Full text of DOI:10.1111/0022-1082.00341

THE JOURNAL OF FINANCE • VOL. LVI, NO. 2 • APRIL 2001
Corporate Bond Trading Costs:
A Peek Behind the Curtain
PAUL SCHULTZ*
ABSTRACT
In this paper, I use institutional corporate bond trade data to estimate transac-
tions costs in the over-the-counter bond market. I find average round-trip trading
costs to be about $0.27 per $100 of par value. Trading costs are lower for larger
trades. Small institutions pay more to trade than large institutions, all else being
equal. Small bond dealers charge more than large ones. I find no evidence that
trading costs more for lower-rated bonds.
TRANSACTIONS COSTS IN EQUITY MARKETS have been studied extensively by finan-
cial economists in recent years. In contrast, almost nothing is known about
the costs of trading bonds. In this paper, I use a large dataset of institutional
corporate bond trades from 1995 to 1997 to estimate trading costs for
investment-grade bonds in the over-the-counter market. I find that round-
trip trading costs average about $0.27 per $100 of par value. Costs are smaller
for large trades. All else being equal, active institutions pay less to trade
than inactive ones, and large dealers charge less than small ones. Bond
ratings seem to have little effect on trading costs.
Two recent working papers also examine institutional bond trading costs.
Hong and Warga ~1998! employ a sample of 1,973 buy and sell trade prices
for the same bond on the same day and estimate an effective spread of $0.13
per $100 of par value for investment-grade bonds and $0.19 per $100 of par
value for non-investment-grade bonds. Chakravarty and Sarkar ~1999! also
estimate trading costs by comparing buy and sell prices for the same bond
on the same day. They find that trading costs are highest for municipal
bonds. Corporate bonds are the next most expensive and treasury bonds are
the cheapest to trade. Chakravarty and Sarkar ~1999! report that trading
costs have declined from 1995 to 1997, at least for municipal bonds. How-
ever, they are unable to find evidence of differences in trading costs across
institutions, perhaps because of their small sample size.
*
University of Notre Dame. Financial support from TIAA-CREF, The University of Chicago,
and The University of Notre Dame made this paper possible. Useful comments were provided
by Shane Corwin, Amy Edwards, Douglas Fore, Jeffrey Harris, Francis Longstaff, Andrew Nybo,
Joseph Sinkey, René Stulz ~the editor!, Arthur Warga, and seminar participants at the Univer-
sity of Georgia, Washington University, and the 1999 Western Finance Association Meetings.
Steve Sterman provided helpful background information on the corporate bond market. An
anonymous referee provided numerous helpful comments. I am responsible for any remaining
blunders and so forth.
6
77

6
78
The Journal of Finance
The methodology employed here differs from the methodology used in Hong
and Warga ~1998! or Chakravarty and Sarkar ~1999!. Rather than using
matched buy and sell trades in the same bond on the same day, I estimate
the bid quote at the time of the trade and regress the difference between the
trade price and the estimated bid price on a dummy variable that takes a
value of one for buys and zero for sells. The coefficient on this dummy vari-
able is an estimate of the difference between the buy and sell prices, or, equiv-
alently, an estimate of the round-trip trading costs. One advantage of using this
technique is that it provides a much larger set of observations and thus per-
mits an examination of the factors that determine trading costs. A second
advantage is that it allows us to examine a more representative cross section
of bonds. If attention is restricted to bonds with both buy and sell orders on
the same day, transactions cost estimates reflect only the most active bonds.
The remainder of the paper is organized as follows: In the first section, I
provide a description of how the corporate bond market works. In Section II,
I provide information on the data used here. Section III reports estimates of
trading costs for corporate bonds. In Section IV, I examine determinants of
trading costs in the bond market. Section V offers a summary and conclusions.
I. How the Corporate Bond Market Works
The corporate bond market is primarily an institutional market. Almost
all trading takes place over the counter, where the potential bond trader
cannot observe all quotes in a central location or on a computer screen. In-
stead, the institution must call several dealers for quotes or broadcast a list
of bonds to sell ~or buy! to various dealers through Bloomberg, a vendor of
quotes and financial information that is popular with institutions. Alterna-
tively, dealers may broadcast quotes on lists of bonds in their inventory to
potential institutional customers. When dealers broadcast quotes, they typ-
ically provide bid or ask quotes but not both. Quotes are indicative, not firm.
If an institution wants to buy or sell a bond that is not quoted, they can
contact dealers to provide quotes. Most of the variation in the prices of
investment-grade bonds comes from fluctuations in interest rates, so bid
and ask quotes for investment-grade bonds are given in terms of additional
basis points of yield over a treasury security of similar maturity. Thus deal-
ers can usually provide a quote by comparing the bond with other bonds
with similar maturities, coupon rates, bond ratings, and call provisions and
then estimating the yield spread between the bond and the treasury secu-
rity. High-yield bonds are quoted in dollars rather than in terms of a pre-
mium from treasury yields because they, like equities, are affected primarily
by firm-specific factors. As a result, dealers cannot easily determine an ap-
propriate quote by comparing the bond to others of similar rating, coupon,
and maturity.
Most bonds trade infrequently. Alexander, Edwards, and Ferri ~1998! cite
anecdotal evidence that bonds initially trade often but that trading declines
as the bonds fall into the hands of institutions who hold them to maturity.

Corporate Bond Trading Costs
679
Also contributing to the lack of volume is the fact that institutions buy bonds
in sufficiently large quantities that even the largest issues can be held en-
tirely by 200 or fewer institutions.
II. Data
Data on corporate bond trades from January 1995 through April 1997 are
obtained from Capital Access International. These data are also employed by
Chakravarty and Sarkar ~1999! and Hong and Warga ~1998!. The record for
each trade consists of the trade date, the identity of the institution, the
identity of the dealer, the bond’s CUSIP number, the firm issuing the bonds,
a description of the bonds, the coupon rate, the maturity date, whether the
trade is a buy or sell, the par value of the traded bonds, and the actual dollar
value of the trade. Capital Access also provides an indicator variable that
says whether each trade date is the actual trade date or whether the date is
an estimate provided by Capital Access. Trades with estimated dates are not
included in transaction cost estimates.
Capital Access obtains their data from several sources. Insurance compa-
nies are required to file a Schedule D with the National Association of In-
surance Commissioners every quarter. This schedule includes the trade date,
amount bought or sold, the identity of the buyer and seller, and the par
value of the bonds traded for all bond market transactions. The president,
secretary, and controller of the insurance company must sign and attest to
the validity of the information in the statement. Inaccurate or fraudulent
statements leave the signers open to statutory fines and possible criminal
penalties. Capital Access obtains Schedule D information from A. M. Best &
Co. and provides further cleaning and checking of these data. Information
on insurance company trades appears complete, which is important because
life insurance companies by themselves hold almost 40 percent of corporate
bonds. Mutual fund managers are required to file information on bond hold-
ings semiannually and some file quarterly. Capital Access collects mutual
fund information directly and also receives information from Morningstar.
Public pension funds are not required to file holdings but Capital Access
obtains information on their trades through voluntary disclosure and through
the filing of Freedom of Information Act forms.
Several features of these data make them well suited for estimating in-
stitutional bond-trading costs. First, prices of actual trades are provided and
the trades are designated as buys or sells. Also, the institution and counter-
party are identified for each trade. Finally, although not all trades of all
institutions are included in the dataset, there are between 3,400 and 9,549
corporate bond trades each month during the sample period.
These data also have several limitations. First, although information is
complete for institutions included in the database, not all institutions are
included. Second, the dollar value of trades is rounded up to the nearest $1,000.
I expect the rounding to have little effect on my transactions cost estimates
as the median sample trade size is $1,513,000. Also, to estimate trading

6
80
The Journal of Finance
costs, I regress the difference between trade prices and their contempora-
neous bid prices on a dummy variable that takes a value of one for buy orders.
Trade prices are rounded up to the nearest $1,000 for both buys and sells,
and thus any bias from this should end up in the intercept term. The slope
coefficient, which estimates round-trip trading costs, should be unaffected.
In total, the Capital Access dataset contains 192,867 corporate bond trades
from January 1995 through April 1997. I match these data with information
from the beginning of the month ~or month end if it is not available! from
the Fixed Income Database compiled at the University of Houston. This source
provides the issue date, maturity date, coupon yield, duration, convexity,
and Moody’s and S&P ratings for each bond contained in the Lehman Broth-
ers’ bond index. This dataset has been used by Hong and Warga ~1998! and
Blume, Lim, and MacKinlay ~1998! among others. Of the total of 192,867
trades in the Capital Access data, 117,015 ~about 61 percent! could be matched
with information in the Fixed Income Database.
For those matched bonds, the median time to maturity is 8.71 years at the
time of a trade. Most trades took place soon after the bonds’ issuance. The
median percentage of the bond’s life remaining at the time of a trade is 85
percent. Trades are large, with the median trade containing 0.98 percent of
the total number of outstanding bonds in the issue. Most of the bonds are
investment-grade bonds. Specifically, 2.2 percent of the bonds are Aaa rated,
1
0.2 percent are Aa rated, 42.8 percent are rated A, and 25.7 percent are
rated Baa.
It is often reported that the over-the-counter bond market is dominated by
a small number of dealers. In total, 601 dealers appear in the sample, but
the top 12 dealers traded 72 percent of the total. Although trading volume
seems to be dominated by a relatively small number of dealers, trading in
specific bonds seems to be spread across multiple dealers.
III. Estimates of Transaction Costs
A. How I Estimate Trading Costs
Round-trip transactions costs are estimated by regressing the difference
between the trade price and the estimated contemporaneous bid quote on a
dummy variable that takes a value of one for buys and zero for sells. That is,
Buy
⌬i ϭ a ϩ a D
i
ϩ ei ,
~1!
0
1
Buy
i
where ⌬ is the price of trade i minus the estimated bid price and D
i
equals one if trade i is a buy and zero otherwise. The variable of interest is
a , an estimate of the difference in prices for buy and sell orders. Or, in
1
1
other words, a is an estimate of the average round-trip trading costs.
1
1
This is similar to the technique used in Warga ~1991! to estimate NYSE bond trading costs
by comparing NYSE trade prices with Lehman Brothers’ quotes.

Corporate Bond Trading Costs
681
Bid quotes are obtained from the University of Houston’s Fixed Income
2
Database. This dataset contains all bonds included in the Lehman Brothers’
bond index. Over the sample period, this index contains between 9,485 and
1
3,582 bonds, but quotes are not available for the least active and recently
issued bonds. A more important limitation of the data is that quotes are
provided at month end only. This means that quotes within the month have
to be estimated. Most of the day-to-day changes in investment-grade corpo-
rate bond prices are explained by changes in the level of default-free interest
rates, or, equivalently, by changes in the prices of treasury bonds. Hence, I
use the change since the end of the previous month in the price of a treasury
bond with a similar duration to estimate bid quotes within a month.
I obtain closing bid prices each day of the sample period for bonds ~or
notes! with 1, 2, 3, 5, 7, 10, and 20 years to maturity from The Wall Street
Journal. At the beginning of each calendar month a treasury bond ~or note!
is selected that is closest to each maturity. If two or more bonds have the
same maturity, I choose the bond with the price that is closest to par on the
last day of the preceding month.
Using these daily treasury bond prices, I estimate within-month quotes
for the corporate bonds using a three-step procedure. First, for each trade, I
take the bond’s previous end-of-month quote and multiply by the percentage
change in price of treasury bonds over the month to predict the month-end
quote. The percentage change in treasury bond prices is calculated as the
weighted average of the percentage price change of bonds with durations
that bracket the bond’s duration, with the weights chosen so that the weighted
average of the treasury bonds’ durations equals the duration of the corporate
3
bond. The change in the 20-year maturity bond price is used to predict the
change in prices of corporate bonds with longer durations, and the 1-year
treasury bond price change is used to predict price changes for bonds with
shorter durations. The second step is to subtract the predicted quote from
the actual end-of-month quote and divide by the number of days in the month.
This gives an average daily error from predicting that the percentage change
in the corporate bond price is the same as the percentage change in treasury
bond prices. These errors are used to make a rough adjustment for the changes
in the yield spread between corporate and treasury bonds and for the idio-
syncratic changes in the bond’s price over the month. The third step is to
take the previous end-of-month price, multiply by the percentage change in
treasury bonds up to the trade date and add on the average daily error times
the number of days from the previous month end to the trade date. This
2
Hong and Warga ~1998! find that the dealer bid quotes from Lehman Brothers that are
used in this dataset are similar to transaction prices from the NYSE’s Automated Bond System
~
ABS! but appear to be more accurate quotes.
3
Durations are obtained from the Fixed Income Database and make no adjustment for the
bonds’ call provisions. The result of using this simple duration measure rather than one that
adjusts for calls will be that bid estimates will be noisier. The estimate of the expected trans-
action cost will be unaffected as long as the noise introduced by this simple measure is uncor-
related with whether trades are buys or sells.

6
82
The Journal of Finance
provides the estimated bid quote for that trade date. Note that this meth-
odology uses the actual end-of-month quote as the estimate for that date and
that the total impact of estimation errors is negligible near the beginning
and end of the month.
As an example of this three-step technique, suppose that I want to esti-
mate the bid quote for a bond on February 10, 1996. At the end of the Jan-
uary, 1996, the bid price of the bond was priced at $1,000. Suppose that
during February, 1996, treasury bonds with the same duration increased in
price by 1 percent. In the first step I predict a value of $1,000 ϫ 1.01 or
$
1,010 for the value of the corporate bond at the end of February. If the
actual price of the corporate bond is $1,012 at the end of February and there
are 20 trading days in February, in the second step, I divide the error of $2
by 20 to get an average daily error of $0.10. If the treasury bonds with the
same duration increased by 0.25 percent from the end of January to Febru-
ary 10 and that date was the seventh trading day of the month, in the third
step, I obtain an estimated bid price of $1,000 ϫ 1.0025 ϩ 7 ϫ $0.10 ϭ
4
$
1,003.20, or, expressed as per $100 of par value, $100.32.
B. Are My Estimates of Bid Quotes Accurate?
As a check on my methodology, I examine how well treasury bond price
changes predict changes in monthly corporate bond quotes in the Fixed In-
come Database. For each bond with quotes in consecutive months, I predict
that the price change is the same as the change in a treasury bond with the
same duration. To approximate that, I take a weighted average of the changes
in the treasury bonds with durations that straddle the bond in question. The
weights are chosen so that the weighted average of the durations equals the
duration of the corporate bond. The prediction error is the difference be-
tween the actual change in the quote and the change predicted by assuming
that the percentage change in the bond price is the same as the percentage
change in the treasury bonds.
Results are described in Table I. The table breaks down price changes and
prediction errors by bond rating. The first column of that table gives the
bond rating and the second gives the number of monthly quote changes avail-
able for bonds with that rating. The next two columns report the mean ab-
solute price change and the mean absolute prediction error. A comparison of
these two columns provides a clear indication that treasury bond price changes
are useful for predicting changes in the corporate bond prices. For investment-
grade bonds, mean absolute price changes range from $0.76 to $1.27 per
$
100 par value. Typically, the mean absolute prediction error is one-third to
one-half of the mean absolute price change. Thus, most of the price change
4
Note that I use Treasury bond price changes to directly calculate the corporate bond price
changes. Dealers quote corporate bonds in terms of an additional yield over Treasuries. An
alternative methodology would be to calculate changes in Treasury bond yield and use this to
estimate changes in the corporate bond yields and use corporate bond yield changes to calculate
bond price changes.

Corporate Bond Trading Costs
683
Table I
Month-to-Month Price Changes in Corporate Bonds and
the Ability of Treasury Bond Price Changes to Explain Them
For every bond in the Fixed Income Database with bid quotes in two consecutive month ends
between January 1995 and March 1997, I calculate the absolute value of the bond’s price change
over the month. I also calculate a prediction of the price change that equals the previous month-
end price multiplied by the price change of Treasury bonds with the same duration over the
month. The prediction error is the difference between the actual quote and the predicted quote.
Errors and price changes are expressed in dollars per $100 of par value. The percentage of bond
predictions improved is the percentage of all monthly quotes that are closer to their predicted
values than the previous month’s quote. The average percentage change explained by treasury
bond changes is calculated by subtracting the mean absolute prediction error from the mean
absolute price change and dividing by the mean absolute price change for each quote and then
averaging across all quotes.
Number
of Bond
Months
Mean
Absolute
Price Change
Mean Absolute
Prediction
Error
% of Bond
Predictions
Improved
Avg. % Change
Explained by
T-bond Changes
Rating
Aaa
Aa
A
Baa
Ba
B
133,981
23,030
69,231
36,350
10,316
14,058
10,843
$0.7527
$1.1739
$1.1865
$1.2696
$1.3045
$1.5912
$1.2501
$0.3369
$0.3837
$0.3640
$0.4665
$1.0775
$1.8009
$1.0848
70.19%
78.65%
80.00%
77.94%
56.30%
39.22%
55.78%
55.24%
67.32%
69.32%
63.26%
17.40%
Ϫ13.18%
13.22%
,
B
for investment-grade bonds is captured by changes in treasury bond prices.
In contrast, very little of the month-to-month price changes of high-yield
bonds are captured by treasury bond prices.
The last two columns show the percentage of bond price changes that are
predicted more successfully with the change in treasury bonds than by the
naive prediction that bond prices will not change at all. For bonds rated Baa
and above, predictions based on treasury bond changes beat the naive pre-
diction about 75 percent of the time. On average, treasury bond price changes
explain 50 to 65 percent of the price changes for investment-grade bonds.
This is calculated by subtracting the mean absolute prediction error from
the mean absolute price change and then dividing by the mean absolute
price change. For B-rated bonds, the naive prediction that the bond price
will be the same at the end of the month as at the beginning is usually more
accurate than prediction obtained from treasury bond price changes. Hence,
by this measure, the percentage of the high-yield bond price changes ex-
plained by treasury bond changes is negative.
It is not surprising that changes in high-yield bond prices are not well
explained by changes in treasury bond prices. Practitioners claim that price
changes in investment-grade bonds are usually caused by changes in inter-
est rates whereas changes in high-yield bond prices are more often due to
changes in firm-specific factors. Therefore, I do not attempt to estimate trad-
ing costs for high-yield bonds with this methodology.

6
84
The Journal of Finance
The results in Table I suggest that using treasury bond price changes to
estimate quotes a month away works well. On average, 61 percent of the
month-to-month changes in investment grade bond quotes can be explained
by changes in treasury bond prices. Mean absolute values of the differences
between actual and predicted quotes range for $0.3369 per $100 par value
for Aaa bonds to $0.4665 per $100 par value for Baa bonds. I expect my
estimates of within-month quotes to be much more accurate for two reasons.
First, I am predicting corporate bond quotes that will occur closer in time to
the previous quote than at the end of the month. Absolute price changes
from the previous quote should be much smaller. Second, in the three-step
procedure I use to estimate trading costs, I adjust prices within the month
for shifts in the yield spread and idiosyncratic changes in individual bond
prices by assuming that the month-long difference between the bid change
and predicted bid change occurs in equal daily increments.
C. Estimates of Trading Costs Based on Trade Prices
and Estimated Bid Quotes
In total, 117,015 trades in the Capital Access database are matched with
quotes from the Fixed Income Database. Of these, 21,837 are discarded be-
cause they do not have quotes both at the beginning and end of the month.
An additional 11,122 trades with a trade date of June 30, 1995, or June 30,
1
996, are discarded because a large number of these trades appear to have
taken place on a different date. A further 6,290 trades were discarded be-
cause their trade date was uncertain. I only estimate trading costs for
investment-grade bonds, so an additional 12,659 trades of bonds rated below
Baa are omitted. I also discard 2,172 trades that take place at prices that
differ by more than 5 percent from the beginning and end-of-month quote
and are thus likely to be data errors. Note that the requirement that the
bond is included in the Fixed Income Database at the end of the previous
month eliminates almost all primary trades. I also throw out the 1,607 re-
maining trades that are flagged as primary trades by Capital Access so that
transactions cost estimates are for secondary market trading only. A total of
6
1,328 clean secondary market trades in investment-grade bonds remain.
Table II describes the distribution of ⌬ , the difference between the price
i
of trade i and the estimated bid quote, for buy and sell trades. The mean ⌬,
expressed in terms of dollars per $100 of par value, is 0.3368 for buys and
0
.0646 for sells. These results are not surprising. If institutions pay more to
buy bonds than they receive when they sell them, the mean ⌬ should be
larger for buys than sells. The median ⌬ is $0.248 for buy orders and $0.064
for sell orders. The interquartile range is from Ϫ$0.108 to $0.749 for buy
orders and from Ϫ$0.295 to $0.449 for sell orders. It is worth observing that
3
1.3 percent of buys occur at prices below the estimated bid whereas 44.4
percent of sells occur at prices below the estimated bid price. This suggests
that in practice, ⌬, the difference between the trade price and the bid price,
is measured with error.

Corporate Bond Trading Costs
685
Table II
The Distributions for Buy and Sell Trades of
D,
the Difference Between the Trade Price and Estimated Bid
The difference is expressed in dollars per $100 of par value. The sample includes all investment-
grade corporate bond trades from Capital Access International from January 1995 through
March 1997 that ~1! could be matched with quotes from the Fixed Income Database at both the
beginning and end of the month, ~2! had a trade price that did not differ from the beginning and
end of the month quote by more than $5 per $100 par value, ~3! was a secondary market trade,
and ~4! did not occur on June 30, 1995 or June 30, 1996. Trades for these dates are not included
because many of them appear to have occurred on different dates. Estimated bids are calculated
through a three-step procedure. First, for each trade, I take the bond’s previous end-of-month
quote and multiply by the percentage change in price of Treasury bonds with similar duration
over the month to predict the month-end quote. Second, I subtract the forecasted quote from the
actual end-of-month quote and divide by the number of days in the month. This gives an av-
erage daily error from predicting that the percentage change in the corporate bond price is the
same as the percentage change in Treasury bond prices. Third, I take the previous end-of-
month price, multiply by the percentage change in Treasury bonds up to the trade date and add
on the average daily error times the number of days from the previous month end to the trade
date.
Percentiles
Buy Order ⌬s
Sell Order ⌬s
1
2
0th
5th
Ϫ0.655
Ϫ0.108
0.248
Ϫ0.915
Ϫ0.295
0.064
Median
7
9
5th
0th
0.749
1.470
0.449
1.045
Mean
0.3368
0.0646
Number of trades
Number of ⌬s , 0
Number of ⌬s ϭ 0
Number of ⌬s . 0
36,060
11,274
40
25,268
11,223
19
24,746
14,026
⌬
could be measured with error if either the estimated bid price or trade
price is inaccurate. Bid prices may be imprecise either because Lehman Broth-
ers’ bids may not reflect the actual bids available in the market or because
the estimates of within-month quotes that I employ are imprecise. Trade
prices may be inaccurate as a result of Capital Access’ practice of rounding
the dollar amount of a trade up to the next $1,000.
The effects of these errors can be seen by observing that I am actually
estimating
Buy
⌬D
ϭ P Ϫ B ϩ h Ϫ y ϭ a ϩ a D
i
ϩ ei ,
~1a!
i
i
i
i
i
0
1
where P is the price of trade i, B is the bid price at the time of trade i, h
i
i
i
is the error from misestimation of P , and y is the error from misestimation
i
i
of B . The expectation of the ordinary least squares estimate of a , the round-
i
1
trip costs of trading, is

6
86
The Journal of Finance
Cov~P Ϫ B ϩ h Ϫ y , D iB uy!
i
i
i
i
E~a ! ϭ E
1
ͩ
Var ~D iB uy!
ͪ
Cov~h , D iB uy!
Var ~D iB uy!
Cov~y , D iB uy!
i
i
ϭ a ϩ E
Ϫ E
.
~2!
1
ͩ
ͪ ͩ
Var ~D iB uy!
ͪ
The estimate of the trading costs is unbiased as long as the covariance be-
tween the trade type and the error in measuring the bid price or the error in
measuring the trade price is zero. Although I am unable to provide direct
evidence on these covariances, it is reasonable to expect both to be close to
zero. Consider first the covariance between errors in my estimates of the bid
and the trade type. Errors in bid price estimates could arise from adjusting
bid prices for interest rate changes with treasury bonds with the wrong
durations or because the month-end bid prices from the Fixed Income Data-
base are wrong. But are institutions more likely to buy ~or sell! when my
estimates of the bid price are too high? There is no reason to expect that.
Now consider the covariance between errors in the trade price and trade
type. Trade price estimates are biased upwards by Capital Access’ practice of
rounding up the dollar value of trade to the next $1,000. This will affect the
estimate of the intercept term a . However, it will only affect the estimate of
0
trading costs if the error is correlated with the trade type. For a trade of
$
500,000, the maximum error from rounding is $0.02 per $100 of par value.
For a trade of $2,000,000, the maximum error is $0.005. So, if buy and sell
order are different sizes, estimate of a can be biased. Order sizes are sim-
1
ilar though. The median size of all buy trades used in the transaction cost
estimates is $1.11 million and the median size of sell orders is $1.87 million.
Inclusion of the trade size in the regression should minimize any remaining
bias.
Regression estimates of trading costs are reported in Table III. The first
row of the table reports estimates based on all 61,328 trades of investment-
grade bonds. The mean difference between the sale price and the estimated
quote ~the regression intercept! is $0.0646. One explanation for the positive
intercept is that on average, the prices dealers pay for bonds exceed the
Lehman Brothers’ bid quotes by 6.464 per $100 of par value. This is quite
possible, as there are a number of dealers and some may be willing to pay
more for a bond than the Lehman Brothers’ bid quote. An alternative expla-
nation for the positive intercept coefficient is that there is a bias in the
methodology. This could arise, for example, if actual bid prices usually ex-
ceeded my estimates. However, as long as the bias is the same for both buys
and sells, the estimate of the slope coefficient on the dummy variable for
buy orders provides an unbiased estimate of the round-trip trading costs.
The estimate for round-trip trading costs that is obtained from the regres-
sion is $0.2721 per $100 par value. Trading costs are estimated with a sur-
prising degree of accuracy. The 95 percent confidence interval for the dummy
variable for buy orders is from $0.2559 to $0.2883 and the t statistic on the

Corporate Bond Trading Costs
687
Table III
Regression Estimates of Round-trip Trading Costs
Differences between the prices of institutional bond trades and the estimated contemporaneous
bid price, ⌬ , for each trade i are regressed on an intercept and a dummy variable that takes a
i
value of one for buys and zero for sells. The coefficient on the dummy variable provides an
estimate of round-trip trading costs. Month-end quotes are obtained from the Fixed Income
Database. Within-month quotes are estimated in three steps. First, the bid quote from the
previous month end is multiplied by the percentage change in the price of Treasury bonds of
similar duration to obtain an estimate of the end of month price for the corporate bond and the
estimated price is compared with the actual end-of-month quote. Second, the difference between
estimated end-of-month bid quote and the actual end-of-month bid quote is divided by the
number of days in the month to obtain an estimate of the average daily change in the yield
curve. Finally, within-month quotes are estimated by multiplying the previous month-end quote
by the change in treasury bonds of similar duration between the quote date and the Trade date
and then adding on the average daily change in the yield curve multiplied by the number of
days from the previous month end to the trade date. The sample includes all investment-grade
corporate bond trades from Capital Access International from January 1995 through March
1
997 that ~1! could be matched with quotes from the Fixed Income Database at both the be-
ginning and end of the month, ~2! had a trade price that did not differ from the beginning and
end of the month quote by more than $5 per $100 par value, ~3! was a secondary market trade,
and ~4! did not occur on June 30, 1995 or June 30, 1996. Trades for these dates are not included
because many of them appear to have occurred on different dates.
Panel A: Estimates of Round-trip Trading Costs in Dollars per $100 Par Value
Mean Sale Price Ϫ
Lehman Quote
~dollars!
Mean Buy Price Ϫ
Mean Sale Price
~dollars!
Adj.
R
2
Bond Rating
t stat.
t stat.
N
All Baa–Aaa
Aaa
Aa
A
Baa
$0.0646
$0.1287
$0.0522
$0.0334
$0.1150
10.19
3.44
3.03
4.01
9.38
$0.2721
$0.3444
$0.2395
$0.2925
$0.2525
32.92
6.83
10.74
27.31
15.28
61,328
1,875
8,127
33,677
17,649
0.0174
0.0238
0.0139
0.0216
0.0130
Panel B: Estimates of Round-trip Trading Costs as a Percentage of the Bond Price
Mean Sale Price Ϫ
Lehman Quote
~Percent!
Mean Buy Price Ϫ
Mean Sale Price
~Percent!
Adj.
R
2
Bond Rating
t stat.
t stat.
N
All Baa–Aaa
Aaa
Aa
A
Baa
0.0606
0.1307
0.0473
0.0297
0.1100
9.60
2.97
2.78
3.58
9.16
0.2701
0.3218
0.2416
0.2927
0.2461
32.84
5.42
10.97
27.47
15.22
61,328
1,875
8,127
33,677
17,649
0.0173
0.0149
0.0145
0.0219
0.0129
buy-order dummy is 32.92. The large sample that is made possible with the
regression methodology permits precise estimates of round-trip trading costs.5
5
The trading cost estimate of $0.2721 per $100 par value that could be obtained more simply
by taking the difference between the mean ⌬ for buys and the mean ⌬ for sells as given in
Table II. However, regressions will be needed as continuous variables like trade size are added.

6
88
The Journal of Finance
The regression is re-estimated separately for bonds with different ratings.
For every rating class, the point estimate of round-trip trading costs is be-
tween $0.2395 and $0.3444 per $100 par value. It could be expected that
transaction costs would be higher for lower-rated bonds, but the transaction
cost estimates do not change in a systematic way across ratings classes. It
appears that any differences in trading costs are small.
Panel B of Table III repeats the analysis of Panel A, but now percentage
differences between trade and estimated bid prices are regressed on a dummy
for buy orders. Trading costs are estimated to be 0.270 percent for investment-
grade bonds, which is almost identical to the $0.272 per $100 of market
value estimated using differences between trade and bid prices in dollars.
Results are also similar to Panel A when different ratings classes are exam-
ined separately. It appears that sample bond prices are sufficiently close to
par value that nearly identical estimates of trading costs are obtained by
calculating percentage or dollar costs.
D. How Do These Cost Estimates Compare with Others?
Chakravarty and Sarkar ~1999! and Hong and Warga ~1998! estimate trad-
ing costs by comparing buy and sell prices for the same bond on the same
day. To evaluate my methodology, I construct a subsample of the trades in-
cluded in the regressions consisting of buy and sell trades of the same bond
on the same day. This leaves 4,387 trades. If there is more than one buy
~
~
sell! trade during a day, I take the simple average of them to estimate a buy
sell! price for the bond on that day. Thus the 4,387 trades become 1,646
matched pairs of buy and sell prices for the same bond on the same day. I
use this subsample to estimate trading costs in two ways. First, I use a
regression with estimated bid prices as in equation ~1!. Second, I use the
mean difference between buy and sell prices for the same bond on the same
day as in Chakravarty and Sarkar ~1999! and Hong and Warga ~1998!.
The regression methodology yields an estimate of $0.1781 per $100 par
value. The coefficient has a t statistic of 4.61. When I calculate trading costs
by averaging differences between buy and sell prices for the same bond on
6
the same day, my estimate is $0.2001 per $100 par value. The t statistic is
1
3.23. This suggests three conclusions. First, the trading cost estimates ob-
tained from my methodology are 2.2¢ lower—a difference that may not be
economically significant. Second, t statistics indicate that calculating a sim-
ple difference between buy and sell trade prices provides more accurate es-
timates of trading costs than the regressions when the same observations
are used for both techniques. This is not surprising, as the bid prices esti-
mates I use are noisy. Finally, we see that when we restrict our sample to
6
If the bond price did not change during the day, we would expect each individual matched
pair to provide a positive spread estimate. Instead, of the 1,646 matched pairs, 856 have pos-
itive estimated bid-ask spreads, 556 have zero estimated spreads, and 234 have negative esti-
mated spreads. Although this dictates the need for a large sample size, an unbiased average
spread estimate can still be obtained if prices are equally likely to rise or fall after each trade.

Corporate Bond Trading Costs
689
bonds with both buy and sell trades on the same date we lose about 93
percent of our observations. Thus, as a result of the much larger sample
sizes, the regression methodology can produce more precise estimates of trad-
ing costs than matching buy and sell trades even though the matching tech-
nique works better when the same sample is used. Even more important is
that only transactions in the most actively traded bonds are used when buy
and sell trades are matched. Extrapolating transactions cost estimates from
these bonds to the general population is problematic.
As a whole, my estimates are comparable to estimates in other studies. We
would expect trading costs to be lower for treasuries than for corporates
given their lower risk and greater liquidity. Elton and Green ~1998! provide
information on bid-ask spreads for treasury securities from GovPX from June
1
$
1, 1996, through June 13, 1996. They report a median quoted spread of
0.0625 and a mean quoted spread of $0.0529 per $100 of par value. Chakra-
varty and Sarkar ~1999! estimate mean round-trip trading costs of $0.2150
per $100 par value for corporate bonds from 1995 through 1997. Their esti-
mate is somewhat lower than mine, but they estimate trading costs by com-
paring buy and sell prices for the same bond on the same day. Thus, their
sample is tilted toward the bonds that trade most frequently. Hong and Warga
~
1998! also estimate trading costs for investment-grade bonds over the same
time by computing differences between buy and sell prices for the same bond
on the same day. Their estimate of the mean spread for investment-grade
bonds is $0.1328. This is lower than my regression estimate and also lower
than the $0.2001 that I obtain using Hong and Warga’s technique of match-
ing buy and sell trades. This is not surprising. They discard trades of less
than $500,000 whereas I retain them for the sample of paired buys and sells.
As the next section will show, smaller trades are more expensive. In addi-
tion, restricting estimates to bonds with buys and sells on the same day
means cost estimates reflect only the most active bonds, and thus estimates
obtained in this way should be lower than the estimates I obtain through the
regressions.
IV. The Determinants of Trading Costs
Thus far I have estimated the average costs of trading corporate bonds
across a wide variety of trades. However, we would expect characteristics of
the bonds and the trade itself to affect transactions costs. In this section I
examine the effects of three factors on trading costs. The first factor is trade
size. Some models of market making imply that large trades should cost
more. In Easley and O’Hara ~1987!, informed traders prefer to trade larger
amounts at any given price. In a resulting separating equilibrium, investors
are charged more for large trades because they are more likely to be moti-
vated by information. Grossman and Miller ~1988! demonstrate that trading
costs are likely to be higher for large trades of volatile assets if market
makers are risk averse. On the other hand, trading costs could decrease with
trade size if the fixed costs of executing a trade are significant.

6
90
The Journal of Finance
A second factor that could affect trading costs is whether or not the insti-
tution involved trades actively. Institutions who trade frequently call deal-
ers for quotes often and have a better sense of what is being bid or offered
than less active institutions. Similarly, dealers are more likely to take the
time to fax offers to buy or sell to institutions that they feel are inclined to
trade. Without a convenient central source for quotes and trade reporting,
inactive institutions are at an informational disadvantage compared to more
active traders. Thus inactive institutions may pay more to trade as a result
of the lack of transparency in the over-the-counter market. For the tests to
follow, I categorize the 20 sample institutions with the largest dollar volume
during the period as active institutions.
A third factor that may affect trading costs is whether the dealer is a large
dealer. Small dealers usually obtain bonds from another dealer through an
interdealer broker. In this case, the price that the institution pays ~receives!
would reflect the markups of both dealers and the interdealer broker. It is
also possible that the larger and more active dealers may enjoy economies of
scale that allow them to charge less for trading. I define large dealers as the
1
2 dealers with the largest dollar volume of trades during the sample period.
To examine the effects of trade size and the dealer and institutional size
on bond trading costs I run the following regression:
Buy
i
Inst
i
Deal
i
Inst
D iD eal
⌬i ϭ a ϩ a D
ϩ a S ϩ a D
ϩ a D
ϩ a D
i
0
1
2
i
3
4
5
Inst
Deal
i
ϩ a Di S ϩ a D
S ϩ ei
i
~3!
6
i
7
where ⌬ is the difference between the price of trade i and the estimated bid
i
Buy
price, Di
is one if trade i is a buy and zero if it is a sell, S is the natural
i
logarithm of the size ~in thousands of dollars! of trade i multiplied by one for
Inst
buy orders and negative one for sell orders, Di is one if trade i is a buy and
negative one if it is a sell for the 20 active institutions and it is zero if the
Deal
institution is not active, and Di
is one if trade i is a buy and negative one
if it is a sell if the dealer is one of the 12 active dealers and zero otherwise.
For the interaction term for active dealers and institutions, the product of
the dummies is positive for buy orders and negative for sell orders when
both the dealer and institution are active, and zero otherwise.
Table IV reports the estimate of equation ~3! and of regressions that in-
clude a subset of the variables in equation ~3!. In each case, the regression
is run for all investment-grade bonds together, but results are similar when
trading costs are estimated separately for bonds with different ratings ~not
shown!. The first column of Table IV shows the coefficient estimates of a
regression of the difference between the trade price and estimated bid price
on an intercept and a dummy variable that takes a value of one for buy
orders. This is the same regression estimated in Table III and is reported for
comparison with the other regressions in the table. The number in paren-
theses under the coefficient estimate is the t statistic. So, for example, the
coefficient of 0.2721 on the buy order dummy in the first regression means

Corporate Bond Trading Costs
691
Table IV
Trading Cost Determinants
The dependent variable is the difference between the trade price and the bid quote for the
bond. Independent variables include a dummy variable that takes a value of one for buy
orders and negative one for sell orders; the natural logarithm of the trade size ~in thousands
of dollars! multiplied by one for buy orders and negative one for sell orders; a dummy vari-
able that takes a value of one if the trade is a buy by a large institution, negative one if the
trade is a sell by a large institution, and zero otherwise; a variable that takes a value of one
if the trade is a buy from an active dealer, negative one if the trade is a sell to an active
dealer, and zero otherwise; an interaction term between the active dealer and active institu-
tion variables; and interaction terms between the trade size variables and the active dealer
and active institution variables. Active institutions are defined as the 20 institutions with the
largest dollar volume of trades in the sample and active dealers are the 12 dealers with the
largest dollar volume of trades. Month-end quotes are obtained from the Fixed Income Data-
base. An estimate of within-month quotes is obtained in three steps. First, the bid quote from
the previous month end is multiplied by the percentage change in the price of Treasury bonds
of similar duration to obtain an estimate of the end-of-month price for the corporate bond.
Second, the estimated price is compared with the actual end-of-month quote. The difference
between estimated end-of-month bid quote and the actual end-of-month bid quote is divided
by the number of days in the month to obtain an estimate of the average daily bond-specific
price change. Finally, within-month quotes are estimated by multiplying the previous month-
end quote by the change in Treasury bonds of similar duration between quote date and the
trade date and then adding on the average daily bond-specific price change multiplied by the
number of days from the previous month end to the trade date. t statistics are shown in
parentheses. The sample includes all investment-grade corporate bond trades from Capital
Access International from January 1995 through March 1997 that ~1! could be matched with
quotes from the Fixed Income Database at both the beginning and end of the month, ~2! had
a trade price that did not differ from the beginning- and end-of-the-month quote by more
than $5 per $100 par value, ~3! was a secondary market trade, and ~4! did not occur on June
3
0, 1995 or June 30, 1996. Trades for these dates are not included because many of them
appear to have occurred on different dates.
Variable
Regression
Intercept
0.0646
10.19!
Ϫ0.5226
~Ϫ25.09!
Ϫ0.5140
~Ϫ23.59!
Ϫ0.5240
~Ϫ24.27!
Ϫ0.7427
~Ϫ23.40!
~
~
Buy order dummy
Log of order size
Active institution dummy
Active dealer dummy
0.2721
32.92!
1.4233
~35.78!
1.4022
~33.50!
1.4210
~34.33!
1.8591
~29.82!
Ϫ0.0806
Ϫ0.0715
~Ϫ4.91!
Ϫ0.7208
~Ϫ24.30!
Ϫ0.1056
~Ϫ22.75!
~Ϫ29.58!
Ϫ0.0473
Ϫ0.1230
~Ϫ7.88!
Ϫ0.4397
~Ϫ8.15!
~
Ϫ4.91!
Ϫ0.0508
Ϫ6.04!
Ϫ0.0819
~Ϫ8.45!
Ϫ0.3581
~Ϫ8.69!
~
Dealer–institution
interaction
0.1173
~6.20!
0.0465
~2.29!
Institution–trade
size interaction
0.0464
~6.63!
Dealer–trade size
interaction
0.0417
~7.02!
2
Adjusted R
0.0173
0.0312
0.0322
0.0328
0.0343

6
92
The Journal of Finance
that the average round-trip cost of trading is $0.2721 per $100 par value.
The t statistic of 32.92 means that this coefficient estimate is significantly
different from zero at any conventional confidence level.
The second column of the table provides regression results when the nat-
ural logarithm of the order size times one for buy orders and negative one
for sell orders is included. The variable is negative and highly significant,
indicating that trade prices are lower ~higher! for large buy ~sell! orders
2
then they are for small buy ~sell! orders. The regression’s R increases from
0
.0173 to 0.0312 when the trade size variable is added.
The next two columns of the table provide estimates of regressions when
both the active institution and the active dealer dummy variables are in-
cluded with and without an interaction term. Dummy variables for active
dealers and institutions take values of one for buys, negative one for sells,
and zero if the trade is not conducted by an active institution ~dealer!. Both
variables are negative and highly significant, indicating that active institu-
tions face lower trading costs and active dealers charge less even after ad-
justing for trade size. A positive coefficient on the interaction term indicates
that it is inactive institutions that realize the largest savings from trading
with active dealers.
The last column of Table IV reports estimates of the regression when vari-
ables for interactions between the institution and trade size and the dealer
and trade size are included. Of particular interest is that the coefficients on
the active institution and active dealer variables remain negative and sig-
nificant and that the interaction between the active dealer and active insti-
tution dummies although positive, is of much smaller magnitude than these
variables. This indicates that active institutions pay less to trade bonds af-
ter adjustment for their propensity to trade with large dealers. The coeffi-
cient on the log of order size is negative. Trading costs decline with trade
size. The coefficients on the interactions between trade size and dummy
variables for active institutions and large dealers are positive but the sum is
smaller than the coefficient on trade size. This means that trading costs also
decrease with size for active institutions and trades with large dealers but
the decline is not as steep as it is for inactive institutions or trades with
small dealers.
Figure 1 plots the round-trip trading cost estimates from the last regres-
sion in Table IV for trade sizes ranging from $100,000 to $5,000,000. Two
cases are shown: active institutions trading with large dealers and inactive
institutions trading with small dealers. Trading costs for the inactive insti-
tutions trading with small dealers are larger over the entire range of trade
sizes. The cost difference narrows to almost nothing for trade sizes around
$
5,000,000, but there are few trades of this size between inactive institu-
tions and small dealers. For both groups, trading costs decrease with trade
size, but the effect is especially pronounced for the inactive institutions.
The evidence so far suggests that inactive institutions pay more to trade
than their more active counterparts. It is possible, though, that the differ-
ence in costs can be attributed to differences in the types of bonds that active

Corporate Bond Trading Costs
693
Figure 1. Estimates of round-trip trading costs for different size trades for active
institutions trading with large dealers and inactive institutions trading with small
dealers.
and inactive institutions trade, or to a misspecification of the regression
used to estimate trading costs. Thus, as an alternative, I create a matched
sample. A secondary trade by one of the 20 active institutions is matched
with a secondary trade by an inactive institution that ~1! occurred in the
same month, ~2! differed in dollar value by no more than 10 percent, ~3!
involved bonds that differed in duration by no more than 10 percent, and ~4!
involved bonds with the same Moody’s letter rating. Each trade is matched
with only one other trade. A total of 11,321 pairs or 22,642 trades remain in
the matched sample.
Table V reports the characteristics of the matched sample. The average
dollar amount of the active institutions’ trades is $3,607,000, whereas the
average trade size for the other institutions is $3,592,000. Thus, trade sizes
of both are more than twice as large as the mean trade size in the sample as
a whole. The matching process eliminates most of the smaller trades. The
mean duration of the bonds traded by the active institutions is 6.817 years
whereas the average duration for the bonds traded by the inactive institu-
tions is 6.805 years. Neither the dollar value of the trades or the durations
of the bonds traded are significantly different across institutions. In con-

6
94
The Journal of Finance
Table V
Characteristics of the Matched Sample of Trades
of Active Institutions and Other Institutions
The sample includes all investment-grade corporate bond trades from Capital Access Inter-
national from January 1995 through March 1997 that ~1! could be matched with quotes from
the Fixed Income Database at both the beginning and end of the month, ~2! had a trade price
that did not differ from the beginning- and end-of-the-month quote by more than $5 per $100
par value, ~3! was a secondary market trade, and ~4! did not occur on June 30, 1995 or June 30,
1
996. Trades for these dates are not included because many of them appear to have occurred on
different dates. A trade by one of the 20 most active institutions is matched with a trade from
another institution if ~1! both trades occurred in the same month, ~2! the dollar value of the
bonds traded did not differ by more than 10%, ~3! the duration of the of the two bonds differed
by no more than 10%, and ~4! the bonds had the same letter rating from Moody’s. No trade is
matched more than once. A total of 11,321 pairs of trades were found for the period January
1
995 through February 1997.
Panel A: Means of Variables
Standard
Mean
Deviation
Minimum
Maximum
Value of trades ~$000s!
Active institutions
Other institutions
t statistic for difference
3,607
3,592
~0.35!
3,271
3,251
10
10
30,602
28,350
Duration of bonds ~years!
Active institutions
Other institutions
t statistic for difference
6.817
6.805
~0.34!
2.681
2.665
1.433
1.406
17.523
17.523
Total value of outstanding bonds ~$000s!
Active institutions
Other institutions
t statistic for difference
258,292
266,693
~Ϫ2.41!
185,709
195,155
NA
NA
1,500,000
1,500,000
Panel B: Differences in Trade Categories
Active
Inactive
t Statistic
Institutions
Institutions
for Difference
Percent buys
54.39
68.68
0.78
7.98
56.24
35.01
57.97
56.66
0.78
7.98
56.24
35.01
~Ϫ0.88!
~18.84!
~0.00!
~0.00!
~0.00!
~0.00!
Percent large dealers
Percent rated Aaa
Percent rated Aa
Percent rated A
Percent rated Baa
trast, the mean dollar value of outstanding bonds in issues traded by the
inactive institutions is $266,693,000 whereas the mean dollar value of the
outstanding bonds in issues traded by the active institutions is $258,292,000.
This difference is statistically significant but economically trivial.

Corporate Bond Trading Costs
695
Panel B of Table V reports differences in trade categories across the active
and inactive institution’s trades. Of the active institutions’ trades, 54.39 per-
cent are buys compared to 57.97 percent of the other institutions’ trades.
This difference is not significant. On the other hand, 68.68 percent of the
active institutions’ trades are with active dealers, compared to only 56.66
percent of the other institutions’ trades. This difference is highly significant.
Finally, by design, there is no difference in the ratings of the bonds traded
by the active and inactive institutions in the matched sample.
To test whether there are differences in trading costs for active and inac-
tive institutions in the matched sample, the following regression is run:
Buy
i
Inactive
i
⌬i ϭ a ϩ a D
ϩ a D
ϩ ei ,
~4!
0
1
3
where ⌬ ϭ the difference between the price of trade i and the estimated bid
i
Buy
Inactive
price, Di ϭ one for buy orders and zero for sell orders, Di
ϭ one if the
institution is not one of the 20 most active and the trade is a buy, negative
one if the institution is not one of the 20 most active and the trade is a sell,
and zero if the institution is an active one.
Results are reported in Table VI. In Panel A, estimates of equation ~4! are
reported for all investment-grade bonds, and for Aaa, Aa, A, and Baa rated
bonds. In the regression that includes all investment-grade bonds, the coef-
ficient on the variable for inactive institutions is positive and highly signif-
icant. Less active institutions pay much more to trade than the most active
ones. When all investment-grade bonds are considered, the active institu-
tions pay round-trip costs of $0.0853 per $100 of par value. The dummy
variable for inactive institutions in the regression takes a value of one for
buys, negative one for sells, and zero for active institutions. Thus an inactive
institution pays $0.0819 more to buy than an active one and receives $0.0819
less on a sale. In total the inactive institutions pay round-trip trading costs
of $0.0853 ϩ 2 ϫ $0.0819 or $0.2491 per $100 of par value—almost three
times as much as the active institutions. When the matched sample is bro-
ken down by the bond’s rating, the regressions imply that less active insti-
tutions pay more to trade bonds in every rating class. The difference is
significant for Aa, A, and Baa rated bonds. The matched sample of Aaa bond
trades has only 176 trades.
In Panel B of Table VI, the regression is rerun but now includes the nat-
ural logarithm of the trade size, duration, the natural logarithm of the value
of the outstanding bond’s value, and a dummy variable for a large dealer. All
of these variables are multiplied by one for buy orders and negative one for
sell orders. This regression is run to insure that small differences in the
matched variables or differences in the propensity to use active dealers or in
the amount of the bond issued do not explain the differences in trading costs
paid by active and inactive institutions. As shown in Table VI, inactive in-
stitutions still pay significantly more to trade, and they pay significantly
more regardless of the bond rating.

6
96
The Journal of Finance
Table VI
Regression Estimates of the Effect of Institutional
Trading Activity on Trading Costs
The sample includes all investment-grade corporate bond trades from Capital Access Inter-
national from January 1995 through March 1997 that ~1! could be matched with quotes from
the Fixed Income Database at both the beginning and end of the month, ~2! had a trade price
that did not differ from the beginning- and end-of-the-month quote by more than $5 per $100
par value, ~3! was a secondary market trade, and ~4! did not occur on June 30, 1995 or June 30,
1
996. Trades for these dates are not included because many of them appear to have occurred on
different dates. A trade by one of the 20 active institutions is matched with a trade from an-
other institutions if ~1! both trades occurred in the same month, ~2! the dollar value of the
bonds traded did not differ by more than 10%, ~3! the duration of the of the two bonds differed
by no more than 10%, and ~4! the bonds had the same letter rating from Moody’s. A total of
11,321 pairs of trades were found for the period January 1995 through February 1997. The sign
of the trade is one for buys and negative one for sells. The active dealer dummy takes a value
of one if the dealers was one of the 12 with the largest in-sample dollar volume of trades and
zero otherwise.
Investment-
grade
Aaa
Bonds
Aa
Bonds
A
Baa
Bonds
Bonds
Panel A: The Effect of Institutional Trading Activity on Trading Costs
Intercept
0.1694
14.82!
0.1931
~1.42!
0.2154
~6.00!
0.1498
~10.46!
0.1891
~8.83!
~
Buy order dummy
0.0853
4.71!
0.1134
~0.54!
Ϫ0.0181
~Ϫ0.31!
0.0542
~2.40!
0.1593
~4.69!
~
Inactive institution ϫ
0.0819
0.0962
0.0882
0.0973
0.0586
sign of trade
~6.40!
~0.66!
~2.15!
~6.10!
~2.44!
2
Adjusted R
0.0091
22,642
0.0027
176
0.0030
1,806
0.0096
12,734
0.0107
7,926
Observations
Panel B: The Effect of Institutional Trading Activity on Trading Costs
after Adjustment for Bond Characteristics
Buy order dummy
1.2304
4.78!
2.5736
~0.99!
Ϫ1.9227
~Ϫ3.55!
1.7321
~5.26!
1.1503
~2.23!
~
Inactive institution ϫ
0.0776
0.0965
0.0789
0.0945
0.0548
sign of trade
~6.04!
~0.66!
~1.92!
~5.87!
~2.28!
Log trade size
Ϫ0.0571
Ϫ0.0675
~Ϫ0.96!
Ϫ0.0637
~Ϫ3.16!
Ϫ0.0601
~Ϫ7.99!
Ϫ0.0551
~Ϫ4.67!
~
Ϫ9.32!
0.0160
6.62!
Ϫ0.0171
Duration
0.0314
~1.34!
0.0248
~3.16!
0.0151
~5.18!
0.0166
~3.45!
~
Log outstanding bonds value
Active dealer dummy
Ϫ0.0793
~Ϫ0.80!
0.1090
~4.24!
Ϫ0.0368
~Ϫ2.88!
Ϫ0.0118
~Ϫ0.58!
~Ϫ1.70!
Ϫ0.0423
0.0902
~0.54!
Ϫ0.1177
~Ϫ2.73!
Ϫ0.0315
~Ϫ1.85!
Ϫ0.0550
~Ϫ2.14!
~
Ϫ3.11!
0.0153
22,642
2
Adjusted R
Ϫ0.0040
0.0215
1,806
0.0172
12,734
0.0153
7,926
Observations
176

Corporate Bond Trading Costs
697
The results of Table VI indicate that even when trade size, duration, and
bond rating are similar, institutions that trade less frequently pay much
more per trade than active institutions. One explanation for this is the lack
of transparency in the bond market. In a market with little transparency,
information about quotes and trades may be expensive and time consuming
to acquire. Active institutions like Prudential Capital Management, Metro-
politan Life Insurance, and TIAA-CREF acquire information as a conse-
quence of their frequent trading. Monitoring of the market by these institutions
can pay off in better prices on many trades. Other institutions in the sample,
such as the Vermont National Bank, The Polish Women’s Alliance, and the
Slavonic Benevolent Order trade infrequently. The benefits of acquiring in-
formation about quotes and trade prices are likely to be smaller because
they trade less often and thus these institutions probably spend less time
and effort on finding information about current market conditions.7
Some evidence on the importance of having information about dealer quotes
is provided by Saunders, Srinivasan, and Walter ~1998!. They study the bids
~
and offers! received by one anonymous asset manager who solicited offers to
buy or sell from bond dealers on behalf of institutional clients from January
to November 1997. Typically, these quotes are received within two minutes
of a request for a price. About 70 percent of the time, more than one bid ~or
offer! was received. On average, for investment-grade bonds, the winning
bid price was 12.0 basis points better than the second best price and 20.5
basis points better than the average price.
Lack of transparency is not the only possible explanation for the differ-
ences in trading costs paid by active and inactive institutions. Active insti-
tutions may provide liquidity to dealers on many of their trades whereas
inactive ones consume liquidity. Frequent traders may also pursue different
investment strategies than less active traders. Note though that the differ-
ences in trading frequency are generally attributable to differences in the
institution’s size. Further examination of why inactive institutions pay more
to trade bonds is needed.
V. Summary and Conclusions
In this paper, I examine corporate bond trades by insurance companies,
mutual funds, and pension funds from January 1995 through March 1997.
In total, my sample includes over $600 billion in trades. I estimate round-
trip transactions costs for secondary trades of investment-grade bonds by
regressing the difference between the trade price and the estimated bid price
7
The lower trading costs paid by institutions do not appear to be a result of relationships
that develop between institutions and dealers. I am unable to find any difference between the
costs that active institutions pay when they trade with dealers with whom they frequently
trade and the costs they pay when they trade with dealers they seldom use. These results are
available from the author on request.

6
98
The Journal of Finance
on a dummy variable that takes a value of one for buy orders and zero for
sell orders. The average round-trip transactions costs across all trades is
$
0.2721 per $100 of par value.
I find that trading costs decline with trade size, are lower when a large
bond dealer is used, and are lower for active institutions than inactive ones.
When trades are matched on the bond’s duration, the dollar value of the
trade, the bond rating, and the month in which the trade took place, inactive
institutions pay more than twice as much to trade. One explanation for this
is that obtaining information on quotes and trades in the bond market is
time consuming and expensive as a result of the lack of transparency in the
bond market. Only larger institutions that can amortize these costs over
many trades find it worthwhile to bear them.
REFERENCES
Alexander, Gordon, Amy K. Edwards, and Michael G. Ferri, 1998, Trading volume and liquidity
in Nasdaq’s high-yield bond market, Working paper, Securities and Exchange Commission.
Blume, Marshall, Felix Lim, and A. Craig MacKinlay, 1998, The declining credit quality of U.S.
corporate debt, Journal of Finance 53, 1389–1413.
Chakravarty, Sugato, and Asani Sarkar, 1999, Liquidity in U.S. fixed income markets: Com-
parative evidence from transactions in corporate, government and municipal bond markets,
Staff report, Federal Reserve Bank of New York.
Easley, David, and Maureen O’Hara, 1987, Price, trade size, and information in securities mar-
kets, Journal of Financial Economics 19, 69–90.
Elton, Edwin, and T. Clifton Green, 1998, Tax and liquidity effects in pricing government bonds,
Journal of Finance 53, 1533–1562.
Grossman, Sanford, and Merton Miller, 1988, Liquidity and market structure, Journal of Fi-
nance 43, 617–633.
Hong, Gwangheon, and Arthur Warga, 1998, An empirical study of bond market transactions,
Working paper, University of Houston.
Saunders, Anthony, Anand Srinivasan, and Ingo Walter, 1998, Price formation in the OTC cor-
porate bond markets: A field study of the inter-dealer market, Working paper, New York
University.
Warga, Arthur, 1991, Corporate bond price discrepancies in the dealer and exchange markets,
Journal of Fixed Income 1, 7–16.