Reference
Bradley, Jordan and Ritter (2003)
Background
After completion of an IPO, a firm is still considered to be in registration for an additional period of time. During this time, 25 days during our sample period which is termed the post-IPO "quiet" period, firms may make statements of fact regarding business developments and may respond to inquiries from analysts and shareholders regarding factual matters.
However, U.S. Securities and Exchange Commission (SEC) regulations generally prohibit firms and their underwriters from publishing opinions (known as recommendation) concerning valuation and from making forward-looking statements regarding earnings, revenues, and similar items.
We will using event study to see how the first recommendation for IPO/the end of the "quiet period" make effects to the stock return. Watch out that the two event may not be on the same day.
Summary
- Keep only the earliest recommendation for each IPO
- Supposed that the event is the first recommendation date for each IPO. Generate a table similar to Table II in the paper per type of recommendation by 3 groups: one for strong buys (IRECCD=1), one for buys (IRECCD=2) and one for the others (IRECCD=3, 4 or 5)
- Repeat 2) using the event as the end of the quiet period i.e. OFFER_DATE+25.
- Repeat 2) with two groups of IPOs, based on whether there are recommendations in the [-2,+2] window around the end of the quiet period.
- Interpret your figure with the results you had for parts 4) and 5)
Data Discription
IPOs_1996_2000: PERMNO (permno of the company), CUSIP (of the company), and OFFER_DATE (the date of the IPO).(p.s. PERMNO and CUSIP are both id of stocks in the U.S.A, but CUSIP would change depending on the company delisting, etc.,while PERMNO would never change)
Recommendation: CUSIP (identification of the company the recommendation refers to), ANNDATS (the date the recommendations was issued), and IRECCD (a code for the type of recommendations, defined as following: 1=STRONG BUY, 2=BUY, 3=HOLD, 4=UNDERPERFORM, and 5=SELL)
AssignmentReturns & Market_returns: return data of the stocks and value-weigthed index as the market portfolio return
Code
1
sort by announce date, and drop the repeating stocks by CUSIP
cd D:\Janis\Desktop\资产定价（投资学）\dataimport pandas as pd
import numpy as np
from scipy import stats
from scipy.stats import ttest_1samp
from datetime import timedelta
import matplotlib.pyplot as pltd_ipo = pd.read_sas('IPOs_1996_2000.sas7bdat')
d_rcm = pd.read_sas('Recommendation.sas7bdat')#按announce date排序，并去掉重复的CUSIP，就能得到股票首次被anno的日期的列表
d_rcm_by_date = d_rcm.sort_values(by = 'ANNDATS', ascending=True).drop_duplicates(subset=['CUSIP'], keep='first')
#把推荐等级IRECCD从b string转换成int，方便后续筛选
d_rcm_by_date['IRECCD'] = [int(s.decode()) for s in d_rcm_by_date['IRECCD']]#加入Offer_Date 股票发行日
d_rcm_by_date = pd.merge(pd.DataFrame(d_ipo, columns=['CUSIP', 'Offer_Date']), pd.DataFrame(d_rcm_by_date, columns=['CUSIP', 'IRECCD', 'ANNDATS']), on = 'CUSIP')
#后续分别以第一次anno、Offer_Date+25为event
d_rcm_by_date| CUSIP | Offer_Date | IRECCD | ANNDATS | |
|---|---|---|---|---|
| 0 | b’78668010′ | 1996-01-11 | 1 | 1996-02-23 |
| 1 | b’65539810′ | 1996-01-18 | 2 | 1996-05-23 |
| 2 | b’75642210′ | 1996-01-22 | 2 | 1996-02-21 |
| 3 | b’02109K10′ | 1996-01-23 | 1 | 1996-03-19 |
| 4 | b’30271210′ | 1996-01-23 | 1 | 1996-02-27 |
| … | … | … | … | … |
| 1846 | b’84749R10′ | 2000-12-07 | 2 | 2001-01-02 |
| 1847 | b’41690610′ | 2000-12-07 | 1 | 2001-02-26 |
| 1848 | b’02888410′ | 2000-12-07 | 1 | 2001-01-03 |
| 1849 | b’37246C10′ | 2000-12-11 | 2 | 2001-04-20 |
| 1850 | b’76122Q10′ | 2000-12-14 | 1 | 2001-01-09 |
1851 rows × 4 columns
2
calculate the delta days from the event day to the day to fill the table, like bucket sort
d_ret = pd.read_sas('AssignmentReturns.sas7bdat')
d_mret = pd.read_sas('Market_returns.sas7bdat')#d_merge_ret中存储全部日期、全部股票的return数据
d_merge_ret = pd.merge(pd.DataFrame(d_ret, columns=['CUSIP', 'PERMNO', 'RET', 'DATE']), pd.DataFrame(d_mret, columns=['date', 'VWRETD']), left_on = "DATE", right_on = "date")
d_merge_ret = d_merge_ret.drop(columns = 'date')
#计算market adjusted returns
d_merge_ret['mar'] = d_merge_ret['RET'] - d_merge_ret['VWRETD']#d_calc中存储股票编号、mar、IRECCD
d_calc = pd.merge(pd.DataFrame(d_merge_ret, columns = ['CUSIP', 'DATE', 'mar']), pd.DataFrame(d_rcm_by_date, columns = ['CUSIP', 'IRECCD', 'ANNDATS']), left_on = "CUSIP", right_on = "CUSIP")
#计算股票mar日期和第一次anno的相差天数delta，并转换为int
#第一个event
d_calc['delta'] = (d_calc['DATE'] - d_calc['ANNDATS']).dt.days
#因题目要求计算-5~5天内的，去除所有时间范围以外的数据
d_calc = d_calc.drop(d_calc[abs(d_calc['delta']) > 5].index)
#后续按IRECCD分组，按delta打印表格，所以可以去除日期
#相当于中心化操作
d_calc = d_calc.drop(columns = ['DATE', 'ANNDATS'])
d_calc
| CUSIP | mar | IRECCD | delta | |
|---|---|---|---|---|
| 27 | b’78668010′ | -0.008855 | 1 | -3 |
| 28 | b’78668010′ | 0.009342 | 1 | -2 |
| 29 | b’78668010′ | -0.024849 | 1 | -1 |
| 30 | b’78668010′ | -0.000660 | 1 | 0 |
| 31 | b’78668010′ | -0.019343 | 1 | 3 |
| … | … | … | … | … |
| 1025184 | b’76122Q10′ | 0.046886 | 1 | -1 |
| 1025185 | b’76122Q10′ | -0.029103 | 1 | 0 |
| 1025186 | b’76122Q10′ | -0.010368 | 1 | 1 |
| 1025187 | b’76122Q10′ | 0.034579 | 1 | 2 |
| 1025188 | b’76122Q10′ | 0.131998 | 1 | 3 |
def calc(k):
#储存题目要求的输出表格
d_ans = pd.DataFrame(columns = ['days', 'amar', 't', 'n'])
d_ans['days'] = range(-5,6)
d_ans.loc[:,:] = 0
#初始化days为-5~5
d_ans['days'] = range(-5,6)
#i计数目前处理到第几列，实际上就是row.index
i = 0
#对表格进行循环
for row in d_ans.iterrows():
#条件k是按推荐等级分组，delta == days是按距离event天数分组，取出符合条件的mar序列
d_temp = d_calc[(k) & (d_calc['delta'] == row[1]['days'])]['mar']
#求average mar
d_ans.loc[i, 'amar'] = d_temp.dropna().mean()
#求符合条件的个数n
df_bool = (k) & (d_calc['delta'] == row[1]['days'])
d_ans.loc[i, 'n'] = df_bool.sum()
#对取出的mar序列做t检验，原假设μ = 0
#注意ttest_1samp输出的[t, p]，这里取t
d_ans.loc[i, 't'] = ttest_1samp(d_temp.dropna(), 0)[0]
i = i+1
return d_ans
#分别传入对IRECCD的判断，以分成{1}，{2}，{3，4，5}三组
print(calc(d_calc['IRECCD'] == 1))
print(calc(d_calc['IRECCD'] == 2))
print(calc(d_calc['IRECCD'] > 2)) days amar t n
0 -5 0.004576 1.792604 504
1 -4 0.010044 3.246565 405
2 -3 0.009570 2.656499 316
3 -2 0.000710 0.229541 309
4 -1 0.010485 3.910741 500
5 0 0.004649 1.626139 690
6 1 0.002323 1.010052 611
7 2 -0.004036 -1.682755 502
8 3 0.001402 0.587696 411
9 4 -0.001336 -0.463270 316
10 5 -0.002402 -0.759221 309
days amar t n
0 -5 0.009697 3.146405 474
1 -4 0.014181 3.851839 369
2 -3 0.011604 2.456671 268
3 -2 0.000684 0.216472 281
4 -1 -0.000216 -0.071135 473
5 0 0.003992 1.329914 632
6 1 -0.001126 -0.395615 566
7 2 -0.001162 -0.392480 472
8 3 0.000293 0.089909 375
9 4 -0.007425 -1.801691 270
10 5 -0.000873 -0.236744 287
days amar t n
0 -5 0.030962 1.548338 23
1 -4 -0.020598 -0.935432 19
2 -3 0.012714 0.891346 19
3 -2 0.004483 0.486314 15
4 -1 -0.017472 -1.948986 23
5 0 -0.039365 -2.226424 33
6 1 -0.036419 -3.233579 29
7 2 -0.008507 -0.515271 24
8 3 -0.011686 -0.890740 20
9 4 0.008818 0.724564 19
10 5 -0.004680 -0.297206 173
#第二个event
d_calc = pd.merge(pd.DataFrame(d_merge_ret, columns = ['CUSIP', 'DATE', 'mar']), pd.DataFrame(d_rcm_by_date, columns = ['CUSIP', 'IRECCD', 'Offer_Date']), left_on = "CUSIP", right_on = "CUSIP")
#delta是有数据的日期与Offer_Date+25之差
d_calc['delta'] = (d_calc['DATE'] - d_calc['Offer_Date']).dt.days - 25
#去除不在[-5,5]内的数据
d_calc = d_calc.drop(d_calc[abs(d_calc['delta']) > 5].index)
d_calc = d_calc.drop(columns = ['DATE', 'Offer_Date'])
#第5题不要求对IRECCD进行分组，故直接传入k = True
print(calc(1)) days amar t n
0 -5 0.002639 1.548408 1172
1 -4 0.007390 4.409964 1329
2 -3 0.012567 7.104972 1129
3 -2 0.009144 3.996107 720
4 -1 0.014305 5.747104 584
5 0 0.007941 3.367472 732
6 1 0.004713 1.973140 915
7 2 0.000736 0.369180 1169
8 3 0.001751 1.047002 1315
9 4 -0.000756 -0.463374 1128
10 5 -0.001592 -0.764718 7205
#加入Offer_Date 股票发行日
d_rcm_by_date = pd.merge(pd.DataFrame(d_ipo, columns=['CUSIP', 'Offer_Date']), pd.DataFrame(d_rcm, columns=['CUSIP', 'IRECCD', 'ANNDATS']), on = 'CUSIP').drop_duplicates(subset=['CUSIP'], keep='first')
#后续分别以第一次anno、Offer_Date+25为event
d_rcm_by_date| CUSIP | Offer_Date | IRECCD | ANNDATS | |
|---|---|---|---|---|
| 0 | b’78668010′ | 1996-01-11 | b’1′ | 1996-02-23 |
| 5 | b’65539810′ | 1996-01-18 | b’2′ | 1996-05-23 |
| 9 | b’75642210′ | 1996-01-22 | b’2′ | 1996-02-21 |
| 25 | b’02109K10′ | 1996-01-23 | b’1′ | 1996-03-19 |
| 44 | b’30271210′ | 1996-01-23 | b’1′ | 1996-02-27 |
| … | … | … | … | … |
| 29828 | b’84749R10′ | 2000-12-07 | b’2′ | 2001-01-02 |
| 29836 | b’41690610′ | 2000-12-07 | b’1′ | 2001-02-26 |
| 29842 | b’02888410′ | 2000-12-07 | b’1′ | 2001-01-03 |
| 29851 | b’37246C10′ | 2000-12-11 | b’2′ | 2001-04-20 |
| 29855 | b’76122Q10′ | 2000-12-14 | b’2′ | 2001-01-10 |
be careful with which is the event!
d_calc = pd.merge(pd.DataFrame(d_merge_ret, columns = ['CUSIP', 'DATE', 'mar']), pd.DataFrame(d_rcm_by_date, columns = ['CUSIP', 'IRECCD', 'Offer_Date', 'ANNDATS']), left_on = "CUSIP", right_on = "CUSIP")
#delta还是用来代表和event的相差天数
d_calc['delta'] = (d_calc['DATE'] - d_calc['Offer_Date']).dt.days - 25
#题目要求判断推荐日anndats是否在距离发行日窗口期[-2,2]，以此分组
#设置in_window，在窗口期内的为True
d_calc['in_window'] = (((d_calc['ANNDATS'] - d_calc['Offer_Date']).dt.days - 25 <= 2) & ((d_calc['ANNDATS'] - d_calc['Offer_Date']).dt.days - 25 >= -2))
#与按IRECCD分组同理
print(calc(d_calc['in_window'] == True))
print(calc(d_calc['in_window'] == False)) days amar t n
0 -5 0.007484 2.365433 441
1 -4 0.013627 3.915532 439
2 -3 0.023586 5.659326 348
3 -2 0.016189 2.719038 169
4 -1 0.029160 4.005638 135
5 0 0.006211 1.283090 258
6 1 0.010800 2.563634 383
7 2 -0.002639 -0.668781 444
8 3 0.001917 0.530497 431
9 4 -0.001357 -0.421846 350
10 5 -0.003517 -0.721198 166
days amar t n
0 -5 -0.000283 -0.145392 731
1 -4 0.004314 2.378810 890
2 -3 0.007657 4.423083 781
3 -2 0.006983 2.954376 551
4 -1 0.009838 4.184526 449
5 0 0.008883 3.528722 474
6 1 0.000331 0.120076 532
7 2 0.002803 1.322881 725
8 3 0.001670 0.950041 884
9 4 -0.000486 -0.259439 778
10 5 -0.001015 -0.445447 5545
plt.figure(figsize = (16, 5))
plt.title("Figure 1")
cov = calc(d_calc['in_window'] == True)
un_cov = calc(d_calc['in_window'] == False)
plt.plot(cov.iloc[:,0], cov.iloc[:,1], 'r:', label = 'cov')
plt.plot(un_cov.iloc[:,0], un_cov.iloc[:,1], 'y--', label = 'un_cov')
plt.legend()
plt.show()Views: 8