.001193	-0.001857	0.001200	0.000747	0.001627	6.821462e-04
std	0.016128	0.022321	0.024197	0.019176	0.019264	0.015136	0.020374	0.015235	0.015437	0.015616	0.012688	0.012278	1.770710e-02
min	-0.036996	-0.070913	-0.060817	-0.057160	-0.082707	-0.042739	-0.063340	-0.037100	-0.049191	-0.037819	-0.033623	-0.030070	-8.270743e-02
max	0.038305	0.053526	0.050713	0.053708	0.042889	0.038466	0.051846	0.034384	0.039796	0.038334	0.031538	0.035925	5.370838e-02

ds

	偏度	峰值	JB检验	P值	count	mean	std	min	max
1	1.639012	36.181934	4169.198151	0.000000e+00	92.0	0.000918	0.016128	-0.036996	0.038305
2	1.679215	33.155897	3299.021498	0.000000e+00	86.0	0.001794	0.022321	-0.070913	0.053526
3	1.083555	26.000192	1467.689222	0.000000e+00	90.0	-0.001471	0.024197	-0.060817	0.050713
4	4.808498	40.821891	5266.979476	0.000000e+00	82.0	0.000430	0.019176	-0.057160	0.053708
5	6.327881	50.633985	8906.937585	0.000000e+00	75.0	0.000300	0.019264	-0.082707	0.042889
6	4.094596	34.286906	3878.669472	0.000000e+00	81.0	0.004687	0.015136	-0.042739	0.038466
7	0.746986	24.804115	1612.073478	0.000000e+00	89.0	0.001271	0.020374	-0.063340	0.051846
8	1.155379	22.347403	1186.442555	0.000000e+00	88.0	-0.001193	0.015235	-0.037100	0.034384
9	3.695878	29.039659	2503.398098	0.000000e+00	83.0	-0.001857	0.015437	-0.049191	0.039796
10	5.440282	40.069552	5597.018945	0.000000e+00	66.0	0.001200	0.015616	-0.037819	0.038334
11	5.579888	43.347194	6279.564813	0.000000e+00	86.0	0.000747	0.012688	-0.033623	0.031538
12	4.809296	33.516392	3967.098890	0.000000e+00	90.0	0.001627	0.012278	-0.030070	0.035925
All	-0.289843	4.182104	72.875276	1.110223e-16	1008.0	0.000682	0.017707	-0.082707	0.053708

for i in range(1,13):
    bel=['January','February','March','April','May',
       'June','July','August','September','October',
       'November','December']
    fig, ax = plt.subplots()
    plt.axhline(y=0,c="orange",ls="--",lw=2)
    data=hs300[hs300['month']==i]
    w_close=data["close"]
    ret = np.log(w_close) - np.log(w_close.shift(1))  # 这里用对数收益率，因为对数收益率的统计特性比较好
    y=ret
    ax.plot(y, linewidth=2.0,label=bel[i-1])
    ax.legend(loc=2) 
    save=['m1.svg','m2.svg','m3.svg','m4.svg','m5.svg','m6.svg','m7.svg','m8.svg',
         'm9.svg','m10.svg','m11.svg','m12.svg']
    plt.grid()
    plt.savefig(save[i-1])
    plt.show()

在这里插入图片描述

3.虚拟变量回归

import statsmodels.api as sm

# 设置虚拟变量
all_data = []
for x in hs300["month"]:
    item={
    "Jan": 1 if x==1 else 0,
    "Feb": 1 if x==2 else 0,
    "Mar": 1 if x==3 else 0,
    "Apr": 1 if x==4 else 0,
    "May": 1 if x==5 else 0,
    "June": 1 if x==6 else 0,
    "July": 1 if x==7 else 0,
    "Aug": 1 if x==8 else 0,       
    "Sep": 1 if x==9 else 0,        
    "Oct": 1 if x==10 else 0,        
    "Nov": 1 if x==11 else 0,        
    "Dec": 1 if x==12 else 0                
    }
    all_data.append(item)

all_data = pd.DataFrame(data=all_data)
r=pd.DataFrame(hs300_lgreturn)
X=all_data[['Jan','Feb','Mar','Apr','May',
       'June','July','Aug','Sep','Oct',
       'Nov','Dec']]
X.index = r.index
df=pd.concat([r,X],axis=1)

# 处理空值
import missingno
# missingno.matrix(df)
# 直接剔除
df = df.dropna()
missingno.matrix(df)
plt.show()

在这里插入图片描述

X=df[['Jan','Feb','Mar','Apr','May',
       'June','July','Aug','Sep','Oct',
       'Nov','Dec']]
Y=df[["close"]]
from statsmodels.compat import scipy
import statsmodels.api as sm
X2 = sm.add_constant(X)
est = sm.OLS(Y*100, X2).fit()
print(est.summary())

                            OLS Regression Results                            
==============================================================================
Dep. Variable:                  close   R-squared:                       0.009
Model:                            OLS   Adj. R-squared:                 -0.002
Method:                 Least Squares   F-statistic:                    0.8154
Date:                Sat, 27 May 2023   Prob (F-statistic):              0.625
Time:                        03:07:54   Log-Likelihood:                -2001.2
No. Observations:                1008   AIC:                             4026.
Df Residuals:                     996   BIC:                             4085.
Df Model:                          11                                         
Covariance Type:            nonrobust                                         
==============================================================================
                 coef    std err          t      P>|t|      [0.025      0.975]
------------------------------------------------------------------------------
const          0.0650      0.052      1.257      0.209      -0.037       0.167
Jan            0.0268      0.178      0.151      0.880      -0.322       0.375
Feb            0.1144      0.183      0.624      0.533      -0.245       0.474
Mar           -0.2121      0.179     -1.182      0.238      -0.564       0.140
Apr           -0.0220      0.187     -0.118      0.906      -0.390       0.346
May           -0.0350      0.195     -0.179      0.858      -0.418       0.348
June           0.4037      0.188      2.143      0.032       0.034       0.773
July           0.0620      0.180      0.344      0.731      -0.292       0.416
Aug           -0.1843      0.181     -1.016      0.310      -0.540       0.172
Sep           -0.2508      0.186     -1.346      0.179      -0.616       0.115
Oct            0.0550      0.207      0.265      0.791      -0.352       0.462
Nov            0.0097      0.183      0.053      0.958      -0.350       0.369
Dec            0.0977      0.179      0.544      0.586      -0.254       0.450
==============================================================================
Omnibus:                       38.999   Durbin-Watson:                   1.998
Prob(Omnibus):                  0.000   Jarque-Bera (JB):               73.064
Skew:                          -0.268   Prob(JB):                     1.36e-16
Kurtosis:                       4.205   Cond. No.                     1.48e+15
==============================================================================

Notes:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.
[2] The smallest eigenvalue is 5.01e-28. This might indicate that there are
strong multicollinearity problems or that the design matrix is singular.

4.arch-garch模型拟合波动率

import arch
am = arch.arch_model(y=Y*100,x=X,mean='LS',lags=0,vol='GARCH') 
res1 = am.fit()
res1.summary()

Least Squares - GARCH Model Results
Dep. Variable:	close	R-squared:	0.006
Mean Model:	Least Squares	Adj. R-squared:	-0.006
Vol Model:	GARCH	Log-Likelihood:	-1966.25
Distribution:	Normal	AIC:	3964.50
Method:	Maximum Likelihood	BIC:	4043.15
		No. Observations:	1008
Date:	Sat, May 27 2023	Df Residuals:	995
Time:	03:07:57	Df Model:	13

Mean Model
	coef	std err	t	P>\|t\|	95.0% Conf. Int.
Const	0.0797	0.115	0.691	0.490	[ -0.146, 0.306]
Jan	0.1310	0.241	0.542	0.588	[ -0.342, 0.604]
Feb	-0.1013	0.272	-0.373	0.709	[ -0.634, 0.432]
Mar	-0.2149	0.278	-0.773	0.440	[ -0.760, 0.330]
Apr	0.2005	0.255	0.786	0.432	[ -0.300, 0.701]
May	-0.0408	0.149	-0.274	0.784	[ -0.333, 0.251]
June	0.4049	0.200	2.025	4.291e-02	[1.292e-02, 0.797]
July	-0.0220	0.164	-0.134	0.893	[ -0.344, 0.300]
Aug	-0.1780	0.199	-0.897	0.370	[ -0.567, 0.211]
Sep	-0.2264	0.217	-1.043	0.297	[ -0.652, 0.199]
Oct	0.0848	0.184	0.462	0.644	[ -0.275, 0.445]
Nov	-9.8481e-03	0.147	-6.683e-02	0.947	[ -0.299, 0.279]
Dec	0.0521	0.167	0.312	0.755	[ -0.275, 0.379]

Volatility Model
	coef	std err	t	P>\|t\|	95.0% Conf. Int.
omega	0.0841	3.683e-02	2.284	2.240e-02	[1.192e-02, 0.156]
alpha[1]	0.0813	2.055e-02	3.956	7.629e-05	[4.102e-02, 0.122]
beta[1]	0.8936	2.287e-02	39.081	0.000	[ 0.849, 0.938]

Covariance estimator: robust

res1.plot()
plt.plot(Y*100)
plt.savefig("residuals.svg")

在这里插入图片描述

am = arch.arch_model(y=Y*100,x=X,mean='AR',lags=0,vol='GARCH') 
res1 = am.fit()
res1.summary()
res1.hedgehog_plot()
plt.savefig("heg2.svg")

在这里插入图片描述

5.使用arch-garch进行波动性消除以进一步验证星期效应的稳健性

Y=res1.conditional_volatility
df=pd.concat([Y,X2],axis=1)
missingno.matrix(df)
plt.show()
df = df.dropna()
df=df.values
X=df[:,1:14]
Y=df[:,0]
from statsmodels.compat import scipy
import statsmodels.api as sm
est2 = sm.OLS(Y, X).fit()
print(est2.summary())

在这里插入图片描述
P值部分小于0.05 说明模型部分参数是显著的存在影响，当前三、四、五、十、十一月份效应是初步存在的。但R2依然不高后续可考虑使用滚动样本检验进一步进行稳健性分析。

参考文献：
[1]张兵.中国股市日历效应研究:基于滚动样本检验的方法[J].金融研究,2005(07):33-44.
[2]焦璇琨,李从欣.中国股票市场周内效应的实证研究[J].统计与管理,2020,35(02):64-68.DOI:10.16722/j.issn.1674-537x.2020.02.012.