Module PHY6002 Inorganic Semiconductor Nanostructures Lectur.docx

Module PHY6002 Inorganic Semiconductor Nanostructures
Lectures 7, 8, 9 and 10
1
Lecture 7 – The fabrication of semiconductor
nanostructures I
Introduction
In this lecture we will look at the techniques used to fabricate
semiconductor
nanostructures. The well-established epitaxial methods used to
produce
quantum wells will be described. The main techniques applied
to produce
quantum wires and quantum dots will be discussed, with a
comparison of their
relative advantages and disadvantages. In the next lecture we
will look in
detail at the most successful technique used to produce quantum
dots, self-
organisation.
Epitaxial techniques
There are two well established epitaxial growth techniques used
to produce
high quality quantum wells: molecular beam epitaxy (MBE) and
metal organic
vapour phase epitaxy (MOVPE).
The following figure shows the main components of an MBE
reactor.

The reactor consists of an ultra-high vacuum chamber with a
number of
effusion cells, each containing a different element. Each cell
has a mechanical
shutter placed in front of its opening. In operation the cells are
heated to a
temperature where the elements start to evaporate, producing a
beam of
atoms which leave the cells. These beams are aimed at a heated
substrate
which consists of a thin wafer of a suitable bulk semiconductor.
The incident
beams combine at the surface of the substrate and a
semiconductor is
deposited atomic-layer by atomic-layer. The substrate is rotated
to ensure
even growth over its surface. By opening the mechanical
shutters in front of
certain cells it is possible to control which semiconductor is
deposited. For
example opening the shutters in front of the Ga and As cells
results in the
growth of GaAs. Shutting the Ga cell and opening the Al cell
switches to the
growth of AlAs. Because the shutters can be operated very
rapidly in
comparison to the rate at which material is deposited, it is
possible to grow
An MBE reactor

2
very thin layers with very sharp interfaces between layers. The
following figure
shows a transmission electron microscope image of a quantum
well sample
containing five wells of different thicknesses. The thinnest well
has a
thickness of only 1nm. Other cells in the MBE reactor may
contain elements
used to dope the semiconductor and it is possible to monitor the
growth as it
proceeds by observing the electron diffraction pattern produced
by the
surface.
The second epitaxial growth technique is metal organic vapour
phase epitaxy
(MOVPE). In this technique the required elements are carried,
as a
component of gaseous compounds, to a suitable chamber where
they mix as
the gases flow over the surface of a heated substrate. The
compounds
breakdown to deposit the semiconductor on the surface of the
substrate with
the remaining waste gases being removed from the chamber.
Valves in the
gas lines leading to the chamber allow the gases flowing into
the reactor to be
switched on and off. A suitable switching sequence allows
layered structures
to be deposited. Because it is difficult to switch a gas flow

quickly, and
because the growth rate with MOVPE is faster than for MBE,
the latter
technique is generally capable of growing thinner layers with
more abrupt
interfaces. However the faster growth rate of MOVPE has
advantages in
commercial production where it is necessary to deposit the
material as quickly
as possible. MOVPE has a number of safety implications as the
gases are
highly toxic. The following figure shows a schematic diagram
of the main
components of a MOVPE system.
A cross sectional transmission electron microscopy (TEM)
image of an InGaAs-
InP quantum well structure containing five wells of different
thicknesses.
Main components of a MOVPE system (From Davies)
3
Requirements for semiconductor nanostructures
Before we look at the various techniques that have been used to
produce
quantum wires and dots, it is useful to consider what properties
ideal

structures should exhibit. This will help in analysing the
relative advantages
and disadvantages of each technique.
The main requirements of a semiconductor nanostructure can be
summarised
as follows
• Size. For many applications we require all the electrons and
holes to be in
their lowest energy state, implying negligible thermal excitation
to higher
states. The amount of thermal excitation is controlled by the
ratio of the
energy spacing between the confined states and the thermal
energy, given
by kT. At room temperature the thermal energy is 25meV and a
rule of
thumb is that the level separation should be at least three times
this value
(~75meV). As the spacing between the states is controlled by
the size of
the structure (see lecture 5 for the case of a quantum well) this
places
requirements on the size of the nanostructure.
• Quality. Defects may increase the probability of carriers
recombining non-
radiatively. Structures with a large number of defects may be
very
inefficient light producers. For optical applications
nanostructures with low
defect numbers are required.
• Uniformity. Devices generally contain a large number of
nanostructures.
Ideally all the nanostructures should be identical otherwise they

will all emit
light at slightly different energies.
• Density. It should be possible to produce dense arrays of
nanostructures.
• Growth compatibility. Industry uses MBE and MOVPE
extensively.
Nanostructures will find more applications if they can be
produced using
either or both of these techniques.
• Confinement potential. The depth of the potential wells which
confine the
electrons and holes must be relatively deep. If this is not case
then at room
temperature carriers will be thermally excited out of the
nanostructure.
• Electron and/or hole confinement. For electrical applications
it is
generally only necessary for either electrons or holes to be
trapped
(confined) within the nanostructure. For electro-optical
applications it is
necessary for both types of carrier to be confined.
• p-i-n structures. Many applications require the electrical
injection of
carriers into the nanostructure or the transfer of carriers,
initially created in
a nanostructure, to an external electrical circuit. This can be
achieved if the
nanostructure can be incorporated within the intrinsic region of
a p-i-n
structure.

Fabrication of semiconductor quantum wires and quantum dots
Lithography and etching
This starts with an epitaxially grown two dimensional system to
provide
confinement along the growth direction. Lithography (etch
resist, optical
lithography with a mask or electron beam lithography) is then
used to define a
pattern on the surface consisting of either wires or dots. These
are
subsequently etched using a plasma, resulting in free standing
dots or wires.
The structure can subsequently be returned to a growth reactor
to be
4
overgrown and incorporated in a p-i-n device. The main stages
of this
technique are shown in the following figure. The main
disadvantage of this
technique is that the surface is damaged during the etching
stage. The
resultant defects produce an optically dead layer where non-
radiative
recombination is the dominant electron-hole recombination
process. This

dead layer has an almost constant width so becomes
increasingly important
as the size of the structure decreases. For the small sizes
required for
practical nanostructures the dead layer occupies all of the
structure which is
consequently optically dead.
Cleaved edge overgrowth
A quantum well is initially grown and then the sample is
cleaved in the growth
reactor along a plane parallel to the growth direction. The
sample is then
rotated through 90° and a second quantum well followed by a
barrier is grown.
The growth sequence is shown in the following figure.
The two quantum wells form a T-shaped structure. At the
intersection of the
two wells the effective well width is slightly larger. Because the
confined
energy levels depend on the inverse of well width squared (see
Lecture 5) the
intersection region has a slightly lower potential and hence
electrons and
holes become trapped there – a quantum wire is formed. If
during the initial
growth multiple wells are grown then the overgrowth of the
final well results in
a linear array of wires. A second cleave followed by a further
overgrowth can
be used to produce quantum dots.
The surfaces produced by cleaving are clean, in contrast to the
dirty surface

formed by etching. Hence cleaved edge overgrowth dots and
wires have a
(a) (b) (c) (d)(a) (b) (c) (d)
The main stages in forming lithographically defined dots. (a)
growth of a 2D quantum
well. (b) surface coating with etch resist. (c) exposure of resist
to form pattern (d)
etching to form dot or wire.
(a) (b) (c) (d)(a) (b) (c) (d)
The steps involved in the cleaved edge overgrowth of a quantum
wire. (a) initial
quantum well growth (b) cleavage to form a perfect surface (c)
rotation (d) growth
of the second quantum well.
5
high optical quality. Their main disadvantage is that the
potential at the
intersection of the wells is not much smaller than in the wells.
The carriers are
only weakly confined in the intersection region and at room
temperature their
thermal energy is sufficient to allow them to escape. These
structures are

therefore generally suitable for studying physics at low
temperatures but not
for device applications, which need to work at room
temperature. In addition
the cleaving step is a difficult, non-standard process.
Growth on Vicinal Substrates
Semiconductors are crystalline materials with a periodic
structure. Only when
a semiconductor crystal is cut in certain directions will it have a
flat surface.
For other directions the surface will consists of a series of steps
(think about a
brick wall). Epitaxial growth is usually performed on flat
surfaces. However the
use of stepped surfaces (so-called vicinal surfaces) can be used
to produce
quantum wires. The size of the steps is determined by the
direction along
which the surface is formed but are typically ~20nm or less.
The above figure shows the main steps in the growth of vicinal
quantum
wires. Starting with the stepped surface (a) the wire
semiconductor is initially
deposited epitaxially (b). Growth tends to occur in the corner of
the steps as it
here that the highest density of atomic bonds occurs. As the
growth proceeds
the semiconductor spreads out from the initial corner. When
approximately
half of the step width has been covered growth is switched to
the barrier
material (c) which is used to cover the remainder of the step.
Growth can then

be switched back to the wire semiconductor to increase the
height of the wire
(d). This growth cycle is repeated until the desired vertical
height is obtained.
Finally the wire is overgrown with a thick layer of the barrier
material (e).
Although very thin wires can be produced using this technique
the growth has
to be very well controlled so that exactly the same fraction of
the step is
covered during each cycle. In addition the coverage on different
steps varies
and it is difficult to ensure that the original steps are uniform.
The resultant
wires tend not to exhibit good uniformity.
Growth on patterned substrates
This starts with a flat semiconductor substrate which is coated
with an etch
resist and then exposed using either optical or electron beam
lithography to
produce an array of parallel stripes. The regions between the
stripes are then
etched in a suitable acid. Because the acid etches different
crystal directions
at different rates, a v-shaped groove is obtained. The patterned
substrate is
then cleaned and transferred to a growth reactor.
(a) (b) (c) (d) (e)(a) (b) (c) (d) (e)
The main steps in the growth of vicinal quantum wires (a)
original stepped surface
(b) growth occurs in corners of steps, sufficient material
deposited to cover ~1/2
of step (c) remainder of step filled in with first material (d)
more wire material

deposited to increase thickness of wire (e) final over growth of
wire.
6
Quantum wires are usually formed from GaAs, with AlGaAs as
the barrier
material. Initially the AlGaAs barrier is deposited. This grows
uniformly over
the whole structure and may sharpen the bottom of the groove
which, after
the etching, has a rounded profile. Next a thin layer of GaAs is
deposited.
Although this again grows over the whole surface, the growth
rate at the
bottom of the groove is faster than that on the sides of the
grooves due to the
different crystal surfaces. A quantum well is formed with a
spatial modulation
of its thickness, being thicker at the bottom of the groove. In a
similar manner
to cleaved edge overgrowth, this thicker region results in a
potential minimum
forming a quantum wire. A second AlGaAs barrier layer can
now be grown;
this re-sharpens the groove after the formation of the wire, after
which further
wires can be grown. The main steps of this technique, resulting
in v-groove
quantum wires, are shown in the above figure.

The following figure shows a cross sectional transmission
electron
microscope image of a multiple v-groove quantum wire
structure. The wires
have a crescent cross section.
(a) (b) (c) (d)
The main steps in the formation of v-groove quantum wires (a)
original patterned
substrate, (b) growth of barrier semiconductor (c) growth of
wire semiconductor,
greater growth at bottom of groove (d) growth of second barrier,
re-sharpening of
groove.
A cross sectional transmission electron micrograph of three v-
groove quantum
wires. The wires have a maximum thickness of approximately
8nm.
7
Because the quantum wire is not next to the original etched
surface, v-groove
quantum wires exhibit good optical efficiencies. However it is
difficult to
control the inplane size of the wires as this is mainly
determined by the shape

of the groove. The uniformity of the wire along its length is
also influenced by
the original groove quality. For achievable wire sizes the energy
level
spacings are typically 20~30meV, some what less than required
for room
temperature operating devices. However in some cases careful
control of the
groove cross-section has lead to slightly larger level spacings.
A further
disadvantage of v-groove quantum wires is their complicated
structure. In
addition to the wire there are quantum wells formed on the sides
of the groove
(side wall wells) and on the region between the grooves (top
wells). These
wells may capture carriers, reducing the fraction which
recombine in the wire
and also producing additional features in the emission spectra.
Although the
top wells and some of the side wells can be removed by etching
after growth
this requires a further fabrication step and the structure may
need to be
returned to the reactor to complete the growth of a p-i-n
structure.
By initially patterning the substrate not with a single array of
stripes but with
two perpendicular arrays to give a two dimensional array of
squares, the
subsequent etching forms an array of pyramidal shaped pits.
Epitaxial growth
now results in the formation of quantum dots at the bottom of
each pit.
Strain induced dots and wires

If a semiconductor is subjected to strain its band structure is
modified. In
particular by applying the correct sign of strain the band gap
may be reduced.
If strain is only applied to a small region of the semiconductor
then a local
reduction of the band gap may occur, resulting in the formation
of a wire or
dot. In practise a local strain is produced by depositing a thin
layer of a
different material (e.g. carbon) on the surface of the
semiconductor. This will
have a very different atomic spacing to the semiconductor so to
fit together
both the atomic positions in the carbon layer and the surface
region of the
semiconductor will alter. This alteration constitutes a strain. If
the carbon layer
is patterned by lithography and then etched to leave only stripes
or dots, the
local strain field produces a wire or dot in the underlying
semiconductor. The
remaining isolated pieces of carbon are known as stressors. It is
necessary to
place a quantum well near to the surface of the semiconductor
to provide
confinement along the growth direction. The steps in the
production of strain
induced dots and wires are shown in the following figure.
(a) (b) (c)(a) (b) (c)
Steps in the formation of strain induced nanostructures (a)
initial quantum well (b)
deposition of carbon layer (c) formation of stressors by

lithography and etching.
The resultant, localised strain field (dashed lines) forms a wire
or dot in the
quantum well.
8
Although this technique involves an etching step, only the
carbon layer is
etched, the etching is kept away from the optically active
quantum well. Hence
defect formation is not a problem as is the case for the etched
dots and wires
described above. However the strain fields only produce a weak
modulation of
the band gap and so the confinement potential is relatively
small. At room
temperature carriers are thermally excited from the dots or
wires.
Electrostatically induced dots and wires
If a thin metal layer is deposited on the surface of a
semiconductor (a
Schottky contact) then a voltage can be applied between the
metal and the
semiconductor. This voltage has the effect of either raising or
lowering the
energies of the conduction and valence bands near the surface,
with respect
to their energies deeper in the semiconductor. If the bands are

raised then a
potential minimum is created for holes near to the surface.
Alternatively if the
bands are lowered a potential minimum for electrons is created.
This is shown
in the following figure.
If the metal layer used to make the Schottky contact is patterned
using
lithography and etching, then the resultant shapes can be used to
locally
modulate the conduction and valence bands, forming quantum
wires or
quantum dots. An added sophistication is to form two slightly
separated metal
strips on the semiconductor surface, a so-called split gate. By
applying
appropriate voltages a potential minimum is created in the
region between the
gates, the width of which is determined by the size of the
applied voltage.
Hence a wire of variable width is created.
Electrostatically induced nanostructures form clean systems as
only the metal
needs to be etched, not the semiconductor. However the
potential minima are
not very deep and the spacing between the energy levels is
small, they are
hence only suitable for low temperature operation. Their main
limitation
however is that only electrons or holes are confined in a given
structure, they
are hence not suitable for optical applications.
V

V
The effect of applying a voltage to a Schottky contacted
semiconductor
9
Quantum well width fluctuations
The width of a quantum well is not constant but exhibits a
spatial fluctuation
(see the following figure). Because the confined energy levels
depend upon
the well width, potential minima are formed for electrons and
holes at points
where the well width is above its average value. These
fluctuations confine
the carriers within the plane of the dot (the well provides
confinement along
the growth direction) to give a quantum dot. Although these
dots have good
optical properties their confining potential is very small, as are
the spacings
between the confined levels. The inplane size of the dots is
virtually
impossible to control (the well width fluctuations are essentially
random) and
the spread of dot sizes is very large. These dots have no device
prospects.

Thermally annealed quantum wells
A GaAs-AlGaAs well is grown using standard epitaxial
techniques. A very
finely focussed laser beam is then used to locally heat the
surface. This
produces a diffusion of Al from the AlGaAs into the GaAs well,
causing an
increase in the band gap. By scanning the beam round the edges
of a square
a potential barrier is produced surrounding the unilluminated
centre of the
square. Carriers optically excited within this square are
confined by the
potential barrier and the quantum well, forming a quantum dot.
Quantum wires
can also be formed by scanning the laser beam along the edges
of a
rectangle. Because the minimum size of the focussed laser beam
is ~1µm the
minimum size of the dots is fairly large (~100nm). This results
in very closely
spaced energy levels and, in addition, the annealing processes
can affect the
optical quality of the semiconductor. This technique also
requires specialised,
non-standard equipment.
Semiconductor nanocrystals
Very small semiconductor particles, which act as quantum dots,
can be
formed in a glass matrix by heating the glass with a small
percentage of a
suitable semiconductor. Dots with radii between 1~40nm are
formed, the
radius being a function of the temperature and heating time. The

main
limitation of these dots is that, because they are formed in an
insulating glass
matrix, the electrical injection of carriers is not possible.
Quantum well width fluctuations. The electrons and holes are
localised in
regions where the well width is above its average value (blue
dashed line).
10
Colloidal quantum dots
These are formed by injecting organometal reagents into a hot
solvent.
Nanoscale crystallites grown in the solution with sizes in the
range 1~10nm.
Subsequent chemical and physical processing can be used to
select a subset
of the crystallites with good size uniformity. The dots are
formed from II-IV
semiconductors, including CdS, CdSe and CdTe. The dots
exhibit good
optical properties but as they are free standing the electrical
injection of
carriers is not possible.
Summary and conclusions
In this lecture we have looked briefly at the two established

epitaxial
techniques (MBE and MOVPE) used to grow two dimensional
quantum wells.
We then considered the main requirements for the properties of
semiconductor nanostructures, before discussing the various
techniques
which have been developed to produce quantum wires and
quantum dots. Of
the techniques used to produce wires the most important are the
v-groove
and electrostatic induced ones. Only the former technique has
been applied to
room temperature device applications (mainly lasers) although
it still has a
number of disadvantages. For quantum dots, growth on
patterned substrates,
strain induced structures, electrostatic induced structures,
quantum well width
fluctuations, quantum well thermal annealing and colloidal dots
have all been
used to study physics in zero-dimensional systems (generally at
very low
temperatures). However none of these techniques has so far
been suitable for
room temperature device applications. We will see in the next
lecture that self-
organised techniques come the closest to producing ideal dots.
Further reading
The epitaxial techniques of MBE and MOVPE are discussed in
Davies ‘The
Physics of Low-Dimensional semiconductors’. Bimberg,
Grundmann and
Ledentsov ‘Quantum Dot Heterostructures’ discuss some of the
requirements
for semiconductor nanostructures. Some of the numerous

fabrication
techniques developed to produce wires and dots are described in
the
previously mention books and in the book by Weisbuch and
Vinter ‘Quantum
Semiconductor Structures’
More information can be obtained from a number of research
papers.
Suggestions are
• A close look on single quantum dots, A Zrenner, Journal of
Chemical
Physics Volume 112 page 7790 (2000). Provides an overview of
many of
the techniques used to prepare quantum dots. Many useful
references.
• Photoluminescence from a single GaAs/AlGaAs quantum dot,
K Brunner
et al Physical Review Letters Volume 69 Page 3216 (1992).
Thermally
annealed dots.
• Quantum size effect in semiconductor microcrystals, A
Ekimov et al Solid
State Communications Volume 56 Page 921 (1985).
Semiconductor
nanocrystals.
• Luminescence from excited states in strain induced InGaAs
quantum dots,
H Lipsanen et al, Physical Review B Volume 51 page 13868
(1995). Strain
induced dots.

11
• One-dimensional conduction in the two-dimensional electron
gas in a
GaAs-AlGaAs heterojunction, T J Thornton et al, Physical
Review Letters
Volume 56 Page 1198 (1986). Electrostatically induced wires.
• Synthesis and characterisation of nearly monodispersive CdE
(E=S, Se,
Te) semiconductor nanocrystallites, C B Murray et al, Journal
of the
Americal Chemical Society Volume 115 Page 8706 (1993).
Colloidal
quantum dots.
• Formation of a high quality two-dimensional electron gas on
cleaved
GaAs, L N Pfeiffer et al, Applied Physics Letters Volume 56
Page 1697
(1990). Cleaved edge overgrowth of quantum wires.
• Patterned quantum well heterostructures grown by OMCVD on
non-planar
substrates - applications to extremely narrow SQW lasers, R
Bhat et al
Journal of Crystal Growth Volume 93 Page 850 (1988). V-
groove quantum
wires.
• Molecular beam epitaxy growth of tilted GaAs AlAs

superlattices by
deposition of fractional monolayers on vicinal (001) substrates,
J M Gaines
et al, Journal of Vacuum Science and Technology B Volume 6
Page 1381
(1988). Growth of quantum wires on vicinal surfaces.
• Self-limiting growth of quantum dot heterostructures on
nonplanar {111}B
substrates, A Hartmann et al Applied Physics Letters Volume 71
Page
1314 (1997). Growth of quantum dots on patterned substrates.
• Homogeneous linewidths in the optical spectrum of a single
gallium
arsenide quantum dot, D Gammon et al, Science Volume 273
Page 87
(1996). Dots formed from quantum well width fluctuations.
12
Lecture 8 – The fabrication of semiconductor
nanostructures II
Introduction
In this lecture we will look at the most successful technique
developed so-far
to fabricate semiconductor quantum dots – self-assembly. The
use of this

technique will be described and some of the properties of
resultant dots will
be discussed.
The growth of strained semiconductor layers
Generally when growing quantum wells it is arranged that the
well, barrier and
substrate semiconductors have the same atomic spacing (lattice
constant).
For example GaAs and AlGaAs have almost identical lattice
constants. GaAs
quantum wells with AlGaAs barriers can therefore be grown on
GaAs
substrates. If we try to grow a semiconductor which has a very
different lattice
constant to that of the substrate, then initially it adjusts its
lattice constant to fit
that of the substrate and the semiconductor will be strained.
However to strain
a material requires energy. Hence as the thickness of the
semiconductor
increases energy will build up. Eventually there is sufficient
energy to break
the atomic bonds of the semiconductor and dislocations (a
discontinuity of the
crystal lattice) form. Beyond this point the semiconductor can
grow with its
own lattice constant, strain energy no longer builds up. The
thickness of
semiconductor which can be grown before dislocations form is
known as the
critical thickness. The critical thickness is a function of the
semiconductor
being grown and also the degree of lattice mismatch between
this
semiconductor and the underlying semiconductor or substrate.

Dislocations provide a very efficient mechanism for non-
radiative carrier
recombination. Hence a structure which contains dislocations
will, in general,
have a very poor optical efficiency. When growing strained
semiconductor
layers it is therefore important not to exceed the critical
thickness.
A good example of a strained semiconductor system is InxGa1-
xAs-GaAs.
When growing quantum wells InxGa1-xAs forms the wells, as it
has the smaller
band gap, with GaAs forming the barriers. As the In
composition of InxGa1-xAs
increases the lattice mismatch between InxGa1-xAs and GaAs
also increases.
Because InxGa1-xAs-GaAs quantum wells are generally grown
on a GaAs
substrate the InxGa1-xAs wells are strained to fit the GaAs
lattice constant.
For low In compositions (x~0.2) it is possible to grow quantum
wells with
thicknesses up to a few 10s nm before the critical thickness is
reached.
However for higher x the critical thickness decreases rapidly.
Self-assembled growth of quantum dots
The lattice mismatch between InAs and GaAs is very large (7%)
and the
critical thickness for the growth of an InAs layer on GaAs is
expected to be
very small (of the order of a few atomic layers). When InAs is
first deposited
on GaAs it grows as a highly strained, flat layer (two
dimensional growth).
However for certain growth conditions before dislocations start

to form the
growth changes to three dimensions in the form of small
islands. These
islands form the quantum dots and sit on the original two
dimensional layer,
which is known as the wetting layer.
13
This behaviour in which the growth transforms from two to
three dimensional
is known as the Stranski-Krastanow growth mode. It is caused
by a trade off
between elastic and surface energy. All surfaces have an
associated energy
because of their incomplete atomic bonds. The surface energy is
directly
proportional to the area of the surface. Hence the surface after
the islands
start to form has a greater energy than the original flat surface.
However
within the islands the lattice constant of the semiconductor can
start to shift
back to its bulk value, hence reducing the elastic energy (note
this shift is
gradually and increases with distance along the growth
direction, there are no
dislocations formed - see following figure). Because the
reduction in elastic
energy is greater than the increase in surface energy the
transformation to

three dimensional growth represents the lowest energy, and
hence most
favourable, state. Following the growth of the dots they are
generally
overgrown by the barrier semiconductor GaAs. The following
figure shows the
main steps in the formation of self-assembled quantum dots.
InAs
GaAs
(a)
(b)
(c)
(d)
InAs
GaAs
(a)
(b)
(c)
(d)
LHS - change in the lattice spacing for atoms in a self-
assembled quantum dot.
RHS the main stages in the formation of a self assembled dot:

(a) GaAs substrate
(with buffer layer), (b) initial 2D growth of InAs (c)
transformation above critical
thickness to 3D island-like growth (d) over growth of dots with
GaAs.
14
The Physical Properties of Self-Assembled Dots
The physical properties of self assembled dots (e.g. size, shape
and density)
depend to some extent on the conditions used to growth them
(e.g.
temperature and growth rate). Typically they have a base size
between
10~30nm, a height of 5~20nm and a density of
1x1010~1x1012cm-2. However
values outside this range may be possible by carefully
controlling the growth
conditions. Because of their small size the energy separation
between their
confined levels is relatively large (40~70meV). They contain no
dislocations
and so exhibit excellent optical properties. They have a high
two dimensional
density and multiple layers can be grown (see below). They are
grown entirely
by an epitaxial process and can easily be incorporated within
the intrinsic
region of a p-i-n structure. Their confinement potential is
relatively deep (100-

300meV) and both electrons and holes are confined. Uniformity
is reasonable
but could be better (see below). The following figure shows a
cross-sectional
transmission electron microscope (TEM) image of a typical
quantum dot. This
is a bare dot which has not been over grown with GaAs (it is
difficult to obtain
similar images of over grown dots as there is very little contrast
between InAs
and GaAs in the TEM images).
The following figure shows an AFM image of quantum dot
sample. Again the
dots have not been overgrown with GaAs.
A cross-sectional TEM image of an InAs quantum dot grown on
GaAs. The base of
the dot is approximately 18nm.
An AFM image of a quantum dot sample. Note the
exaggerated vertical scale.
15
The shape and composition of self assembled quantum dots
Although extensively studied there is still considerable
uncertainty as to the

precise shape of self assembled quantum dots. Various shapes
have been
claimed including pyramids, truncated pyramids, cones and
lenses (part of a
sphere). One problem in determining the shape is that it is
difficult to study
dots which have been overgrown. Although bare dots can be
studied using
AFM and related surface techniques, there is some evidence that
the dot
shape may change when they are overgrown. It may be that the
shape of self
assembled quantum dots depends upon the precise growth
conditions.
A further complication is the composition of the dots. The dots
can either be
grown using pure InAs or the alloy InGaAs. However even when
grown with
InAs there is evidence that the dots consist of InGaAs
indicating the diffusion
of Ga into the dots from the surrounding GaAs. The Ga
composition in the
dots is unlikely to be uniform leading to a highly complicated
system which is
difficult to model theoretically (see below).
Multiple quantum dot layers
Once one layer of dots has been deposited and overgrown with
GaAs a flat
surface is formed upon which a second layer can be deposited.
It is hence
possible to grow multiple layers of dots. When the first dot
layer is deposited
the positions of the dots are reasonably random. As the InAs in
the dots
gradually returns to its bulk lattice constant as the dot height

increases, the
initial GaAs deposited on top of the dot will be slightly
strained. A strain field
will be produced in the GaAs above each dot, although this will
gradually
decrease to zero as the thickness of the GaAs is increased.
However if, when
the next dot layer is deposited, these strain fields are still
present (only a thin
GaAs layer has been grown) they may act as nucleation sites for
the next
layer of dots. In this case the dots are vertically aligned and
stacks of aligned
dots may be formed with 10 or more dots in a stack. This
alignment only
occurs when successive dot layers are separated by very thin
GaAs layers
(<10nm). For thicker GaAs layers the strain field is essentially
zero when the
next layer is deposited and the dots form at random positions.
The following
figure shows a cross sectional transmission electron microscope
image of a
sample containing 10 dot layers with each layer separated by
9nm of GaAs.
The vertical alignment of the dots into stacks can be clearly
seem. This
alignment may be important for the electronic and optical
properties as it is
possible that electrons and holes may be able to move between
the dots in a
stack.
A cross sectional TEM image of vertically aligned quantum
dots.

16
Dot uniformity
The growth of self assembled dots is a semi-random process.
Dots at different
positions on the surface will start to form at slightly different
times as the
amount of InAs deposited will not be totally uniform. This
results in the final
shape and size (and possibly composition) varying slightly from
dot to dot. As
the energies of the confined energy states are a function of the
dot size,
shape and composition these will also vary from dot to dot.
The emission from a single dot will consist of a very sharp line
(similar to the
emission from an atom). However most experiments on self
assembled
quantum dots probe a large number of dots. For example a
typical
photoluminescence experiment will use a laser beam focussed to
a diameter
of 250µm. If the dot density is 1x1011cm-2 the area of the laser
beam will
contain ~50 million dots, each of which will contribute to the
measured
spectrum. As each dot will emit light at a slightly different
energy the sharp
emission from each dot will merge into a broad, featureless
emission. This is

known as inhomogeneous broadening. Only if the number of
dots probed can
be reduced significantly (e.g. by reducing the diameter of the
laser beam - see
later lectures) will the individual sharp emission lines be
observed.
The non-uniformity of self-assembled quantum dots and the
resultant
inhomogeneous broadening of the optical spectra is a
disadvantage for a
number of potential device applications. For example the
absorption is spread
out over a wide energy range instead of being concentrated at a
single
energy. The inhomogeneous broadening also complicates
fundamental
physics studies; as will be discussed in later lectures. However
there are
some applications (e.g. optical memories) which make use of
the
inhomogeneous broadening. The following figure shows
photoluminescence
spectra of different numbers of quantum dots. This is achieved
by evaporating
an opaque metal mask on the sample surface in which holes of
different sizes
are formed. By shining the laser beam through these different
size holes,
different numbers of dots can be probed.
Photoluminescence spectra of different numbers of quantum
dots.
From Gammon MRS Bulletin Feb. 1998 Page 44

17
Theoretical modelling of self-assembled quantum dots
Self assembled quantum dots have a high degree of strain and
this strain is
non-uniform. In addition they have a complicated shape. This
makes the
calculation of the confined energy levels very difficult. The
following figures
show the distribution of strain, calculated for pyramidal shaped
dots, and the
shapes of the wavefunctions for the lowest energy electron and
hole states.
As we will see in later lectures the optical spectra of the
quantum dots are
very complicated and difficult to interpret. Hence it is still not
possible to test
the predictions of the various available theoretical models. In
addition many of
the input parameters required for the models (e.g. the exact dot
size, shape
and composition) are still not well known.
The strain distribution in self assembled quantum dots: (a)
through the wetting
layer, (b) through the dot. From Stier et al PRB 59, 5688
(1999).

Electron and hole wavefunctions for the lowest energy confined
quantum dot
states. From Stier et al ibid.
18
Different self assembled quantum dot systems
The most commonly studied self assembled system consists of
InAs or
InGaAs dots grown within a GaAs matrix. The band gap of bulk
InAs is 0.4eV
but quantum confinement and strain increase this to between
0.95 and 1.4eV,
the precise value being dependent on the shape and size of the
dots. This
energy range correspond to wavelengths 1300~900nm, which is
in the near
infrared region of the electromagnetic spectrum.
The emission energy can be increased if InAs or InGaAs dots
are grown in an
AlGaAs matrix. This allows energies up to ~1.8eV (≡690nm) to
be obtained. Al
can also be added to the dots to increase their emission energy
(AlInAs-
AlGaAs dots).
Self assembled dots have also been fabricated from other
semiconductor
combinations where there is sufficient lattice mismatch.
Examples include InP

dots in GaInP (emission energy ~1.6-1.9eV [~775-650nm]), Ge
dots in Si and
InSb, GaSb or AlSb dots in GaAs (emission energy ~1.0-1.3eV
[~1200-
950nm]). More recently there have been attempts to grow dots
in the wide
band gap nitride semiconductors GaN, InN and AlN.
Summary and Conclusions
In this lecture we have looked at the most promising method for
producing
quantum dots suitable for electro-optical applications. The main
properties of
quantum dots prepared using the self-assembly technique are
compared with
other types of dots and wires in the following table. Self-
assembled dots
satisfy the majority of requirements for device applications,
possibly with the
exception of uniformity. As we will see in later lectures, a
number of devices
based on self assembled quantum dots have now been
demonstrated.
Further reading
'Quantum Dot Heterostructures' by Bimberg et al provides a
comprehensive
overview of the self-assembly technique including a discussion
of optical,
electrical and structural studies and devices based on these
quantum dots.

19
Fa
br
ic
at
io
n
te
ch
ni
qu
e
S
pa
ci
n
g
of
en
er
gy

le
ve
ls
(s
iz
e)
O
pt
ic
al
Q
ua
lit
y
U
ni
fo
rm
ity
D
en
si

ty
E
pi
ta
xi
al
G
ro
w
th
D
ep
th
o
f c
on
fin
in
g
po
te
nt
ia
l

E
le
ct
ro
n
&
h
ol
e
co
nf
in
em
en
t
p-
i-n
st
ru
ct
u
re
s

Li
th
og
ra
pg
hy
a
nd
e
tc
hi
ng
po
or
V
er
y
po
or
G
oo
d
re
as

on
ab
le
-
co
nt
ro
lla
bl
e
Y
es
b
ut
e
xt
ra
s
te
ps
re
qu
ire

d
H
ig
h
Y
es
Y
es
C
le
av
ed
e
dg
e
ov
er
gr
ow
th
po
or
G
oo

d
V
er
y
go
od
H
ig
h
-
co
nt
ro
lla
bl
e
Y
es
b
ut
e
xt
ra
s

te
ps
re
qu
ire
d
S
m
al
l
Y
es
Y
es
V
ic
in
al
s
ur
fa
ce

s
Y
es
G
oo
d
po
or
H
ig
h
Y
es
H
ig
h
Y
es
Y
es
P
at
te
rn

ed
s
ub
st
ra
te
s
R
ea
so
na
bl
e
G
oo
d
R
ea
so
na
bl
e
M
ed

iu
m
-
co
nt
ro
lla
bl
e
Y
es
b
ut
e
xt
ra
s
te
ps
re
qu
ire

d
H
ig
h
Y
es
Y
es
S
tr
ai
n
in
du
ce
d
R
ea
so
na
bl
e
G
oo

d
R
ea
so
na
bl
e
m
ed
iu
m
-
co
nt
ro
lla
bl
e
Y
es
b
ut
e
xt

ra
s
te
ps
re
qu
ire
d
R
ea
so
na
bl
e
Y
es
N
o
E
le
ct
ro
st

at
ic
al
ly
in
du
ce
d
R
ea
so
na
bl
e
N
ot
ap
pl
ic
ab
le
go
od

M
ed
iu
m
-
co
nt
ro
lla
bl
e
Y
es
b
ut
e
xt
ra
s
te
ps
re

qu
ire
d
Lo
w
N
o
N
o
T
he
rm
al
ly
a
nn
ea
le
d
po
or
re
as
on

ab
le
po
or
po
or
Y
es
b
ut
e
xt
ra
s
te
ps
re
qu
ire
d
R
ea
so

na
bl
e
Y
es
no
na
no
cr
ys
ta
ls
go
od
go
od
re
as
on
ab
le
go
od
N

o
H
ig
h
ye
s
no
C
ol
lo
id
al
d
ot
s
go
od
go
od
go
od
go
od

N
o
H
ig
h
ye
s
no
S
el
f-
as
se
m
bl
y
go
od
go
od
re
as
on
ab

le
hi
gh
-
s
em
i
co
nt
ro
lla
bl
e
Y
es
hi
gh
ye
s
ye
s

20
Lecture 9 – Modulation doping and transport
phenomena in semiconductor nanostructures
Introduction
Using a technique known as modulation doping it is possible to
obtain
extremely high carrier mobilities in semiconductor
nanostructures. This has a
number of practical applications and also leads to the
observation of a
number of highly novel transport related phenomena.
Modulation Doping
We saw in Lecture 2 that in a bulk semiconductor the carrier
mobility is limited
by phonon scattering at high temperatures and scattering from
charged
impurity atoms at low temperatures. The temperature
dependence of the
electrical mobility hence has the following form.
Although the low temperature mobility can be increased by
reducing the
impurity density this lowers the electrical conductivity as it is
these impurities
which provide the free carriers (doping).
In a semiconductor nanostructure however it is possible to
spatially separate
the dopant atoms and the resultant free carriers, significantly
reducing this
scattering mechanism. This leads to very high low temperature
carrier

mobilities. This arrangement, which is known as remote or
modulation doping,
is shown schematically for n-type doping of a quantum well in
the following
figure. In this case the donor atoms are placed only in the wider
band gap
barrier material, the quantum well is undoped1. However the
electrons
released by the donor atoms in the barrier transfer into the
lower energy well
states, resulting in a spatial separation of the free electrons and
the charged
donor atoms. The confined electrons in the quantum well are
said to form a
two-dimensional electron gas (2DEG); a two-dimensional hole
gas can
similarly be formed by doping the barriers p-type. The non-zero
charge
1 This is simply achieved during MBE growth by only opening
the shutter in front of the cell
containing the dopant atoms during growth of the barriers. In
MOVPE the gas carrying the dopant
atoms is similarly switched.
M
ob
ili
ty
Temperature
Phonon
scattering

Impurity
scattering
M
ob
ili
ty
Temperature
Phonon
scattering
Impurity
scattering
Temperature dependence of electrical mobility for a
semiconductor
21
present in both the barriers and the well2 adds an electrostatic
potential
energy which results in a bending of the band edges, as
indicated in figure
(b). This band bending allows the formation of a modulation
doping induced
2DEG at a single interface (a single heterojunction) between
two different

semiconductors, as shown in figure (c). Here the combined
effects of the
conduction band offset and the band bending result in the
formation of a
triangular shaped potential well which restricts the motion of
the electrons to
two dimensions.
In a modulation doped structure the barrier region immediately
adjacent to the
well is generally undoped, forming a spacer layer, which further
separates the
charged dopant atoms and the free carriers. By optimising both
the width of
this spacer layer and the structural uniformity of the interface,
and by
2 The total charge of the structure remains zero but there are
equal and opposite charges in the well and
barriers.
(a) (b) (c)
Donor atom Free electron
(a) process of n-type modulation doping in a quantum well, (b)
as (a) but also showing the
effects on the band edges of the non-zero space charges, (c)
modulation doping of a single
heterostructure.
0.1 1 10 100
1

10
100
1000
1980
1982
1989
GaAs-AlGaAs
single heterojunctions
Clean bulk GaAs
Bulk GaAs
El
ec
tro
n
M
ob
ili
ty
(c
m
2 V

-1
s-1
)
Temperature (K)
Temperature dependence of the mobility of bulk GaAs (standard
and clean) and three GaAs-
AlGaAs single heterostructures (numbers give the
corresponding years). Data taken from
Stanley et al (Appl. Phys. Lett. 58, 478 (1991)) and Pfeiffer et
al (ibid 55, 1888 (1989))
22
minimising unintentional background impurities, it is possible
to achieve
extremely high low temperature mobilities. The previous figure
compares the
temperature variation of the electron mobility of standard bulk
GaAs, a very
clean bulk specimen of GaAs and a series of GaAs-AlGaAs
single
heterojunctions. At high temperatures, where mobility is limited
by phonon
scattering, the mobilities of the different structures are very
similar. At low
temperatures the mobility of bulk GaAs is increased in the
cleaner material

where a lower impurity density reduces the charged impurity
scattering.
However the absence of doping results in a low carrier density
and, as a
consequence, a low electrical conductivity. It is therefore not
possible to
achieve both a high conductivity and high mobility in a bulk
semiconductor.
Modulation doping however results in both high free carrier
densities and low
temperature mobilities more than two orders of magnitude
larger than those of
clean bulk GaAs and almost four orders of magnitude larger
than ‘standard’
bulk GaAs. The data for the different heterojunctions presented
in the figure
demonstrates how the low temperature mobility of a single
heterojunction has
increased over time, reflecting optimisation of the structure, the
use of purer
source materials and cleaner MBE growth reactors. The ability
to produce
2DEGs of extremely high mobility has allowed the observation
of a range of
interesting physical processes, a number of which will be
discussed later in
this lecture and the following lecture.
Modulation doping is now used extensively to provide the
channel of field
effect transistors (FETs), particularly for high frequency
applications. Such
devices are known as high electron mobility transistors
(HEMTs) or
modulation doped field effect transistors (MODFETs). Although
the use of
modulation doping provides negligible enhancement of the room

temperature
carrier mobility, the free carriers are confined to a two
dimensional sheet in
contrast to a layer of non-zero thickness for conventional
doping. This precise
positioning of the carriers results in devices exhibiting more
linear
characteristics and, for still unclear reasons, these devices also
exhibit lower
noise. III-V semiconductor HEMTs or MODFETs operating up
to ~300GHz are
achievable with applications including mobile communications
and satellite
signal reception.
The Hall effect in bulk semiconductors
The following figure shows the geometry used to study the Hall
effect. A
current Ix flows along a semiconductor bar to give a current
density Jx (=Ix/wh).
A magnetic field B applied normal to the axis of the bar
produces a magnetic
force on each moving charge carrier given by qvB, where q is
the charge and
v the carrier drift velocity. This force causes the carrier motion
to be deflected
in a direction perpendicular to both the field and the original
motion as shown
in the figure. As a consequence of this deflection there is a
build up of the
charge carriers, and hence a non-zero space charge, along the
side of the
bar, which results in the creation of an electric field along the
y-axis, Ey. This
so-called Hall field produces an electrostatic force (qEy) on the
charge carriers

which opposes the magnetic force. Equilibrium is quickly
reached where the
two forces balance to give a zero net force.
/( ) / 1/( )y y x y x HqE qvB E vB J B nq or E J B nq R= ⇒ = =
= =
23
where the last step follows from the relationship Jx=nqv (see
Lecture 2). The
ratio Ey/(JxB) is known as the Hall coefficient and has a value
1/(nq). As Ey
produces a voltage between the sides of the bar, given by
Vy=wEy, all three
quantities Ey, Jx and B are easily determined allowing RH and
hence the
product nq to be found. A Hall measurement of a bulk
semiconductor hence
allows the carrier density n to be determined as well as the
majority carrier
type (electrons or holes) from the sign of RH.
The Quantum Hall Effect
The Hall effect can also be observed in a nanostructure
containing a 2DEG.
Experimentally the electric field along the sample, Ex, can also
be determined

by measuring Vx as shown in the previous figure. This allows
two resistivities
to be determined, defined as:
ρ ρxx
x
x
xy
y
x
E
J
E
J
= =
Because RH=Ey/(BJx), for a bulk semiconductor ρxy=RHB,
which increases
linearly with increasing magnetic field, with ρxx remaining
constant. However
for a two-dimensional system a very different behaviour is
observed, as
shown in the following figure. In this case although ρxy
increase with
increasing field, it does so in a step-like manner. In addition
ρxx oscillates
between zero and non-zero values, with zeros occurring at fields

where ρxy
forms a plateau. This surprising behaviour of a two-dimensional
system is
known as the Quantum Hall effect and was discovered in 1980
by Klaus von
Klitzing, for which he was awarded the 1985 Nobel Physics
Prize. The
Quantum Hall effect arises as a result of the form of the density
of states of a
two-dimensional system in a magnetic field. This corresponds to
that of a fully
quantised system, with quantisation in one direction resulting
from the
physical structure of the sample and quantisation in the
remaining two
directions provided by the magnetic field. Diagram (a) of the
following figure
shows the discrete energy levels for a perfect system. However
in any real
system the levels are broadened by carrier scattering events and
the energy
levels have the form given by the right hand diagrams. These
‘bands’ of states
VXVY
IX
JX
B
w
h Ex
Ey

The geometry of the Hall effect
0 1 2 3 4 5 6 7 8 9
0
2000
4000
6000
8000
10000
12000
14000
ρ
XY resistance (h/e
2)
1/7
1/6
1/5
1/4
1/3
1/2
(x60)ρxx

ρ
xy
R
es
is
ta
nc
e
(Ω
)
Magnetic Field (T)
An example of the integer quantum Hall
effect. Data taken from Paalanen et al,
Phys. Rev. B. 25, 5566 (1982)
24
have similarities with the energy bands in a solid (see Lecture
1) and as in
that case the electronic properties are a very sensitive function
of how the
charge carriers occupy the bands. Each band formed by the

magnetic field is
known as a Landau level and it can be shown that the
degeneracy of each
Landau level is given by
eB
h
Hence as the field is increased the degeneracy of each level also
increases.
Therefore for a given carrier density in the structure the number
of occupied
levels decreases with increasing field. In (c) the Landau level
degeneracy is
such that only the lowest two levels are occupied. This
corresponds to the
case of an insulator with completely filled bands followed by
completely empty
bands. In this case the structure has a zero conductivity
(σxx=0). In (b) the
field has been increased so that now the second Landau level is
only half
filled. Conductivity is possible for the electrons in this level
and hence σxx≠0.
Under conditions of high magnetic field the following
relationships relate the
conductivity and resistivity components
2
1xx
xx xy H

xy xy
R B
σ
ρ ρ
σ σ
≈ ≈ =
The first relationship shows that the zero conductivity values
obtained when
exactly an integer number of Landau levels are occupied results
in a zero
value for ρxx.
The plateau values of ρxy can be found by noting that if exactly
j Landau levels
are fully occupied then
S
eB
N j
h
=
where NS is the two dimensional carrier density. From the
above definition of
ρxy
(a) (b) (c)
Quantised energy levels of a two dimensional system placed in a

magnetic field (a) case of
zero level broadening (b) and (c) with level broadening and for
different occupations of the
levels up to the dashed line.
25
2
1 25812.8
xy H
S
B h
R B
N e j e j
ρ = = = = Ω
The plateau values of ρxy are sample independent and are
related to the
fundamental constants h and e. Values for ρxy can be measured
to very high
accuracy and are now used as the basis for the resistance
standard and also
to calculate the fine structure constant α=µ0ce2/2h, where the
permeability of
free space, µ0, and the speed of light, c, are defined quantities.

The parameter j is known as the filling factor The quantum Hall
effect
discussed previously occurs for integer values of j and is
therefore known as
the integer quantum Hall effect. However, in samples with very
high carrier
mobilities, plateaus in ρxy and minima in ρxx are also observed
for fractional
values of j, giving rise to the fractional quantum Hall effect.
The discovery and
theoretical interpretation of the fractional quantum Hall effect,
which results
from the free carriers behaving collectively rather than as single
particles, lead
to the award of the 1998 Nobel Physics prize to Stormer, Tsui
and Laughlin.
An example of the fractional quantum Hall effect is given in the
above figure
which was recorded at very low temperatures for a very high
mobility GaAs-
AlGaAs single heterostructure. In addition to minima in ρxx and
plateaus in ρxy
for integer values of the filling factor, similar features are also
observed for
non-integer values, for example 3/5, 2/3, 3/7 etc.
Ballistic Carrier Transport
The carrier transport considered so far is controlled by a series
of random
scattering events (see Lecture 2). However the high carrier
mobilities which
can be obtained by the use of modulation doping correspond to
very long path
lengths between successive scattering events, lengths that can
significantly

An example of the fractional quantum Hall effect which where
the filling factor j has non
integer values. The integer quantum Hall effect is still observed
at low fields. Figure from
R Willet et al Phys. Rev. Lett. 59, 1776 (1987).
26
exceed the dimensions of a nanostructure. In this case a carrier
can pass
through the structure without experiencing a scattering event, a
process
known as ballistic transport. Ballistic transport conserves the
phase of the
charge carriers and leads to a number of novel phenomena, two
of which will
now be discussed.
When carriers travel ballistically along a quantum wire there is
no dependence
of the resultant current on the energy of the carriers. This
results from a
cancellation between the energy dependence of their velocity
(v=(2E/m*)1/2)
and the density of states, which in one dimension varies as E-
1/2 (see Lecture
6). For each subband occupied by carriers, a conductance equal
to 2e2/h is
obtained, a behaviour known as quantised conductance. If the
number of
occupied subbands is varied then the conductance of the wire

will exhibit a
step-like behaviour, with each step corresponding to a
conductance change of
2e2/h. Quantum conductance is most easily observed in
electrostatically
induced quantum wires (see Lecture 7). The gate voltage
determines the
width of the wire, which in turn controls the energy spacing
between the
subbands. For a given carrier density, reducing the subband
spacing results
in the population of a greater number of subbands and hence an
increased
conductance. The following figure shows quantum conductance
in a 400nm
long electrostatically induced quantum wire. These
measurements are
generally performed at very low temperatures to obtain the very
high
mobilities required for ballistic transport conditions. In contrast
to the plateau
values observed for ρxy in the quantum Hall effect, which are
independent of
the structure and quality of the device, the quantised
conductance values of a
quantum wire are very sensitive to any potential fluctuations
which result in
scattering events. This sensitivity prevents the use of quantum
conductance
as a resistance standard.
The inset to the above figure shows a structure in which a
quantum wire splits
into two wires which subsequently rejoin after having enclosed
an area A.
Under ballistic transport conditions the wavefunction of an
electron incident on

the loop will split into two components which, upon
recombining at the far side
-1.6 -1.4 -1.2 -1.0 -0.8 -0.6
0
2
4
6
8
10
12 Split gate
2D EG
O hmic co ntacts
Split gate
2D EG
O hmic co ntacts
C
on
du
ct
an
ce

(u
ni
ts
2
e2
/h
)
Split Gate Bias Voltage (V)
Example of quantum conductance in a
quantum wire defined electrostatically
from a 2DEG. The inset shows the
sample geometry. Data from Hamilton et
al, Appl. Phys. Lett. 60, 2782 (1992).
0 10 20 30 40 50 60 70 80
50
100
150
200
250
300
AA

R
es
is
ta
nc
e
(Ω
)
Magnetic Field (mT)
An example of the Aharonov-Bohm effect in
an electrostatically defined quantum ring.
The inset shows the sample geometry. Data
from Timp et al, Phys. Rev. B. 39, 6227
(1989).
27
of the loop, will interfere. If a magnetic field is now applied
normal to the plane
of the loop an additional phase difference is acquired or lost by
the
wavefunctions, depending upon the sense in which they traverse
the loop.
The phase difference increases by 2π when the magnetic flux
through the

loop, given by the area multiplied by the field (BA), changes by
h/e. Hence as
the magnetic field is increased the system will oscillate between
conditions of
constructive interference (corresponding to a high conductance)
and
destructive interference (corresponding to low conductance).
The change in
field (∆B) between two successive maxima (or minima) is given
by the
condition ∆BA=h/e, resulting in the conductance of the system
oscillating
periodically with increasing field. An example of this
behaviour, known as the
Aharonov-Bohm effect is shown in the previous figure for a
loop of diameter
1.8µm formed from the 2DEG of a GaAs-AlGaAs single
heterostructure by
patterning the surface with metal gates defined by electron
beam lithography.
In this lecture we have shown how modulation doping allows
the attainment of
very high carrier mobilities at low temperatures. This allows the
observation of
a number of novel effects including the integer and fractional
quantum Hall
effects. The high mobilities correspond to long average
distances between
scattering events and carriers may be able to pass through a
nanostructure
ballistically without undergoing a single scattering event. In
this case
processes which include quantised conductance and the
Aharonov-Bohm

effect are observable.
Further reading
The paper by Pfeiffer et al (Appl. Phys. Lett. 55, 1888 (1989))
describes the
optimisation of the MBE technique to give very high electron
mobilities.
Carrier scattering processes are discussed in detail in ‘The
Physics of Low
Dimensional Semiconductors’ by J H Davies. The discussion of
the integer
quantum Hall effect give in this lecture is relatively non-
mathematical. A more
detailed treatment which includes the importance of disorder is
given in ‘Band
theory and Electronic Properties of Solids’ by J Singleton
(OUP).
28
Lecture 10 Tunnelling and related processes in
semiconductor nanostructures
Introduction
Quantum mechanical tunnelling, in which a particle passes
through a
classically forbidden region, is the mechanism by which α
particles escape
from the nucleus during α decay and electrons escape from a
solid in

thermionic emission. Tunnelling can also be observed in
semiconductor
nanostructures where the ability to deposit very thin layers
permits the easy
production of tunnelling barriers. Tunnelling can be observed
either through a
single barrier or through two barriers separated by a quantum
well or quantum
dot. A range of novel physical processes are observed with a
number of
practical applications.
Tunnelling through a single square barrier
Consider the single square barrier of potential height V0 and
thickness a as
shown in the following figure. Such a structure can be easily
fabricated by
depositing a thin layer of a wide band gap semiconductor
between thicker
layers of a narrower band gap semiconductor. Away from the
barrier, and on
both sides, would normally be doped regions to provide a
reservoir of carriers.
By fabricating a suitable device an applied voltage can be used
to vary the
energy of the carriers and their ability to pass through the
barrier is indicated
by the magnitude of current flowing through the device.
The following figure shows the calculated transmission
probability for an
electron of energy E incident on a barrier of height 0.3eV and
thickness 10nm.
The classical result has a value of zero when the electron energy
is less than
the barrier height and one otherwise. In contrast the quantum

mechanical
result is non-zero for energies below that of the barrier height
indicating that
the electron can quantum mechanically tunnel through the
barrier, a region
where classically it would have negative kinetic energy. The
oscillations of the
probability for energies which exceed the barrier height are a
result of the
interference between waves which are reflected from the two
sides of the
barrier.
For electron energies less than the barrier height the
transmission probability
T can be approximated to
Vo
a
E
Schematic diagram of a single barrier tunnelling structure.
29
*
0

2
0
2 ( )16
exp( 2 )
m V EE
T a where
V
κ κ
−
≈ − =
Because of the exponential function the transmission probability
is very
sensitive to both the energy of the electron and the width and
height of the
barrier.
Double barrier resonant tunnelling structures
Of greater practical interest than a single barrier tunnelling
structure is the
case of two barriers separated by a thin quantum well, known as
a double
barrier resonant tunnelling structure (DBRTS). A schematic
diagram of a
DBRTS is shown in the following figure. Quantised energy
levels are formed
in the quantum well as described in Lecture 5.

Calculated transmission coefficient as a function of electron
energy for a single barrier of height
0.3eV. taken from J H Davies ‘The Physics of Low-dimensional
semiconductors’ CUP
I
V
I
I
V
V
(a)
(b)
(d)
(c)
A double barrier resonant tunnelling structure.
30

The previous figure also shows a DBRTS for various applied
voltages. For the
sign of voltage shown electrons travel from left to right.
Electrons are first
incident on the left most barrier through which they must
tunnel. However at
low applied voltages their energy when they have tunnelled into
the well is
below that of the lowest confined state and the two barriers plus
the well
therefore behave as one effectively thick barrier; the tunnelling
probability and
hence the current is very low. As the voltage is increased the
energy of the
electrons tunnelling through the first barrier comes into
resonance with the
lowest state in the well. The effective barrier width is now
reduced and it
becomes much easier for the electrons to pass through the
structure. As a
result the current increases significantly. For further increase in
voltage the
resonance condition is lost and the current decreases. However
additional
resonances may be observed with higher energy confined states.
The figure
also shows the expected current-voltage characteristic of a
DBRTS indicating
the relationship between specific points on the characteristic
and the different
voltage conditions.
The previous figure shows experimental results obtained for a
DBRTS
consisting of a 20nm GaAs quantum well confined between

8.5nm AlGaAs
barriers. Resonances with five confined quantum well states are
observed.
Beyond each resonance a DBRTS exhibits a negative
differential resistance,
a region where the current decreases as the applied voltage is
increased.
Such a characteristic has a number of applications including the
generation
and mixing of microwave signals. Very high frequencies are
possible because
of the rapid transit time of the electrons through the structure.
DBRTS can also exhibit hysteresis in their current-voltage
characteristics,
particularly when the thicknesses of the two barriers are
asymmetrical. A
thinner first barrier allows carriers to tunnel easily into the well
but a thicker
second barrier impedes escape, resulting in charge build up in
the well. This
charge build up modifies the voltage dropped across the initial
part of the
structure and maintains the resonance condition to higher
voltages than would
0
10
20
30
40
50

60
0 1 2 3
0
10
20
30
40
x35
E4
E3
E2
E1
C
ur
re
nt
(m
A
)

Bias Voltage (V)
x100
C
ur
re
nt
(m
A
)
Bias Voltage (V)
Measured current voltage characteristics of a double barrier
resonant tunnelling
structure. Data supplied by P Buckle and W Tagg (University of
Sheffield).
31
occur in the case of an empty well. This broadened resonance is
only
observed as the voltage is increased allowing charge to

accumulate in the
well. If the voltage is taken above the resonance condition the
well empties
and decreasing the voltage results in a narrower resonance as
there is now
no charge accumulation. For such a structure the current follows
a different
path depending upon the direction in which the voltage is
varied; the current-
voltage characteristics exhibit a hysteresis. The inset to the
previous figure
shows the characteristics of an asymmetrical DBRTS with 8.5
and 13nm thick
Al0.33Ga0.67As barriers and a 7.5nm In0.11Ga0.89As quantum
well.
Two important figures of merit for a resonant tunnelling
structure are the
widths of the resonance and the ratio of the current at the peak
of the
resonance to that immediately after the resonance, the peak-to-
valley-ratio.
Once resonance has been reached with the lowest energy
confined quantum
well state it might be expected that current would continue to
flow for higher
voltages because of the continuum of states which exist as a
result of inplane
motion (see Lecture 5). However when an electron tunnels
through the first
barrier not only must energy be conserved but also the two
components of the
inplane momentum or wavevectors kx and ky. Conservation of
kx and ky
prevents tunnelling into higher continuum states as these
correspond to high
values of kx and ky whereas the tunnelling electrons will

generally have
relatively small inplane wavevectors. In fact the electrons to the
left of the first
barrier will have a range of initial energies, a result of their
density and the
Pauli exclusion principle, and hence a range of kx and ky
values. This range of
inplane wavevectors contributes to the width of the resonance.
That the current immediately after a resonance does not fall to
zero indicates
that additional non-resonant tunnelling is occurring. The precise
nature of
these additional processes is still unclear but may include
tunnelling via
impurity states in the barriers or phonon scattering which allows
electrons of
an initially incorrect energy to tunnel via the quantum well
states. In general
the peak-to-valley-ratio decreases as the device temperature is
increased.
Tunnelling via quantum dots – Coulomb blockade
The quantum well of a double barrier resonant tunnelling
structure can be
replaced by a quantum dot. In addition to the modification of
the energy level
structure the small size of a typical quantum dot results in a
new effect. A
small quantum dot will posses a relatively large capacitance. If
a quantum dot
already contains one or more electrons then a significant energy
is required to
add an additional electron as a result of the work that must be
done against
the repulsive electrostatic force between like charges. This
charging energy,

given by e2/2C where C is the dot capacitance, modifies the
energies of the
confined dot states which would occur for an uncharged system.
Charging
effects are most easily understood by referring to a structure of
the form
shown in the inset to the following figure, which consists of a
quantum dot
placed close to a reservoir of free electrons. Applying a voltage
to the metal
gate on the surface of the structure allows the energy of the dot
to be varied
with respect to the reservoir. If a given energy level in the dot
is below the
energy of the reservoir then electrons will tunnel from the
reservoir into the dot
level. Alternatively if the energy level is above the reservoir
then the level will
be unoccupied. Hence by varying the gate voltage the dot states
can be
sequentially filled with electrons. This filling can be monitored
by measuring
32
the capacitance of the
device which will exhibit a
characteristic feature each
time an additional electron
is added to the dot.
The main part of the

previous figure shows the
capacitance trace recorded
for a device containing an
ensemble of self assembled
quantum dots. These dots
have two confined electron
levels; the lowest (ground
state) able to hold two
electrons (degeneracy of
two) with the excited level
able to hold four electrons
(degeneracy of four). In the
absence of charging effects
only two features would be
observed in the capacitance
trace, one at the voltage
corresponding to the filling of the ground state, the other when
the voltage
reaches the value required for electrons to tunnel into the
excited state.
However once one electron has been loaded into the ground
state charging
effects result in an additional energy, and a higher voltage,
being required to
add the second electron. This leads to two distinct capacitance
features
corresponding to the filling of the ground state. Similarly four
distinct features
are expected as electrons are loaded into the excited state
although in the
present case inhomogeneous broadening prevents these being
individually
resolved. This charging behaviour is known as Coulomb
blockade and is
observed experimentally when the charging energy exceeds the

thermal
energy, kT.
Coulomb blockade effects can also be observed in transport
processes where
carriers tunnel through a quantum dot. Suitable dots may be
formed
electrostatically using split gates to define the dot and to
provide tunnelling
barriers between the dot and the surrounding 2DEG which forms
a reservoir
of carriers. An additional gate electrode allows the energy of
the dot to be
varied with respect to the carrier reservoirs. The relatively large
dot size
results in Coulomb charging energies that are much larger than
the
confinement energies. The former therefore dominate the
energetics of the
system. The inset to the following figure shows a schematic
diagram of the
structure where a small bias voltage has been applied between
the left and
right two-dimensional carrier reservoirs. The dot initially
contains N electrons
resulting in an energy indicated by the lower horizontal line. An
additional
electron can tunnel into the dot from the left hand reservoir but
this increases
the dot energy by the charging energy. Hence this process is
only
energetically possible if the energy of the dot with N+1
electrons lies below
the maximum energy of the electrons in the left hand reservoir.
Tunnelling of
this additional electron into the right hand reservoir may
subsequently occur

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4
2.10
2.12
2.14
2.16
2.18
2.20
2.22
2.24
f r e e e l e c t r o n s
q u a n t u m d o t
b l o c k in g b a r r ie r
g a te
Excited state
Ground state
C
ap
ac

ita
nc
e
(n
F)
Voltage (V)
Structure and results from a device in which a
controllable number of electrons can be loaded on to
a quantum dot. Figure redrawn from Fricke et al
Europhysics Lett. 36, 197 (1996).
33
but only if the N+1 dot energy
lies above the maximum
energy of this reservoir. If
these two conditions are
satisfied, requiring that the
N+1 dot energy lie between
the energy maxima of the two
reservoirs, a sequential flow
of single electrons through
the structure occurs; the
system exhibits a non-zero
conductance. As the gate
voltage is used to vary the
dot energy, the condition for
sequential tunnelling will be

satisfied for different values
of N and a series of
conductance peaks will be
observed, an example is
shown in the above figure for
a dot of radius 300nm. This
large dot size results in a
large capacitance and a correspondingly small charging energy
(0.6meV for
the present example). Hence measurements must be performed
at very low
temperatures in order to satisfy the condition e2/2C>>kT. Two
practical
applications of Coulomb blockade will be described in a later
lecture.
In this lecture we have seen that it is possible to fabricate
tunnelling structures
based on semiconductor nanostructures. Double barrier resonant
tunnelling
structures give very non-linear current-voltage characteristics
and display
negative differential resistance. Because the transit time of
carriers through
such a structure is very short they have a number of applications
including
high frequency microwave oscillators and mixers. Tunnelling
structures
containing a quantum dot display an added complication due to
the charge of
the carriers; the Coulomb blockade effect.
Further reading
For a fuller, mathematical treatment of Coulomb blockade the

following
articles may be useful, ‘Artificial Atoms’ by M A Kastner,
Physics Today 24
January 1993 and ‘Single electron charging effects in
semiconductor quantum
dots’ by L P Kouenhoven et al Zeitschrift für Physik B
Condensed Matter 85,
367 (1991).
The generally mathematics of quantum mechanical tunnelling is
described in
quantum mechanics text books and also with respect to the
present subject in
‘The Physics of Low-Dimensional semiconductors’ by J H
Davies CUP. Finally
‘Low-Dimensional Semiconductors materials, physics,
technology, devices’ by
M J Kelly OUP discusses applications of resonant tunnelling
structures.
-0.60 -0.58 -0.56 -0.54 -0.52 -0.50
0.0
0.5
1.0
N
N+1
eV
C
on

du
ct
an
ce
(e
2 /h
)
Gate Voltage (V)
Coulomb blockade effect observed for tunnelling
through an electrostatically defined quantum dot.
The measurement temperature is 10mK. The inset
shows the carrier tunnelling steps and the energy
levels of the system. Data redrawn from L P
Kouwenhoven, et al Z. Phys. B. 85, 367 (1991).
Financial Management • Autumn 2004 • pages 5 - 37
Why Has IPO Underpricing
Changed Over Time?
Tim Loughran and Jay Ritter*
In the 1980s, the average first-day return on initial public
offerings (IPOs) was 7%. The
average first-day return doubled to almost 15% during 1990-
1998, before jumping to 65%

during the internet bubble years of 1999-2000 and then
reverting to 12% during 2001-2003.
We attribute much of the higher underpricing during the bubble
period to a changing issuer
objective function. We argue that in the later periods there was
less focus on maximizing IPO
proceeds due to an increased emphasis on research coverage.
Furthermore, allocations of
hot IPOs to the personal brokerage accounts of issuing firm
executives created an incentive
to seek rather than avoid underwriters with a reputation for
severe underpricing.
What explains the severe underpricing of initial public offerings
in 1999-2000, when the average
first-day return of 65% exceeded any level previously seen
before? In this article, we address
this and the related question of why IPO underpricing doubled
from 7% during 1980-1989 to
almost 15% during 1990-1998 before reverting to 12% during
the post-bubble period of 2001-
2003. Our goal is to explain low-frequency movements in
underpricing (or first-day returns) that
occur less often than hot and cold issue markets.
We examine three hypotheses for the change in underpricing: 1)
the changing risk composition
hypothesis, 2) the realignment of incentives hypothesis, and 3) a
new hypothesis, the changing
issuer objective function hypothesis. The changing issuer
objective function hypothesis has
two components, the spinning hypothesis and the analyst lust
hypothesis.
The changing risk composition hypothesis, introduced by Ritter
(1984), assumes that riskier

IPOs will be underpriced by more than less-risky IPOs. This
prediction follows from models
where underpricing arises as an equilibrium condition to induce
investors to participate in the
IPO market. If the proportion of IPOs that represent risky stocks
increases, there should be
greater average underpricing. Risk can reflect either
technological or valuation uncertainty.
Although there have been some changes in the characteristics of
firms going public, these
changes are found to be too minor to explain much of the
variation in underpricing over time if
there is a stationary risk-return relation.
The realignment of incentives and the changing issuer objective
function hypotheses both
We thank Hsuan-Chi Chen, Harry DeAngelo, Craig Dunbar,
Todd Houge, Josh Lerner, Lemma Senbet and James
Seward (the Editors), Toshio Serita, Ivo Welch, Ayako Yasuda,
and Donghang Zhang; seminar participants at the
2002 Chicago NBER behavioral finance meetings, the 2002
Tokyo PACAP/APFA/FMA meetings, the 2003 AFA
meetings, Boston College, Cornell, Gothenburg, Indiana,
Michigan State, Penn State, Stanford, the Stockholm
School of Economics, Vanderbilt, NYU, SMU, TCU, and the
Universities of Alabama, California (Berkeley), Colorado,
Houston, Illinois, Iowa, Notre Dame, and Pennsylvania, and
several anonymous referees; and especially Alexander
Ljungqvist for useful comments. Chris Barry, Laura Field, Paul
Gompers, Josh Lerner, Alexander Ljungqvist, Scott
Smart, Li-Anne Woo, and Chad Zutter generously provided IPO
data. Bruce Foerster assisted us in ranking underwriters.
Underwriter ranks are available online at
http://bear.cba.ufl.edu/ritter/rank.htm. Donghang Zhang
supplied useful

research assistance.
*Tim Loughran is a Professor of Finance at the University of
Notre Dame. Jay Ritter is the Cordell Professor of
Finance at the University of Florida.
Financial Management • Autumn 2004 6
posit changes over time in the willingness of issuing firms to
accept underpricing. Both
hypotheses assume that underwriters benefit from rent-seeking
behavior that occurs when
there is excessive underpricing.
The realignment of incentives hypothesis, introduced by
Ljungqvist and Wilhelm (2003),
argues that the managers of issuing firms acquiesced in leaving
money on the table during
the 1999-2000 bubble period. (Money on the table is the change
between the offer price and
the first closing market price, multiplied by the number of
shares sold.) The hypothesized
reasons for the increased acquiescence are reduced chief
executive officer (CEO) ownership,
fewer IPOs containing secondary shares, increased ownership
fragmentation, and an
increased frequency and size of “friends and family” share
allocations. These changes made
issuing firm decision-makers less motivated to bargain for a
higher offer price.
The realignment of incentives hypothesis is similar to the
changing risk composition
hypothesis in that it is changes in the characteristics of

ownership, rather than any
nonstationarities in the pricing relations, that are associated
with changes in average
underpricing. It differs from the changing risk composition
hypothesis, however, in that
underpricing is not determined solely by the investor demand
side of the market.
In our empirical work, we find little support for the realignment
of incentives hypothesis as
an explanation for substantial changes in underpricing. We find
no relation between the
inclusion of secondary shares in an IPO and underpricing. And
although CEO fractional
ownership was lower during the internet bubble period, the CEO
dollar ownership (the market
value of the CEO’s holdings) was substantially higher, resulting
in increased incentives to
avoid underpricing. Furthermore, it is possible that changes in
the characteristics of
ownership may be partly a response to higher underpricing as
well as a cause. Ljungqvist
and Wilhelm (2003) do not provide an explanation for why
these changes occurred.
The changing issuer objective function hypothesis argues that,
holding constant the
level of managerial ownership and other characteristics, issuing
firms became more willing to
accept underpricing. We hypothesize that, during our sample
period, there are two reasons
for why issuers became more willing to leave money on the
table. The first reason is an
increased emphasis on analyst coverage. As issuers placed more
importance on hiring a lead
underwriter with a highly ranked analyst to cover the firm, they

became less concerned
about avoiding underwriters with a reputation for excessive
underpricing. We call this desire
to hire an underwriter with an influential but bullish analyst the
analyst lust hypothesis.
This results in each issuer facing a local oligopoly of
underwriters, no matter how many
competing underwriters there are in total, because there are
typically only five Institutional
Investor all-star analysts covering any industry. As Hoberg
(2003) shows, the more market
power that underwriters have, the more underpricing there will
be in equilibrium.
The second reason for a greater willingness to leave money on
the table by issuers is the
co-opting of decision-makers through side payments. Beginning
in the 1990s, underwriters
set up personal brokerage accounts for venture capitalists and
the executives of issuing
firms in order to allocate hot IPOs to them. By the end of the
decade, this practice, known as
spinning, had become commonplace. The purpose of these side
payments is to influence the
issuer’s choice of lead underwriter. These payments create an
incentive to seek, rather than
avoid, underwriters with a reputation for severe underpricing.
We call this the spinning
hypothesis. In the post-bubble period, increased regulatory
scrutiny reduced spinning
dramatically. This is one of several explanations why
underpricing dropped back to an average
of 12%. The reduction in spinning removed the incentive for
issuers to choose investment
bankers who underprice. Investment bankers responded by
underpricing less in the post-

bubble period.
Loughran & Ritter • Why Has IPO Underpricing Changed Over
Time? 7
The contributions of our research are three-fold. First, we
develop the changing issuer
objective function hypothesis for the increased underpricing of
IPOs during the 1990s and
the bubble periods. Second, we document many patterns
regarding the evolution of the US
IPO market during the last two decades. Much of the data has
been or will be posted on a
website for other researchers to use. Many, although not all, of
these patterns have been
previously documented, especially for the first two subperiods.
Third, we formally test the
ability of the changing risk composition, realignment of
incentives, and changing issuer
objective function hypotheses to explain the changes in
underpricing from 1980-1989 (“the
1980s”) to 1990-1998 (“the 1990s”), 1999-2000 (“the internet
bubble”), and 2001-2003 (“the
post-bubble period”).
Much of the increased underpricing in the bubble period is
consistent with the predictions
of the changing issuer objective function hypothesis. In multiple
regression tests, the
changing risk composition and the realignment of incentives
hypotheses have little success
at explaining the increase in first-day returns from the 1980s to
the 1990s, to the bubble
period, or to the post-bubble period. The regression results

show that only part of the
increase in the bubble period is attributable to the increased
fraction of tech and internet
stocks going public. Consistent with the changing issuer
objective function hypothesis,
underpricing became much more severe when there was a top-
tier lead underwriter in the
latter time periods. These conclusions are not substantially
altered after controlling for the
endogeneity of underwriter choice.
The rest of this article is as follows. In Section I, we present
our changing issuer objective
function hypothesis. In Section II, we describe our data. In
Section III, we report year-by-
year mean and median first-day returns and valuations. In
Section IV, we report average first-
day returns for various univariate sorts. In Section V, we report
multiple regression results
with first-day returns as the dependent variable. Section VI
discusses alternative explanations
for the high underpricing of IPOs during the internet bubble
period. Section VII presents our
conclusions. Four appendices provide detailed descriptions of
our data on founding dates,
post-issue shares outstanding, underwriter rankings, and
internet IPO identification.
I. Causes of a Changing Issuer Objective Function
Most models of IPO underpricing are based on asymmetric
information. There are two
agency explanations of underpricing in the IPO literature. Baron
(1982) presents a model of
underpricing where issuers delegate the pricing decision to
underwriters. Investment bankers

find it less costly to market an IPO that is underpriced.
Loughran and Ritter (2002) instead
emphasize the quid pro quos that underwriters receive from
buy-side clients in return for
allocating underpriced IPOs to them. The managers of issuing
firms care less about
underpricing if they are simultaneously receiving good news
about their personal wealth
increasing. This argument, however, does not explain why
issuers hire underwriters who will
ex post exploit issuers’ psychology. Neither does the
realignment of incentives hypothesis.
One can view issuers as seeking to maximize a weighted
average of IPO proceeds, the
proceeds from future sales (both insider sales and follow-on
offerings), and side payments
from underwriters to the people who will choose the lead
underwriter:
α
1
IPO Proceeds + α
2
Proceeds from Future Sales + (1 - α
1
- α
2
)Side Payments (1)

The changing issuer objective function hypothesis states that
issuers choosing an
underwriter in some periods put less weight on IPO proceeds
and more weight on the proceeds
from future sales and side payments.
In Equation (1), IPO proceeds are a function of the choice of
underwriter and underwriting
contract (auction or bookbuilding) at the start of the process
and, several months later, the
bargaining at the pricing meeting for IPOs when bookbuilding is
used. Loughran and Ritter
(2002) provide a prospect theory analysis of the bargaining at
the pricing meeting. The
Ljungqvist and Wilhelm (2003) realignment of incentives
hypothesis can also be viewed as
a theory of the bargaining at the pricing meeting. Neither of
these theories, though, explains
why an issuing firm would choose an underwriter that would, at
the pricing meeting, propose
an offer price that leaves more money on the table than
necessary. In contrast, the changing
issuer objective function hypothesis does provide a theory for
the choice of underwriter at
the start of the process. Before discussing the analyst lust and
spinning hypotheses in more
detail, we explain why underwriters want to underprice.
A. Why Underwriters Want to Underprice IPOs
Underwriters, as intermediaries, advise the issuer on pricing the
issue, both at the time of
issuing a preliminary prospectus that includes a file price range,
and at the pricing meeting
when the final offer price is set. If underwriters receive

compensation from both the issuer
(the gross spread) and investors, they have an incentive to
recommend a lower offer price
than if the compensation was merely the gross spread.
Bookbuilding is the mechanism used to price and allocate IPOs
for 99.9% of our sample,
with auctions used for the other 0.1%. In the case of
bookbuilding, underwriters can decide
to whom to allocate shares if there is excess demand.
Benveniste and Wilhelm (1997) and
Sherman and Titman (2002) emphasize that underwriter
discretion can be used to the benefit
of issuing firms. Underwriters can reduce the average amount of
underpricing, thereby
increasing the expected proceeds to issuers, by favoring regular
investors who provide
information about their demand that is useful in pricing an IPO.
Shares can be allocated to
those who are likely to be buy-and-hold investors, minimizing
any costs associated with
price support.
Underwriter discretion can completely eliminate the winner’s
curse problem if underwriters
allocate shares in hot issues only to those investors who are
willing to buy other IPOs. As
Ritter and Welch (2002) note, if underwriters used their
discretion to bundle IPOs, problems
caused by asymmetric information could be nearly eliminated.
The resulting average level of
underpricing should then be no more than several percent. Thus,
given the use of
bookbuilding, the joint hypothesis that issuers desire to
maximize their proceeds and that
underwriters act in the best interests of issuers can be rejected

whenever average
underpricing exceeds several percent.
Although underwriter discretion in allocating IPOs can be
desirable for issuing firms, it
can also be disadvantageous if conflict of interest problems are
not controlled. Underwriters
acknowledge that in the late 1990s IPOs were allocated to
investors largely on the basis of
past and future commission business on other trades. In 1998-
2000, for example, Robertson
Stephens allocated IPOs to institutional clients almost
exclusively on the basis of the amount
of commission business generated during the prior 18 months,
according to its January 9,
2003 settlement with the NASD and SEC. Credit Suisse First
Boston (CSFB) received
commission business equal to as much as 65% of the profits that
some investors received
Time? 9
from certain hot IPOs, such as the December 1999 IPO of VA
Linux.1 The VA Linux IPO was
priced at $30 per share, with a 7% gross spread equal to $2.10
per share. For an investor who
was allocated shares at $30, and who then sold at the closing
market price of $239.25, the
capital gains would have amounted to $209.25 per share. If the
investor then traded shares to
generate commissions of one-half of this profit, the total
underwriter compensation per
share was $2.10 plus $104.625, or $106.725.

The receipt of commissions by underwriting firms in return for
hot IPO allocations violates
NASD Rule 2110 on “Free Riding and Withholding.” Because
the underwriter has an economic
interest (a share of the profits) in the IPO after it has been
allocated, there is not a “full
distribution” of the security. This is economically equivalent to
withholding shares and
selling them at a price higher than the offer price, in violation
of Rule 2110. But if the NASD
(a self-regulatory organization) did not enforce its rules,
underwriters might find it optimal to
violate the rules. Evidence consistent with commission business
affecting IPO allocations is
contained in Reuter (2004).
The willingness of buy-side clients to generate commissions by
sending trades to integrated
securities firms depends on the amount of money left on the
table in IPOs. Underwriters have an
incentive to underprice IPOs if they receive commission
business in return for leaving money on
the table. But the incentive to underprice presumably would
have been as great in the 1980s as
during the internet bubble period, unless there was a “supply”
shift in the willingness of firms to
hire underwriters with a history of underpricing. We argue that
such a shift did indeed occur,
resulting in increased underpricing.
B. The Analyst Lust Explanation of Underpricing
We hypothesize that issuing firms have increasingly chosen
their lead underwriter largely
on the basis of expected analyst coverage. Providing research

coverage is expensive for
investment bankers; the largest brokerage firms each spent close
to $1 billion per year on
equity research during the bubble (Rynecki, 2002). These costs
are covered partly by charging
issuers of securities explicit (gross spread) and implicit
(underpricing) fees. The more that
issuing firms see analyst coverage as important, the more they
are willing to pay these costs.
There are several reasons for our opinion that analyst lust was
more important during the
1990s and bubble period than in the 1980s. First, the investment
bankers and venture
capitalists we have talked to are unanimous in their agreement.
Supporting this, in the early
1970s Morgan Stanley had “no research business to speak of,”
even though it was a major
IPO underwriter (Schack, 2002). As we will show, the number
of managing underwriters in
1See the January 22, 2002 SEC litigation release 17327 and
news release (available on the SEC website at
http://www.sec.gov), and the NASD Regulation news release
(available at http://www.nasdr.com). The NASD
Regulation news release states that “For example, after a CSFB
customer obtained an allocation of 13,500 shares
in the VA Linux IPO, the customer sold two million shares of
Compaq and paid CSFB $.50 a share—or $1
million—as a purported brokerage commission. The customer
immediately repurchased the shares through other
firms at normal commission rates of $.06 per share at a loss of
$1.2 million on the Compaq sale and repurchase
because of the $1 million paid to CSFB. On that same day,
however, the customer sold the VA Linux IPO shares,
making a one-day profit of $3.3 million.”

According to paragraphs 48 and 49 of the SEC complaint, for
the July 20, 1999 IPO of Gadzoox, which CSFB
lead managed, “at least 261,025 shares were allocated to
customers that were willing to funnel a portion of their
IPO profits to CSFB.” CSFB distributed approximately 3.4
million of the 4.025 million offer, which went from
an offer price of $21 to a closing price of $74.8125, up 256%.
The following day, July 21, 1999, CSFB was the
lead manager on MP3, which was priced at $28 and closed at
$63.3125, up 126%. “CSFB distributed 7.2 million
of the 10.35 million MP3 shares offered through underwriters.
Of the 7.2 million MP3 shares distributed by
CSFB, at least 520,170 shares were allocated to customers that
were willing to funnel a portion of their trading
profits to CSFB.”
IPO syndicates has increased over time. Investment bankers
note that co-managers are
included in a syndicate almost exclusively to provide research
coverage. Indeed, by 2000 co-
managers were generally not even invited to participate in road
shows and the pricing meeting
at which the final offer price is determined.
Second, as valuations have increased, changes in growth rates
perceived in the financial
markets represent more dollars. Firm value can be decomposed
into the value of existing
assets in place plus the net present value of growth
opportunities. As the value of growth
opportunities increases relative to the value of assets in place,
issuing firms come to place

more importance on analyst coverage. In 1982, for example,
when the market price-earnings
(PE) ratio was about 8, the difference in valuation for a firm
with forecasted growth of 10%
versus 15% might translate into a difference in PEs of 8 versus
12. In 1999, when the market
PE was about 25, the difference in valuation for forecasted
growth of 10% versus 15% might
translate into a difference in PEs of 25 versus 40. For a firm
with $1.00 in earnings per share,
in 1982 the difference in values would have been $4 per share,
but in 1999 it would be $15.
A final reason for the increased importance of analyst coverage
in the bubble period is the
greater visibility of analyst recommendations because of the
internet and cable television
stations such as CNBC. Consistent with this statement, Busse
and Green (2002, Table 5)
report that trading volume for Nasdaq stocks during June
through October 2000 increased
by an average of 300,000 shares in the four minutes after an
analyst mentioned a stock
favorably on CNBC’s Midday Call segment.
The analyst lust hypothesis does not necessarily assume any
conflict of interest between
managers and other pre-issue shareholders. If favorable analyst
coverage results in a higher
market price, all pre-issue shareholders benefit.
There is ample supporting evidence for this analyst lust
hypothesis. Dunbar (2000) presents
evidence that underwriters in 1984-1994 subsequently increased
their IPO market share if
they had an analyst who was highly ranked in the Institutional

Investor (II) annual survey.
Clarke, Dunbar, and Kahle (2003, Table 2) report that
investment banks gaining an II all-star
analyst subsequently boosted their market share of IPOs in the
analyst’s industry; the
changes were greater in 1995-1999 than in 1988-1994. The
Krigman, Shaw, and Womack
(2001) survey of issuing firms finds that one of the most
important reasons to switch
underwriters in a seasoned offering is to seek additional and
influential analyst coverage
from the new banker. Ljungqvist, Marston, and Wilhelm (2003)
analyze the determinants of
lead underwriter choice for debt and follow-on equity offerings
conducted during December
1993 through June 2002. They report that the presence of an II
all-star analyst in the issuing
firm’s industry increases the probability of that underwriter
being chosen as the lead, holding
constant that bank’s fraction of the issuer’s equity deals during
the prior five years.
Hong and Kubik (2003) report that analysts making optimistic
forecasts are more likely to
move to a higher-status brokerage firm if they change jobs.
Furthermore, analysts whose
employer underwrites stocks that they cover are more likely to
be forced out, the less optimistic
their forecasts are. Hong and Kubik report that these biases
became even stronger in the
1999-2000 period. Discussions with executives of firms going
public in 2001-2003 suggest
that analyst coverage is still an important determinant of
underwriter choice, in spite of the
Global Settlement restrictions on analyst participation in IPOs.

Cliff and Denis (2004) test the analyst lust hypothesis using a
sample of 1,050 US firms
conducting IPOs during 1993-2000 that subsequently conducted
at least one follow-on equity
offering during 1993-2001. They find that issuers are less likely
to switch underwriters for
their first SEO if there had been greater underpricing, and if the
IPO underwriter’s analyst
covered the stock one year after the IPO. In their Table 6
regression with an analyst coverage
Time? 11
instrument, they report that having an all-star analyst in the
industry of the issuing firm at
the time of the IPO is associated with first-day returns that are
16.3% higher. Furthermore,
their subperiod results show higher incremental underpricing
associated with hiring an
underwriter with an II all-star covering the firm in the bubble
period than earlier.
The evidence in all these studies is consistent with the analyst
lust hypothesis, and those
that report subperiod results find that the effects were stronger
in the late 1990s when
valuations were highest, just as we predict.
C. The Spinning Explanation of Underpricing
In 1999-2000, the average amount of money left on the table of
$85 million per IPO adds up
to $68 billion (in dollars of 2003 purchasing power), which

seems way too high to be justified
as equilibrium compensation for purchasing analyst coverage.
This raises two questions.
First, if issuing firms wanted to purchase analyst coverage, why
did they pay for it by
leaving money on the table, rather than paying a higher gross
spread? Second, why did they
leave so much money on the table?
Our answers are as follows. First, money on the table is state-
contingent compensation;
the deals leaving a lot of money on the table were the deals
where the managers of issuing
firms found themselves facing a substantial increase in their
personal wealth (Loughran and
Ritter, 2002). Second, with bookbuilding, underwriters have
discretion over the allocation of
hot IPOs. Some shares went to “friends and family” of the
issuing firm, as Ljungqvist and
Wilhelm (2003) show. But some shares also went to the
executives of issuing firms and their
venture capitalists through personal brokerage accounts
(Siconolfi, 1997).
In this article, we introduce a new agency explanation for IPO
underpricing, the spinning
hypothesis, which is based on a conflict of interest between
decision-makers and other pre-IPO
shareholders. It posits that decision-makers are willing to hire
underwriters with a history of
underpricing because the decision-makers receive side
payments.2 The decision-makers are the
individuals who choose the managing underwriters, especially
the lead underwriter, for an IPO.
These decision-makers are the general partners of the lead
venture capital firm (if a firm is financed

with venture capital money) and the top managers of the issuing
firm. The other pre-issue
shareholders are the limited partners of venture capital firms
and other minority shareholders.
Elkind and Gimein (2001) describe the “Friend of Frank”
brokerage accounts set up for decision-
makers by CSFB, where Frank Quattrone, head of technology
investment banking, worked:
[I]n the 1990s firms also began offering shares to potential
clients... by setting up brokerage
accounts specifically for hot IPOs. Under these arrangements,
VCs and entrepreneurs made
a moderate deposit (perhaps $250,000) and signed over
discretionary authority to the brokers
whose firms were seeking their favor. Typically, IPO shares
would be flipped for a quick—
and riskless—windfall. “The stock would go into the hands of
venture capitalists and the
managements of companies that were going to go public next,”
notes a Silicon Valley fund
manager. “This was the closest thing to free money that there
was. It may not be all that much
different from a briefcase filled with unmarked tens and
20s.”...Indeed, two Silicon Valley
CEOs, who asked that their names not be used, said that because
several competing
investment banks were offering them cheap IPO shares, they
could not have been influenced
2On April 28, 2003, the “global settlement” between ten top
investment banking firms and the NASD, NYSE,
SEC, and the states, coordinated by New York Attorney General
Eliot Spitzer, imposed a “no spinning” rule that
prohibits officers and directors who are in a position to “greatly

influence” investment banking decisions from
receiving IPO allocations. Proposed NASD Rule 2712 addresses
spinning and both clarifies and strengthens
NASD Rule 2710.
when choosing between them.
The March 7, 2003 San Jose Mercury News lists, by name and
company affiliation, 63
Silicon Valley executives who had “Friends of Frank” accounts
at CSFB. The median executive
received first-day capital gains of $538,000 from IPO
allocations.3
Payments like this to individuals motivate the managers of an
issuing firm to choose an
underwriter with a reputation for leaving money on the table.
This spinning theory of IPO
underpricing explains why underwriters and issuing firm
managers prefer to forego net
proceeds by leaving money on the table, rather than pay a
higher gross spread. Money on
the table is the currency by which underwriters can influence
other venture capitalists and
issuing firm executives; gross spread revenue cannot be
redistributed except in a more
transparent manner.
If spinning is an important reason for underpricing in the bubble
period, why wasn’t it
important a decade earlier? In the 1980s, relatively little money
was left on the table in IPOs

because valuations were low and analyst coverage was not
perceived to be as important as
it became in the 1990s. As IPO underpricing increased over
time, we hypothesize that the use
of hot IPOs to reward decision-makers created an incentive for
decision-makers to seek out
underwriters known to leave money on the table, rather than to
avoid such underwriters.
Allocating these hot IPOs to the decision-makers of issuing
companies and their friends
(through friends and family accounts) allowed underwriters to
underprice even more. In
other words, underpricing fed on itself. In this regard, both our
changing issuer objective
function and Ljungqvist and Wilhelm’s (2003) realignment of
incentives hypotheses are
similar: Underpricing creates incentives for even more
underpricing. What constrains
underpricing from increasing without limit is that raising money
is still a goal for an issuer.
II. Data
Our primary data source for IPOs over 1980-2003 is the
Thomson Financial Securities Data
(also known as Securities Data Co.) new issues database. We
have made hundreds of
corrections to this database, and we have collected missing
information for thousands of
observations from a number of sources, including prospectuses;
Howard and Co.’s Going
Public: The IPO Reporter for IPOs over 1980-1985; Dealogic
for IPOs after 1990; and the
SEC’s Electronic Data Gathering and Retrieval (EDGAR)
system for IPOs after 1996 (final
prospectuses are identified on EDGAR as document 424B at

http://www.sec.gov).4
In all of our analysis, we exclude best efforts offers (typically
very small offerings, these
are not covered by Thomson Financial Securities Data); ADRs
(American Depository
Receipts, issued by foreign firms that list in at least one other
market outside the US);
closed-end funds; REITs (real estate investment trusts); banks
and savings and loans (S&Ls);
partnerships; and firms not covered by CRSP within six months
of the offering. We also
exclude IPOs with an offer price below $5.00 per share. What
remains are almost all IPOs of
3Descriptions and evidence regarding spinning are presented in
a number of additional sources. Smith (2002)
describes the allocation of IPOs to top executives by Goldman
Sachs. Smith, Grimes, Zuckerman, and Scannell
(2002) describe the allocations to venture capitalists, and
Sherburne (2002) lists the allocations to WorldCom
officers and directors and to other telecom executives by
Citigroup’s Salomon Smith Barney unit.
4While Thomson Financial’s database is missing some assets
and sales data, and many founding dates, we find no
evidence of any backfilling bias. That is, there is no evidence
that subsequent “winners” are more comprehensively
or accurately covered than other IPOs, so researchers using this
database should not worry about introducing a
survivorship bias.
Time? 13

domestic operating companies that are large enough to be of
interest to institutional investors.
The sample size is 6,391 firms, although in some of the tables
we are missing up to 6% of the
sample because of incomplete information.
The main source of information on venture capital backing is
Thomson Financial.
Supplemental data on venture capital backing has been provided
by Chris Barry, Paul Gompers,
and Josh Lerner.
Information on the founding date of companies comes from a
variety of sources, discussed
in more detail in Appendix A. Laura Field, Alexander
Ljungqvist, and Li-Anne Woo provided
many of the founding dates. We are missing a reliable founding
date for 120 firms.
The original file price range for IPOs over 1980-1982 is
transcribed from Howard and Co.’s
Going Public: The IPO Reporter. The file price range for IPOs
from 1983 and later comes
from Thomson Financial. We are missing the file price range for
11 firms in the early 1980s.
To calculate the market value of an IPO, we use the offer price
multiplied by the post-issue
number of shares outstanding. For firms with a single class of
shares outstanding, the
primary source of data on the post-issue number of shares is
CRSP. For firms with more than
one class of shares outstanding (dual-class firms), we use data
from a variety of sources, as
described in Appendix B.

Information on assets, sales, and earnings per share (EPS) in the
year prior to going
public comes mainly from Thomson Financial. When figures are
available, we use sales and
earnings per share for the most recent 12 months prior to going
public. Otherwise, we use the
most recent fiscal year numbers. Additional sources of
information include Dealogic for
post-1990 IPOs, Howard and Co.’s Going Public: The IPO
Reporter for 1980-1985 IPOs, and
EDGAR. If a firm has zero trailing sales, we assign a sales
value of $0.01 million, since in our
empirical work we use logarithms, and the logarithm of zero is
undefined. If we are unsure
whether sales are zero or are missing, we treat the value as
missing. We are missing sales
numbers for 85 firms and assets numbers for 223 firms.
We use Thomson Financial Securities Data as our source for
information on lead
underwriters and the number of managing underwriters for each
IPO. For underwriter prestige
rankings, we start with the Carter and Manaster (1990) and
Carter, Dark, and Singh (1998)
rankings, and then create rankings for 1992-2003 in the spirit of
their methodology. Appendix
C provides a detailed description of the procedures. The
underwriter prestige rankings are
on a 0 to 9 scale, and are based on the pecking order seen in
“tombstone” advertisements. In
our empirical work, if there is more than one lead underwriter,
we use the rank of the bookrunner
or the highest-ranking joint bookrunner.
Appendix D describes how we identify internet IPOs and lists
the SIC codes that we use

to categorize IPOs as a technology (tech) firm or not.
III. Time-Series of First-Day Returns and Valuations
Figure 1 plots the annual volume and average first-day return on
IPOs over 1980-2003.
Table I reports the means (Panel A) and medians (Panel B) of
the first-day returns by year of
issue and by subperiod. In all of our analysis, we split the
sample into four subperiods:
January 1980-December 1989 (“the 1980s”), January 1990-
December 1998 (“the 1990s”),
January 1999-December 2000 (“the internet bubble”), and
January 2001-December 2003 (“the
post-bubble period”).
In the 1980s, the average first-day return was slightly over 7%.
The average first-day
return increased to almost 15% in the 1990s, and then jumped to
65% during the internet
Figure 1. Number of IPOs (Bars) and Average First-Day Returns
(Diamonds) by
Cohort Year
IPOs with an offer price below $5.00 per share, unit offers,
REITs, closed-end funds, banks and S&Ls,
ADRs, partnerships, and IPOs not listed on CRSP within six
months of the offer date are excluded. Data
are from Thomson Financial Securities Data and other sources,
with corrections by authors. The first-day

return is defined as the percentage change from the offer price
to the closing price. The data plotted are
reported in Panel A of Table I.
0
100
200
300
400
500
600
700
800
19
80
19
82
19
84
19
86
19

88
19
90
19
92
19
94
19
96
19
98
20
00
20
02
Calendar Year
N
u
m
b
e
r
o
f

IP
O
s
0
10
20
30
40
50
60
70
80
A
v
e
ra
g
e
F
ir

s
t-
D
a
y
R
e
tu
rn
s
, %
bubble. In the post-bubble period, annual IPO volume dropped
to 80 issues or fewer with a
mean first-day return of approximately 12%.
Table I shows that from 1980 through 1994 the underpricing of
IPOs was typically quite
modest, as was the amount of money left on the table. In every
year from 1995 through 2000,
the average first-day return was higher than in any year between
1981 and 1994. Underpricing
took a big jump in the bubble period, as did the amount of
money left on the table. The
number of managing underwriters increased steadily until 2003,
with a rapid acceleration in
the late 1990s. The conventional wisdom is that the growth in
the number of managing
underwriters is associated with greater emphasis on analyst
coverage.
For IPOs in the 1980s, Panel B reports that the median

valuation of $72 million using the
offer price was less than twice the annual sales of $38 million.
In the 1990s, the market-to-
sales ratio increased to 2.7 (median valuation of $122 million
relative to median sales of $46
million). During the internet bubble period, the median
valuation using the offer price jumped
to $387 million while the median sales fell to $15 million, for a
market-to-sales ratio of 26.
Using the valuation implied by the first closing market price,
the market-to-sales ratio is even
higher, at 38. This rapid escalation in market-to-sales ratios
suggests that valuation
uncertainty played a role in increased underpricing over time. In
the post-bubble period, the
market-to-sales ratio fell back to 2.4, approximately what it was
in the 1990s.
Time? 15
Table I. Number of IPOs, First Day Returns, Number of
Managing Underwriters,
Amount of Money Left on the Table, Valuation Levels, and
Sales by Cohort Year
IPOs with an offer price below $5.00 per share, unit offers,
REITs, closed-end funds, banks and S&Ls, ADRs,
and IPOs not listed on CRSP within six months of issuing are
excluded. Data are from Thomson Financial
Securities Data, with supplements from Dealogic and other
sources, and corrections by authors. The first-day
return is defined as the percentage change from the offer price
to the closing price. The number of domestic

managing underwriters includes both lead underwriters and co-
managers. Money on the table is defined as the
first-day price change (offer price to close) times the number of
shares issued (global offering amount,
excluding overallotment options). Both valuation calculations
use the post-issue number of shares outstanding.
Valuations are computed by multiplying either the offer price or
the first closing market price by the post-issue
shares outstanding. Sales are for the last 12 months prior to
going public, as reported in the prospectus. The
mean and median sales are computed for the 6,306 firms for
which a sales number is available. All dollar
values are in dollars of 2003 purchasing power adjusted using
the Consumer Price Index.
Panel A. Means
Millions of 2003 Dollars
Post-Issue
Valuation
Year
Number
of IPOs
First-Day
Return

Number of
Managing
Underwriters
Money
on the
Table
Offer
Price
Market
Price
Sales
1980 70 14.5% 1.4 $5.6 $145 $181 $77
1981 191 5.8% 1.3 $1.4 $102 $109 $55
1982 77 11.4% 1.4 $3.3 $111 $126 $41
1983 442 10.1% 1.5 $3.5 $151 $165 $92
1984 172 3.6% 1.5 $0.5 $89 $91 $84
1985 179 6.3% 1.5 $2.0 $188 $194 $202
1986 378 6.3% 1.5 $2.9 $182 $194 $171
1987 271 6.0% 1.8 $3.9 $219 $234 $248
1988 97 5.4% 1.7 $2.0 $306 $315 $300
1989 105 8.1% 1.6 $3.3 $229 $245 $241
1990 104 10.8% 1.9 $4.4 $206 $225 $365
1991 274 12.0% 2.0 $6.6 $211 $236 $237
1992 385 10.2% 2.0 $5.8 $217 $237 $222
1993 484 12.8% 2.1 $8.4 $269 $304 $263
1994 387 9.8% 2.0 $4.5 $179 $193 $204
1995 434 21.5% 2.3 $12.1 $268 $320 $211
1996 623 16.7% 2.4 $12.3 $330 $392 $160
1997 437 14.0% 2.5 $11.3 $287 $334 $181
1998 268 22.2% 2.9 $21.1 $540 $652 $332

1999 457 71.7% 3.4 $86.2 $890 $1,519 $368
2000 346 56.1% 3.7 $82.8 $963 $1,635 $270
2001 80 13.5% 4.4 $30.9 $2,084 $2,239 $2,130
2002 67 8.9% 4.7 $17.3 $1,147 $1,239 $1,137
2003 63 12.2% 4.0 $16.0 $575 $645 $380
1980-1989 1,982 7.3% 1.5 $2.8 $170 $181 $149
1990-1998 3,396 14.8% 2.3 $10.0 $281 $325 $222
1999-2000 803 65.0% 3.6 $84.7 $921 $1,569 $326
2001-2003 210 11.7% 4.4 $22.1 $1,332 $1,442 $1,289
Total 6,391 18.7% 2.3 $17.5 $361 $474 $248
Table I. Number of IPOs, First Day Returns, Number of
Managing Underwriters,
Amount of Money Left on the Table, Valuation Levels, and
Sales by Cohort Year
(Continued)
Panel B. Medians
Millions of 2003 Dollars
Post-Issue
Valuation
Year
Number

of IPOs
First-
Day
Return
Number of
Managing
Underwriters
Money on
the Table
Offer
Price
Market
Price
Sales
1980 70 8.0% 1 $0.8 $65 $77 $43
1981 191 0.0% 1 $0.0 $64 $65 $26
1982 77 3.7% 1 $0.4 $57 $64 $20
1983 442 2.6% 1 $0.5 $81 $86 $26
1984 172 0.0% 1 $0.0 $49 $51 $37
1985 179 2.5% 1 $0.6 $66 $66 $47
1986 378 1.3% 1 $0.2 $71 $75 $48
1987 271 1.4% 2 $0.4 $83 $84 $48
1988 97 2.5% 2 $0.5 $109 $117 $93
1989 105 4.3% 2 $1.2 $100 $113 $55
1990 104 5.4% 2 $1.5 $111 $121 $55
1991 274 7.5% 2 $2.5 $120 $135 $67

1992 385 4.2% 2 $1.1 $111 $120 $55
1993 484 6.3% 2 $1.9 $106 $117 $58
1994 387 4.5% 2 $1.2 $87 $93 $46
1995 434 13.3% 2 $4.5 $127 $150 $37
1996 623 10.0% 2 $3.6 $136 $156 $33
1997 437 9.4% 2 $3.3 $128 $143 $41
1998 268 9.0% 3 $3.4 $178 $213 $45
1999 457 37.5% 3 $29.8 $345 $529 $18
2000 346 27.4% 3 $23.3 $436 $607 $11
2001 80 10.0% 4 $10.3 $442 $465 $140
2002 67 8.0% 4 $8.6 $479 $506 $194
2003 63 9.8% 4 $10.3 $335 $369 $165
1980-1989 1,982 1.9% 1 $0.4 $72 $76 $38
1990-1998 3,396 7.8% 2 $2.4 $122 $134 $46
1999-2000 803 32.3% 3 $27.1 $387 $563 $15
2001-2003 210 8.8% 4 $9.7 $394 $459 $164
Total 6,391 6.3% 2 $1.7 $123 $136 $40
IV. Univariate Sorts
Can the changing characteristics of IPOs, a realignment of
incentives, and changing
issuer objectives explain the increase in underpricing over time?
In this section, we first
provide some evidence based on univariate sorts. Table II
reports the mean first-day
returns on IPOs after several simple sorts for four subperiods:
the 1980s, the 1990s, the
internet bubble, and the post-bubble period. One can see that
some of the cross-sectional
patterns in the 1980s reversed in the 1990s. In the 1990s, larger
offers were underpriced
more than smaller ones, and IPOs with a prestigious lead
underwriter were underpriced

Time? 17
Table II. Average First-day Returns on IPOs Categorized by
Proceeds, Assets, Sales,
Age, Industry, VC-backing, Share Overhang, and Underwriter
Prestige
Unit offers, REITs, closed-end funds, banks and S&Ls, ADRs,
IPOs with an offer price below $5.00, and
IPOs not listed on CRSP within six months of the offer date are
excluded. Data are from Thomson
Financial Securities Data and other sources, with corrections by
the authors. The sample size is 6,391
IPOs for 1980-2003. High-prestige underwriters are those with a
Carter and Manaster (1990) ranking of 8
or higher on a 9-point scale. Rankings for 1985-1991 are based
upon the Carter et al. (1998) rankings.
Rankings for 1992-2003 are by the authors. Further descriptions
of how age, industry, and underwriter
prestige are defined are in the appendices. Firms are classified
by proceeds on the basis of whether the
global gross proceeds are higher or lower than the median issue
size in the prior calendar year, with no
adjustment for inflation. Firms with pre-issue assets of less than
$40 million (2003 purchasing power) are
classified as small. Firms with trailing 12 month sales of $40
million or less (2003 purchasing power) are
classified as low sales firms. Share overhang is the ratio of
retained shares to the public float. Low share
overhang IPOs have an overhang ratio lower than 2.333
(representing a global offer size of 30% or more

of the post-issue shares outstanding, if all of the shares in the
IPO are issued by the firm). The file price
range is missing for 11 firms. Sales is missing for 85 firms. Age
is missing for 120 firms, and assets is
missing for 223 firms.
1980-1989 1990-1998 1999-2000 2001-2003
Segmented by Return N Return N Return N Return N
Proceeds
Small 7.4% 880 12.1% 1,551 32.7% 232 12.4% 77
Large 7.3% 1,102 17.0% 1,845 78.1% 571 11.3% 133
Assets
Small 9.0% 1,095 16.8% 1,519 71.0% 458 12.0% 50
Large 4.5% 717 13.1% 1,825 57.2% 344 11.6% 160
Sales
Low 9.2% 1,003 18.3% 1,545 73.0% 560 12.5% 52
High 5.2% 944 11.7% 1,805 46.6% 240 11.5% 157
Age
Young (0-7 years old) 9.0% 1,003 17.1% 1,640 75.2% 536
14.6% 72
Old (8 years and older) 5.8% 942 12.7% 1,681 45.2% 263
10.1% 134
Industry
Tech and internet-related 10.2% 576 22.2% 1,081 80.6% 585
16.4% 60
Non-technology 6.2% 1,406 11.3% 2,315 23.1% 218 9.8%
150
Segmented by venture capital backing
Non VC-backed 7.1% 1,437 13.8% 2,000 38.5% 316 9.4%
125
VC-backed 8.0% 545 16.1% 1,397 82.2% 487 15.0% 85
Segmented by source of shares offered
Exclusively sold by firm 7.7% 868 13.8% 1,999 69.4% 681
11.7% 147

Including secondary shares 7.1% 1,114 16.1% 1,396 40.4%
122 11.7% 63
Segmented by share overhang
Low 7.8% 885 11.8% 1,846 26.1% 134 7.2% 87
High 7.0% 1,097 18.3% 1,550 72.7% 669 14.8% 123
Segmented by underwriter prestige
Low-prestige 9.1% 1,119 12.9% 1,302 35.1% 151 12.2% 45
High-prestige 5.1% 863 15.9% 2,094 71.9% 652 11.5% 165
Segmented by the offer price relative to the file price range
Revised up 20.5% 246 32.0% 777 119.0% 362 24.3% 42
OP within range 7.8% 1,181 12.3% 1,750 26.8% 296 10.3%
116
Revised down 0.5% 544 4.3% 867 7.9% 145 4.5% 52
All 7.3% 1,982 14.8% 3,396 65.0% 803 11.7% 210
more than those without.5 In the 1990s and internet bubble
years, IPOs had high returns
when a relatively small fraction of the firm was sold in the IPO,
as measured by the ratio of
retained shares to issued shares, called share overhang by
Bradley and Jordan (2002). But
this pattern was not present in the 1980s. Several other patterns
have increased in magnitude
over time. Going across each row in Table II, underpricing
uniformly increased until the post-
bubble period.
In Table II, during the 1980s, tech stock IPOs had an average
first-day return of 10.2%.
This is the highest average first-day return of any category

during the 1980s except for the
set of IPOs whose offer price was revised upward from the file
price maximum. If the changing
characteristics of IPOs explained all the changes in
underpricing across time, it would be
hard to imagine that the average first-day return in the 1990s
would have increased to much
more than 10.2% if the first-day returns were drawn from a
stationary distribution.
Barry (1989), Habib and Ljungqvist (2001), and Ljungqvist and
Wilhelm (2003) argue that,
because the dilution effect hurts selling shareholders more than
if they retain their shares,
there will be more severe underpricing of pure primary
offerings than of IPOs with secondary
shares. Table II reports that pure primary offerings were
associated with greater underpricing
during the internet bubble period, a pattern not present in any
quantitatively important
manner in the 1980s, 1990s, or the post-bubble period. We now
look at some of the patterns
in more detail.
A. Age
Figure 2 graphs the average first-day return in each subperiod
after classifying firms by
their age at the time of going public. In each subperiod, there is
more underpricing of young
firms than of old firms, although the relation is not strictly
monotonic. Our results for the
1980s are consistent with those reported by Muscarella and
Vetsuypens (1990).
Even more noteworthy is the increased underpricing, holding

age constant, as one moves
from the 1980s to the 1990s to the internet bubble period.6
Thus, Figure 2 shows that the
increase in underpricing over time does not occur merely
because younger firms are going
public. Instead, the relation between age and first-day returns is
nonstationary.
Figure 3 plots the 25th, 50th, and 75th percentiles of the age
distribution for the IPOs in each
cohort year over 1980-2003. Four patterns stand out. First, in
the early 1990s, the proportion
of young firms dropped. This decline is associated with an
increase in the number of “reverse
LBOs,” firms going public again after a leveraged buyout.
Second, in 1999, more young firms
went public. This increase is associated with the internet
bubble. Third, after the bubble
burst, few young firms went public. Fourth, there is no strong
secular trend in the age
distribution of firms going public. With only temporary
aberrations, the median age has
stayed remarkably constant at about 7 years. The median age of
an issuing firm was 7 years
in the 1980s and 8 years in the 1990s, before falling to 5 years
during the internet bubble, and
5The difference in underpricing of 7.4% for small firms and
7.3% for large firms in the 1980s is lower than found
in other studies because we screen out IPOs with an offer price
below $5.00 per share. These low price IPOs had
an average first-day return of 20.5%, and their inclusion would
boost the average return on small IPOs during the
1980s to 8.8%. Low priced IPOs are historically subject to fraud
and have been avoided by institutional investors.
There has been a decrease in these issues over time partly due

to tighter listing requirements on Nasdaq, and partly
due to greater regulatory pressures on this part of the IPO
market.
6The greater variation of average first-day returns during the
internet bubble period is due to two features of the
data. First, the internet bubble period has a smaller sample size,
so each age group has fewer firms in it. Second,
within each age group, the standard deviation of first-day
returns is higher. The post-bubble period patterns are
also affected by a very small sample size in most age
categories.
Time? 19
Figure 2. Average First-day Returns by Age of Firm at Time of
IPO
Average first-day returns on IPOs during 1980-1989 (N =
1,945), 1990-1998 (N = 3,321), 1999-2000 (N
= 799), and 2001-2003 (N = 206) by age of firm at the time of
its IPO. IPOs with trailing 12-month sales
of over $200 million (2003 purchasing power) that are less than
two years old are not included, for these
are typically spinoffs or reverse LBOs or have the founding
dates incorrectly listed as the date of
reincorporation in Delaware. The age of the firm is defined as
the calendar year of the IPO minus the
calendar year of the founding.
0
10

20
30
40
50
60
70
80
90
100
0 1 2 3 4 5 6 7 8 9 10 11 12 13 15 17 20 30 40 50 60 70
AGE
A
v
er
ag
e
F
ir
s
t-

D
ay
R
et
u
rn
%
1999-2000
1990-1998
1980-1989
2001-2003
then rising dramatically to 12 years during the post-bubble
period.
B. CEO Ownership
The realignment of incentives hypothesis posits that issuing
firm executives will not
bargain as hard for a higher offer price if the CEO owns less of
the firm. Ljungqvist and
Wilhelm (2003) present regression evidence consistent with this
prediction, using the
percentage of shares owned by the CEO as the measure of
ownership. It is not obvious,
however, that CEO percentage ownership is as important as the
market value of these shares
if we want to measure the managerial benefits of a higher offer
price. For a pure primary

offering, the opportunity cost to a pre-issue shareholder of
underpricing is the dollar value
of money left on the table multiplied by the pre-issue fraction
of the firm owned by that
shareholder. Holding the amount of money left on the table
from the sale of primary shares
constant, the fractional ownership is the correct measure of the
opportunity cost to a CEO.
But as our Table I shows, the amount of money left on the table
was not constant during
Figure 3. 25th, 50th, and 75th Percentiles of Firm Age at Time
of Going Public by
Year of IPO
Each year, companies going public are ranked by firm age. The
25th, 50th (median), and 75th percentiles
of this age distribution are then plotted. For example, in 1980,
25% of IPOs were 2 years old or younger,
50% were 6 years old or younger, and 75% were 11 years old or
younger. For each subperiod, the 25th,
50th, and 75th percentiles of the age distribution are 3, 7, and
16 years old (the 1980s); 4, 8, and 16 years
old (the 1990s); 3, 5, and 9 years old (the internet bubble); and
6, 12, and 26 years old (the post-bubble
period). The 25th, 50th, and 75th percentiles of the age
distribution at the time of going public for the entire
sample of 6,271 IPOs are 4, 7, and 15 years old.
0

5
10
15
20
25
30
35
40
1
9
8
0
1
9
8
1
1
9
8
2
1
9

8
3
1
9
8
4
1
9
8
5
1
9
8
6
1
9
8
7
1
9
8
8
1
9

8
9
1
9
9
0
1
9
9
1
1
9
9
2
1
9
9
3
1
9
9
4
1
9

9
5
1
9
9
6
1
9
9
7
1
9
9
8
1
9
9
9
2
0
0
0
2
0

0
1
2
0
0
2
2
0
0
3
Years
A
g
e
1996-2000.
To be explicit, the dollar value of the opportunity cost of
underpricing to a CEO, if the
offering is entirely primary, is:
(2)
where N
ceo
is the number of shares owned by the CEO, N
o

is the pre-issue number of shares
outstanding, N
n
is the number of newly issued (primary) shares, P is the first
closing market
price, and OP is the offer price per share. Ljungqvist and
Wilhelm (2003) emphasize that the
CEO ownership fraction N
ceo
/N
o
was lower during the bubble period than in previous years.
But it is also the case that N
n
was much higher, while the distribution of nominal offer prices
did not change much.
Table III tabulates the median pre-issue CEO percentage
ownership reported by Ljungqvist
and Wilhelm (2003) for 1996-2000 and an estimate of the pre-
issue number of shares owned
Time? 21
Year

Number
of IPOs
Median Pre-Issue
Number of
CEO Shares
Median
Offer Price
Median CEO
Pre-Issue
Dollar Value,
Millions
Median Pre-Issue
% CEO
Ownership
1996 623 723,591 $12.00 $8.68 m 10.4%
1997 437 880,401 $11.75 $10.34 m 12.8%
1998 268 1,188,677 $12.50 $14.86 m 11.8%
1999 457 1,394,336 $14.00 $19.52 m 8.0%
2000 346 1,554,172 $14.00 $21.76 m 5.3%
by the CEO for the median company going public in a year,
computed as the product of the
median CEO fractional ownership times the median pre-issue
shares outstanding. We also

report the median offer price in each year and an approximation
of the median dollar value of
shares owned by CEOs, valued at the offer price.7
Inspection of Table III shows that, while CEO percentage
ownership decreased during
1996-2000, the number of shares owned more than doubled
because of the number of shares
outstanding quadrupled. This dramatic increase in pre-issue
shares outstanding is attributable
to the substantial increase in valuations along with a relatively
constant offer price. Thus,
the median CEO’s market value of equity rose, even though the
fractional holdings fell. If
one were to focus on the market value of the shares owned by
the CEO when the firm went
public, the realignment of incentives hypothesis predicts a
decrease in underpricing during
the bubble period due to the incentive effect. Wealth effects
associated with the higher
market value of the shares might dominate substitution effects,
however, making predictions
hazardous, as Ljungqvist and Wilhelm acknowledge. In any
case, the substantial increase
during 1996-2000 in CEO dollar holdings is in sharp contrast to
the decline in CEO holdings
when ownership is measured as a percentage of shares
outstanding.
C. Prestigious Underwriters
In general, underwriters with a Carter and Manaster rank of 8.0
to 9.0 (on a scale of 0 to 9)
are considered to be prestigious national underwriters. Those
with a rank of 5.0 to 7.9 are
considered to be quality regional or niche underwriters.

Underwriters with a rank of 0 to 4.9
7Alexander Ljungqvist has computed the value of the median
CEO’s pre-issue market value of equity, using the
Ljungqvist and Wilhelm sample, which is virtually identical to
ours for the 1996-2000 period. His numbers for
the median market value each year show the same trend that we
report in Table III, where we multiply the product
of several medians. Ljungqvist’s pre-issue market value of
equity for the median CEO increases from $6.76
million in 1996 to $20.64 million in 1999 before declining to
$16.86 million in 2000, while our Table III medians
increase from $8.68 million in 1996 to $21.76 million in 2000.
Table III. Pre-Issue CEO Ownership in Dollar Values and
Percentage, 1996-2000
The median pre-issue number of CEO shares is computed as the
product of the median pre-issue number
of shares outstanding and the median pre-issue % CEO
ownership. This should be viewed as an approxi-
mation of the actual median pre-issue number of CEO shares.
The median pre-issue % CEO ownership
is from Ljungqvist and Wilhelm (2003, Table III). The median
CEO pre-issue dollar value is computed
as the product of the prior two columns, and is also an
approximation of the actual median. Neither the
median offer price nor the median market value (median pre-
issue number of CEO shares times the
median offer price) is adjusted for price level changes
(inflation). Inflation averaged less than 3% per
year during this period.

are generally associated with penny stocks; many with ranks of
3.0 or lower have been
charged by the SEC with market manipulation. In Table IV, we
categorize IPOs on the basis of
lead underwriter prestige. Inspection of the sample sizes shows
that prestigious lead
underwriters increased their market share from under 50% in the
1980s to over 60% in the
1990s, and then to about 80% during the internet and post-
bubble periods.8
Beatty and Welch (1996), Cooney, Singh, Carter, and Dark
(2001), and others have
documented that a negative relation between underwriter
prestige and underpricing in the
1980s reversed itself in the 1990s, although the authors offer no
explanation for the reversal.
Our Table IV findings confirm this reversal. To rationalize the
pattern of the 1980s that
prestigious underwriters are associated with less underpricing,
Carter and Manaster (1990)
and Carter et al. (1998) argue that IPOs taken public by
prestigious underwriters benefit from
superior certification. Because of the greater reputation capital
that is committed, investors
do not demand as large a discount on these offers. The higher
underpricing associated with
prestigious underwriters in the 1990s and the internet bubble
period is inconsistent with the
joint hypothesis that underwriters are attempting to maximize
issuer proceeds and that
certification is an important determinant of the required amount
of money left on the table.
Instead, it is consistent with the changing issuer objective
function hypothesis.

If issuers became more willing to hire underwriters with a
history of underpricing after the
1980s, this could occur either because of a shift in which
underwriters were hired, or a shift
in the behavior of the underwriters. That is, underwriters,
especially those with influential
analysts and a willingness to allocate hot IPOs to the personal
brokerage accounts of issuing
firm decision-makers, could have changed their pricing policies
in order to leave more money
on the table. The evidence suggests that most of the shifts
occurred via changes in the
behavior of individual underwriters, rather than shifting market
shares. For example, for IPOs
with Goldman Sachs as the bookrunner, the average
underpricing was 5.0% in the 1980s,
23.8% in the 1990s, 99.8% during the bubble, and 11.0% during
the post-bubble period.
Table IV shows that over time, especially in the internet bubble
period, prestigious
underwriters relaxed their underwriting standards and took
public an increasing number of
very young and unprofitable companies. The median sales of
firms taken public by prestigious
underwriters dropped from $80 million in the 1980s to just $17
million during the internet
bubble period.
Tables II and IV also report changes over time in the fraction of
IPOs with upward revisions
of the offer price relative to the file price range. Table II
reports that, in the 1980s, it was twice
as likely to see a downward revision as an upward revision, and
in the bubble period, the

proportion of upward revisions was much higher. This cannot be
accounted for by differences
in returns on the Nasdaq Composite in the three weeks prior to
issuing. In the first three
subperiods, the average three-week return immediately prior to
issuing was about 1%,
although in the post-bubble period it was only 0.54%.
Our analyst lust hypothesis can explain the changes over time
that are documented in
Table IV. In the 1980s, investment bankers competed for IPO
underwriting mandates more on
the basis of implied valuations and less on the basis of analyst
coverage (because α
1
of
Equation (1) was higher in the 1980s). If an underwriter
indicated it would price a firm higher
than the competition, it was likely to be chosen. But in winning
the mandate, the underwriter
implicitly committed to a higher file price range, leaving less
room to avoid a downward
revision if market conditions weakened. Investment bankers tell
us that managing “issuer
8Since in all subperiods the biggest deals are more commonly
managed by prestigious underwriters, if market share
is computed using gross proceeds rather than the number of
IPOs, the market share of prestigious underwriters
would be uniformly higher.

Time? 23
Table IV. Mean and Median First-day Returns, Median Age,
Sales, EPS, and Share
Overhang, and Industry Representation Categorized by
Underwriter Prestige
Unit offers, REITs, closed-end funds, banks and S&Ls, ADRs,
and IPOs not listed on CRSP within six months
of the offer date are excluded. Data are from Thomson Financial
Securities Data, Dealogic, and other sources.
High-prestige underwriters are those with a Carter and Manaster
(1990) ranking of 8 or higher on a 9-point
scale. Rankings for 1984 and later are based upon the Carter et
al. (1998) rankings and updates by the authors
of this paper. See Appendix C for details. Sales are measured in
millions of dollars of year 2003 purchasing
power, using the Consumer Price Index. Share overhang is the
ratio of retained shares to the public float.
Percentage tech is the percentage of IPOs that are classified as
technology or internet-related, as defined in
Appendix D. The sample size is 6,391 IPOs over 1980-2003,
except for age, sales, EPS, and offer price
revision, where some observations are lost due to missing
information.
1980-1989 1990-1998 1999-2000 2001-2003
Return N Return N Return N Return N
Mean first-day returns
Low prestige 9.1% 1,119 12.9% 1,302 35.1% 151 12.2% 45
High prestige 5.1% 863 15.9% 2,094 71.9% 652 11.5% 165
Median first-day returns
Low prestige 2.5% 1,119 7.1% 1,302 12.2% 151 11.1% 45
High prestige 1.2% 863 8.7% 2,094 37.5% 652 8.3% 165

Median Age
Low prestige 6 years 1,115 7 years 1,298 5 years 151 9 years
45
High prestige 9 years 849 8 years 2,050 5 years 649 14
years 162
Median trailing sales (millions)
Low prestige $21.5 1,086 $25.8 1,268 $9.1 150 $44.1 45
High Prestige $80.2 861 $71.7 2,082 $17.3 650 $269.4 164
Median trailing 12-month EPS
Low prestige $0.38 1,099 $0.25 1,302 -$0.58 151 -$0.25 45
High prestige $0.59 855 $0.27 2,094 -$1.18 652 $0.02 165
Median share overhang
Low prestige 2.28 1,119 1.96 1,302 2.91 151 2.00 45
High prestige 2.82 863 2.44 2,094 4.31 652 2.97 165
Percentage with an offer price above the maximum of the file
price range
Low prestige 9% 1,110 11% 1,302 28% 151 9% 45
High prestige 17% 861 30% 2,094 49% 652 23% 165
Percentage tech and internet-related
Low prestige 30.6% 1,119 28.3% 1,302 68.9% 151 33.3% 45
High prestige 27.1% 863 34.0% 2,094 72.8% 652 27.3% 165
All 7.3% 1,982 14.8% 3,396 65.0% 803 11.7% 210
expectations” is part of their job. In the 1990s, underwriters
with star analysts could win a
mandate without committing to a high valuation that issuers
would anchor their expectations
on. In the bubble period, this was taken to an extreme; many
issuers accepted a low file price
range because they were more focused on choosing an
underwriter with an influential analyst
or with underpriced IPOs to allocate to an executive’s personal
brokerage account than on
getting a high valuation.

The academic literature generally views the midpoint of the file
price range as an unbiased
estimate of the offer price, and revisions in the offer price as
reflecting unanticipated strong
or weak demand. Houston, James, and Karceski (2004) report
that during the bubble period,
the file price range was low-balled relative to the value implied
by comparable firm multiples.
During the internet bubble, Donaldson Lufkin Jenrette and
Goldman Sachs, among others,
low-balled the file price range on some IPOs in what DLJ refers
to as a “walkup” strategy in
its “pitchbook” for the August 2000 Viasource IPO.
In the early 1980s, many underwriters were thinly capitalized
firms where risk-sharing was
important. On a $50 million deal with a 7% gross spread, the
underwriters shared $3.5 million
in fees. The lead underwriter might get 20% of this, or $0.7
million. As underwriters grew
larger, the lead manager was able to keep 60% of the fees, or
$2.1 million. Furthermore, with
more money left on the table, the lead underwriter could get
quid pro quos that might be
worth another $2.1 million. So it became a lot more lucrative to
be the lead underwriter. To get
this business, it was important to have an analyst who would be
bullish. Issuers were willing
to pay higher indirect fees due to both the analyst lust
hypothesis and the spinning
hypothesis. The time series evidence is consistent with this

story, but what about cross-
sectional implications?
V. Multiple Regression Results
One explanation for the cross-sectional pattern between age and
first-day returns is that
younger firms are riskier firms, and investors need to be
compensated for this risk. The
negative relationship between sales and first-day returns
reported in Table II also can be
interpreted as demonstrating a relation between the risk of an
IPO and underpricing. The
univariate sorts in Tables II and IV, however, are not
independent. Tech firms are much more
likely to be young firms, for instance. Thus, to examine
marginal effects, we report multiple
regression results with first-day return as the dependent
variable. Our explanatory variables
are chosen on the basis either of their association with first-day
returns in our univariate
sorts, or to test the changing risk composition, realignment of
incentives, and changing
issuer objective function hypotheses.
A. Ordinary Least Squares Regression Results
In the first and second rows of Table V, we use a total of 15
explanatory variables: a Carter-
Manaster top-tier underwriter dummy (set equal to one if the
lead underwriter has a rank of
8 or more, and zero otherwise), the logarithm of assets, a tech
stock dummy, an internet stock
dummy, the logarithm of (1 + age), share overhang (the ratio of
retained shares to issued
shares), a VC dummy, a pure primary offering dummy, the

logarithm of sales, a dummy variable
for IPOs in 1990-1998, a dummy variable for IPOs in 1999-
2000, a dummy variable for IPOs in
2001-2003, and interaction terms between the Carter-Manaster
top-tier underwriter dummy
and the time period dummy variables. Both assets and sales are
measured in millions of
dollars of 2003 purchasing power. The regression is:
First-Day Return
i
= a
0
+ a
1
Top-Tier Underwriter Dummy
i
+ a
2
ln(Assets)
i
+ a
3
Tech Dummy
i
+ a
4

Internet Dummy
i
+a
5
ln(1 + Age)
i
+ a
6
Overhang
i
+ a
7
VC Dummy
i
+ a
8
Pure Primary Dummy
i
+ a
9
ln(Sales)
i
+ a

10
Top-Tier Dummy·Nineties Dummy
i
+ a
11
Top-Tier Dummy·Bubble Dummy
i
+ a
12
Top-Tier Dummy·Post Dummy
i
+ a
13
Nineties Dummy
i
+ a
14
Bubble Dummy
i
+ a
15
Post Dummy
i

+ e
i
The variables ln(assets), tech stock dummy, internet dummy,
ln(1 + age), and ln(sales) measure
changing risk composition. The pure primary dummy is a
measure of the realignment of incentives,
Time? 25
P
e
r
io
d
In
te
r
.
T
o
p
-T

ie
r
U
n
d
e
rw
r
it
e
r
D
u
m
m
y
L
n
(A
s
s

e
ts
)
T
e
c
h
D
u
m
m
y
In
te
r
n
e
t
D
u
m
m

y
L
n
(1
+
a
g
e
)
S
h
a
r
e
O
v
e
r
h
a
n
g

V
C
D
u
m
m
y
P
u
r
e
P
r
im
a
r
y
D
u
m
m
y

L
n
(S
a
le
s
)
T
o
p
-
T
ie
r
D
u
m
m
y
·
N
in

e
ti
e
s
D
u
m
m
y
T
o
p
-
T
ie
r
D
u
m
m
y

·
B
u
b
b
le
D
u
m
m
y
T
o
p
-
T
ie
r
D
u
m

m
y
·
P
o
s
t
D
u
m
m
y
N
in
e
ti
e
s
D
u

m
m
y
B
u
b
b
le
D
u
m
m
y
P
o
s
t
D
u
m
m
y
R

2
a
d
j
(1
)
A
ll
2
.3
3
(1
.5
7)
3
.7
6
(3
.8
7)
-2

.4
0
(-5
.2
1)
5
.8
3
(4
.8
6)
34
.4
3
(7
.0
9)
-1
.2
8
(-3

.5
3)
3.
67
(9
.0
8)
1
.9
2
(1
.8
1)
-1
.1
7
(-1
.5
7)

0
.4
2
(1
.1
2)
-
-
-
8
.8
6
(1
4.
04
)
33
.4
9
(1
1.
04
)

5.
39
(3
.6
0)
0.
27
9
(2
)
A
ll
5
.4
5
(3
.6
8)
-2
.5
0

(-2
.6
6)
-2
.3
6
(-5
.1
3)
5
.5
6
(4
.6
7)
34
.8
8
(7
.2
2)

-1
.1
8
(-3
.3
1)
3.
54
(8
.9
7)
1
.7
8
(1
.6
9)
-1
.6

0
(-2
.1
7)
0
.4
2
(1
.1
3)
5.
05
(4
.6
0)
36
.2
5
(6
.1

7)
5.
29
(1
.5
3)
6.
82
(7
.7
7)
6
.6
6
(1
.2
8)
3.

34
(1
.0
7)
0.
29
0
(3
)
19
80
-1
98
9
12
.4
2
(8
.8
0)

-1
.7
7
(-2
.2
4)
-0
.7
5
(-2
.0
0)
2
.4
4
(2
.6
2)
-4
.0
6

(-0
.9
0)
-0
.8
9
(-2
.4
6)
0.
16
(0
.7
9)
0
.3
0
(0
.3
6)

-0
.7
5
(-0
.9
8)
-0
.2
5
(-0
.6
9)
-
-
-
-
-
0.
03
7

(4
)
19
90
-1
99
8
15
.4
3
(1
1.
39
)
3
.3
2
(3
.3
9)
-1
.7
1

(-4
.6
5)
5
.4
6
(5
.2
1)
31
.2
9
(3
.4
7)
-1
.3
1
(-3
.9
4)

2.
76
(7
.5
7)
-1
.9
0
(-1
.8
3)
-3
.2
5
(-3
.9
9)
0
.0

5
(0
.1
9)
-
-
-
-
-
0.
13
0
(5
)
19
99
-2
00
0
-1
6.
68

(-1
.5
9)
21
.2
2
(3
.1
1)
-5
.8
9
(-2
.9
8)
16
.9
8
(2
.8

4)
25
.0
2
(3
.7
2)
-0
.9
7
(-0
.2
9)
9.
35
(6
.6
5)
21

.4
8
(3
.4
2)
5
.5
9
(1
.0
6)
2
.3
8
(1
.3
2)
-
-
-
-

-
0.
19
9
(6
)
20
01
-2
00
3
8
.4
1
(2
.1
2)
-1
.2
9
(-0
.4

6)
-0
.9
7
(-1
.4
9)
5
.2
6
(1
.9
1)
-3
.2
5
(-0
.5
7)
-1

.3
5
(-1
.1
5)
2.
23
(4
.0
1)
4
.0
0
(1
.6
5)
-2
.4
1
(-1

.1
3)
1
.0
9
(2
.4
2)
-
-
-
-
-
0.
11
9
T
a
b
le
V

. R
e
g
re
s
s
io
n
s
o
f
P
e
rc
e
n
ta
g
e
F
ir
s
t-
D

a
y
R
e
tu
rn
s
o
n
S
e
v
e
ra
l V
a
ri
a
b
le
s
a
n
d

In
te
ra
c
ti
o
n
T
e
rm
s
T
he
s
am
pl
e
in
ro
w
1
in
cl

ud
es
5
,9
90
U
S
o
pe
ra
ti
ng
fi
rm
IP
O
s
ov
er
1
98
0-
20
03

w
he
re
th
e
of
fe
r p
ri
ce
is
a
t l
ea
st
$
5.
00
a
nd
c
om
pl
et

e
da
ta
o
n
al
l o
f t
he
v
ar
ia
bl
es
is
a
va
il
ab
le
. T
he
su

bp
er
io
ds
h
av
e,
r
es
pe
ct
iv
el
y,
1
,7
52
, 3
,2
35
, 7
97
, a
nd

2
06
o
bs
er
va
ti
on
s.
T
he
d
ep
en
de
nt
v
ar
ia
bl
e
in
a
ll

r
eg
re
ss
io
ns
is
th
e
pe
rc
en
ta
ge
f
ir
st
-d
ay
r
et
ur
n
fr

om
th
e
of
fe
r
pr
ic
e
to
th
e
fi
rs
t-
da
y
cl
os
in
g
pr
ic

e.
T
he
to
p-
ti
er
u
nd
er
w
ri
te
r
du
m
m
y
ta
ke
s
a
va
lu

e
of
o
ne
if
th
e
le
ad
u
nd
er
w
ri
te
r
ha
s
an
u
pd
at
ed
C

ar
te
r
an
d
M
an
as
te
r
(1
99
0)
r
an
k
of
8
o
r
m
or
e,
a

nd
ze
ro
o
th
er
w
is
e.
L
n(
as
se
ts
)
is
th
e
na
tu
ra
l l
og

ar
it
hm
o
f
th
e
pr
e-
is
su
e
bo
ok
v
al
ue
o
f
as
se
ts
, e
xp

re
ss
ed
in
m
il
li
on
s
of
d
ol
la
rs
o
f
20
03
p
ur
ch
as
in
g

po
w
er
u
si
ng
th
e
C
P
I.
T
he
te
ch
du
m
m
y
ta
ke
s
a

va
lu
e
of
o
ne
(
ze
ro
o
th
er
w
is
e)
if
th
e
fi
rm
is
in
th
e

te
ch
no
lo
gy
b
us
in
es
s,
a
nd
th
e
in
te
rn
et
d
um
m
y
is
s

im
il
ar
ly
d
ef
in
ed
(
in
du
st
ri
es
a
re
d
ef
in
ed
in
A
pp
en

di
x
D
).
L
n(
1
+
a
ge
)
is
th
e
na
tu
ra
l l
og
o
f
1
pl

us
th
e
ye
ar
s
si
nc
e
th
e
fi
rm
’s
f
ou
nd
in
g
da
te
a
s
of

th
e
IP
O
. S
ha
re
o
ve
rh
an
g
is
th
e
ra
ti
o
of
r
et
ai
ne
d

sh
ar
es
to
th
e
pu
bl
ic
f
lo
at
(
th
e
nu
m
be
r
of
s
ha
re
s

is
su
ed
).
T
he
V
C
d
um
m
y
ta
ke
s
a
va
lu
e
of
o
ne
(
ze

ro
o
th
er
w
is
e)
if
th
e
IP
O
is
b
ac
ke
d
by
v
en
tu
re
c
ap

it
al
. T
he
p
ur
e
pr
im
ar
y
du
m
m
y
eq
ua
ls
o
ne
(
ze
ro
o

th
er
w
is
e)
if
th
e
of
fe
ri
ng
is
1
00
%
p
ri
m
ar
y
(i
.e

.,
no
s
ec
on
da
ry
s
ha
re
s
so
ld
).
L
n(
sa
le
s)
is
th
e
na
tu

ra
l l
og
o
f t
he
tr
ai
li
ng
a
nn
ua
l f
ir
m
s
al
es
, e
xp
re
ss
ed

in
m
il
li
on
s
of
d
ol
la
rs
o
f 2
00
3
pu
rc
ha
si
ng
po
w
er

. T
he
n
in
et
ie
s
du
m
m
y
ta
ke
s
a
va
lu
e
of
o
ne
(z
er
o

ot
he
rw
is
e)
if
th
e
IP
O
o
cc
ur
re
d
du
ri
ng
1
99
0-
19
98
. T

he
b
ub
bl
e
du
m
m
y
ta
ke
s
a
va
lu
e
of
o
ne
(z
er
o
ot
he

rw
is
e)
if
th
e
IP
O
oc
cu
rr
ed
d
ur
in
g
19
99
-2
00
0.
T
he

p
os
t
du
m
m
y
ta
ke
s
a
va
lu
e
of
o
ne
(
ze
ro
o
th
er
w

is
e)
i
f
th
e
IP
O
o
cc
ur
re
d
du
ri
ng
2
00
1-
20
03
. T
he
i

nt
er
ac
ti
on
t
er
m
s
m
ul
ti
pl
y
th
e
to
p-
ti
er
un
de
rw

ri
te
r
du
m
m
y
by
th
e
ti
m
e
pe
ri
od
d
um
m
ie
s.
T
he
t-

st
at
is
ti
cs
(
in
p
ar
en
th
es
es
)
ar
e
ca
lc
ul
at
ed
u
si
ng

W
hi
te
’s
(
19
80
)
he
te
ro
sk
ed
as
ti
ci
ty
-c
on
si
st
en
t m
et

ho
d.
F
ir
st
-D
ay
R
et
ur
n i
=
a
0
+
a
1
T
op
-T
ie
r
U
nd

er
w
ri
te
r
D
um
m
y i
+
a
2l
n(
A
ss
et
s)
i +
a
3T
ec
h
D
um

m
y i
+
a
4
In
te
rn
et
D
um
m
y i
+
a 5
ln
(1
+
a
ge
) i
+
a
6O

ve
rh
an
g
i +
a
7
V
C
D
um
m
y i
+
a
8P
ur
e
P
ri
m
ar

y
D
um
m
y i
+
a
9l
n(
S
al
es
) i
+
a
10
T
op
-T
ie
r
D
um
m

y·
N
in
et
ie
s
D
um
m
y i
+
a
11
T
op
-T
ie
r
D
um
m
y·
B
ub

bl
e
D
um
m
y i
+
a 1
2T
op
-T
ie
r
D
um
m
y·
P
os
t
D
um
m

y i
+
a 1
3N
in
et
ie
s
D
um
m
y i
+
a
14
B
ub
bl
e
D
um
m
y i

+
a
15
P
os
t
D
um
m
y i
+
e
i
with a predicted positive coefficient. The changing issuer
objective function hypothesis is tested
by the change over time in the coefficients on the top-tier
underwriter dummy. Our hypothesis is
that, ceteris paribus, IPOs underwritten by top-tier underwriters
were underpriced more in the
1990s and, especially, in the bubble period because of spinning
and because they had more
highly ranked analysts. We use a top-tier Carter-Manaster
ranking as a proxy for all-star analyst
presence and the ability and willingness to spin. It should be
noted that the vast majority of

Institutional Investor all-star analysts are employed by top-tier
underwriters, which we define as
investment bankers with a Carter-Manaster rank of 8 or higher.
Several variables capture the predictions of multiple
hypotheses. For example, all three
hypotheses are consistent with a positive coefficient on
overhang, because the opportunity
cost of underpricing is lower, the lower is the fraction of the
firm sold (and thus the greater
the overhang), and small proportionate offerings are associated
with high valuations.
The slope coefficients in the row 1 regression are generally
consistent with the univariate
results reported earlier, although the lack of significance for the
VC dummy and ln(sales)
suggests that correlations between variables drive some of the
univariate patterns. The
negative coefficients on ln(assets) and ln(1 + age), and the
positive coefficients on the tech
and internet dummies, are consistent with the changing risk
composition hypothesis, given
that the bubble period saw a higher proportion of IPOs by young
tech and internet firms
than other periods. The negative and insignificant coefficient on
the pure primary dummy is
not consistent with the realignment of incentives hypothesis.
Recall that the average first-day return increased from 7.3% in
the 1980s to 14.8% in the
1990s, 65.0% during the internet bubble, and 11.7% in the post-
bubble period. We seek to
explain these increases: 7.5 percentage points from the 1980s to
the 1990s, 57.7 percentage
points from the 1980s to the internet bubble, and 4.4 percentage

points from the 1980s to the
post-bubble period. In Table V, the row 1 coefficient on the
nineties dummy of 8.86, or 8.86%,
suggests that none of the increase in underpricing from the
1980s to the 1990s has been
explained. The coefficient on the bubble dummy variable of
33.49 implies that only some of
the 57.7% difference in underpricing between the 1980s and the
internet bubble period is
accounted for. And the coefficient of 5.39 on the post-bubble
dummy variable suggests that
the variables are not adequate to explain the difference in
underpricing between the 1980s
and the post-bubble period as well.
In row 2, we add three explanatory variables allowing a shift in
the top-tier underwriter
dummy coefficient over time. Specifically, we add three
interaction terms by multiplying the
top-tier underwriter dummy by the time period dummies. As the
changing issuer objective
function hypothesis would predict, all three of these interaction
variables have positive
coefficients in row 2, and the shifts in the 1990s and bubble
periods are statistically significant.
In row 2, the coefficient on the nineties dummy of 6.82 (6.8%)
indicates that we are still
unable to explain the unconditional difference in underpricing
between the 1980s and 1990s
of 7.5%. Most importantly, however, the coefficient on the
bubble dummy falls to a statistically
insignificant 6.66 (6.7%). Since the unconditional difference in
underpricing between the
1980s and the bubble period is 57.7%, the row 2 regression is
able to account for the vast
majority of the extra underpricing associated with the bubble

period. The same is true for the
post-bubble dummy, where the coefficient of 3.34 (3.3%) is
statistically indistinguishable
from both zero and the unconditional difference in underpricing
of 4.4%. Hence, the shift in
the top-tier underwriter variable can explain all of the increase
in first-day returns between
the 1980s and the bubble and post-bubble time periods.
Thus, the coefficients on the time period dummies in row 1
suggest that neither the
changing risk composition hypothesis nor the realignment of
incentives hypothesis is able
Time? 27
to explain much of the unconditional changes in underpricing
over time. On the other hand,
when we introduce variables suggested by the changing issuer
objective function hypothesis
in row 2, the bubble period dummy drops to a statistically
insignificant 6.66%, although the
inability to explain the higher underpricing in the 1990s
remains.
Rows 3-6 present subperiod results. The top-tier underwriter
dummy coefficient is reliably
negative in the 1980s, positive in the 1990s, very positive in the
bubble period, and insignificant
in the post-bubble period. In the bubble period, the coefficient
on the top-tier underwriter
dummy is 21.22. This implies that IPOs with a top-tier lead
underwriter had 21.2% higher first-

day returns than IPOs with less prestigious bankers, after
adjusting for other factors.
This increase in underpricing associated with prestigious
underwriters in the 1990s and
the bubble period is a test of the changing issuer objective
function hypothesis. Also
consistent with this hypothesis is the increasing market share of
top-tier underwriters reported
in Tables II and IV. As we have argued, issuer decision-makers
were willing to pay for their
services by leaving money on the table because of the side
payments and the positive
analyst coverage that they or their companies received.
Inspection of the subperiod results in rows 3-6 of Table V
shows that the parameter
estimates on all of the explanatory variables except ln(1 + age)
have changed over time. This
nonstationarity suggests that increased underpricing over time is
not attributable entirely to
an increase in the fraction of IPOs by riskier companies or a
realignment of incentives,
unless, for example, an omitted variable bias has different
effects in different subperiods.
B. Instrumental Variable Regression Results
Habib and Ljungqvist (2001), Fernando, Gatchev, and Spindt
(2004), and others argue that
the prestige of the lead underwriter is endogenous in regressions
with underpricing as the
dependent variable. Habib and Ljungqvist (2001) deal with this
by running two-stage least
squares regressions for underpricing, where rather than using
the Carter-Manaster underwriter

prestige rank, they use the predicted rank from a first-stage OLS
regression. In Table VI, we
report underpricing regression results after controlling for the
endogeneity of underwriter
choice by using an instrument for the Carter-Manaster
underwriter rank. Our qualitative
conclusions are not substantially altered.
In Panel A of Table VI, the first-stage OLS regression for
underwriter rank has as explanatory
variables ln(assets), a tech dummy, an internet dummy, ln(1 +
age), share overhang, a VC
dummy, a pure primary dummy variable, ln(sales), and
age/assets.9 In rows 1 and 2 of our
Table V regressions, the pure primary dummy, the VC dummy,
and ln(sales) were weakly
related at best to first-day returns. In Panel A of Table VI, these
three variables are strongly
related to underwriter rank, except for the post-bubble
subperiod, where a small sample size
limits the statistical significance of all variables.
In Panel B of Table VI, we report regression results with
underpricing as the dependent
variable. In row 6, we report OLS regression coefficients. In
row 7, we report regression
coefficients from the second-stage regression using the
predicted value of underwriter rank
to construct the top-tier underwriter dummy instrument. That is,
if the predicted Carter-
Manaster rank is 8 or higher, the predicted value of the top-tier
underwriter dummy is one,
and zero otherwise.
Both rows 6 and 7 use the entire 24-year sample period, and a
comparison of the two rows

shows that controlling for the endogeneity of underwriter choice
does not substantially alter
the conclusions drawn from Table V. Both the 1990s and the
post-bubble dummy variables
9For IPOs with an age-to-assets ratio higher than one, we set
the ratio value at one.
T
a
b
le
V
I.
R
e
g
re
s
s
io
n
s
o
f
P

e
rc
e
n
ta
g
e
F
ir
s
t-
D
a
y
R
e
tu
rn
s
o
n
a
n
In

s
tr
u
m
e
n
t
fo
r
T
o
p
-t
ie
r
U
n
d
e
rw
ri
te
r
D

u
m
m
y,
L
n
(A
s
s
e
ts
),
T
e
c
h
D
u
m
m
y,
I
n

te
rn
e
t
D
u
m
m
y,
L
n
(1
+
a
g
e
),
a
n
d
S
h
a
re

O
v
e
rh
a
n
g
V
ar
ia
bl
es
a
re
d
ef
in
ed
in
T
ab
le
V

. V
al
ue
s
of
th
e
ag
e/
as
se
ts
ra
ti
o
gr
ea
te
r t
ha
n
on
e
ar

e
se
t e
qu
al
to
o
ne
. T
he
re
gr
es
si
on
s
us
e
5,
99
0
IP
O
s

fr
om
1
98
0-
20
03
, a
nd
1
,7
52
, 3
,2
35
, 7
97
,
an
d
20
6
IP
O

s,
r
es
pe
ct
iv
el
y,
i
n
th
e
su
bp
er
io
ds
.
P
an
el
A
r
ep

or
ts
f
ir
st
-s
ta
ge
O
L
S
r
eg
re
ss
io
ns
w
it
h
th
e
C
ar

te
r-
M
an
as
te
r
un
de
rw
ri
te
r
ra
nk
(
as
u
pd
at
ed
)
as
t

he
d
ep
en
de
nt
va
ri
ab
le
. I
n
P
an
el
B
, r
ow
6
re
po
rt
s

an
O
L
S
re
gr
es
si
on
w
it
h
th
e
fi
rs
t-
da
y
re
tu
rn
a
s

th
e
de
pe
nd
en
t v
ar
ia
bl
e.
R
ow
s
7-
11
re
po
rt
s
ec
on
d-
st

ag
e
re
gr
es
si
on
s
w
he
re
th
e
pr
ed
ic
te
d
va
lu
e
of
th
e

to
p-
ti
er
u
nd
er
w
ri
te
r d
um
m
y
fr
om
th
e
co
rr
es
po
nd
in

g
fi
rs
t-
st
ag
e
re
gr
es
si
on
is
u
se
d
as
a
n
in
st
ru
m
en

t.
T
hu
s,
if
th
e
pr
ed
ic
te
d
C
ar
te
r-
M
an
as
te
r r
an
k
is

8
o
r a
bo
ve
,
th
e
to
p-
ti
er
u
nd
er
w
ri
te
r
du
m
m
y
in

st
ru
m
en
t i
s
se
t e
qu
al
to
o
ne
, a
nd
z
er
o
ot
he
rw
is
e.
F

or
e
xa
m
pl
e,
th
e
ro
w
8
r
eg
re
ss
io
n
fo
r
th
e
19
80
s

us
es
th
e
pr
ed
ic
te
d
va
lu
e
of
th
e
ro
w
2
re
gr
es
si
on

a
s
th
e
un
de
rw
ri
te
r r
an
k
in
st
ru
m
en
t.
T
he
to
p-
ti
er

in
st
ru
m
en
ts
in
ro
w
7
o
f p
an
el
B
u
se
th
e
su
bp
er
io
d-

sp
ec
if
ic
re
gr
es
si
on
s
in
ro
w
s
2-
5
of
p
an
el
A
. T
he
t-

st
at
is
ti
cs
(i
n
pa
re
nt
he
se
s)
a
re
c
al
cu
la
te
d
us
in

g
W
hi
te
’s
(
19
80
)
he
te
ro
sk
ed
as
ti
ci
ty
-c
on
si
st
en
t m

et
ho
d.
U
W
R
an
k
i =
a
0
+
a
1
ln
(A
ss
et
s)
i +
a
2
T
ec

h
D
u
m
m
y i
+
a 3
In
te
rn
et
D
u
m
m
y
i
+
a
4
ln
(1
+

a
g
e)
i
+
a
5
O
v
er
h
an
g
i
+
a
6
V
C
D
u
m
m
y
i +
a

7
P
u
re
P
ri
m
ar
y
D
u
m
m
y
i +
a 8
ln
(S
al
es
) i
+
a

9
A
g
e/
A
ss
et
s i
+
e i
F
ir
st
-D
ay
R
et
u
rn
i =
a
0
+

a
1
T
o
p
T
ie
r
D
u
m
m
y
i +
a 2
ln
(A
ss
et
s)
i
+
a
3
T

ec
h
D
u
m
m
y
i +
a
4
In
te
rn
et
D
u
m
m
y
i +
a
5
ln
(1
+

a
g
e)
i +
a 6
O
ve
rh
an
g i
+
e
i
P
a
n
e
l
A
.
F
ir

s
t-
S
ta
g
e
O
L
S
R
e
g
r
e
s
s
io
n
s
w
it
h
U
n
d
e
r
w

r
it
e
r
R
a
n
k
a
s
t
h
e
D
e
p
e
n
d
e
n
t
V
a
r
ia
b
le

P
e
r
io
d
In
t
e
r
.
L
n
(
A
s
s
e
t
s
)
T
e

c
h
D
u
m
m
y
In
te
r
n
e
t
D
u
m
m
y
L
n
(
1

+
a
g
e
)
S
h
a
r
e
O
v
e
r
h
a
n
g
V
C

D
u
m
m
y
P
u
r
e
P
r
im
a
r
y
D
u
m
m
y
L
n

(
S
a
le
s
)
A
g
e
/
A
s
s
e
t
s
R
2
a
d
j
(1

)
A
ll
4.
55
(3
9.
14
)
0
.4
2
(1
3.
56
)
0.
28
(5
.7

1)
0
.8
1
(1
0.
82
)
0.
26
(8
.2
5)
0.
10
(8
.6
4)
0.

99
(2
1.
50
)
-0
.5
2
(-
11
.4
9)
0.
09
(4
.8
0)
-1
.2
7

(-
9.
57
)
0.
43
0
(2
)
19
80
-1
98
9
4
.3
5
(2
3.
20
)
0
.5

5
(8
.9
8)
0.
31
(3
.3
6)
-0
.4
4
(-
2.
14
)
0.
19
(3
.1

2)
0.
01
(0
.7
0)
0.
80
(9
.1
0)
-0
.5
8
(-
6.
63
)
0.

10
(2
.4
8)
-0
.8
5
(-
4.
06
)
0.
46
7
(3
)
19
90
-1
99
8

4
.3
2
(2
3.
81
)
0.
44
(8
.9
8)
0.
20
(2
.7
7)
0.
57

(2
.9
3)
0.
23
(5
.4
0)
0.
15
(6
.6
6)
1.
04
(1
6.
33

)
-0
.6
4
(-
10
.4
0)
0.
11
(4
.4
9)
-1
.3
1
(-
6.
45
)
0.

41
5
(4
)
19
99
-2
00
0
5
.9
9
(1
8.
51
)
0.
17
(3
.4
9)

0.
32
(3
.2
5)
0.
01
(0
.1
1)
0.
26
(3
.0
6)
0.
08

(5
.0
3)
0.
97
(8
.1
0)
-0
.0
2
(-
0.
15
)
0.
07
(2
.4
9)

-1
.3
8
(-
4.
22
)
0.
34
8
(5
)
20
01
-2
00
3
7.
29
(1

6.
89
)
0.
11
(1
.3
4)
0.
19
(0
.9
8)
-0
.1
9
(-
0.
64
)

0.
30
(2
.9
0)
0.
09
(2
.0
0)
0.
28
(1
.3
0)
-0
.0
4

(-
0.
19
)
-0
.0
7
(-
1.
79
)
-2
.9
3
(-
4.
40
)
0.
38
8

Time? 29
P
a
n
e
l
B
.
F
ir
s
t-
d
a
y
R
e
tu
r
n
a
s
D

e
p
e
n
d
e
n
t
V
a
r
ia
b
le
,
w
it
h
S
e
c
o
n
d
-s
ta

g
e
R
e
g
r
e
s
s
io
n
s
i
n
R
o
w
s
7
-1
1
P
e
r

io
d
In
te
r
.
T
o
p
-T
ie
r
U
n
d
e
rw
r
it
e
r

D
u
m
m
y
L
n
(A
s
s
e
ts
)
T
e
c
h
D
u
m
m
y

In
te
r
n
e
t
D
u
m
m
y
L
n
(
1
+
a
g
e
)
S
h

a
r
e
O
v
e
r
h
a
n
g
T
o
p
-
T
ie
r
D
u
m
m

y
·
N
in
e
ti
e
s
D
u
m
m
y
T
o
p
-
T
ie
r

D
u
m
m
y
·
B
u
b
b
le
D
u
m
m
y
T
o
p
-
T

ie
r
D
u
m
m
y
·
P
o
s
t
D
u
m
m
y
N
in

e
ti
e
s
D
u
m
m
y
B
u
b
b
le
D
u
m
m
y
P
o

s
t
D
u
m
m
y
R
2
a
d
j
(6
)
A
ll,
O
LS
4.
91
(3

.5
6)
-1
.8
8
(-2
.1
0)
-2
.0
4
(-6
.9
5)
6.
18
(5
.6
1)

34
.9
2
(7
.2
5)
-0
.9
6
(-2
.7
6)
3.
48
(8
.8
9)
5.
10

(4
.6
7)
36
.0
5
(6
.1
3)
4.
53
(1
.3
1)
6
.4
8
(7
.6
2)

5
.9
9
(1
.1
5)
3
.3
3
(1
.0
7)
0.
29
0
(7
)
A
ll
6

.5
8
(4
.5
8)
1
.0
4
(0
.8
9)
-2
.6
8
(-8
.0
1)
5
.6
3
(5

.0
7)
33
.7
5
(6
.9
5)
-0
.9
7
(-2
.8
0)
3.
35
(8
.9
5)

3.
89
(3
.4
1)
27
.0
0
(4
.8
4)
9.
99
(3
.0
0)
8
.5
2

(1
2.
21
)
18
.5
1
(4
.8
7)
-0
.1
1
(-0
.0
4)
0.
29
0
(8
)
19

80
-1
98
9
12
.5
4
(8
.8
9)
0
.9
6
(1
.0
0)
-1
.3
6
(-4
.1

9)
2
.3
6
(2
.6
9)
-4
.5
9
(-0
.9
8)
-0
.9
0
(-2
.7
1)
0.

12
(0
.6
0)
-
-
-
-
-
0.
03
6
(9
)
19
90
-1
99
8
14

.1
6
(9
.6
7)
3
.7
9
(3
.1
9)
-1
.6
9
(-5
.4
8)
5
.4
1
(5

.4
4)
31
.1
7
(3
.3
8)
-1
.2
2
(-3
.7
2)
2.
57
(7
.3
3)

-
-
-
-
-
0.
12
5
(1
0)
19
99
-2
00
0
14
.9
3
(1
.6
6)

23
.8
5
(3
.6
1)
-7
.2
5
(-4
.3
0)
17
.1
0
(2
.8
7)
24
.3

6
(3
.5
7)
-1
.0
0
(-0
.3
2)
9.
47
(6
.6
7)
-
-
-
-
-

0.
18
8
(1
1)
20
01
-2
00
3
7
.2
0
(2
.0
9)
1
.0
8
(0
.3

2)
-0
.5
2
(-0
.8
3)
6
.3
3
(2
.3
6)
-0
.6
4
(-0
.1
2)
-0

.4
8
(-0
.4
1)
2.
01
(3
.8
1)
-
-
-
-
-
-
0.
09
8
T
a

b
le
V
I.
R
e
g
re
s
s
io
n
s
o
f
P
e
rc
e
n
ta
g
e
F

ir
s
t-
D
a
y
R
e
tu
rn
s
o
n
a
n
In
s
tr
u
m
e
n
t
fo

r
T
o
p
-t
ie
r
U
n
d
e
rw
ri
te
r
D
u
m
m
y,
L
n
(A
s

s
e
ts
),
T
e
c
h
D
u
m
m
y,
I
n
te
rn
e
t
D
u
m
m

y,
L
n
(1
+
a
g
e
),
a
n
d
S
h
a
re
O
v
e
rh
a
n
g
(

C
o
n
ti
n
u
e
d
)
have approximately the same coefficient values in rows 6 and 7,
but there is a difference
between the rows in terms of the bubble dummy coefficient.
Row 6 (OLS) reports an insignificant
coefficient of 5.99 while the row 7 second-stage regression has
a coefficient of 18.51. This is
still much closer to zero than the 57.7% unconditional
difference in underpricing, however.
Rows 8-11 report subperiod results for the instrumented
regressions, which are analogous
to the OLS regressions in rows 3-6 of Table V. Controlling for
the endogeneity of underwriter
choice has no impact on our qualitative conclusions, except that
for the 1980s the coefficient
on the top-tier underwriter dummy changes from negative to
insignificantly positive.

VI. Alternative Explanations for the Underpricing of Internet
Stocks
Many alternative explanations have been advanced for the
severe underpricing of IPOs
during the internet bubble.10 One view is that many issuers
were more concerned with the
market price at lockup expiration than with what the offer price
was. Developing this idea,
Aggarwal, Krigman, and Womack (2002) argue that severe
underpricing generates
“information momentum,” resulting in a higher market price at
the end of the lockup period
when insiders typically sell some of their shares. While this
may be true, it is not clear that
the benefits to the issuing firm exceed the opportunity cost
associated with the increased
dilution from underpricing the IPO. Nevertheless, we are
comfortable with the notion that
during the internet bubble issuers placed less weight on IPO
proceeds and more weight on
the proceeds from future insider sales and follow-on offerings
than they did in prior periods.
This, after all, is part of the analyst lust hypothesis.
During the internet bubble, there were widespread concerns
about the valuation of internet
stocks. One explanation for the severe underpricing of internet
IPOs is that underwriters
were unwilling to price these offerings at the level that the
market was willing to pay out of
concern about lawsuits and damage to reputation if and when
the stocks eventually dropped
in price. The argument is that unsophisticated day traders and
others were bidding up the
price to unjustified levels, and the underwriters were unwilling

to price the IPOs at the
market price determined by these “noise traders.”
In untabulated results, we do not find a negative relation
between first-day returns and
subsequent performance in either the 1980s or the 1990s, but we
do find reversals during the
internet bubble.11 For example, of the 19 IPOs with a first-day
return of more than 300%
during the internet bubble, the average buy-and-hold return
from the first closing price until
the end of December 2002 is –95.0%.12 Measured from the
offer price, the average return
through December 2002 (or the delisting date, if earlier) is –
73.7% for these 19 IPOs, compared
to –43.5% for the other bubble period IPOs. This evidence is
consistent with the idea that
10DuCharme, Rajgopal, and Sefcik (2001), Schultz and Zaman
(2001), and Ofek and Richardson (2003), among
others, examine various hypotheses for the high underpricing of
US internet stocks. Arosio, Giudici, and Paleari
(2001) present evidence for the severe underpricing of European
internet stocks.
11Logue, Rogalski, Seward, and Foster-Johnson (2002) for IPOs
in 1988-1995 and Houge, Loughran, Suchanek,
and Yan (2001) for IPOs in 1993-1996 find a slight negative
relation between first-day returns and subsequent
three-year stock performance. Lowry (2003) finds no relation
for IPOs in 1973-1996, and Loughran and Ritter
(2002) find no relation for IPOs in 1990-1998.
12The only one of these 19 IPOs that did not decline by more
than 90% from the first-day close through
December 2002 is Cobalt Networks, which was acquired in
December 2000 after falling by 65.1%. Measured from
the first closing price to 180 calendar days later, the average
return was –46.8%. The bookrunners (with partial

credit given for joint bookrunners) on these 19 IPOs were SG
Cowen for 1, CSFB for 3, Deutsche Bank for 1.5,
Donaldson Lufkin Jenrette for 0.5, Goldman Sachs for 1.5,
Merrill Lynch for 2, Morgan Stanley for 8.5, and
Robertson Stephens for 1.
Time? 31
overoptimistic investor sentiment temporarily inflated the
market prices on these IPOs.
We are skeptical of this explanation for severe underpricing,
however, for if underwriters
were concerned that the market prices on internet stocks were
too high, presumably their
analyst recommendations after the end of the quiet period would
have been bearish. Bradley,
Jordan, and Ritter (2003), Cliff and Denis (2004), and Houston,
James, and Karceski (2004)
find this is in fact not the case.
The poor subsequent performance of IPOs with high first-day
returns in the bubble period
is also consistent with a less innocuous explanation, however.
As is typical in the academic
IPO literature, we have taken the first closing market price as
exogenous. Yet Smith and
Pulliam (2002) state that:
[T]he Securities and Exchange Commission is examining
whether some securities firms
coerced investors who got hot IPO shares into placing orders for
the same stocks at

higher prices on the first day of trading, as a condition of
getting the IPOs. That
practice, known as “laddering,” contributed to the huge one-day
run-ups in many IPOs
during the tech-stock mania. The SEC’s laddering probe has
focused on firms including
Goldman Sachs Group Inc., Morgan Stanley, Robertson
Stephens and J.P Morgan Chase.
On October 1, 2003, J.P. Morgan Chase settled with the SEC,
paying a $25 million fine.
Investors would be willing to buy these additional shares in the
aftermarket if the sum of
the profits from the IPO allocation they received and the
aftermarket purchases were positive
(calculated using the weighted average purchase price). In many
cases, the sales would
occur on the day the quiet period ends, which is when the
underwriters’ analysts typically
initiate coverage, almost always with “buy” ratings. Thus,
tainted analyst recommendations,
which unsuspecting individual investors paid attention to, allow
an exit at an inflated price.
Laddering would contribute to a negative correlation between
first-day returns and long-
run returns because the extra buying pressure on the first day
from these purchase orders
would create subsequent selling pressure when these shares
were sold. Unless the market
price is unaffected by buying and selling pressure, there will be
price impacts. The evidence
of stock price effects for analyst initiations at the end of the
quiet period (Bradley et al., 2003
and Ofek and Richardson, 2003), and for selling pressure at the

end of the lockup period
(Bradley, Jordan, Roten, and Yi, 2001; Brav and Gompers,
2003; and Field and Hanka, 2001)
suggests that such effects are present for IPOs.
VII. Conclusions
Why has underpricing changed over time? We explore three
non-mutually exclusive
explanations: changing risk composition, a realignment of
incentives, and a changing issuer
objective function.
A small part of the increase in underpricing can be attributed to
the changing risk
composition of the universe of firms going public. The physical
riskiness of firms going
public, as measured, for example, by age or assets, did not
change very much between the
1980s and the 1990s, although the bubble period saw a high
proportion of very young firms
go public, and the post-bubble period saw a high proportion of
older firms.
The realignment of incentives hypothesis argues that managerial
incentives to reduce
underpricing have decreased over time because of, among other
reasons, reduced CEO
ownership and a higher fraction of IPOs with no secondary
shares. When we look at the
whole sample period, however, there are only weak cross-

sectional relations between
underpricing and both the fraction of the firm sold and a dummy
variable for a pure primary
offering. Furthermore, CEO ownership, as measured by the
dollar value of holdings at the
offer price, was twice as high during the bubble period as
during the 1996-1998 period. This
measure of CEO incentives suggests that underpricing should
have decreased during the
bubble period.
The changing issuer objective function hypothesis posits two
reasons for why issuers
became more complacent about underpricing in the 1990s and
internet bubble period. First,
the analyst lust hypothesis states that analyst coverage became a
more important factor for
issuers choosing a lead underwriter, due to higher valuations
than in the 1980s. Since
underwriters do not charge explicit fees for providing analyst
coverage, issuers pay through
the indirect cost of underpricing. Second, the spinning
hypothesis argues that venture
capitalists and the executives of issuing firms were co-opted
through the allocation of hot
IPOs to their personal brokerage accounts. This gave these
decision-makers an incentive to
choose a lead underwriter with a reputation for leaving money
on the table in IPOs. Although
the excessive dilution that results from underpricing their own
IPO lowers their wealth, they
gain on personal account when other hot IPOs are allocated to
them. Since the profits from
these other IPOs are imperfectly correlated with their
undiversified paper wealth in their own
company, the decision-makers are willing to accept excessive

underpricing when their own
firm goes public.
Multiple regressions with underpricing as the dependent
variable yield evidence that
supports the changing issuer objective function hypothesis.
Specifically, top-tier underwriters
are associated with more underpricing in the 1990s, and
especially in the bubble period. This
is the result in both OLS and two-stage procedures that control
for the endogenous choice
of the lead underwriter. This is consistent with issuers choosing
top-tier underwriters who
have both influential analysts and, until spinning was
prohibited, many other hot IPOs to
allocate to important decision-makers. Furthermore, there is
strong corroborating evidence
in recent academic studies examining the relation between
Institutional Investor all-star
analysts and both IPO underpricing and changes in underwriter
market share, and in regulatory
settlements regarding spinning. We know of no evidence that is
inconsistent with the testable
implications of the spinning and analyst lust hypotheses.
We also document patterns in the US IPO market. The universe
of companies going public
in the US has changed over time. For example, we document
that there has been a pronounced
shift towards technology stocks and firms with negative
earnings. How firms are brought
public has changed over time, too. The market share of the
prestigious national underwriters
has increased, and regional investment banking firms are
increasingly shut out of lead
underwriter positions.

The reasons that IPOs are underpriced vary, depending on the
environment. In the 1980s, it
is conceivable that the winner’s curse problem and dynamic
information acquisition were the
main explanations for underpricing that averaged 7% in the US.
During the internet bubble, we
think that these were not the main reasons for underpricing.
Instead, analyst coverage and
side payments to CEOs and venture capitalists became of
significant importance.�
Appendix A. Founding Dates
The founding date is generally defined as the date of
incorporation. We try to find the date
of original incorporation, rather than a later date if the firm has
reincorporated in Delaware or
Time? 33
changed its name. Founding dates for 1980-1984 generally come
from inspection of the
prospectus. For 1985-1995, most of the founding dates were
provided by Laura Field. For 1985-
1987, Moody’s is the main source of data. For 1988-1992, the
prospectus is the main source. For
1993-1995, Disclosure and S&P Corporate Descriptions are the
main sources. For 1993, some of
the founding dates have come from Renaissance Capital. For
1996-2003, founding dates come
from a variety of sources: Securities Data Co., Moody’s, Dun &
Bradstreet’s Million Dollar

Directory, and inspection of the prospectuses on Edgar, and
were collected primarily by Laura
Field (Field and Karpoff, 2002) and Li-Anne Woo. Some
founding dates for 1999-2003 are from
Thomson Financial’s The IPO Reporter, an industry newsletter.
According to Laura Field, for
1988-1992, founding dates are earlier than the date of the most
recent incorporation for 48% of
the firms. An example of this is from the April 2000 prospectus
of Krispy Kreme doughnuts.
The firm going public was incorporated in 1999, but the
predecessor corporation was
incorporated in 1982. Elsewhere in the prospectus one finds the
statement that their first
doughnut shop was opened in 1937. We use 1937 as the
founding date.
For 1996-2000, we have used some of the founding dates that
Alexander Ljungqvist and
William Wilhelm have tabulated for their paper (Ljungqvist and
Wilhelm, 2003). They inspected
prospectuses and made judgments on many spinoffs.
Firms with inflation-adjusted (2003 purchasing power) sales in
the last 12 months prior to
going public of $200 million or more and younger than 2 years
are frequently “reverse LBOs”
or divisional spinoffs. For spinoffs, the founding date of the
division is used, when possible.
This may be the founding date of the parent corporation. For
example, Lucent Technologies
(a 1996 IPO) is the former Bell Labs division of AT&T. Its
founding date is given as the
founding date of Bell Labs. In general, “roll-ups” are given a
founding date corresponding
to the founding date of the parent firm (frequently a year before

the IPO).
Age is defined as the calendar year of offering minus the
calendar year of founding. Thus,
a 2-year old firm may be anywhere from 13 months old to 35
months old.
Because some years (1980-1984, 1988-1993, and 2000-2003)
have founding dates that are
primarily from the prospectus, rather than dates of incorporation
from Moody’s, etc., some
of the variation over time may be due to different data sources.
Appendix B. Post-Issue Shares Outstanding and Dual-class
Shares
Of the 6,391 IPOs in our sample, 433 have multiple classes of
shares outstanding after the
IPO. Most of these are firms whose IPO is composed of Class A
shares. Class B shares with
superior voting rights are owned by pre-issue shareholders, and
are not publicly traded.
These firms present a problem for computing the market
capitalization. CRSP reports shares
outstanding only for share classes that are publicly traded on
Nasdaq, the Amex, or the
NYSE. Thus, using the CRSP-reported shares outstanding to
compute the market capitalization
captures only part of the market value. To take an extreme
example, the United Parcel Services
IPO of November 9, 1999 issued 109 million shares of Class A
stock, but over 1 billion shares
of Class B stock also existed. Using only the Class A shares
outstanding would underestimate
the market value by 91%. The December 9, 1998 IPO of Infinity
Broadcasting is another

example. 140 million Class A shares were issued. CRSP reports
this as the number of shares
outstanding. But there were also 700 million Class B shares
outstanding, giving a market cap
six times as high for all the shares. In all our calculations of
market capitalization, we assume
that non-traded shares have the same price per share as the
publicly traded class.
Thomson Financial Securities Data has many errors in reporting
the number of post-issue
shares outstanding, although the firm attempts to capture all
classes. For single-class IPOs,
CRSP is much more reliable. For dual-class IPOs, Thomson
Financial is more reliable.
Ljungqvist and Wilhelm (2003), in their analysis of IPOs from
1996-2000, also report substantial
error rates in Thomson Financial’s data on, e.g., post-issue
shares outstanding, EPS, venture
capital backing, and founding dates.
If we use just the CRSP-reported shares outstanding, the median
market cap figure that we
calculate is 4% lower than the Table I, Panel B numbers
reported. The mean market cap using
CRSP data is 17% lower than the numbers reported in Table I,
Panel A.
Scott Smart and Chad Zutter supplied us with a list of 258 dual-
class IPOs for 1990-1998,
along with the post-issue shares outstanding. CRSP does not

identify all the IPOs that
involve dual-class shares that Smart and Zutter (2003) identify.
The post-issue shares
outstanding number that Smart and Zutter have recorded is the
same as the Thomson Financial
number only a little over 50% of the time. For discrepancies
where we could check the
prospectus using EDGAR (beginning in 1996), we found that
Smart and Zutter were correct
over 90% of the time. For dual-class IPOs where we could not
verify the number, we use the
Smart and Zutter number as the first choice and the maximum of
the Thomson Financial and
the CRSP number as the second choice. We use Dealogic’s
number if we cannot inspect the
prospectus on EDGAR.
Appendix C. Underwriter Rank for IPOs over 1992-2003
For underwriter prestige rankings, we start with the Carter and
Manaster (1990) and Carter
et al. (1998) rankings. When a firm goes public, the
underwriting section of the prospectus
lists all the investment banking firms that are part of the
underwriting syndicate, along with
the number of shares that each underwrites. Lead underwriters
are listed first, followed by
co-managing underwriters, and then other syndicate members.
More prestigious underwriters
are listed first in the non-managing underwriting section, in
brackets, with underwriters in
higher brackets underwriting more shares. If an underwriter
always appears in the highest
bracket among non-managing underwriters, it is assigned the
top ranking of 9 on a 0-9 scale.

For underwriters in the 1992-2003 period, we assign a ranking
as follows. The May 1999
Goldman Sachs prospectus lists over 120 underwriters, with
numerous brackets. Managing
and co-managing underwriters are assigned a ranking of 9; other
underwriters are given a
ranking based on their bracket, with a few minor adjustments.
Other underwriters not included
in the Goldman Sachs prospectus are assigned a ranking of 1 or
2 if they were penny stock
underwriters that had been subject to enforcement actions by the
SEC during 1995-1999 (the
information on enforcement actions was provided by the
Chicago office of the SEC’s Division
of Enforcement).
The numerical reputation ranking of remaining underwriters was
determined by Bruce
Foerster of South Beach Capital in Miami. Foerster has been an
investment banker for close
to 30 years, participating in the underwriting of 150 IPOs and
hundreds of other transactions
while a managing director at A.G. Becker Paribas, Paine
Webber, Lehman Brothers, and South
Beach Capital. He is also the editor of the Securities Industry
Association’s Capital Markets
Handbook (Foerster, 2004), and has an encyclopedic knowledge
of the investment banking
industry during the last few decades. For the handful of other
underwriters that Bruce
Foerster was not familiar with and that were not identified in
our other procedures, we assign
a rank based on the offer price of IPOs that they underwrote,
with penny stock underwriters
earning the lowest ranks.

Time? 35
We made several alterations to the Carter and Manaster
rankings for 1980-1984 and the
Carter, Dark, and Singh rankings for 1985-1991. Carter, Dark,
and Singh assign Hambrecht &
Quist a 9.0, which we lower to 8.1. Carter and Manaster assign
a rank of 2.0 to D.H. Blair in the
1980-1984 period, and Carter, Dark, and Singh assign it a rank
of 8.0 to D.H. Blair during 1985-
1991. We assign a 4.1 to D.H. Blair for all years. A potential
flaw in the Carter and Manaster
methodology is that a penny stock underwriter that is never
allowed into a syndicate of
reputable underwriters might never be in a low bracket. Our
judgment methodology avoids
this problem. Note that we make very few changes in rankings.
All of the rankings we assign are integers followed by a 0.1 (1.1
up to 9.1). We attach a 0.1
to all our rankings so that other researchers can easily
distinguish between our rankings and
those from Carter and Manaster and Carter, Dark, and Singh,
which never end with a 0.1. To
use our rankings in empirical work, we recommend using “if
then” commands to covert the
x.1 rankings to x.0.
Appendix D. Internet and Technology Firms
To identify IPOs that are internet-related at the time of their
offer, we merge the internet
identifications of Thomson Financial Securities Data, Dealogic,

and IPOMonitor.com. In
1998, Securities Data classified only 18 IPOs as internet stocks,
omitting such firms as uBID,
Ticketmaster Online/Citysearch, NetGravity, and Verio.
IPOMonitor.com classified 27 IPOs
from 1998 as internet stocks, but omitted Cdnow and Interactive
Magic, among others. Since
these sources generally did not backdate the identification of
early internet companies, we
assign a “1” value to America On-Line, Spyglass, and Netscape.
The classifications have some inherent arbitrariness. For
example, Storage Area Network
(SAN) companies and telecommunications companies are not
internet stocks; nor are such
IPOs as VA Linux and Perot Systems.
SDC identifies two IPOs from the 1980s as internet firms: IPC
Communications, a
manufacturer of telecommunications equipment, and
McClatchey Newspapers, which offered
on-line services.
Tech stocks are defined as those in SIC codes 3571, 3572, 3575,
3577, 3578 (computer
hardware), 3661, 3663, 3669 (communications equipment),
3671, 3672, 3674, 3675, 3677, 3678,
3679 (electronics), 3812 (navigation equipment), 3823, 3825,
3826, 3827, 3829 (measuring and
controlling devices), 3841, 3845 (medical instruments), 4812,
4813 (telephone equipment),
4899 (communications services), and 7371, 7372, 7373, 7374,
7375, 7378, and 7379 (software).
References

Aggarwal, R., L. Krigman, and K. Womack, 2002, “Strategic
IPO Underpricing, Information Momentum,
and Lockup Expiration Selling,” Journal of Financial Economics
66, 105-137.
Arosio, R., G. Giudici, and S. Paleari, 2001, “Why Do (or Did?)
Internet-stock IPOs Leave So Much Money
on the Table?” Politecnico di Milano (Italy) Working Paper.
Baron, D., 1982, “A Model of the Demand of Investment
Banking Advising and Distribution Services for
New Issues,” Journal of Finance 37, 955-976.
Barry, C., 1989, “Initial Public Offerings Underpricing: The
Issuer’s View—A Comment,” Journal of Finance
44, 1099-1103.
Beatty, R.P., and I. Welch, 1996, “Issuer Expenses and Legal
Liability in Initial Public Offerings,” Journal of
Law and Economics 39, 545-602.
Benveniste, L.M., and W.J. Wilhelm, 1997, “Initial Public
Offerings: Going by the Book,” Journal of Applied
Corporate Finance 10, 98-108.
Bradley, D.J. and B. Jordan, 2002, “Partial Adjustment to
Public Information and IPO Underpricing,”
Journal of Financial and Quantitative Analysis 37, 595-616.
Bradley, D.J., B. Jordan, and J.R. Ritter, 2003, “The Quiet
Period Goes Out with a Bang,” Journal of
Finance 58, 1-36.

Bradley, D.J., B. Jordan, I.C. Roten, and H. C. Yi, 2001,
“Venture Capital and IPO Lockup Expiration: An
Empirical Analysis,” Journal of Financial Research 24, 465-494.
Brav, A. and P. Gompers, 2003, “The Role of Lock-ups in
Initial Public Offerings,” Review of Financial
Studies 16, 1-29.
Busse, J.A. and T.C. Green, 2002, “Market Efficiency in Real
Time,” Journal of Financial Economics 65,
413-437.
Carter, R.B., F.H. Dark, and A.K. Singh, 1998, “Underwriter
Reputation, Initial Returns, and the Long-run
Performance of IPO Stocks,” Journal of Finance 53, 285-311.
Carter, R.B. and S. Manaster, 1990, “Initial Public Offerings
and Underwriter Reputation,” Journal of
Finance 45, 1045-1068.
Clarke, J.E., C. Dunbar, and K.M. Kahle, 2003, “All-star
Analyst Turnover, Investment Bank Market Share,
and the Performance of Initial Public Offerings,” Georgia Tech
Working Paper.
Cliff, M. and D.J. Denis, 2004, “Do IPO Firms Purchase
Analyst Coverage with Underpricing?” Journal of
Finance (Forthcoming).
Cooney, J.W., A.K. Singh, R.B. Carter, and F.H. Dark, 2001,
“IPO Initial Returns and Underwriter Reputation:
Has the Inverse Relationship Flipped in the 1990s?” University
of Kentucky, Case Western Reserve
University, and Iowa State University Working Paper.

DuCharme, L.L., S. Rajgopal, and S.E. Sefcik, 2001, “Why Was
Internet IPO Underpricing so Severe?”
University of Washington Working Paper.
Dunbar, C.G., 2000, “Factors Affecting Investment Bank Initial
Public Offering Market Share,” Journal of
Financial Economics 55, 3-41.
Elkind, P. and M. Gimein, 2001, “The Trouble with Frank,”
Fortune (September 3).
Fernando, C.S., V.A. Gatchev, and P.A. Spindt, 2004, “Wanna
Dance? How Firms and Underwriters Choose
Each Other,” Tulane University Working Paper.
Field, L.C. and G. Hanka, 2001, “The Expiration of IPO Share
Lockups,” Journal of Finance 56, 471-500.
Field, L.C. and J. Karpoff, 2002, “Takeover Defenses of IPO
Firms,” Journal of Finance 57, 1857-1889.
Foerster, B., 2004, Securities Industry Association’s Capital
Markets Handbook, 5th Ed., Aspen, CO,
Aspen Publishers.
Habib, M. and A. Ljungqvist, 2001, “Underpricing and
Entrepreneurial Wealth Losses: Theory and Evidence,”
Review of Financial Studies 14, 433-458.
Hoberg, G., 2003, “Strategic Underwriting in Initial Public
Offerings,” Yale University Working Paper.
Hong, H. and J.D. Kubik, 2003, “Analyzing the Analysts:
Career Concerns and Biased Earnings Forecasts,”
Journal of Finance 58, 313-351.

Houge, T., T. Loughran, G. Suchanek, and X. Yan, 2001,
“Divergence of Opinion, Uncertainty, and the
Quality of Initial Public Offerings,” Financial Management 30,
5-23.
Houston, J., C. James, and J. Karceski, 2004, “What a
Difference a Month Makes: Stock Analyst Valuations
Following Initial Public Offerings,” University of Florida
Working Paper.
Krigman, L., W. Shaw, and K. Womack, 2001, “Why Do Firms
Switch Underwriters?” Journal of Financial
Economics 60, 245-284.
Time? 37
Ljungqvist, A.P., F. Marston, and W.J. Wilhelm, 2003,
“Competing for Securities Underwriting Mandates:
Banking Relationships and Analyst Recommendations,” NYU
Working Paper.
Ljungqvist, A.P. and W.J. Wilhelm, 2003, “IPO Pricing in the
Dot-com Bubble,” Journal of Finance 58, 723-752.
Logue, D.E., R.J. Rogalski, J.K. Seward, and L. Foster-Johnson,
2002, “What Is Special about the Roles of
Underwriter Reputation and Market Activities in Initial Public
Offerings?” Journal of Business 75, 213-243.
Loughran, T. and J.R. Ritter, 2002, “Why Don’t Issuers Get
Upset About Leaving Money on the Table in
IPOs?” Review of Financial Studies 15, 413-443.

Lowry, M., 2003, “Why Does IPO Volume Fluctuate So Much?”
Journal of Financial Economics 67, 3-40.
Muscarella, C. and M. Vetsuypens, 1990, “Firm Age,
Uncertainty, and IPO Underpricing: Some New
Empirical Evidence,” SMU Working Paper.
NASD Regulation, 2002, “NASD Regulation Charges Credit
Suisse First Boston with Siphoning Tens of
Millions of Dollars of Customers’ Profits in Exchange for “Hot”
IPO Shares,” news release available at
http://www.nasdr.com/news (January 22).
Ofek, E. and M. Richardson, 2003, “Dotcom Mania: The Rise
and Fall of Internet Stock Prices,” Journal of
Finance 58, 1113-1138.
Reuter, J., 2004, “Are IPO Allocations for Sale? Evidence from
the Mutual Fund Industry,” University of
Oregon Working Paper.
Ritter, J.R., 1984, “The ‘Hot Issue’ Market of 1980,” Journal of
Business 57, 215-240.
Ritter, J.R. and I. Welch, 2002, “A Review of IPO Activity,
Pricing, and Allocations,” Journal of Finance 57,
1795-1828.
Rynecki, D., 2002, “Rebuild the Chinese Wall,” Fortune (June
24).
Schack, J., 2002, “Promises, Promises,” Institutional Investor
(October), 39-45.
Schultz, P. and M. Zaman, 2001, “Do the Individuals Closest to
Internet Firms Believe They Are Overvalued?”

Journal of Financial Economics 59, 347-381.
Securities and Exchange Commission, 2002, “SEC Sues CSFB
for IPO Violations; CSFB Will Pay $100
Million,” Litigation release 17327 available at
http://www.sec.gov/litigation.
Sherburne, J.C., 2002, “Letter to the Honorable Michael G.
Oxley,” available at http://
financialservices.house.gov/media/pdf/citiresp2_001.pdf.
Sherman, A.E. and S. Titman, 2002, “Building the IPO Order
Book: Underpricing and Participation Limits
with Costly Information,” Journal of Financial Economics 65, 3-
29.
Siconolfi, M., 1997, “Underwriters Set Aside IPO Stock for
Officials of Potential Customers,” Wall Street
Journal (November 12), A1.
Smart, S. and C.J. Zutter, 2003, “Control as a Motivation for
Underpricing: A Comparison of Dual- and
Single-class IPOs,” Journal of Financial Economics 69, 85-110.
Smith, R., 2002, “Goldman Gave Hot IPO Shares to Top
Executives of Its Clients,” Wall Street Journal
(October 3), A1.
Smith, R., A. Grimes, G. Zuckerman, and K. Scannell, 2002,
“Something Ventured and Something Gained?”
Wall Street Journal (October 17), C1.
Smith, R. and S. Pulliam, 2002, “Two More Wall Street Firms
Are Targeted in Trading Probe,” Wall Street
Journal (April 25), A1.

White, H., 1980, “A Heteroskedasticity-consistent Covariance
Matrix Estimator and a Direct Test for
Heteroskedasticity,” Econometrica 48, 817-838.
Assessed questions
Assessed questions
· Hints: Be careful with the different units. You will generally
need to work in m, m3, kg etc. Also energies given in eV will
generally have to be converted into Joules (by multiplying by
the electronic charge) before they can be used in the various
equations.
Electron Mobility (m
2
V
-1
s
-1
) 0.1 1 10 100
Temperature (K)
1. Using data taken from the above figure calculate the change
in the electron scattering time for the clean bulk GaAs sample
as the temperature is reduced from ~300K to 5K. What is the
corresponding change for the best single heterojunction between
~300K and 1K. Using your results calculate the average
distance travelled by electrons between collisions in clean bulk
GaAs at 5K and the best heterojunction at 1K. The electron

effective mass in GaAs is 0.067m0.
2. The semiconductors AlAs, InSb and GaP are to be grown on
an InP substrate. Using the following figure calculate the
approximate lattice mismatch between these semiconductors and
InP, expressing your results as a percentage of the InP lattice
constant. For each case state if the epitaxial layer will be
subjected to compressive or tensile strain.
3. The exciton binding energy in a semiconductor is given by
the equation
e m4 */32π2ħ2 εr2 ε02
where m* is the carrier effective mass and εr is the relative
permittivity of the semiconductor. In a quantum well made from
this semiconductor the exciton binding energy is enhanced by a
factor of 1.9. Calculate a value for the binding energy of an
exciton in this quantum well where m*=0.09mo and εr=10.
Calculate the temperature corresponding to this energy.
4. The exciton binding energy of a semiconductor is 8meV and
when used to form a quantum wire this binding energy is
increased by a factor of three. If the semiconductor has a bulk
band gap of 1.520eV and the confinement energies for the
lowest electron and hole states are 140 and 25meV respectively,
calculate the energy of the lowest excitonic transition.
5. A quantum wire has a rectangular cross section with
dimensions 4nm and 6nm. If the effective mass of the electrons
is 0.08m0 calculate the energies of the first 6 confined electron
states, giving the values of the two quantum numbers for each
state.
6. An edge emitting semiconductor laser is found to have facets
mirrors of reflectivity R=0.35. If the laser is surrounded by air
of refractive index 1 what is the refractive index of the
semiconductor?
Useful constants

Charge of an electron e = 1.6x10-19C Mass of an electron m =
9.1x10-31Kg
Planck's constant h = 6.6x10-34Js
Planck's constant /2πℏ = 1.0x10-34Js
Boltzmann's constant k = 1.38x10-23JK-1
Permittivity of free space ε0=8.85x10-12Fm-1
Speed of light c=3.0x108ms-1
1
1
1

Module PHY6002 Inorganic Semiconductor Nanostructures Lectur.docx

More Related Content

Similar to Module PHY6002 Inorganic Semiconductor Nanostructures Lectur.docx (20)

More from moirarandell (20)

Recently uploaded (20)

Module PHY6002 Inorganic Semiconductor Nanostructures Lectur.docx