Professional Documents
Culture Documents
=
=
T
t t
t
V
R
T
AMIH
1
000 , 1
1
(2),
where R
t
is the stock return measured in logarithm on closing prices at date t, V
t
is the trading
volume on date t, and T is the number of trading days in the observation period.
For the sub-sample of continuously-traded stocks, we compute duration-weighted average
quoted spreads,
=
=
K
k k
k k
k
K
k
k
mid
bid ask
d
d
QS
1
1
1
(3),
and average effective spreads,
=
=
N
n n
n n
mid
mid P
N
ES
1
1
(4),
where
k
bid ,
k
ask ,
k
mid and
k
d are respectively the best bid quote, the best ask quote, the
mid quote and the duration of the best quotes observed at the time of the k
th
quoted spread in
the observation period; K is the total number of quoted spreads observed for the stock in the
9
observation period; P
n
is the transaction price for the n
th
transaction in the observation period;
mid
n
is the mid-quote prevailing at the time of the n
th
trade; and N is the total number of trades
in the period.
2.3. Measures of information asymmetry
As for spreads, measures of information asymmetry are derived over six months of
continuous trading from the 1
st
of October following the IPO date until the 31
st
of March of
the next year. For firms listed after the 1
st
of October, the start of the observation period is
postponed to 5 days after the IPO date. We then estimate the magnitude of information
asymmetry with four methodologies: the alpha coefficient of the Huang and Stoll (1997)s 3-
way spread decomposition, denoted
hs
, the average 30-minute price impact denoted PI
30min
,
the alpha coefficient of Lin, Sanger, and Booth (1995) denoted
lsb
, and the PIN measure
denoted PIN.
The 3-way decomposition method of Huang and Stoll (1997) can be conducted using two
different GMM estimation procedures. The first one requires the estimation of four
parameters, s, , , and , the probability of a reversal in trade sign. The second one uses the
mid-quote variation instead of the transaction price variation, and the observed posted spread
that prevails at the time of a trade instead of the constant spread s, which reduces to three (,
, and ) the number of parameters to estimate. We utilise the second methodology as it
produces a higher frequency of convergence of the GMM iteration algorithm, and we bunch
trades la Huang and Stoll (1997) to correct biases in the estimation of and due to order-
splitting practices.
We also conduct the decomposition of the effective spread in a realized spread and a price
impact within a 30-mn interval in the manner of Bessembinder and Kaufman (1997)
approach. Price impacts at a 30-mn interval are calculated as follows:
=
+
=
N
n t
t t
mid
mid mid
T
PI
1
min 30
min 30
1
(5),
Then, the Lin, Sanger and Booths adverse selection components (LSB) are estimated as the
sensitivity
LSB
of mid price revisions to trade sizes with the following regression model for
each stock i:
( )
1 1 + +
+ =
t t t t lsb t t
e Q mid P mid mid (6),
where
t
Q is the sign of trade t.
Notations used:
10
n
mid is the mid price at the time of the n
th
spread quoted in the observation
period;
N is the number of spreads quoted for the stock over the considered period;
t
P is the trade price of the t
th
transaction;
t
mid is the mid price prevailing before the t
th
transaction in stock i;
min 30 + t
mid is the mid price prevailing 30 minutes after the t
th
transaction;
T is the total number of trades for the stock over the observation period.
Finally, we implement the PIN measure following Easley, Kiefer, OHara, and Paperman
(1996). The probability of observing B buys and S sells on a given day can be implemented as
follows:
( )( ) ( ) ( )
( ) ( )
( )
( )
( ) ( )
( )
( )
( )
( ) ( )
! !
1
! !
! !
1 , , , ,
B
B T
e
S
T
e
S
T
e
B
T
e
S
T
e
B
T
e S B L
T
S
T
S
T
B
T
S
T
B
T
+
+
+
+
=
+ +
(7),
where is the probability of an information event which is bad news with probability and
good news with probability 1-. The arrival rate of informed trade arrival is . is the rate of
uninformed buy and sell trade arrivals.
On T days which are independent by assumption, the likelihood of observing ( )
T
t t t
S B
1
,
=
Buys
and sells over T days, corresponds to the product of likelihoods:
( ) ( ) ( )( ) ( )
=
=
=
T
t
t t
T
t
t t
S B L S B L
1
1
, , , , , , , ,
(8),
In order to get the probability of informed trading (PIN), we maximise the likelihood defined
in equation (8):
2 +
= PIN
(9).
2.4. Measures of ownership structure
We estimate the ownership concentration by observing several variables of ownership
structure for the whole sample during the period 1995-2004. These variables have been
extracted from DAFSA Liens to measure ownership concentration after the IPO at the date of
the 31
st
of December just after the IPO. We extracted the percentage of shares held by the
managers (MAN). We measure the percentage of shares held by institutional investors (INST)
and the managers family (FAM). In order to estimate the ownership concentration, we
11
identified all the blockholders who possess at least 5 percent of the firm shares and computed
their total holding in percentage (BLOCK). We also calculated the Herfindhal index (HERF),
by summing squared shareholdings of the five largest shareholders:
=
=
5
1
2
i
i
s HERF
(10),
where
i
s is the part that belongs to the i
th
largest shareholder (i=1,,5).
3. Test design
The relationships between initial underpricing, ownership concentration and secondary
markets liquidity are analysed running a three-stage multivariate analysis that combines
logistic and OLS regressions in the Heckman style to avoid endogeneity biases. The same
methodology is used to test the links between information asymmetry, initial underpricing,
and ownership concentration. All tests are conducted on the whole sample first. Then, they
are repeated on the sub-sample of firms that went public by using a book-building procedure.
As a matter of fact, the discretion provided by the book-building mechanism in the share
allocation process may result in a more effective effect of IPO underpricing on ownership
structure and post-listing liquidity.
3.1. First-stage logistic regression: estimation of the probability of underpricing
In a first stage, the probability for an issue to be underpriced, denoted ( ) 0 U P > , is modelled
as a function of the IPO size
5
:
( )
1 1 0
~
0 + + = > SIZE a a U P
(11),
where SIZE is the logarithm of the issue size calculated as the number of shares on sale in the
IPO multiplied by the subscription price.
5
Others factors such as earnings, P/E ratio, book-to-market have been inserted in the model, but none of them has been
proved to influence the probability of underpricing.
12
3.2. Second-stage OLS regressions
3.2.1. Post-listing liquidity and initial underpricing
In a second stage, liquidity measures are regressed on the level of underpricing. For the whole
sample, the average daily turnover (TURN) and the Amihud illiquidity ratio (AMIH) are
regressed on the level of underpricing predicted in the first-stage model, after controlling for
volatility, market value, and price level:
( ) ( )
2 4 0 3 2 1 0
~
) 0 (
> is the predicted value of the initial underpricing measured as the predicted
probability from model (9) multiplied by initial return measured over the first five days of
trading.
For the sub-sample of IPO stocks that were continuously traded during the observation period,
the relationship between spreads and initial underpricing is also tested. Control variables used
are volatility, trading volumes, and price level:
( ) ( )
3 4 0 3 2 1 0
~
) 0 (
ln ln + > + + + + = U U P c P c V c c c ES or QS (13),
where ln(V) is the logarithm of the average daily trading volume over the six-months period.
3.2.2. Ownership structure and initial underpricing
Measures of ownership structure (HERF, BLOCK and INST) are modelled as a function of the
IPO size, the family ownership, and the underpricing level predicted at the first stage:
4 3 2 1 0
~
) 0 (
, LOCK B
, and NST I
and LOCK B
>
5.362.10
-5
*** -0.0135** -0.0026 -6.4719.10
-4
4.049.10
-5
*** -0.0112 -0.0035 -0.0010
(<.0001) (0.0448) (0.3302) (0.7659) (0.0035) (0.1618) (0.2507) (0.6781)
Cox-Snell R 1.34%
Adjusted R 37.67% 6.82% 64.64% 66.46% 45.76% 4.94% 63.93% 65.03%
Note: The second column of the table reports the results of the first stage Logit regression in which the dependent variable is a dummy equal to 1 if the issue is underpriced, 0
otherwise. The probability of an issue to be underpriced is modelled as a function of SIZE, the logarithm of the issue proceeds in million euros. The rest of the table reports the
results of the 2-stage least square regressions of liquidity measures onto the initial underpricing predicted in the Heckman style, i.e. calculated as ( ) 0
LOCK B
NST I
# obs. Adj. R
HERF 1.2320*** -0.0594*** 0.2572*** -6.3214.10
-4
224 29.47%
(<.0001) (<.0001) (<.0001) (0.2026)
BLOCK 1.1893*** -0.0315*** 0.1748*** 8.922.10
-5
224 23.64%
(<.0001) (0.0010) (<.0001) (0.8070)
INST -0.5088*** 0.0394*** -0.1130*** 9.962.10
-5
224 13.96%
(0.0031) (0.0002) (<.0001) (0.8241)
TURN -0.0203*** 0.0016*** 6.2276.10
-4
* 0.0022*** 0.0065** 206 33.16%
(0.0038) (<.0001) (0.0722) (0.0003) (0.0268)
TURN -0.0263*** 0.0016*** 6.0947.10
-4
* 0.0022*** 0.0115*** 206 33.82%
(0.0011) (<.0001) (0.0674) (0.0003) (0.0086)
TURN -0.0209*** 0.0017*** 8.7176.10
-4
** 0.0022*** -0.0166*** 206 33.98%
(0.0018) (<.0001) (0.0209) (0.0003) (0.0065)
AMIH 10.8218*** 0.0225 -0.4682** -0.7050** 0.6684 206 5.39%
(0.0045) (0.7979) (0.0135) (0.0322) (0.6727)
AMIH 11.2934** 0.0149 -0.4936*** -0.7098** 0.3378 206 5.31%
(0.0102) (0.8648) (0.0072) (0.0311) (0.8875)
AMIH 11.0549 0.0207 -0.4613 -0.7072 -1.1837 206 5.36%
(0.0026) (0.8141) (0.0258) (0.0316) (0.7217)
QS 9.7481*** 0.1455*** -0.6757*** 0.2066 -0.1008 96 64.54%
(<.0001) (0.0012) (<.0001) (0.2592) (0.8779)
QS 9.8626*** 0.1452*** -0.6755*** 0.2059 -0.2031 96 64.55%
(<.0001) (0.0013) (<.0001) (0.2609) (0.8394)
QS 9.6875*** 0.1447*** -0.6774*** 0.2075 0.4145 96 64.57%
(<.0001) (0.0012) (<.0001) (0.2568) (0.7417)
ES 8.2036*** 0.1385*** -0.5715*** 0.2085 -0.0025 96 66.66%
(<.0001) (0.0002) (<.0001) (0.1571) (0.9963)
ES 8.2047 0.1385 -0.5714 0.2085 -0.0029 96 66.66%
(<.0001) (0.0002) (<.0001) (0.1573) (0.9971)
ES 8.2009*** 0.1381*** -0.5720*** 0.2086 0.0729 96 66.66%
(<.0001) (0.0001) (<.0001) (0.1569) (0.9425)
23
Note: The first three lines of the table report the estimations for the second-stage lest square regressions of ownership variables on underprincing. The remainder of the table displays the results of the third-stage
regressions of liquidity measures onto ownership variables. HERF is the Herfindhal index of ownership concentration. BLOCK is the percentage of shares controlled by the blockholders. INST is the percentage of
shares controlled by institutional investors. All ownership measures are measured after the IPO (at the end of the year following the IPO). Liquidity measures are computed over a six-month period surrounding the
end of the IPO year. Liquidity is measured by the average daily turnover (TURN) and the Amihud illiquidity ratio (AMIH) for all stocks. Duration-weighted average quoted spreads (QS) and average effective
spreads (ES) are calculated for continuously traded stocks. ERF H
, LOCK B
, and NST I
are the respective value of HERF, BLOCK, and INST as predicted by the second-stage models. is the closing return
volatility over the six-month observation period. lnMV is the market value in logarithm. lnV represents the logarithm of the average daily trading volume in euros. P
0
is the IPO price. P-values are in parentheses.
*
,
**
,
***
denote significance at 10, 5 and 1% levels respectively.
24
Table 5. After-market information asymmetry and underpricing
All IPOs Book-built IPOS only
lsb
hs
PI
30min
PIN
lsb
hs
PI
30min
PIN
0.6111*** -0.1432 1.5610*** 0.7397** 0.5407*** -0.0346 1.6393** 0.3439
intercept
(<.0001) (0.7840) (0.054) (0.0100) (0.0001) (0.9546) (0.0261) (0.2477)
-0.0228*** 0.0236 -0.0336 -0.0213 -0.0158** 0.0247 -0.0317 -0.0001
lnMV
(0.0014) (0.3954) (0.3134) (0.1509) (0.0269) (0.4408) (0.4153) (0.9941)
0.00189 -0.0504 -0.1447** -0.0147 -0.0172 -0.0908* -0.1788*** -0.0163
ln
0
P
(0.8701) (0.2729) (0.0135) (0.5488) (0.1369) (0.0978) (0.0080) (0.5251)
-0.0096 0.0701 -0.0491 0.0517 -0.0130 0.0279 -0.0553 0.0529
MAN
(0.5822) (0.2494) (0.5552) (0.1634) (0.4516) (0.6943) (0.5583) (0.1742)
-0.0095 0.0107 -0.0015 -0.0013 -0.0202 0.0051 -0.0185 6.568.10
-5
NM
(0.4744) (0.8502) (0.9812) (0.9623) (0.1079) (0.9334) (0.7924) (0.9981)
0.0346*** -0.0867* 0.1380** -0.0394 0.0361*** -0.0995* 0.1243* -0.0257
NTIC
(0.0047) (0.0690) (0.0197) (0.1229) (0.0020) (0.0932) (0.0511) (0.3111)
-3.8412.10
-4
** -3.4537.10
-4
-0.0027*** -9.3203.10
-4
** -3.7802.10
-4
** 2.2623.10
-4
-0.0025** -9.9402.10
-4
**
( ) U 0 U P
>
(0.0444) (0.6687) (0.0036) (0.0219) (0.0401) (0.8670) (0.0126) (0.0164)
Number of
observations
95 33 95 95 82 28 82 82
Adjusted R 13.15% 4.60% 21.73 7.65% 16.86% 4.49% 20.95% 6.26%
Note: This table displays the estimations for the second-stage regressions of information asymmetry measures on initial underpricing. Dependant variables are the alpha coefficient
of Lin, Sanger, and Booth (1995), the alpha coefficient of Huang and Stoll (1997), the average 30-minute price impact, and the PIN measure, denoted
lsb
,
hs
, PI
30min
, and PIN
respectively. The dependent variable ( ) U U P > 0
is the probability for an issue to be underpriced as predicted in the first-stage logit regression multiplied by the actual level of
underpricing. Control variables comprise the market value in logarithm (lnMV), the IPO price in logarithm (lnP
0
), the managers holdings after the IPO (MAN), a binary variable
equal to 1 for New Market issues (NM), a binary variable equal to 1 for new technologies firms (NTIC). ***,**,* indicate that the coefficient is significantly positive or negative
respectively at the 1%, 5%, 10% level. P-values are reported in brackets.
25
Table 6. After-market information asymmetry and ownership structure
All IPOs Book-built IPOs only
lsb
hs
PI
30min
PIN
lsb
hs
PI
30min
PIN
Number of
observations
95 33 95 95 82 28 82 82
ERF H
coefficient -0.0406 -0.0570 0.0931 -0.0661 -0.0435 -0.3032 -0.0977 0.1978
P-value 0.5160 0.7781 0.7240 0.6210 0.5375 0.3182 0.7596 0.2137
Adj. R 9.48% 4.21% 22.15% 2.22% 12.47% 8.89% 29.59% 0.81%
LOCK B
coefficient -0.1016 -0.1257 -0.1217 -0.1516 -0.1258 -0.4920 -0.0829 0.1664
P-value 0.2794 0.6713 0.7952 0.4508 0.2287 0.2692 0.8898 0.4828
Adj. R 10.25% 4.59% 13.46% 2.58% 13.72% 9.89% 13.67% 0.60%
NST I
coefficient 0.0846 0.0403 -0.0244 0.2729 0.1162 0.3407 -0.0468 -0.1802
P-value 0.5141 0.9217 0.9697 0.3236 0.3406 0.5186 0.9463 0.5140
Adj. R 9.48% 3.95% 13.39% 3.03% 13.09% 6.28% 13.65% -0.69%
Note: This table reports the results of third-stage regressions, in which measures of information asymmetry (
lsb
,
lsb
, PI
30min
and PIN) are regressed on the predicted values of the ownership variables ERF H
, LOCK B
, and NST I
alternatively. For
each regression, the table provides the coefficient of the ownership variable used, the P-value associated, and the adjusted R
of the regression. Control variables included in the regressions comprise market value, price level, insiders holdings, market
segment (traditional or new market), and industrial sector (new technologies vs traditional industries and services). ***,**,*
indicate that the coefficient is significantly positive or negative respectively at the 1%, 5%, 10% level.