Secondary Transaction Costs in Bonds

1
Secondary Transaction Costs in Bonds

• Larry Harris

2
Formal Disclaimer

• The Securities and Exchange Commission, as a
  matter of policy, disclaims responsibility for
  any private publication or statement by any of
  its employees.
• The views expressed herein are those of the
  author and do not necessarily reflect the views
  of the Commission or of the authors colleagues
  upon the staff of the Commission.

3
Secondary Bond Markets

• Corporate bonds.
• Municipal bonds.
• Government bonds.

4
Bond Market Characteristics

• Many securities.
• Infrequently traded.
• Almost no contemporaneous price transparency.
• Almost no quotes.

5
The Main Policy Issue

• How does market opacity affect liquidity?
• New car dealer comparison.
• Comparison to equity markets.

6
Important Issues

• What are secondary transaction costs in the bond
  markets?
• What determines these costs?
• How does bond complexity affect these costs?

7
The Research Program

• Examine all municipal (MSRB) and corporate
  (TRACE) bond trades.
• Measure average transaction costs for each bond.
• Identify cross-sectional determinants of these
  costs.
• Identify how costs change when bond trades become
  more transparent.

8
The Samples
9
The MRSB Sample

• Broker-dealers report all municipal bond trades
  to the MRSB.
• Price, time, size, dealer, customer side.
• Our one-year sample periodNovember 1999
  October 2000.
• These data are now available on the next day on
  the Internet.

10
The TRACE Sample

• Broker-dealers report all corporate bond trades
  to the NASD.
• Price, time, size, dealer, customer side.
• Our one-year sample periodJanuary 2003
  December 2003.

11
MRSB Sample Selection (from Section 3.1)
Deleted Unknown securities Derivatives Varia
ble rate bonds Missing data Unidentified cost
Pricing errors regressions
12
TRACE Sample Selection (from Table 1)
Same deletion criteria as applied to the MSRB
sample.
13
MSRB Bond Characteristics
14
TRACE Bond Characteristics(from Table 2)
15
MSRB Characteristics (from Table 1, Panel B)
Credit Quality
16
TRACE Characteristics (from Table 1, Panel B)
Credit Quality
17
Municipal Bond Complexity Features

• Callable
• Sinking fund
• Extraordinary call
• Nonstandard interest payment frequency
• Nonstandard interest accrual method
• Credit enhanced

18
MSRB Characteristics (from Table 1, Panel D)
Bond Complexity
19
MRSB Transparency

• During most of the sample period, bond trades
  were made public on the next day if the bond
  traded four times.
• Transparency and trade activity therefore are
  correlated.

20
Corporate Transparency

• NYSE ABS bond trades are completely transparent.
• Trades for TRACE-transparent bonds were reported
  with a 45 minute lag.
• Bonds have been made TRACE-transparent based on
  credit quality and original issue size (IOS).

21
TRACE-Transparent Bonds

• Throughout 2003 All bonds rated A and above
  with original issue sizegt1B.
• March 1, 2003 All bonds rated A and above with
  100MgtOISgtB.
• April 14, 2003 120 bonds rated BBB with
  stratified original issue sizes.

22
2003 Corporate Transparency (from Table 1, Panel
D)
23
Transaction Cost Measurement Methods
24
Benchmark Methods

• Most transaction cost measures require price
  benchmarks.
• Quotes
• Average price Warga and others
• Closing or opening prices
• Without benchmarks, we must use econometric
  methods.

25
Econometric Approaches

• Bid/ask bounce is due to transaction costs.
• Measure the bounce.
• The Roll Serial covariance spread estimator.
• Regression methods useful when we know the side
  trade initiators (customers) are on.

26
A Constructive Introduction to Our Econometric
Method
27
Price and Value

• Log Price Log Value /- trade cost
• Let Qt indicate with values 1, or -1 whether
  trade t was initiated by a customer buyer or
  seller.

28
Add Interdealer Trades

• Let It indicate with values 1 or 0 whether trade
  t was an interdealer trade.
• Set Qt to 0 for interdealer trades.
• Let dt be the unknown interdealer price impact.

29
Let Cost Vary with Size

• An average response function plus a random error.

30
Bond Transaction Returns

• Log price change between trades t and s produces
  a regression equation. (The trades need not be
  in order.)

31
Model Value Returns

• Bond value returns have drift, common, and
  idiosyncratic components.
• Random in bond-specific value.

32
The Cost Function

• Municipal bonds
• Corporate bonds

33
The Regression Model

• Combining terms gives

34
The Error Term

• has variance
• where Dts 0, 1, or 2 counts the interdealer
  trades among trades t and s.

35
Estimation Strategy

• Estimate the model without the indices for each
  bond.
• Adjust prices to remove trade costs.
• Use repeat sales methods to compute the indices.
• Involves weighted regressions.
• Re-estimate the model with the indices.

36
Weighted Least Squares

• Estimate the model with OLS for each bond.
• Use pooled constrained WLS to regress the squared
  residuals on independent variables to estimate
  the variance components.
• Re-estimate the model with WLS.
• Iterate until convergence.

37
Cost Estimates

• Estimated cost for a given size is
• The estimate error variance is

38
Mean Cost Estimates

• Compute weighted means across bonds. For
  weights, use estimates of the precision of the
  cost estimate (inverse estimator error variance).
• The data thus tell us where the information is.

39
Results
40
Mean Estimated Municipal Transaction Costs
(Figure 1)
41
Mean Estimated Corporate Transaction Costs
(Figure 1)
42
Alternative Cost Functions(Municipal Figure 2)
43
By Trading Activity (Munis)
44
By Trading Activity (Corps)
45
By Credit Quality (Munis)
46
By Credit Quality (Corps)
47
By Issue Size (Munis)
48
By Issue Size (Corps)
49
By Bond Complexity (Munis)
50
By Time Since Issuance (Munis)
51
By Time To Maturity (Munis)
52
By Transparency (Corps)
53
Cross-sectional Regressions
54
Cross-sectional Regressions

• Cross-sectional regression analyses help isolate
  effects by disentangling conflicting effects.
• Dependent variable Average bond transaction
  cost estimate for a representative trade size.
• Estimate the models with WLS.

55
Information Considerations

• The dependent variable observations are noisy
  estimates for which we have estimates of the
  estimator error variances.
• The model should have an independent, equal
  variance error term.

56
Regression Weights

• Obtain OLS residuals.
• Regress OLS squared residuals on a constant and
  on the error variances to obtain predicted
  variances.
• Use the inverse of the predicted variances as
  weights for the WLS analysis.

57
Regressors

• Inverse Price
• Fixed costs (clearing?)
• Credit Rating Index
• Complexity Features
• Age/Maturity Features
• Size/Scale Features

58
Municipal Results From Table 3, 100,000 Trade
Size
59
Inverse Price and Credit Rating Coefficients
60
A Quick Digression

• Credit is missing for 18 percent of the bonds. We
  set the credit quality index to 0 and the missing
  credit dummy to 1.
• The missing credit coefficient should equal the
  average (missing) credit quality index times the
  credit quality index coefficient.
• The implied average credit quality index is 47
  2.1 22.

61
Complexity Coefficients(in bps)
62
Age/Maturity Coefficients
63
Size/Scale Coefficients
64
Other Municipal Results (From Table 3)

• Generally similar results for other trade sizes.
• However, some evidence that institutional
  investors are less adversely affected by
  instrument complexity than retail investors.

65
Corporate ResultsFrom Table 5, 100,000 Trade
Size
66
Credit Rating Coefficients(in bps)
67
Additional Risk Coefficients
68
Maturity and Age Coefficients
69
Size Coefficients
70
Some Complexity Coefficients(in bps)
71
Transparency Coefficients(in bps)
72
Corporate Cost Determinants (From Table 5)

• Generally similar results for other trade sizes.
• Transparency has the least effect in the smallest
  and largest trade sizes.

73
Time-series Analysis of Corporate Transparency
74
Transparency Changes

• All 3,004 bonds rated A and up with
  100Mltoriginal issue sizelt1B became
  TRACE-transparent on March 1, 2003.
• A size-stratified sample of 120 intermediate
  sized BBB rated bonds became transparent on April
  14.
• What happened to costs?

75
Samples
76
Time-series Method

• For each sample, use a regression model to
  estimate a different pooled average cost response
  function for each day.
• Simultaneously estimate a common factor return
  using repeat sales index estimation method.

77
Sketch of Time-series Model
78
Difference of Differences Comparison Method

• On each day, compute difference in costs between
  the March 1 sample and the three control samples.
• Compare the average cost differences before and
  after March 1.
• Use time-series sample variances to construct
  t-statistics.

79
Results for 100K Trade Size(from Table 6)
80
More Results

• Similar results for other trade sizes.
• Similar, but smaller, results for the 120 BBB
  bonds.
• -5 and -7 bps versus two comparison samples, both
  statistically significant.

81
Learning about Transparency
82
Diffusion of Impact

• The results underestimate the long run benefits
  of transparency because many were unaware that
  prices were available.
• Obtaining last trade prices wasand is
  stilldifficult.
• These observations probably explain why the BBB
  effect is smaller.

83
A Back of the Envelope Calculation

• Cross-sectional effect at 100K trade size -3.8
  bps for TRACE-transparent and -3.5 for
  ABS-listed.
• Time-series effect -10, -11, -15 bps for versus
  various comparisons for the March 1 bonds, and -5
  and -7 for the BBB bonds.
• Safe to say minimum -5 bps.

84
A Back of the Envelope Calculation

• About 2 trillion 2003 volume in non-transparent
  corporate bonds.
• 5 bps of 2 trillion is one billion dollars.
• The estimate is not unrealistic in comparison to
  total dealing profits.

85
Conclusion
86
Summary

• Municipal and corporate bonds are expensive to
  trade.
• Retail investors, and perhaps even issuers, could
  benefit if issuers issued simpler bonds.
• Studies such as this one are essential inputs
  into the regulatory process.

87
A Final Perspective

• A corporate bond can be hedged by a portfolio of
  Treasury bonds and the issuers stock.
• Both trade in fully price-transparent markets!

88
An Important Additional Argument

• Fair valuation of bond funds will be improved by
  greater transparency.

89
Progress

• As of October 1, trades in 17,000 corporate bonds
  are available for dissemination within 30
  minutes.
• 99 percent of all corporate issues will be
  TRACE-transparent with a 15-minute lag by July
  2005.
• Starting in January 2005, all trades in municipal
  issues will available in real time with a
  15-minute lag.

90
Some Predictions

• Retail interest in bonds will surge.
• New trading systems will emerge.
• Volumes will increase.
• Dealers will continue to make moneyperhaps
  morebut it will be more difficult.

91
Time for more sunshine!