III Empirical implications and analysis:
Before presenting the results and empirical
implications of the model, I should describe the different steps and the
methodology I used to obtain the final specification:
Ø 1^{st} Step: using the AAII and II weekly
sentiment indicators, I calculate three investors sentiment measures: bullish
proportion (% of bullish investors / % of bullish and bearish investors),
bullishbearish Spread (% of bullish investors  % of bearish investors) and
the bullish bearish Ratio (% of bullish investors/ % of bearish investors).
From these 3 possible measures, I should find the best measure of sentiment and
the model that best explains the data. I try to construct 3 models, each one
contains all the variables with one of the 3 possible sentiment measures, so
every specification contains 17 explanatory variables with a constant:
Underwriter reputation, Overhang ratio, R&D intensity, VC backed, Firm age,
Firm size Ln(assets), Firm size Ln(sales), Issue size, Firm profitability, ROA,
Issue risk (tech dummy), Insiders' ownership, Institutional ownership,
Blockholders' ownership, AAII sentiment indicator, II sentiment indicator and
Time dummy.
The regression results and the models adjustment quality and
different measures of goodness of fit allow the choice of the model containing
the bullish bearish Ratio as the model that best explains underpricing within
the sample and using these control variables: the highest Rsquared
and adjusted Rsquared which measure the success of the regression in
predicting the values of the dependant variable within the sample, and the
lowest Akaike Information Criterion and Schwarz Criterion
which are measures of the goodness of fit of an estimated statistical
model and they measure the efficiency of the parameterized model in terms of
predicting the data. So in a first step, I continue working with the model
using the Bullish Bearish Ratio as measure of
investor sentiment with all the other explanatory variables.
The tables 6 and 7 present the regression results of the
models containing all the variables with respectively the bullish proportion
and the bullishbearish spread as measures of investors' sentiment. The tables
also contain criterion used to decide which model explains best the dependant
variable within the sample.
Table 6. Ordinary Least Squares Regression Results
(with Bullish proportion as investors' sentiment measure)

Dependent Variable: UNDERPRICING





Method: Least Squares





Sample (adjusted): 1 214 Included observations: 117 after
adjustments




White HeteroskedasticityConsistent Standard Errors &
Covariance





Variable

Coefficient

Std. Error

tStatistic

Prob.

C

0.069774

0.309901

0.225149

0.8223

UNDREWRITER REPUTATION

0.095300

0.071413

1.334489

0.1851

OVERHANG RATIO

0.000577

0.000260

2.218402

0.0288

R D INTENSITY

2.857168

0.928221

3.078112

0.0027

VC BACKED DUMMY

0.014893

0.062503

0.238276

0.8122

FIRM AGE

0.014673

0.031730

0.462419

0.6448

FIRM SIZE LN ASSETS

0.028986

0.026733

1.084293

0.2809

FIRM SIZE LN SALES

0.000771

0.009336

0.082608

0.9343

ISSUE SIZE

0.007820

0.039760

0.196668

0.8445

FIRM PROFITABILITY

0.067138

0.053643

1.251562

0.2137

ROA

0.055279

0.059343

0.931524

0.3538

ISSUE RISK TEH DUMMY

0.000610

0.042783

0.014249

0.9887

INSIDER OWNERSHIP

0.170075

0.086312

1.970474

0.0516

INSTITUTIONAL OWNERSHIP

0.233858

0.231054

1.012134

0.3139

BLOCKHOLDERS OWNERSHIP

0.368274

0.220834

1.667651

0.0985

AAII BULL PROPORTION

0.477753

0.238433

2.003722

0.0478

INVI BULL PROPORTION

0.226409

0.369040

0.613508

0.5409

TIME DUMMY

0.146285

0.053135

2,742582

0.0070

Rsquared

0.286636

Mean dependent var

0.295424

Adjusted Rsquared

0.164139

S.D. dependent var

0.234851

S.E. of regression

0.214713

Akaike info criterion

0.098386

Sum squared resid

4.564085

Schwarz criterion

0.326563

Log likelihood

23.75561

HannanQuinn criter.

0.074138

Fstatistic

2.339949

DurbinWatson stat

2.326671

Prob(Fstatistic)

0.004784




Table 7. Ordinary Least Squares Regression Results
(with Bullish Bearish Spread as investors' sentiment measure)

Dependent Variable: UNDERPRICING





Method: Least Squares





Sample (adjusted): 1 214 Included observations: 117 after
adjustments




White HeteroskedasticityConsistent Standard Errors &
Covariance





Variable

Coefficient

Std. Error

tStatistic

Prob.

C

0.411271

0.171480

2.398364

0.0183

UNDREWRITER REPUTATION

0.095188

0.071482

1.331636

0.1860

OVERHANG RATIO

0.000600

0.000259

2.317958

0.0225

R D INTENSITY

2.842995

0.930156

3.056472

0.0029

VC BACKED DUMMY

0.015700

0.062839

0.249853

0.8032

FIRM AGE

0.014385

0.031341

0.458990

0.6472

FIRM SIZE LN ASSETS

0.027984

0.026783

1.044824

0.2986

FIRM SIZE LN SALES

0.000854

0.009332

0.091548

0.9272

ISSUE SIZE

0.006576

0.040015

0.164343

0.8698

FIRM PROFITABILITY

0.067658

0.053506

1.264489

0.2090

ROA

0.058442

0.059784

0.977542

0.3307

ISSUE RISK TECH DUMMY

0.001121

0.042737

0.026232

0.9791

INSIDER OWNERSHIP

0.167241

0.085970

1.945335

0.0546

INSTITUTIONAL OWNERSHIP

0.246116

0.228617

1.076544

0.2843

BLOCKHOLDERS OWNERSHIP

0.381754

0.219424

1.739803

0.0850

AAII BULL BEAR SPREAD

0.003062

0.001532

1.998317

0.0484

INVI BULL BEAR SPREAD

0.001949

0.002459

0.792373

0.4300

TIME DUMMY

0.144385

0.053118

2.718194

0.0078

Rsquared

0.288315

Mean dependent var

0.295424

Adjusted Rsquared

0.166106

S.D. dependent var

0.234851

S.E. of regression

0.214461

Akaike info criterion

0.100743

Sum squared resid

4.553344

Schwarz criterion

0.324207

Log likelihood

23.89344

HannanQuinn criter.

0.071782

Fstatistic

2.359205

DurbinWatson stat

2.328713

Prob(Fstatistic)

0.004430




Ø 2^{nd} Step: I check the correlation between
the explanatory variables, I use Klein correlation detection criterion:
comparison between Rsquared (R^{²}) and the different
squared simple correlation coefficients (r^{²}) with
r^{²}=Cov^{²}(X_{i},X_{j})/
Var(X_{i})*Var(X_{j})
r^{²}_{Xi,Xj }> R^{² }
for i?j means the presence of correlation between X_{i} and
X_{j}.
For this, I should use the correlation matrix to determine the
different correlation coefficients between all the explanatory variables. For
R^{² }= 0.2897, I find a correlation between issue size and firm
size Ln (assets) (r^{²}= 0.4361) and a correlation between
institutional ownership and blockholders' ownership (r^{²}=
0.8326). I construct all the specifications necessary to decide which control
variables I should eliminate and which explanatory variables I should keep in
the model to give the best explanation of underpricing phenomenon (the
dependant variable of the model). Founding my decision on measures of goodness
of fit and of adjustment quality for the model, I eliminate issue size and
institutional ownership, and I keep all the other control variables. The final
specification contains 15 explanatory variables: Underwriter Reputation,
Overhang Ratio, R&D Intensity, VC Backed, Firm Age, Firm Size Ln(assets),
Firm Size Ln(sales), Firm Profitability, ROA, Issue Risk (tech dummy),
Insiders' Ownership, Blockholders' Ownership, AAII Bullish Bearish Ratio, II
Bullish Bearish Ratio and Time Dummy.
With a new correlation matrix, I check the correlation between
these 15 control variables: r^{²}_{Xi,Xj }<
R^{² }= 0.2833^{ } for i?j
There is no correlation between the control variables after
eliminating issue size and institutional ownership.
The final specification is as follows:
Underpricing = a_{0} + a_{1}Underwriter
Reputation Dummy + a_{2}Overhang Ratio+ a_{3}R&D Intensity
+ a_{4}VC Dummy + a_{5}Ln (1+age) + a_{6}Ln (assets) +
a_{7}Ln (sales) + a_{8}Firm Profitability + a_{9}ROA +
a_{10}Issue Risk Dummy + a_{11}Insiders' Ownership +
a_{12} Blockholders' Ownership + a_{13}Individual Bullish
Bearish Ratio+ a_{14}Institutional Bullish Bearish Ratio +
a_{15}Time Dummy + å_{i}
With underwriter reputation, overhang ratio, R&D intensity
and VC backing used to measure the informational asymmetry and exactly are
proxies for the firm quality. Age, firm size, firm profitability, ROA and issue
risk are measures of risk. Insiders' ownership and blockholders' ownership are
proxies for issuer bargaining power. Individual and Institutional bullish
bearish ratios are measures of investors' sentiment. Finally, time dummy is a
control variable to verify the impact of the beginning of the financial crisis
on IPO underpricing.
I present the Ordinary Least Squares regression results in
table 8. The first result is that the model is statistically significant
(p.value= 0.001935), which means the relevance of the control variables used in
this model to explain the underpricing phenomenon (the dependant variable). I
find acceptable values of Akaike Information Criterion and Schwarz Criterion
which measure the goodness of fit and the efficiency of the parameterized model
in terms of predicting the data.
Let's verify the individual significance of the explanatory
variables. 5 statistically significant variables: Overhang Ratio, R&D
Intensity, Insiders' Ownership, Individual Bullish Bearish Ratio and Time
Dummy.
§ The negative and significant coefficient of insiders'
ownership is consistent with the hypothesis that issue firms with high
bargaining power request higher offer prices conducting lower degree of
underpricing. This negative relation between issuer bargaining power and
underpricing is consistent with the findings of Ljungqvist, Nanda and Singh
(2003), Ljungqvist and Wilhelm (2003), Loughran and Ritter (2004) and many
other researchers who used this variable in their studies.
§ The positive and significant coefficient of Individual
Bullish Bearish Ratio is consistent with the hypothesis that investors with
high and optimistic sentiment are willing to pay higher prices to acquire the
IPO shares leading to higher underpricing. This result is consistent with the
findings of researchers who are interested in the behavioral explanation of IPO
first day price run up using different proxies for sentiment: Lee, Shleifer and
Thaler (1991) with discount on closedend funds, Cornelli, Goldreich and
Ljungqvist (2004) with grey market prices and many other researchers using the
behavioral approach and the investor sentiment.
§ The positive and significant coefficient of Time dummy
presents the positive relation between underpricing and the beginning of the
financial crisis which has an impact on the IPO first day returns.
§ The negative and significant coefficients of overhang
ratio and R&D intensity (both are measures of firm quality) can be
explained by the fact that high quality issue firms request higher offer prices
for their IPOs conducting lower level of underpricing.
§ The lack of significance of the 6 measures of risk
(age, Ln(assets), Ln(sales), firm profitability, ROA and issue Risk (tech
dummy)) is consistent with the findings of Bartov, Mohanram and Seethamraju
(2003): using a dummy variable for risky IPOs, they find that there is no
correlation between risk and underpricing and the notion of risk can not
explain the setting of a lower offer price and then the underpricing
phenomenon.
§ The lack of significance of the VC backing coefficient
confirms the findings of many researchers that used VC dummy as a proxy and as
a signal for firm quality. A question can be asked about the explanatory power
of this variable and its introduction in explanatory models of underpricing
phenomenon.
By regrouping the three theories advanced earlier as
explanations for the underpricing phenomenon in the same model and by using
different proxies for each theory, I show that:
Ø The informational asymmetry theory and particularly
the issuing firm quality proxied by R&D intensity or by the overhang ratio
presents a relevant explanation to the short run IPO anomaly.
Ø The issuer bargaining power proxied by insiders'
ownership is also an important and a reliable determinant of underpricing.
Ø Finally, one of the main goals of this study is to
show the relevance and the importance of investors' sentiment and behaviour as
an explanation to the short run IPO puzzle, and to find the positive
correlation between the investors' optimism and underpricing phenomenon. This
hypothesis is confirmed in this study.
In this model, I distinguish between the two types of
investors: individual and institutional investors using a direct measure of
sentiment for each type calculated from the American Association of
Individual Investors and the Investors Intelligence
survey data, these are two other contributions in this study. The positive and
significant coefficient of individual investors' sentiment and the positive but
insignificant coefficient of institutional investors' sentiment lead to an
important finding: individual investors are those driving the first day closing
prices and then underpricing anomaly and are more conducting the short run IPO
puzzle than the institutional investors. The individual investors' sentiment
and their over optimism and enthusiasm is more important and relevant in
explaining the IPO underpricing anomaly than does the institutional investors'
sentiment. Individual investors are the type more conducting the IPO first day
returns and the short run IPO puzzle.
This finding can be explained by the fact that in some
researches and studies, institutional investors are defined as rational
investors. This can be a possible explanation for the irrelevance of
institutional investors' sentiment in explaining the underpricing anomaly.
Table 8. Ordinary Least Squares Regression
Results

The sample size is 217, R&D expenditures is missing for 90
firms, Underwriter ranking is missing for 6 firms and sales is missing for 4
firms, reducing the number of observations to 117. The dependent variable is
the percentage firstday return from the offer price to the firstday closing
price. The underwriter reputation dummy takes a value of one if the lead
underwriter has an updated Carter and Manaster (1990) rank of 8 or more, and
zero otherwise. Share overhang is the ratio of retained shares to the public
float (the number of shares issued). R&D intensity is the ratio of PreIPO
R&D expenditures to the expected market value.

The VC dummy takes a value of one (zero otherwise) if the IPO
is backed by venture capital. Ln(1 + age) is the natural log of 1 plus the
years since the firm's founding date as of the IPO. The Ln(assets) is the
natural logarithm of the preissue book value of assets expressed in millions
$. Ln(sales) is the natural log of the firm sales a year prior the offering,
expressed in millions $. Firm profitability is the ratio of PreIPO EBITDA to
total assets value.

ROA is the ratio of PreIPO net income to total assets value.
The issue risk is a tech dummy takes a value of one (zero otherwise) if the
firm is in the technology business. Insiders' ownership and blockholders
ownership are proxies to ownership structure and issuer bargaining power.AAII
and II bullish bearish Ratios are the % of bullish investors divided by the %
of bearish investors. Time dummy takes a value of one (zero otherwise) if the
IPO occurred after July 2007. The tstatistics are calculated using White's
(1980) heteroskedasticityconsistent method.

Dependent Variable: UNDERPRICING





Method: Least Squares





Sample (adjusted): 1 214 Included observations: 117 after
adjustments




White HeteroskedasticityConsistent Standard Errors &
Covariance










Variable

Coefficient

Std. Error

tStatistic

Prob.

C

0.292331

0.166898

1.75155

0.0829

UNDERWRITER REPUTATION

0.093909

0.069317

1.354779

0.1785

OVERHANG RATIO

0.000488

0.000221

2.205774

0.0297

R D INTENSITY

2.946844

0.867721

3.396074

0.001

VC BACKED DUMMY

0.029582

0.057948

0.510493

0.6108

FIRM AGE

0.010542

0.030908

0.341071

0.7338

FIRM SIZE LN ASSETS

0.02194

0.02136

1.027144

0.3068

FIRM SIZE LN SALES

0.000269

0.009324

0.028826

0.9771

FIRM PROFITABILITY

0.034211

0.049397

0.692579

0.4902

ROA

0.017825

0.051876

0.343617

0.7318

ISSUE RISK TECH DUMMY

0.004296

0.0414

0.103774

0.9176

INSIDERS OWNERSHIP

0.177436

0.082075

2.161876

0.033

BLOCKHOLDERS OWNERSHIP

0.137413

0.09812

1.400456

0.1644

AAII BULL BEAR RATIO

0.098838

0.04782

2.066853

0.0413

INVI BULL BEAR RATIO

0.034061

0.039904

0.853586

0.3954

TIME DUMMY

0.147677

0.053652

2.752493

0.007

Rsquared

0.283352

Mean dependent var

0.295424

Adjusted Rsquared

0.176919

S.D. dependent var

0.234851

S.E. of regression

0.213066

Akaike info criterion

0.127981

Sum squared resid

4.585098

Schwarz criterion

0.249752

Log likelihood

23.48689

HannanQuinn criter.

0.025374

Fstatistic

2.662259

DurbinWatson stat

2.364719

Prob(Fstatistic)

0.001935




