# What Can One Million Regressions Tell Us About IPO

Document Sample

```					                        Introduction
Data
Estimation Results
Conclusion

What Can One Million Regressions Tell Us

Authors: Alexander Butler, Michael Keefe, and Robert
Kieschnick

The University of Texas at Dallas

UTD Brown Bag
March 11, 2009

Butler, Keefe, and Kieschnick   One Million Regressions
Introduction
Data
Estimation Results
Conclusion

Outline
1   Introduction
Motivation
Fragility
Method & Measures
Approach
2   Data
Explanatory Variables
Sample Construction
Sample Statistics
3   Estimation Results
OLS Results
Outliers and Price Support
OLS vs Quantile
4   Conclusion
Butler, Keefe, and Kieschnick   One Million Regressions
Introduction    Motivation
Data     Fragility
Estimation Results    Method & Measures
Conclusion     Approach

Motivation

In our view, there is no single dominant theoretical
cause for underpricing. Thus, it is not so much a
matter of which model is right, but more a matter of
the relative importance of different models . . . To date,
there has been little empirical work attempting to
quantify the relative importance of different
explanations of underpricing.

Ritter and Welch (2002)

Butler, Keefe, and Kieschnick   One Million Regressions
Introduction    Motivation
Data     Fragility
Estimation Results    Method & Measures
Conclusion     Approach

Testing for fragility to speciﬁcation changes

Explanatory variables based on theoretical and empirical
literature
These variables constitute our priors
Which variables are robust to changes in priors? (ie.
variable selection)
Our tests draw upon work by Leamer(1983,1985) and
Sala-I-Martin(1997)

Butler, Keefe, and Kieschnick   One Million Regressions
Introduction    Motivation
Data     Fragility
Estimation Results    Method & Measures
Conclusion     Approach

InitialReturn = αj + Aβa,j + Z βz,j + X βx,j +

A ≡ Two variables run in every regression
Z ≡ Variable of Interest
X ≡ Remaining variables
M ≡ all possible combinations of 3 variables taken from X
j=1,2, . . . M subsets of variables
Method
1   Choose a variable of interest Z
1   Select a combination of 3 variables from X
2   Perform estimation and save the coefﬁcient and robust
standard errors associated with Z
3   Repeat for all possible combinations of 3 from X variables
(10,660 estimates)
2   Repeat for each Z variable

Butler, Keefe, and Kieschnick   One Million Regressions
Introduction   Motivation
Data    Fragility
Estimation Results   Method & Measures
Conclusion    Approach

Measures of robustness to speciﬁcation change

For each variable of interest, Z, we generate 10,660
ˆ
estimates of βz,j
Generate three robustness measures
1   Extreme Bounds Test - Variable deemed robust if all 10,660
coefﬁcient estimates have the same sign within 2 standard
deviations
2   Extreme Bounds Percent - Percent of the 10,660 coefﬁcient
ˆ
estimates, βz,j , that have the same sign within 2 standard
deviations
3   Sala-I-Martin Probability ≡ SiM
ˆ       1      M ˆ
βz =    M      j=1 βz,j
M
ˆ2
σz =    1
M
ˆ2
j=1 σz,j
Compute probability coefﬁcient doesn’t change sign

Butler, Keefe, and Kieschnick   One Million Regressions
Introduction    Motivation
Data     Fragility
Estimation Results    Method & Measures
Conclusion     Approach

To show economic magnitude as well as account for
outliers, price support and time variation we:
Normalize non-dummy variables to provide economic
magnitudes of coefﬁcients
ˆ
Estimation Methods βz,j using:
1   OLS
2   Least Absolute Deviation (Quantile Regression) - Adjusts
for outliers and mixture distribution.
Winsorize sample in increments of .005 to determine effect
of outliers
Break sample into periods following Loughran and Ritter
(2004)
1   90S (1993-1998)
2   Bubble (1998-2000)
3   Post (2000-2006)
Butler, Keefe, and Kieschnick   One Million Regressions
Introduction
Explanatory Variables
Data
Sample Construction
Estimation Results
Sample Statistics
Conclusion

What variables have been used in the literature to
explain ﬁrst day return

Compiled list of papers and variables
Grouped variables into:
1   Firm Economic Characteristics (e.g. Firm Age)
2   Market Characteristics (e.g. Matched Industry Return)
3   Firm Ownership Characteristics (e.g. Secondary Shares
Sold)
4   Underwriter Characteristics (e.g. Investment Bank
Reputation)
Many variables have slight variations in name and
deﬁnition
Select 44 variables

Butler, Keefe, and Kieschnick   One Million Regressions
Introduction
Explanatory Variables
Data
Sample Construction
Estimation Results
Sample Statistics
Conclusion

Sample Construction: 1993-2006

1   Merge SDC and CRSP ⇒ 5,554 observations
2   Retain Ordinary Common Shares ⇒ 4,631 observations
3   Merge Jay Ritter Data
Founding Date ⇒ 4,319 out of 4,631
Internet Business ⇒ 522/589 out of 4,631
Share Class ⇒ 322/403 out of 4,631

4   Merge Time Series Variables
Rolling 30 Day Return for 49 Industries - Ken French Web Site
Rolling 30 Day Return or CRSP market indexes
Average IPO Return and Number of IPOs - Ritter

5   Merge with Dataset from Kieschnick et al. ⇒ 2,856 out of 4,631
Compustat, TAQ, and LexisNexus
Drop ﬁnancials, IPO with offer prices less than \$5, and foreign ﬁrms

Butler, Keefe, and Kieschnick   One Million Regressions
Introduction
Explanatory Variables
Data
Sample Construction
Estimation Results
Sample Statistics
Conclusion

Variable Changes over Sample Period
Sample Means
Variable                                     90S   Bubble Post
First Day Return                            16.27 68.85 12.51
Ln of 1st Day Avg Trd Size                   7.97   6.70    7.11
Offer Revision from Orgnl Flng               0.29  15.56   -4.03
Ln(1 + Shrs Rtnd/Shrs Ofrd)                  1.18   1.65    1.21
Ln of Headline Instances                     1.73   2.84    2.78
Prior 30 Day Industry Rtrn+                  3.79   4.70    3.36
NASDAQ Dummy                                 0.86   0.93    0.74
Investment Bank Reputation                   7.07   8.16    7.49
VC Backed Dummy                              0.39   0.65    0.44
JR Internet Dummy                            0.03   0.45    0.10
Ln of IPO Firm Age                           2.22   1.99    2.58
SDC High Tech Dummy                          0.50   0.83    0.52
Ln of Firm Sales                             3.45   2.73    4.43

Butler, Keefe, and Kieschnick   One Million Regressions
Introduction
OLS Results
Data
Outliers and Price Support
Estimation Results
OLS vs Quantile
Conclusion

OLS Results for 1993-1998 using 44 variables

1993-1998 (90S)
Extreme
Mean     SiM    Bounds
Variable                               Beta    Prob    Percent
Ln of 1st Day Avg Trd Size            -0.628   1.00     98.2%
Offer Revision from Orgnl Flng        0.462    1.00     89.0%
Ln(1 + Shrs Rtnd/Shrs Ofrd)           0.145    1.00     92.3%
Ln of Headline Instances              0.066    1.00     77.7%
Prior 30 Day Industry Rtrn+           0.159    1.00     93.7%
NASDAQ Dummy                          0.075    1.00     88.7%
Investment Bank Reputation            0.084    0.99     53.2%
VC Backed Dummy                       -0.026   0.68      0.0%
JR Internet Dummy                     1.406    1.00     95.8%
Ln of IPO Firm Age                    -0.038   0.99     70.8%
SDC High Tech Dummy                   0.095    0.98     48.7%
Ln of Firm Sales                      -0.208   0.99     83.0%

Butler, Keefe, and Kieschnick   One Million Regressions
Introduction
OLS Results
Data
Outliers and Price Support
Estimation Results
OLS vs Quantile
Conclusion

OLS Results by period using 44 variables

1993-1998 (90S)      1999-2000 (BUBBLE)        2001-2006 (POST)
Extreme                 Extreme                 Extreme
Mean    Bounds        Mean      Bounds          Mean     Bounds
Variable                           Beta    Percent       Beta      Percent         Beta     Percent
Ln of 1st Day Avg Trd Size        -0.628    98.2%       -0.341     100.0%         -0.235    100.0%
Offer Revision from Orgnl Flng    0.462     89.0%       0.554       99.2%         0.464     99.3%
Ln(1 + Shrs Rtnd/Shrs Ofrd)       0.145     92.3%       0.172       98.1%         0.151     93.8%
Ln of Headline Instances          0.066     77.7%       0.111       54.5%         0.124     92.9%
Prior 30 Day Industry Rtrn+       0.159     93.7%       0.108       33.9%         0.111     89.9%
NASDAQ Dummy                      0.075     88.7%       0.204       93.6%         0.002      8.8%
Investment Bank Reputation        0.084     53.2%       0.145       89.2%         0.091     61.1%
VC Backed Dummy                   -0.026     0.0%       0.216       77.8%         0.323      99.9%
JR Internet Dummy                 1.406     95.8%       0.213       32.3%         0.154      0.0%
Ln of IPO Firm Age                -0.038    70.8%       -0.089      37.6%         -0.071     13.6%
SDC High Tech Dummy               0.095     48.7%       0.126       36.1%         0.083      0.2%
Ln of Firm Sales                  -0.208    83.0%       0.033        0.5%         -0.072     20.2%

Butler, Keefe, and Kieschnick     One Million Regressions
Introduction
OLS Results
Data
Outliers and Price Support
Estimation Results
OLS vs Quantile
Conclusion

The effect of outliers and price support

Table: OLS and Quantile Regressions from 1993 to 2006
OLS using Winsorized Data                   Quantile
Extreme
Mean Beta                       Mean    Bounds
Variable                          W .000       W .010    W .030        W .050      Beta    Percent
Ln of 1st Day Avg Trd Size        -0.249       -0.244     -0.202       -0.171     -0.103     1.000
Offer Revision from Orgnl Flng     0.503        0.550      0.502        0.455     0.342      1.000
Ln(1 + Shrs Rtnd/Shrs Ofrd)        0.171        0.164      0.130        0.110     0.062      1.000
Prior 30 Day Industry Rtrn+        0.106        0.080      0.063        0.053     0.046      0.981
Ln of Headline Instances           0.104        0.097      0.090        0.081     0.050      0.996
NASDAQ Dummy                       0.050        0.048      0.049        0.055     0.039      0.882
Investment Bank Reputation         0.095        0.086      0.079        0.066     0.038      0.966
VC Backed Dummy                    0.119        0.105      0.083        0.074     0.086      0.993
JR Internet Dummy                  0.594        0.541      0.431        0.337     0.379      1.000
Ln of IPO Firm Age                -0.044       -0.032     -0.028       -0.025     -0.011     0.022
SDC High Tech Dummy                0.111        0.104      0.090        0.081     0.054      0.823
Ln of Firm Sales                  -0.100       -0.092     -0.089       -0.078     -0.058     0.884

Butler, Keefe, and Kieschnick       One Million Regressions
Introduction
OLS Results
Data
Outliers and Price Support
Estimation Results
OLS vs Quantile
Conclusion

QLS vs Quantile by period

1993-1998             1999-2000              2001-2006        Mean OLS
Mean       Mean       Mean        Mean       Mean        Mean      Extreme
OLS       Quantile    OLS       Quantile     OLS       Quantile    Bounds
Variable                           Beta        Beta      Beta        Beta       Beta        Beta      Percent
Ln of 1st Day Avg Trd Size        -0.626      -0.248    -0.330      -0.154     -0.235      -0.172      98.3%
Offer Revision from Orgnl Flng     0.429      0.410      0.528       0.538      0.434       0.436      95.5%
Ln(1 + Shrs Rtnd/Shrs Ofrd)        0.124      0.058      0.148       0.061      0.142       0.122      93.0%
Prior 30 Day Industry Rtrn+        0.169      0.069      0.111       0.048      0.114       0.101      74.3%
Ln of Headline Instances           0.061      0.035      0.105       0.043      0.129       0.116      69.2%
NASDAQ Dummy                      -0.004      0.061     0.083        0.033     -0.053      -0.040      63.6%
Investment Bank Reputation         0.104      0.043      0.144       0.060      0.086       0.039      60.1%
VC Backed Dummy                   -0.041      0.061     0.209        0.114     0.318        0.364      54.9%
JR Internet Dummy                  1.436      0.710      0.213       0.115      0.150       0.211      40.8%
Ln of IPO Firm Age                -0.037      -0.019    -0.081      -0.034     -0.071      -0.038      37.2%
SDC High Tech Dummy                0.096      0.057      0.104       0.062      0.078       0.139      20.4%
Ln of Firm Sales                  -0.191      -0.079    0.039       -0.002     -0.066      -0.098      32.7%

Butler, Keefe, and Kieschnick     One Million Regressions
Introduction
Data
Estimation Results
Conclusion

Conclusions
Robust variables
1 First Day Average Trade Size and Number of Headline Instances ⇒
Demand from smaller investors important
2 Shares Retained ⇒ Ownership structure after IPO important
3 Offer Revision ⇒ Demand and Bookbuilding Theories important
4 NASDAQ, Internet Dummy, Industry Return ⇒ Valuation uncertainty
important
Fragile Variables
1 Ln of Firm Size, Ln of Firm Age, EBITDA to Assets ⇒ information
asymmetry between Issuing Firm and other agents may not be important;
however,
2 Investment Bank Reputation (Certiﬁcation) was important during bubble
3 Multi-Class Dummy ⇒ control theories may not be important
4 VC Backed Dummy Changes Sign (Consistent with Ritter and Loughran
2004)
5 Broad indexes such as CRSP EW index ⇒ broad measures not important

Overall our ﬁndings indicate further research should focus on the relationship
between ﬁrst trading day investor demand, market variables and underpricing

Butler, Keefe, and Kieschnick   One Million Regressions

```
DOCUMENT INFO
Shared By:
Categories:
Stats:
 views: 83 posted: 9/3/2009 language: English pages: 15
How are you planning on using Docstoc?