Taxes on Tax-Exempt Bonds

Andrew Ang Columbia University and NBER Vineer Bhansali PIMCO Yuhang Xing Rice University This Version: 6 March, 2007 JEL Classication: G12, G28, H20, H24 Keywords: municipal bonds, income and capital gains tax, de minimis boundary, public nance

thank Rick Green, Gur Huberman, Dan Li, Bob McDonald, and Andrew Schmidt for helpful discussions. We are especially grateful to Jeff Strnad for providing detailed comments. We thank seminar participants at Brigham Young University, Columbia University, Rutgers University, UC Irvine, and UC San Diego. We also thank Philippe Mueller for tabulating some of the data and Jihong Zang for checking some law sources. Columbia Business School, 3022 Broadway 805 Uris, New York, NY 10027. Ph: (212) 854-9154, Email: aa610@columbia.edu, WWW: http://www.columbia.edu/aa610. PIMCO, 840 Newport Center Drive, Suite 100, Newport Beach, CA 92660. Ph: (949) 720-6333, Email: Vineer.Bhansali@pimco.com Jones School of Management, Rice University, Rm 342, MS 531, 6100 Main Street, Houston, TX 77004. Ph: (713) 348-4167, Email: yxing@rice.edu.

We

Electronic copy available at: http://ssrn.com/abstract=1025105

Taxes on Tax-Exempt Bonds

Abstract Individuals must pay tax on the secondary market transactions of tax-exempt bonds. The prots involving changes in bond prices are taxed either as income or as a capital gain. We nd that municipal bonds carrying market discount, which are subject to income tax, command higher yields than municipal bonds not subject to taxes arising from secondary market trades. However, the after-tax yields on municipal bonds with market discount are around 30 basis points higher than yields on comparable municipal securities not subject to market discount taxation. We estimate an implied tax rate of around 80% using trades of municipal bonds entering regions where income tax rates apply.

Electronic copy available at: http://ssrn.com/abstract=1025105

Introduction

The coupon payments and original issue discount (OID) of municipal bonds are exempt from federal income tax and are tax-advantaged nancial instruments for individual investors in high tax brackets. However, the prots from trading municipal bonds in secondary markets are taxable. These taxes are levied on capital gains or market discount and depend not only on the purchase price of the bond, but also on the bonds issue yield, issue price, and original maturity. While most municipal bond trades are not subject to tax, there is an important subset of municipal bond transactions involving bonds subject to income tax. In some years these transactions represent over 30% of all transactions. If individuals are the marginal agents in municipal markets, municipal securities subject to tax should trade at higher yields to compensate individuals for assuming the tax liabilities attached to these bonds, compared to municipal bonds with no tax liabilities. Since 1993, accrued market discount is taxed at regular income tax rates. The Internal Revenue Code (IRC) provides a de minimis exception, which allows small amounts of market discount to be considered zero and be treated as capital gains. That is, below the de minimis boundary, accrued market discount is taxed as income. Above the de minimis boundary, bonds may be subject to capital gains tax. If bonds are trading above par or accreted OID, all bond cashows are not subject to tax. Thus, investors face a discontinuous tax treatment from these different tax boundaries. In particular, bonds trading below the de minimis boundary are the least attractive to individual investors as these bonds carry income tax liabilities, and thus these bonds should carry the highest yields to compensate investors for bearing market discount taxation. In this paper, we study the effect of these no tax, capital gains tax, and income tax boundaries on municipal bond prices. Municipal bonds are an excellent asset class to examine the effects of tax faced by individual investors. Individual investors hold over 70% of all municipal issues, which suggests that they are likely to be the agents setting prices in these markets. In contrast, for other asset markets, institutional investors are likely to dominate. For example, the de minimis rule also holds for taxable Treasury and corporate bonds, but dealers and other nancial institutions are likely to set prices in these markets. In fact, Green and degaard (1993) nd that after the 1986 tax reform, the marginal investor in Treasury bonds has a zero marginal tax rate. Thus, the municipal bond universe is a unique place to observe the effect of individual taxes on prices. Our rst contribution is to demonstrate that taxes matter in determining the cross-sectional

and time-series prices of tax-exempt bonds. We investigate how the tax law affects the relative pricing of municipal bonds with high tax liabilities. We show that these tax effects should be theoretically economically signicant by comparing the yields of municipal bonds subject to tax with the yields of comparable municipal bonds that are fully tax exempt. We also nd this effect in data, so municipal bonds subject to tax command higher yields than fully non-taxable municipal securities. However, the yields of municipal bonds with the highest tax burdens are higher than what can be explained with a present value model of bond after-tax cashows constructed using the zero-coupon municipal yield curve. Specically, investors purchasing these taxable municipal bonds in A-grade credit classes would obtain after-tax yields around 30 basis points higher than yields on comparable securities not subject to tax. These high yields on municipal bonds subject to market discount taxation persist when taking only insured bonds and are especially high, over 60 basis points, for bonds with short 1-2 year maturities. Our results are also robust to considering bonds from the same serial issue trading above or below the de minimis boundary. Our paper is related to Li (2006), who advocates that prices just below the de minimis boundary are dominated and should not be observed in theory. We especially focus on the rst trades of bonds entering below de minimis territory because they reect a changing tax status of the bond and can be used to estimate changing implied tax rates priced by investors on the same security. We show that the prices of bonds entering below de minimis territory have very high implied tax rates. As bonds cross into regions where they are subject to income tax, bond prices decline as if an income tax rate of 79% applied, with priced tax rates as high as 101% for interdealer trades. The largest after-tax yields occur for bonds trading deep below the de minimis threshold. Similarly, investors are also willing to give up large amounts of aftertax yield when bonds leave regions where they were subject to income tax. Thus, our results suggest that investors demand signicantly higher yields to hold below de minimis bonds than what taxes seem to justify. Our paper is related to a long literature that shows how taxes matter for asset prices (see Poterba, 2002, for a summary). Some of these papers address how municipal bonds are priced relative to other assets, like taxable Treasury debt, corporate securities, and equity securities (see, among many others, Auerbach and King, 1983; McDonald, 1983; and more recently Sialm, 2006). Another related literature examines how the spread in tax-exempt bond yields relative to Treasury yields are affected by changes in tax law (see, among others, Poterba 1986, 1989; Kochin and Parks, 1988; Fortune, 1996; Slemrod and Greimel, 1999; Brooks, 2002). Our 2

approach and focus is very different from these papers because we document that taxes affect the relative pricing of tax-exempt bonds themselves.1 The paper is organized as follows. Section 2 presents an overview of the tax treatment of gains in municipal bond transactions. We describe the data and the benchmark yield curve in Section 3. Section 4 contains the main results and shows that below de minimis bonds do carry higher yields than other municipal bonds, but these yields are higher than what the zero-coupon yield curve justies. Finally, Section 5 concludes.

Income and Capital Gains Taxes on Municipal Bonds

We use the term municipal bonds to describe all tax-exempt bonds, which includes bonds issued by municipalities, counties, states, other government authorities, and other entities entitled to issue tax-exempt debt. The interest paid on municipal bonds is not subject to income tax levied by the federal government, but may be subject to state income tax if an investor holds municipal bonds issued by states where the investor is not considered to be a resident.2 Specifically, IRC 103(a) exempts any interest received from municipal bonds for all taxpayers as not counting towards gross investment income that is subject to federal tax. Consistent with the principle that interest on municipal bonds is not taxed, OID is not taxable and the amortization of original issue premiums cannot be deducted. This tax treatment of initial issue premium is consistent with the tax exemption of OID, since a municipal issuer can always change the initial issue price and coupons in opposite directions to produce the same issue yield.
1

A large outstanding puzzle is that the spread of Treasury yields to municipal bond yields declines at long

maturities. This effect is still unexplained, but many common explanations like default risk can be ruled out (see Chalmers, 1998). Green (1993) proposes that the raw Treasury yield is not the right benchmark and constructs a risk-free portfolio of taxable Treasury bonds so that income is transformed and treated as capital gains. However, McDonald (2006) notes that IRC 1258 enacted in 1993 requires that coupon interest on Treasury debt in Greens transaction must be reported as income. Of course, this puzzle persists in both the pre-1993 and post-1993 periods. Our study does not directly address this issue because it focuses only on the tax effects involving the pricing of municipal bonds relative to other municipal bonds. 2 The federal tax-exemption of state and local bonds originally had a constitutional basis, as afrmed by the Supreme Court in 1895 in the case of Pollack v. Farmers Loan and Trust Company. But, in 1988, the Supreme Court overturned the constitutional basis for the tax exemption in South Carolina v. Baker, so the tax exemption of municipal bonds is not protected by the U.S. constitution but now rests with Congress. Recently, in September 2006, the Kentucky Supreme Court declined to review the decision of the Kentucky Court of Appeals that taxing out-of-state bonds, but not in-state bonds, violated the commerce clause of the U.S. constitution (Article I, Sec. 8, cl. 3). The Supreme Court is yet to rule on this issue.

However, while the coupons paid by municipal bonds are tax-exempt, an investor may pay federal income or capital gains tax when a municipal bond is purchased or sold in the secondary market, just as in the case of a sale of a taxable Treasury or corporate bond. That is, while OID is non-taxable, discounts arising from market transactions are taxable. In our analysis, we do not consider the effect of state taxes on municipal bonds, or in effect, we assume that the marginal investor in a particular states bonds is an individual who is a resident of that state (see Cole, Liu and Smith, 1994). Our focus is on the effect of federal income and capital gains taxes on municipal bonds faced by a person purchasing a tax-exempt bond in the secondary market. Investors must pay income tax on any municipal bond purchased or sold with market discount. According to current tax law (IRC 1278(a)(2)(A)), market discount is created when a par bond trades for a price less than par, or an OID bond is sold at a discount to the accrued value of the OID. Bonds with less than one year of maturity are considered to have no market discount. Prior to 1 May, 1993, market discount was treated as a capital gain. Under the Revenue Reconciliation Act of 1993, accreted market discount is now taxed as ordinary income at the time a bond is sold or redeemed (IRC 1276(a)(1)). To explain the taxation of market discount, we rst show how market discount is computed for a par bond. Then, we consider the cases of a premium and an OID bond.

2.1

Case of a Par Bond

First, consider a bond of par value $100 with an original 10-year maturity, paying semi-annual coupons of 10%. We refer to this bond as Bond A. Suppose that two years after issue, with eight years to maturity, Bond A trades at a price of $95. The market discount on this bond is 100 95 = $5. If the bond is held to maturity, the investor owes ordinary income tax on $5, which is paid when the bond matures. For the investor purchasing Bond A, only the nal cashow of the bond is affected, because the investor pays no tax on the coupons of the municipal bond. The tax code provides a de minimis exception in IRC 1278(a)(2)(C), which states: If the market discount is less than
1 4

of 1 percent of the stated redemption price of

the bond at maturity multiplied by the number of complete years to maturity (after the taxpayer acquired the bond), then the market discount shall be considered to be zero. This de minimis rule mandated by legislation imposes a discontinuity between income and 4

capital gains tax rates at the de minimis cut-off. The de minimis boundary is deterministic and is specic to each bond.3 Applied to our example, the de minimis boundary of Bond A is 100(1 0.0025 8) = $98 at time t = 2 as Bond A has eight complete years to maturity. Thus, if Bond A trades for $98.50, say, two years after issue, then this price is above the de minimis boundary and Bond A is considered to have no market discount. If the investor holds Bond A to maturity, the investor would pay capital gains tax on 100 98.50 = $1.50 when Bond A matures. Again, like the market discount case, only the nal cashow of the bond is affected. Naturally, since the top federal income tax rate is currently 35%, which is higher than the current long-term capital gains rate of 15%, once a bond crosses the de minimis boundary, it becomes subject to a more onerous tax treatment. This decreases the cashows received by the purchaser of the bond. In particular, the bond price should jump discretely downwards as the bond price falls through the de minimis boundary to maintain the same after-tax yield to the purchaser of the bond. Similarly, as a bond pushes upwards through de minimis, the nal cashow to the bond purchaser increases, and the value of the bond rises. This effect can be also be interpreted in duration terms: bonds lying near the de minimis boundary have higher durations than what a simple pricing model without tax effects would predict. Around the de minimis boundary, straight bonds have negative convexity. We refer to bonds trading below the de minimis threshold as below de minimis bonds. Suppose that individual investors are the marginal investors in municipal bond markets. Then, since below de minimis bonds are subject to income tax, these bonds should trade at lower prices and higher yields than bonds trading above the de minimis boundary to compensate individuals for bearing these tax liabilities.

2.2

Case of a Premium Bond

Municipal bond premiums as a result of secondary market transactions are not deductible under IRC 171(a)(2), but capital gains are subject to tax.4 This asymmetry in the tax law means that a premium bond has the same tax treatment as a par bond. Thus, for a par or premium bond,
3

There is no tax asymmetry for investors selling bonds with market discount to deduct a capital loss. Consider

an investor holding a par bond from issue for longer than a year. Selling at a price above par generates a taxable capital gain and selling at a price below par generates a deductible capital loss. Capital gains rates apply in both these transactions for the seller. The asymmetric tax treatment applies only to the purchaser of a bond. 4 A holder must amortize the premium on a municipal bond under IRC 1.171-1(c)(1) for reporting purposes and to determine the adjusted basis in the bond, but these are not deductible for individuals under current tax law, unlike taxable bonds (see IRS Publication 550).

there are two important tax boundaries. Above par, all the bond cashows are not subject to tax. The area below par to the de minimis boundary subjects the difference between the purchase price and par value to capital gains tax. Below de minimis bonds purchased in the secondary market have market discounts that are taxed at income tax rates. For bonds originally issued at discount, there are also two important tax regions, but since the bond is not issued at par, the bounds need to be changed to take into account the effect of the OID. We now discuss the treatment of OID bonds.

2.3

Case of an OID Bond

Computing market discount for an OID bond is more complicated than computing market discount for par or premium bonds. In the case of dening market discount on an OID bond, IRC 1278(a)(2)(B) replaces the stated redemption price of the bond at maturity by the revised issue price. Thus, for a bond originally issued at discount, market discount is dened as the difference between the purchase price of a bond and the original issue price of the bond plus accreted OID. From issue date to maturity, OID accretes according to the constant yield accrual method (IRC 1272-1(b)).5 Since OID is original interest, it is not taxable for a municipal bond, and thus the accretion of OID as the bond matures is also not taxable. Market discount is created when an OID bond trades at a price below the bonds original issue price plus accreted OID. The original issue price plus accreted OID is termed the revised issue price (IRC 1278(a)(4)) and can be computed as the present value of the remaining cashows of the bond discounted at the bonds original issue yield. As an example, consider Bond B, which is an OID bond originally issued with a 10-year maturity paying a 10% semi-annual coupon. Bond B was issued at a price of $88.5301 with a par value of 100. The semi-annual initial yield at issue of this bond is 12%. Figure 1 illustrates the accreted OID of this bond in the convex solid line. At any point in time, the revised price of the bond is the value of the remaining payments of the bond discounted at its original 12% yield. Suppose that at year 2, an investor buys Bond B at a price of $84. With eight years remaining, the revised issue price of the bond is the discounted value of 16 coupons of $5 received at 6month intervals at a yield of 6% every six months. This revised issue price is $89.8941, which is equivalent to the original issue price of $88.5301 plus $1.3640 in accreted OID. The market
5

For bonds issued prior to 27 September, 1985, straight line amortization (or the ratable accrual method) can

be used. An investor would never rationally choose straight line amortization because the constant yield method leads to a slower accrual of the market discount.

discount at t = 2 is the difference between the revised issue price and the purchase price, which is 89.8941 84.0000 = $5.8941. This is shown in Figure 1 as the solid vertical line at t = 2. If the investor holds Bond B to maturity, the market discount of $5.8941 is taxed at income tax rates when the bond matures at t = 10. An alternative way to view the calculation of market discount is as follows. According to the original OID schedule of Bond B from t = 2 to maturity at t = 10, Bond B should increase in price from $89.8941 to $100.0000. This increase of 100.0000 89.8941 = $10.1059 in the OID schedule is tax exempt. If Bond B is purchased at t = 2 for $84 and held to maturity, then a portion of the 100 84 = $16 gain is tax-free because some of this increase would have happened under the original accrual schedule. Only the gain in excess of the accreted OID is taxable. Thus, the taxable gain, which is considered income, is 100 84 10.1059 = $5.8941, and the income tax on $5.8941 is payable at maturity if the bond is held to maturity. The de minimis boundary for Bond B is still dened relative to the stated redemption price of the bond. Thus, the de minimis boundary for Bond B at t = 2 is 89.89411000.00258 = $87.8941. In Figure 1, we graph the de minimis boundary in black dots below the accreted OID solid line. Any trade above the de minimis level is considered to have no market discount. Thus, if Bond B trades at t = 2 for a price greater than $87.8941, then the gain is considered to be de minimis and there is no market discount, but the gain may be subject to capital gains tax. For example, suppose that Bond Bs price at t = 2 is $89. The investor would see a gain of 100 89 = $11 if Bond B is held to maturity, and $10.1059 of this gain is tax-free according to Bond Bs accreted OID schedule. Thus, if held to maturity, the investor would pay capital gains tax on 100 89 10.1059 = $0.8941 when Bond B matures. As a nal case, suppose that Bond B is trading above its accreted OID schedule. For example, suppose that at t = 2, Bond B trades for $91, which is greater than the revised price of Bond B of $89.8941. An investor buying Bond B and holding it to maturity would see a gain of 100 91 = $9, but none of this is taxable since under the OID accretion schedule, the investor is entitled to a tax-free gain of 100 89.8941 = $10.1059 from t = 2 to t = 10. Thus, the accreted OID acts as a bound below which the OID bond becomes subject to tax, at least at capital gains rates. In addition, if the bond price is below de minimis, the market discount is taxed at income tax rates. Thus, we can summarize the three areas with different tax treatments for an OID bond held to maturity in the following table:

Region No-Tax Region Capital Gains Tax Region

Bound P RP DM < P < RP

Tax Treatment All cashows tax-exempt No market discount (RP P ) is taxed at capital gains rates

Income Tax Region

P DM

Market discount exists (RP P ) is taxed at income tax rates

We denote the revised issue price of the bond as RP , the purchase price of the bond as P , and the de minimis boundary as DM . The par and premium bonds can be mapped into this table because for these bonds, the revised price of the bond is always par value, that is RP = par. The payment of tax occurs at maturity if the bond is held to maturity. We denote the tax-exempt region where the bond is trading above the revised price of the bond as the No-Tax Region, which occurs for a par or premium bonds trading above par value and an OID bond trading above the revised price. The second boundary is the de minimis (DM ) bound. Bonds trading between the de minimis boundary and the revised price are subject to capital gains. We refer to this as the Capital Gains Region. As bonds decrease further in price to trade below de minimis in the Income Tax Region, they become subject to income tax.

2.4

Computing Municipal Bond Prices

The Tax-Exempt Yield We follow Rule G-33 of the Municipal Securities Rulemaking Board to compute municipal bond prices.6 Since we only consider bonds with at least one year to redemption (as market discount only applies to bonds with maturities greater than one year), the price of a municipal bond on $100 par value is given by:
N

P =
n=1

100 A 100 C/2 + 100 C, n1+w N 1+w (1 + Y /2) (1 + Y /2) 360

(1)

where Y is the semi-annual yield-to-maturity of the bond; C is the semi-annual coupon rate implying a six-month coupon rate of C/2 every six months; and N is the number of remaining
6

Available at http://www.msrb.org/msrb1/rules/ruleg33.htm

coupon payments occurring at 6-month intervals. The fraction w is dened as: w= 180 A , 180

where A is the number of accrued days from the beginning of the interest payment period to the settlement date. We follow the 30/360 convention in the municipal bond market to compute A, so we count 30 days for each complete month to make 180 days in each interest rate period and 360 days in one calendar year. The yield Y is often called the tax-exempt yield because it is the yield computed on tax-exempt municipal bonds. We refer to Y as the yield. In equation (1), the fraction w accounts for the accrued interest on the bonds. For a bond valued on a date when an interest payment is received, w = 0 and the bond has no accrued interest. The rst two terms in equation (1) calculate the present value of all cashows of the bond, including accrued interest. The last term in equation (1) subtracts accrued interest. The accrued interest is added on the bond conrmation at settlement to compute the total amount due. The After-Tax, Tax-Exempt Yield The yield, Y , does not take into account the taxes that must be paid by an investor purchasing a municipal bond in the secondary market. To adjust for taxes, we dene an after-tax yield, Y , on municipal bonds, assuming that the bonds are held to maturity. Using the same notation as equation (1), we implicitly dene Y to solve:
N

P =
n=1

100 tax A 100 C/2 + 100 C, n1+w N 1+w (1 + Y /2) (1 + Y /2) 360

(2)

where tax is the appropriate tax payment payable at maturity, which is given by: 0 if P RP tax = C (RP P ) if DM < P < RP I (RP P ) if P DM ,

(3)

for RP the revised price of the bond, C the capital gains rate, and I the income tax rate. The de minimis boundary is given by DM = RP 100 0.0025 oor(N/2). The number of complete years of maturity is given by oor(N/2), where oor() rounds the number of remaining cashows downwards to the nearest integer. In equations (2) and (3), taxes reduce the nal cashow, and hence increase the after-tax yield, for bonds trading below the revised price. Taxes only affect the last cashow of the bond as the bond coupons are exempt from tax. 9

In computing the after-tax yield, we assume that the income tax rate and the capital gains rate applied at maturity are the top marginal federal tax rates in the year of the trade. For example, a trade in 2006 would use I = 0.35 and C = 0.15. These rates have changed across our sample and start at I = 0.396 and C = 0.28 in 1995.7 Municipal bonds do respond to perceptions of future, and actual, changes in tax rates (see, for example, the summary of Fortune, 1996) and agents may anticipate future changes in the tax schedule. We also compute the tax rates implied from secondary market trades using equation (2). These can be identied by bonds crossing the de minimis boundary and changing their tax rates from I to C or vice versa.

2.5

Other Issues

In our exposition, we considered only the effect of capital gains and income taxes on municipal bonds for the case where a bond is held to maturity, similar to the literature estimating implied tax rates on bond prices like Litzenberger and Rolfo (1984) and Green and degaard (1993). In Appendix A, we discuss the tax treatment of bonds sold before maturity. This issue becomes important because the timing of the tax payment occurs the sooner of when the bond matures or when the bond is sold. Even if investors believe that the expected holding period return on a municipal bond is constant, investors may have an incentive to sell early if there is a tax benet from accelerating the payment of the market discount tax, rather than postponing it to maturity. For an investor buying a par or a premium bond, there is no ex-ante incentive for that investor to sell the bond early assuming that the expected return on the bond is equal to its purchase yield (or a version of the Expectations Hypothesis [EH] holds for purchasing that municipal bond). While the EH is rejected for Treasury yields, the R2 s of predictability regressions are typically very low (see Campbell and Shiller, 1991), and it is not clear that individual investors are able to successfully market time in the municipal bond market to take advantage of these deviations from the EH.8 Early evidence by Brick and Thompson (1978) nds that lagged Treasury yields do not predict long-term municipal yields. However, even if investors expect to receive the purchase yield over the remaining maturity of the bond, there is an incentive for investors to sell an OID bond early because the accretion of OID at the issue yield is different from the accretion of discount at the purchase yield. In Appendix A, we show that this usually amounts to less than
7

These rates are available from the IRS. See, for example, http://www.irs.gov/formspubs/article/0id=

150856,00.html for the 2006 federal tax rate schedule. 8 Retail investors trying to time the market generally underperform aggregate market benchmarks in equity market trades (see, for example, Barber and ODean, 2000).

10

one basis point in normal cases. The value of a municipal bond may also be affected by other issues not easily captured in simple cashow discounting methods. Constantinides and Ingersoll (1984) and Strnad (1995), among others, demonstrate that the value of a bond should also be affected by the tax options that trading the bonds allow to time the realizations of capital gains and losses. Chalmers (2000) notes that these effects are much less prevalent in the municipal bond market because bond premium amortization cannot be deducted as an expense and municipal bonds cannot be directly shorted (because only tax-exempt authorities and institutions can pay tax-exempt interest). Finally, unlike the market for Treasury bonds, municipal bond markets are generally not very liquid. Downing and Zhang (2004), Hong and Warga (2004), Green, Hollield and Schrhoff (2005), and Harris and Piwowar (2006), among others, nd large trading costs, especially for retail customers in the municipal bond market. We remove from our sample all transactions with par amount traded below $10,000 to minimize these effects. Even after removing small trades, liquidity may still be an important determinant in pricing. To partially account for this, we treat inter-dealer transactions separately from dealer transactions with customers. We also separately look at municipal issues only for California and New York issues. Municipal bonds from these states tend to be the most liquid, as noted by Biais and Green (2005), because these states have high income tax rates and have many residents with high marginal tax rates for whom in-state municipal bonds are attractive investments. We will also match bonds trading with market discount with other bonds that have fully tax-exempt cashows on the basis of par value traded.

Data

In Section 3.1, we begin by showing that post-1992, individuals are the main holders of municipal bonds, holding over 70% of all outstanding municipal issues. Section 3.2 describes the dataset, consisting of 5,372,631 transactions from January 1995 to November 2005 on an Agrade credit quality sample of municipal bonds. To facilitate comparison of these bond prices, we construct a municipal bond zero-coupon curve each business day, which we describe in Section 3.3.

11

3.1

Who Holds Municipal Bonds?

Figure 2 shows the main investors in municipal bonds from 1952:Q1 to 2006:Q1 broken down into direct and indirect household investors, banks, and insurance companies. Indirect household investment comprises mutual funds, closed-end funds, and money market funds. The data is computed using the Flow of Funds data from the Federal Reserve by taking the various amounts held by each investor class and dividing by the total number of municipal bonds outstanding each period. The amount of municipal bonds held by foreign investors and pension funds is very small (well below 1%) and is not shown. During the 1970s and 1980s, banks were the main holders of municipal bonds because they were allowed to buy municipal securities, receiving tax-exempt income, and funded these purchases by issuing debt or borrowing for which they received a tax deduction of up to 80% of the interest expense. This changed with the Tax Reform Act of 1986, which disallowed the deduction for the carrying cost of purchasing tax-exempt municipal bonds. This caused the holdings of municipal bonds by banks in Figure 2 to decline substantially from a high of 50% in the mid-1970s to below 10% by 2006.9 The other large institutional holders of municipal securities are insurance companies. Post-1992, insurance companies hold around 12% of all municipal bonds. The proportion of municipal bonds held by insurance companies has uctuated roughly between 10-20% since the 1950s, with a large drop from above 20% in 1980 to slightly above 10% in 1986. During this period, the property and casualty insurance industry experienced large underwriting losses, which caused insurance companies to disinvest in municipal securities. Since the early 1990s, individual investors are the main holders of municipal bonds and now hold well above 70% of all municipal issues. Poterba and Samwick (2002) show that as a households marginal tax rate increases, the probability that a household holds tax-exempt bonds increases and the portfolio share of a households wealth in municipal bonds also increases. This is not surprisingly because municipal bonds are tax-advantaged vehicles for individuals with high marginal tax rates. Households can hold municipal bonds directly or indirectly. At the end of the rst quarter of 2006, individuals directly held 38.1% of all municipal securities
9

One incentive for banks continuing to hold municipal bonds is that under IRC 265(b)(3)(B), banks can

deduct 80% of the carrying cost of a bank-qualied municipal bond. The total volume of these bonds is small as a bank qualied issuer can issue no more than $10 million of tax-exempt bonds in a given year. Most of these bank-qualied issues rarely trade in the secondary market. In our sample, fewer than 6% of all transactions involve bank-qualied bonds. For a history of institutional developments in the municipal market, see Hildreth and Zorn (2005).

12

outstanding and held 33.6% through mutual funds, closed-end funds and other pass-through intermediary vehicles. Thus, at the end of the rst quarter of 2006, individuals directly and indirectly held a total of 71.8% of all outstanding municipal bonds. In summary, since 1992, individuals have held substantial amounts of municipal bonds. This suggests that tax rates of individual investors may play a role in the relative pricing of municipal securities.

3.2

Municipal Bond Transactions Data

Our data on municipal bonds is the Municipal Securities Rulemaking Board (MSRB) dataset, which contains all transactions of municipal bonds involving municipal bond dealers registered with the MSRB. These trades are made available with a one-day lag through the Bond Market Association and with a short lag of 15 minutes through data vendors such as Bloomberg and Reuters. The MSRB database lists a price, a trade date, and the par value traded of each transaction. From January 24, 1995 to August 25, 1998, only interdealer transactions are included in the data. After August 25, 1998, all transactions between dealers and customers are recorded with an indicator denoting whether the transaction is a sale or purchase. Over our sample period from January 1995 to November 2005, the MSRB database contains 58,791,805 individual transactions involving 1,974,798 unique municipal securities, which are identied through a CUSIP number. The MSRB database contains only the coupon, dated date of issue, and maturity date of each security. We obtain other issue characteristics for all the municipal bonds traded in the sample from Bloomberg. Specically, we collect information on the bond type (callable, putable, or sinkable, etc.); coupon type (oating, xed, or OID); the issue price and yield; the tax status (federal and/or state tax-exempt, or subject to the Alternative Minimum Tax (AMT)); the size of the original issue; the S&P rating; and whether the bond is insured. Our S&P rating is collected in May 2006, so the S&P rating is relevant at the time of maturity for bonds that have expired, or at May 2006 for current bonds at that time.10 In our analysis, we focus on bonds issued in the 50 states that are exempt from federal and state income taxes not subject to the Alternative Minimum Tax (AMT). We take bonds rated by S&P with a rating of A- or higher, which we refer to as the A-Grade class of municipal bonds.
10

Neither S&Ps Ratings IQuery or Moodys Ratings Delivery Service provide complete historical rating in-

formation for public nance issues. Both these companies only provide past ratings changes for bonds that they currently cover. That is, for bonds that have defaulted, matured, or are no longer covered by analysts at S&P or Moodys, past rating information on these individual bonds cannot be directly obtained from these companies. However, S&P and Moodys do publish aggregate historical information on the default experience and transitions between each credit class.

13

Over our sample period, there have been zero defaults in A-Grade municipals and the defaults of investment-grade municipal bonds have been much lower than investment-grade corporate bonds (see studies on municipal defaults by, among others, Litvack and Rizzo, 2000; Woodell, Montrone and Brady, 2004).11 We take only straight bonds with maturities one to 10 years because market discount does not apply for bonds with less than one year of maturity and there are relatively few straight bonds with maturities longer than 10 years. Bonds with very long maturities often are issued with call or sinking fund provisions. We also do not take transactions within a month of issue because Green, Hollield and Schrhoff (2006) document signicant aftermarket effects on newly issued bonds. Appendix B contains a detailed descriptions of our data lters. After merging our transactions data with the descriptive data and applying our data lters, we are left with a sample of 5,372,631 transactions on 259,471 unique securities. Thus, each bond trades 21 times, on average, over our 11 year sample. A small fraction (5.65%) of issues trade only once. In the top panel of Figure 3, we plot the total number of trades each month. The large jump in the number of trades in August 1998 is due to the inclusion of all trades between dealers and customers being added at this date. Prior to this date, only interdealer transactions are recorded. In the bottom panel of Figure 3, we plot the proportion of bond transactions each month involving bonds trading below their de minimis boundaries. The gure also overlays the Agrade 5-year zero-coupon yield. Naturally, as interest rates increase, bond prices decline and the number of transactions involving bonds with prices below de minimis increases. This is clearly seen in the large spike of de minimis transactions (over 30%) taking place in 2000. In 1998 and over 2001-2003 as interest rates decreased, the number of de minimis transactions decreases. Nevertheless, because of the large amount of transactions in our database, there are still a sizeable number of de minimis trades in these years. For example, in 2002, there are 12,803 transactions of below de minimis bonds, while there are 31,526 trades of bonds with prices below de minimis in 2003. In Table 1, we report proportions of the unique CUSIPs (Panel A) and proportions of the transactions (Panel B) falling into various categories. Panel A reports that Texas, California,
11

An alternative view is provided by Trzcinka (1982), who argues that the risk premium for municipal bonds is

on average higher than corporates of the same rating. If true, this risk premium is not observed in ex-post defaults of municipal versus corporate bonds of the same S&P rating. The famous defaults on the bonds of the Washington Public Power Supply System and Orange County, California occurred in 1990 and 1994, respectively, before our sample starts in 1995.

14

New York, and Michigan are the states of origin having the largest proportions of bonds issued. Of the 259,471 unique bonds in our sample, only a minority (24.43%) of them are par bonds, most of them are general obligation bonds (48.71%) and approximately three quarters are rated AAA by S&P and are insured.12 Panel B shows that of the 5,372,631 transactions of bonds, most of the trades involve bonds issued in California, New York, Texas, and Florida. Overall, the proportions of trades of each state correspond roughly to the proportions of bonds issued by those states. But, there are some states like New York, whose share of trades (12.81%) is signicantly larger than the number of unique CUSIPs (8.92%) of NY securities. Panel B shows that trades of discount bonds constitute 35.67% of all trades, which is higher than the proportion of bonds issued at discount (30.83% in Panel A). This is unlike corporate and U.S. Treasury issues, which are issued almost exclusively at par. The proportion of transactions from dealers to customers is 67.88% in our sample. Most of these trades (43.42% of all trades) involve sales to customers from dealers at the ask. The median par amount traded is $50,000 with very large trades constituting over $1 million. The remainder of trades (32%) are interdealer trades. The proportion of trades that are ask trades or bid trades are different for transactions above or below the de minimis boundary. This indicates that the bid-ask spread may play a role for the potential different pricing effects of bonds trading above or below de minimis. For all the transactions above de minimis, the percentage of ask, bid, and interdealer prices are 45.28%, 24.12%, and 30.60%, respectively, whereas the corresponding proportions involving below de minimis bonds are 27.5%, 37.19%, and 35.31%, respectively. Hence, to minimize effects from possible bias resulting from the bid-ask spread, we treat bid, ask, and interdealer trades separately in our analysis. In Table 2 we provide some further details on the distribution of original issue discount and premiums of our 259,471 bonds. Most bonds issued at discount are issued just slightly below par, but there are a large number of bonds issued at deep discount. For bonds issued at premiums, many bonds (15% of all CUSIPs) carry substantial premiums of at least $5 above par. The large cross-section of original issue prices is important for our analysis because the main reason why bonds decline in price is through increasing interest rates, since the credit risk in our A-grade municipal bonds is negligible. Bonds issued at different prices will decline at
12

The main insurers are the Municipal Bond Insurance Association (MBIA), AMBAC Indemnity Corporation,

Financial Security Assurance (FSA), and Financial Guaranty Insurance Company (FGIC), which are all rated AAA by S&P. These companies insure 20.49%, 16.13%, 13.03%, 12.98% of the bonds in our sample.

15

different amounts when interest rates rise. Thus, at a given time when interest rates have risen, the bonds trading below de minimis will not all have been issued at one particular point in time. This substantially reduces, but does not eliminate, the xed time effects of the dated dates of issue on our analysis.

3.3

Municipal Zero-Coupon Bond Curves

To provide a benchmark for all municipal bond trades, we construct a daily municipal zerocoupon yield using the trades of all municipal bonds in our A-grade sample. We follow the method of Nelson and Siegel (1987), so the zero-coupon yield for maturity n half-years, rn , is given by: rn = 0 + (1 + 2 ) 1 exp(n/ ) 2 exp(n/ ), n/ (4)

which is determined by the parameters = {0 , 1 , 2 , }. We estimate these parameters daily by tting the Nelson-Siegel curve to all the A-grade bonds traded each business day. In using all bonds, we bias all our results downwards because below de minimis bonds should have higher yields to compensate investors for the tax payments they must make. For each transaction price, we use the zero-coupon rate implied by equation (4) to discount the cashows of the bonds. This gives us a tted price for each bond, P , which we compute by:
N

Pm =
n=1

100 A 100 C/2 + 100 C, (1 + rn /2)n1+w (1 + rN /2)N 1+w 360

(5)

where each cashow is discounted by the zero-coupon yield. We denote bond prices computed using the tted zero yield curve as model-implied (or zero-implied) bond prices P m with corresponding model-implied yields Y m . By denition, Y m is the yield implied by valuing the bond cashows using the municipal zero-coupon yield curve on that trading day. Although strictly speaking the zero yield curve is model-free, we use the term model-implied to denote that it would represent fair value to a valuation model that would take as given, or t exactly, the estimated term structure of municipal zero-coupon bonds. Model-implied yields are the theoretical yields which would apply if the zero-coupon yield curve represents fundamental value. Each trading day, we estimate the parameters to minimize the distance between actual transaction prices and the predicted prices using the zero curve: min
i

(Pim Pi )2 , 16

(6)

where Pim is the price of bond i computed using the Nelson-Siegel zero-coupon curve in equation (5) and Pi is the transaction price of bond i. We take the summation over all bonds traded each day in our sample.13 We nd that the differences between actual yields and zero-implied yields are very small, with the average difference between transactions yields and zero-implied yields over all bonds for our sample being 2 basis points. Figure 4 plots the time-series of our estimated municipal zero-coupon curves for maturities of 1, 5, and 10 years in the top two panels and the bottom left panel. Along with our estimates, we also plot the zero-coupon curve for municipal bonds from Bloomberg, which start in August 2001. Bloombergs method of computing zero rate curves is not made public, but Bloombergs sample includes bonds with callable and sinking fund features. Bloombergs estimation sample also relies on dealer quotations in addition to using transaction prices. Our zero-coupon rates are slightly lower than the Bloomberg data, which is as expected because our sample excludes all bonds with embedded option features. Yields at the 1-year maturity hovered around 1% from 1995 to 1999, and rose to 5% in 2000 before declining to 1% in 2003. At the end of our sample in November 2005, the 1-year zero municipal yield was around 3%. At the long end of the yield curve, 10-year zero yields have been much more stable, around 4-5% over 1995 to 2005. These trends mirror the movements in Treasury yields. In the lower right panel of Figure 4, we plot the average term structure of municipal zerocoupon rates over our sample from 1 to 10 years. It is a well-known stylized fact that municipal yield curves have always been upward sloping (see Fabozzi and Feldstein, 1983). We nd that our estimated zero yield curves are upward sloping in every trading day. The average term spread between 1- and 10-year zero-coupon yields is 1.27%. To compare this number to coupon yields, we also plot the average par coupon yield curve implied from our zero yields. The corresponding par yield spread is 1.18%.

How Taxes Affect Tax-Exempt Bond Prices

In Section 4.1, we show that the theoretical effect of taxes on tax-exempt bond prices may be large. We examine this effect in data in Sections 4.2 to 4.6. In Section 4.2, we show that yields on below de minimis bonds are higher than what the zero curve predicts. Section 4.3 investigates if empirical duration is responsible for this effect and also presents detailed controls for default
13

In our estimations of the zero curve, there are some bonds that trade more than once per day. For these bonds,

there will be more than one trading price per day, but there is only one zero-curve implied model price.

17

risk and transaction size. In Section 4.5 we examine the effect of taxes on bonds crossing into, or out of, taxable regions. We estimate implied tax rates using bonds entering or leaving taxable regions in Section 4.6.

4.1

What the Effects of Taxes on Tax-Exempt Bonds Should Be

We now calibrate the effect of taxes on municipal bond prices. Taxes only affect the nal cashow of a municipal bond subject to tax, so it could be the case that since the tax payment is small relative to the nal coupon and the return of principal, the effect of taxes on municipal bonds should also be small. This will be true under certain situations, but we now demonstrate that we should expect non-negligible tax effects for an average bond using a representative numerical example. Consider a $100 face value bond paying semi-annual coupons of rate C with a maturity of N/2 years. This bond was originally issued at par. If the current municipal yield curve is at at the after-tax yield y, then the price of this bond, assuming the bond is held to maturity, is given by: P = 1 (1 + y/2)N
1 N n=1

100 C/2 100 (1 ) + , (1 + y/2)n (1 + y/2)N

(7)

which is derived by rearranging equation (2). The bond price P and the tax rate depend on each other and must be solved jointly, with 0 if P 100 = C if DM < P < 100 I if P DM and the de minimis boundary DM = 100(1 0.0025 N/2). An investor buying this bond at price P would have an IRR of y. However, the quoted yield on this bond in order to produce an IRR of y must be higher than y because of the effect of taxes. An investor buying this bond at P would be quoted a tax-exempt yield y that satises the equation:
N

P =
n=1

100 100 C/2 + . (1 + y /2)n (1 + y /2)N

For a bond where y C, the bond price is greater than 100 and there are no tax effects ( = 0) and y = y. For bonds where y > C, the tax reduces the nal bond payment and lowers the bond price, consequently raising the tax-exempt yield, y , on these bonds relative to 18

the municipal yield y. Thus, we can compute the additional yield required by bonds subject to tax, y y, for these bonds to have the same required return as a newly issued municipal security with yield y. The yield difference, y y, is the additional yield that a municipal bond subject to tax would need to bear in order for an investor buying that bond to have an IRR of y. To illustrate the effects of taxes, we choose C = 2.35%, which is the median coupon rate in the sample (see Table 1). In Figure 5, we graph the additional yield y y required by this bond to produce an IRR of y, which is the yield on fully non-taxable municipal bonds. We graph y on the x-axis and y y in basis points on the y-axis. We also conservatively assume that I = 0.35 and C = 0.15, the lowest tax rates in our sample. These are conservative choices because tax effects will be larger as the tax rates are higher in the early years of our sample. We consider the case for maturities of 2, 5, and 10 years. Naturally, as maturity shortens, the effect of taxes rises because the nal tax payment at maturity is worth more in present value terms. We vary y from zero to 6%. The effect of taxes increases with y as there is a larger tax payment on the capital gain when the purchase price of the bond decreases as y increases. Figure 5 shows that the effect of taxes should not be negligible and cannot be easily ignored. Below y = 2.35%, there are no tax effects, so y y = 0. As the yield rises above 2.35%, the price of the bond falls below par and the bond rst becomes subject to capital gains taxes. This effect is fairly small, at around 5 basis points. As y further increases, income taxes now apply and there is a discrete jump in y y. For y = 0.03, the additional yield required is already around 20 basis points, even for a 10-year maturity. We should expect to see effects of at least this magnitude in data. As yields reach 6%, the additional yields required are over 100 basis points for a 10-year bond, but this is an extreme case. In summary, if taxes are important determinants in municipal bond prices, we should expect to see non-zero differences in cross-sectional yields for bonds with low prices, relative to their taxable boundaries. We now examine just how large the tax premium is in data.

4.2

Tax Effects in the Cross Section

To characterize the effects of taxes on the cross-section of municipal bonds, we partition all trades into one of seven bins based on the transaction price P based on their tax treatment:

19

Transaction Price Bins 1 2 3 4 5 6 7 P > RP + 1.0 RP + 0.5 < P RP + 1.0 RP P RP + 0.5 DM < P < RP DM 0.5 < P DM DM 1.0 < P DM 0.5 P DM 1.0

Tax Treatment No Tax No Tax No Tax Capital Gains Tax Income Tax Income Tax Income Tax

The rst three bins lie in the no-tax region and the middle bin (DM < P < RP ) contains bonds subject to capital gains taxes at redemption. The last three bins comprise bonds which trade below the de minimis boundary and are subject to income tax. Note that the revised price and de minimis boundary are unique to each bond and change over time. We dene the boundaries of these bins based on bond prices because the IRC denes the tax regions in terms of bond prices. To understand these partitions, consider the case of a par bond, where the revised price is par value. The rst three bins contain transactions where prices are above $100 par value: prices greater than $101, prices between $100.50 and $100, and prices between $100 and $100.50. If a par bond trades below $100, it is subject to tax. In the middle bin containing transactions above de minimis to $100, an investor must pay capital gains tax. The last three bins contain transactions of below de minimis bonds, where income tax rates apply. These bins are specied in relative terms from the de minimis boundary: from 50c below de minimis to de minimis, from $1 below de minimis to 50c below de minimis, and trades more than $1 below de minimis. For an OID bond, revised price and the de minimis boundary depend on the OID accretion schedule, and the bins change boundaries appropriately. Yield Differences Across Bonds Figure 6 reports the averages of transactions yields across the bins (Yield). For each transaction on each trading day, we also compute the theoretical prices of bonds from that days zero-coupon yield curve and also the implied yield from the model price, which we refer to as Model Yields following equation (5). These model yields take as given the zero curve and do not account for tax effects. The dotted lines in each panel represent 95% condence intervals. These condence bands are very tight, on average around 4 basis points. In the top panel of Figure 6, we compute the averages using all trades. The other panels use only ask prices, bid 20

prices, and interdealer trades, respectively. Figure 6 shows that the model-implied yields are fairly at across the buckets. This is as expected because if the zero yield curve does represent fundamental value and there are no tax effects, the model-implied yields should be perfectly horizonal if each bin contains bonds of the same maturity. There is an approximately 20 basis point increase in model-implied yields for bonds trading between revised price and de minimis (DM < P < RP ) compared to the adjoining bins for the model-implied yields. This is because the average maturity of trades in this bin is slightly higher, at 5.68 years, compared to the no-tax and income tax regions with average maturities of 4.60 and 4.53 years, respectively. In sharp contrast to the fairly at model-implied yields, the actual tax-exempt yields increase dramatically as we move from the no-tax region to the income tax region. In the top panel using all transactions, the average yield of bonds where P > RP + 1.0 is 4.00%. As bonds cross into the capital gains region, DM < P < RP , yields increase to 4.23%. For deep below de minimis bonds where P DM 1.0, average yields are 4.78%, which is 78 basis points higher than the bonds with the highest prices. Taxes seem to matter for municipal bond prices! This is consistent with other papers documenting that individual investors react rationally to tax effects in other asset pricing decisions, like allocations to mutual funds or tax-deferred accounts (see, for example, Bergstresser and Poterba, 2002; Bernheim, 2002). Our results show that taxes affect even the relative prices of tax-exempt bonds. These patterns are repeated taking only transactions at the ask and at the bid in the middle two panels, which also illustrate the large bid-ask spreads in the municipal bond market. For ask transactions, the average yield line now lies lower than the model-implied yields. The ask yields start at 3.66% for the P > RP + 1.0 bin and rise to 4.25% for the last P DM 1.0 bin, so there is an increase of 59 basis points going from the rst bin to the last bin. For bid transactions, the actual yields always lie on top of the yields implied by the zero-coupon yield curve and monotonically increase from 3.91% for bonds with the highest prices to 4.69% for deep below de minimis bonds. Finally, the last panel of Figure 6 shows that for interdealer trades, the yield difference between bins 1 and 7 is 43 basis points. Benchmarking to the Zero Curve We take a closer look at the yield spreads across premium and discount bonds in Table 3, which reports the transaction yields relative to the model-implied yields. Panel A reports the average yield spread for different types of trades, where the yield spread is dened as the transaction 21

yield minus the zero-implied yield. Panel A reports the same information as Figure 6, except we report more detailed information for the yield spreads. In computing averages of the yield spreads, we take days where at least one trade takes place in all the bins. Table 3 reports the number of days used in the computation and the average number of bonds in each bin. For all trades, we compute averages across 2,101 trading days, and on a given trading day, the average number of trades in each bin ranges from 32 to 1,528. The number of trading days is smallest taking only ask trades, at 989, but in across all types of trades, there are at least 13 observations per trading day in each bin. We report standard errors in parentheses. Not surprisingly, these standard errors are very small because of the very large number of observations. Panel A shows that for all trades, the average yield spreads in the rst three no-tax bins are less than 4 basis points. There is little effect of a bond trading in the capital gains tax region, with a yield spread in the DM < P < RP bin of 4 basis points. However, once bond prices decline below de minimis, there are increasingly large spreads. Bonds trading up to $0.50 below de minimis have yields 27 basis points higher than what the zero curve predicts. Bonds with prices $1.00 or more below de minimis have yields 62 basis points higher than the model-implied yields. Panel A also reports that these patterns are robust to considering other transactions types. For example, ask transactions trade approximately 16-18 basis points below the municipal curve for bonds above the de minimis boundary, and up to 21 basis points above the curve for deep below de minimis bonds. Interdealer trades are 1-2 basis points above the municipal zero curve for above de minimis bonds and trade up to 44 basis points above the zero curve for bonds trading below de minimis. These are economically very large differences in yields. Our economic analysis suggests that the results in Panel A should not be surprising: taxes should theoretically affect the cross-sectional pricing of tax-exempt bonds and the data show that this is the case. We now turn to the question of whether the higher yields earned by municipal bond subject to tax are fair compensation for bearing that tax. If investors buying taxable municipal bonds correctly anticipate that they must pay tax and assume that the tax payments occur at maturity, then investors buying below de minimis bonds will be receiving the aftertax yield in equation (2). We dene the after-tax yield spread as the transaction after-tax yield minus the model-implied yield. We report these spreads in Panel B of Table 3. Not surprisingly, Panel B shows that the after-tax yield spreads are smaller than the raw yield spreads in Panel A. However, they are not zero, as would be the case if equation (2) perfectly 22

accounted for tax effects. Note that the after-tax yield spreads are exactly the same as the taxexempt yield spreads in Panel A for the rst three price bins in the no-tax region as by denition there are no tax effects. For all trades, bonds in the capital gains region, DM < P < RP , are priced almost at fair value. However, as bonds enter below de minimis territory, the yields of below de minimis bonds are higher than what equation (2) predicts. Specically, for bonds trading more than $1.00 below de minimis, an investor could make an after-tax prot of 28 basis points per annum relative to the zero curve, by buying these bonds, holding them to maturity, and paying the tax on market discount. Panel B shows that there is much less of a difference between deep below de minimis bonds (bin 7) and bonds with high prices (bin 1) for ask trades, with a difference of only 7 basis points. For bid and interdealer trades, there are differences of 27 and 10 basis points between bins 1 and 7. Thus, dealers are potentially earning a large proportion of the extra yield not directly explained by the zero curve in purchasing below de minimis bonds from customers. Retail customers or buy-side institutional investors would be much better off holding bonds with market discount to maturity than by selling them.

4.3

Empirical Duration, Default Risk, and Trade Size

In Table 3, the positive after-tax yield spreads indicate that investors are demanding more compensation for bearing municipal bonds subject to tax than can be explained by simple discounting of after-tax cashows using the zero curve. This taxation premium may be due to the particular characteristics of bonds such as empirical duration, default risk or trade size, which we now investigate. Empirical Duration Besides yield or price levels, municipal bond traders often consider duration risk in the management of their portfolios. The higher yields on below de minimis bonds may represent compensation for bearing bonds with duration that is not captured by the zero curve. Managing these duration exposures may be important for tracking, or benchmarking, against index positions. In Table 4, we compare empirical duration with model-implied durations. We compute the empirical duration for a bond if the previous trade of the bond occurs within the past 30 days and the price level of the last trade occurs in the same tax region as the current trade. Bins 1-3 represent the no-tax region, bin 4 is the capital gains tax region, and bins 5-7 are the income tax

23

regions. The empirical duration is computed as Dt = (Pt Plast )/Plast , Yt Ylast (8)

where Pt is the current trade price of the bond with tax-exempt yield Yt and Plast is the price of the bond within the last 30 days with tax-exempt yield Ylast . Similarly, we compute a modelimplied duration:
m Dt = m m (Ptm Plast )/Plast , m Ytm Ylast

(9)

which instead uses the bond prices and tax-exempt yields implied by the zero coupon yield curve. We report the empirical and model duration for each of the seven price bins. We are particularly interested in the duration spread D Dm , which is the empirical duration not captured by the theoretical zero curve. Table 4 reports that duration in excess of the zero curve is larger for bonds well in the notax region than for deep below de minimis bonds. For bin 1 with prices greater than $1.00 above revised price, D Dm is 2.12, whereas bonds in bin 7 trading at least $1.00 below de minimis have a duration spread of 1.62. These patterns are also observed when only interdealer trades are considered, where bonds in bins 1 and 7 have almost the same duration spreads, with D Dm being 1.86 and 1.70, respectively. Thus, the higher yields for below de minimis bonds is not due to these bonds having higher empirical duration than the zero curve implies compared to municipal bonds without market discount. Controlling for Default Risk Investors may demand an after-tax premium to hold below de minimis bonds because they fear these bonds may carry higher default risk than a typical A-grade municipal security. We rst consider only short maturity bonds with maturities between 1-2 years where cumulative default risk is smallest. Investors with shorter horizons who cannot hold bonds for long periods would also focus on these maturities. Panel A of Table 5 reports that for bonds with 1-2 year maturities, the difference in the after-tax yield spread between bonds with prices greater $1.00 above revised price (bin 1) and deep below de minimis bonds trading at prices lower than $1.00 below de minimis (bin 7) is over 72 basis points, with after-tax yield spreads of -8 basis points and 63 basis points, respectively in bins 1 and 7. This after-tax yield spread is much larger than the after-tax yield spread on all bonds of 31 basis points reported in Table 3. The high after-tax yield spreads on short maturity bonds are especially puzzling because these bonds are most likely to be held to maturity and the investments carry little cumulative default risk. 24

By construction, our sample is specically constructed to minimize default risk by taking only A-grade bonds. Nevertheless, as a second default risk control, we take only insured bonds with the highest AAA credit ratings. Panel B of Table 5 reports the after-tax yield spreads on these bonds. The after-tax yield spread difference between bins 1 and 7 is 21 basis points, indicating that insurance per se does not remove the de minimis premium. The fact that we still observe high yields on bonds with market discount among insured bonds does not rule out a default story if default is a Peso-event and the bond market is pricing in an extremely rare event. To implement a very strict default control, we employ the following strategy. Municipal bonds are usually issued in series, with many bonds of different types and maturities being issued simultaneously by the same issuer. We consider above and below de minimis bonds with different tax treatments in the same series. In Panel C of Table 5, we divide bonds from the same series into three bins: above RP subject to no tax, between RP and DM , and below DM . Table 5 shows that the yield differences on bonds with and without market discount are extremely unlikely to be due to default risk. Bonds in the same series with different tax treatments have an after-tax yield spread difference of 29 basis points between bonds above RP and bonds below DM . This is almost exactly the same as the spread of 31 basis points among all bonds reported in Table 3. Controlling for Liquidity We consider three controls for liquidity. First, we take only New York and California bonds in Panel A of Table 6. These issues are generally more liquid than other states and comprise 26.5% of all trades (see Table 1). Panel A nds that for these states, the after-tax yield spread for bin 7 is smaller, at 12 basis points, compared to 28 basis points for all states in Table 3. However, even in these markets, the high yields of below de minimis bonds persist, with a difference of 18 basis points between the after-tax yield spreads of bins 1 and 7. Second, the high spreads on municipal bonds with market discount may vary as the proportion of below de minimis trades varies. Specically, the path of municipal interest rates in the bottom plot of Figure 3 shows that 5-year interest rates rose from 4 to 5% over 1999 to 2000 and there were a large proportion of trades that involved below de minimis bonds in 2000. There are also large numbers of below de minimis trades in 1995. In Panel B, we remove trades in 1995, 1999, and 2000. Excluding this subsample, we observe slightly larger after-tax yield spreads of 37 basis points for bin 7 excluding the 1995, and 1999-2000 subsample, compared to 31 basis points in the full sample (see Table 3). A larger proportion of below de minimis trades only 25

seems to be weakly related to the high yields on below de minimis bonds. As a nal liquidity control, we match trades by transaction size. For each trade in the price bin with prices more than $1.00 below de minimis (bin 7), we match trades in bin 7 with a trade of exactly the same par amount traded among bonds trading at least $1.00 above their revised prices (bin 1). If no matching trade is found in bin 1, we do not consider that transaction. If there is more than one trade in bin 1 with the same par value traded as the trade in bin 7, we take a random trade of the same transaction size. Panel C shows that controlling for transaction size slightly reduces, but does not change our results. The difference in after-tax yield spreads of 26 basis between bins 1 and 7 is only slightly smaller than the raw 31 basis point spread, without matching for transaction size, in Table 3. In summary, controlling for liquidity slightly lowers the high after-tax yields on tax-exempt bonds subject to market discount tax, but does not eliminate them.

4.4

Characterizing the De Minimis Premium

To further characterize the de minimis premium, we perform a cross-sectional regression in Table 7. We take the spread of actual after-tax transaction yields less model-implied yields, measured in basis points, for bonds trading below de minimis. We regress the after-tax yield spread onto various independent variables. We investigate the effects of par value, maturity, the distance to the de minimis boundary, and issue price. We measure the distance to de minimis as DM P in dollars per $100 par value. We report selected coefcients on xed effects, but also include time dummies for each year and for the eight largest states in terms of transaction volume (CA, NY, FL, TX, NJ, MI, OH, and PA). To establish a base case, consider an original general obligation par bond issued in NY carrying an AAA rating which is insured. This bond sells in an interdealer below de minimis trade with a trade size of $50,000 and has a 5-year maturity at the time of trade. (Table 1 reports that $50,000 is the median trade size in our sample.) Suppose that this bond trades at $0.50 below par. The after-tax yield spread on this trade, from the estimated coefcients in Table 7 is 34 basis points in 2005. That is, an investor buying this bond would receive an extra 34 basis points in yield after tax by buying this bond compared to what the zero yield curve warrants. Table 7 reports that below de minimis bonds with larger trade sizes have lower yield spreads. If the trade size increased to $1 million, the after-tax yield spread would decline, on average, by 10 basis points. This is consistent with our relatively weak effect our liquidity controls

26

produced in Table 6. If this bond instead carried a 1-year maturity, the after-tax yield spread would widen to 42 basis points. Decreasing the bond price from $0.50 below par to $1.50 below par would increase the after-tax yield spread to 35 basis points compared to 34 basis points. Hence, controlling for everything else, it is not so much the distance to de minimis but that the bond is trading below de minimis that causes its high after-tax yield spread. From the coefcients in Table 7, the original issue price of the bond has the weakest effect, implying that there is little economic effect of the bonds original issue price on the after-tax yield spread. We also report the coefcients on selected xed effects in Table 7. While these are highly statistically signicant because of the large number of observations, the economic effects of whether the bond is a general obligation bond or revenue bond is minimal. There is also only a small effect if this bond moved from being AAA and becomes an AA bond, with the after-tax yield spread increasing from 34 to 37 basis points.

4.5

Events When Bonds Cross Taxable Regions

In this section, we track individual bonds as they cross into or out of each of the taxable regions. This event-study approach is useful because it gauges the effects of tax on the same bond, rather than considering the prices of different bonds in the cross section. We examine all events when the RP or DM boundaries are crossed over short periods. Crossing the RP Boundary In Table 8, we examine the crossover events between the no-tax region and the capital gains tax region. The boundary between these regions is the revised issue price, RP . We trace through the effects of bonds crossing down through RP and up through RP . We only consider bonds entering or leaving the capital gains tax region to the no-tax region, with no below de minimis trades counted. We dene the rst trade crossing RP as event time zero. We rst consider these crosses using all transactions in Panel A. In the rst two rows, Crossing Down and Crossing Up, we track trades whose last trade prior to the cross occurred within the last ve trading days, but not on the same trading day, as the event trade. In the last two rows, Crossing Down on the Same Day, and Crossing Up on the Same Day, we consider trades where the last trade prior to the cross and the trade crossing RP occur on the same trading day. Not surprisingly, Panel A shows that yields of bonds entering (leaving) the capital gains tax region increase (decrease). Bonds becoming subject to capital gains tax see their yields increase by 27 basis points if the last trade occurs up to 5 days prior, and 21 basis points if RP is crossed 27

on the same day as the last trade. As bonds leave the capital gains region, their yields decrease by 37 basis points compared to their last trade up to 5 days ago, and bonds which cross up through RP on the same day as the last trade decrease their yields by 25 basis points. The theoretical changes in these yields, reported in the second last column labeled Y0m for fully tax-exempt bonds are two orders of magnitude smaller than the reactions we see in data. In the column labeled Y0, , we report the changes in the after-tax yields between the last trade and the event trade. If we fully account for tax effects using the present value model in equation (2), then these changes in after-tax yields should be close to the changes of the modelimplied yields. The changes in after-tax yields are smaller than the changes in tax-exempt yields, but the differences are minimal, at well below 1 basis point. Thus, agents seem to be demanding larger yields than what the tax effects warrant and give up too much yield when the tax effects are removed. In Panels B-D, we consider ask, bid, and interdealer trades, respectively. In Panel B (C), we take trades where both the trade prior to the event and the event trade crossing RP both occur at the bid (ask), while in Panel D, both the last trade and event trade must be interdealer transactions. In all cases, we nd that the actual changes in yield in data are two orders of magnitude larger than the changes implied by the zero-coupon curve. For interdealer trades, bonds which cross up through RP when the last trade occurred up to 5 days ago decrease their yields by 20 basis points, compared to the model-implied yield change of around 1 basis point. As the change in the after-tax yields is -18 basis points, only 2 basis points of the 20 basis point decrease can be traced to tax effects using equation (2). Crossing the DM Boundary Table 9 examines bonds crossing over the de minimis boundary. These are bonds which cross from the income tax region, but as they cross they may enter either the capital gains tax region, DM < P < RP , or the no-tax region, P RP . Similarly, we track bonds trading above de minimis, which could be either fully tax-free or subject to capital gains tax, and then become subject to income tax. We dene the rst trade crossing DM as the event trade. Panel A considers all such trades, while we consider trades where the event trade and the prior trade are both bid, ask, or interdealer trades in Panels B-D, respectively. Panel A shows that bonds entering the below de minimis region increase their yields by 57 basis points when their last trade occurred up to 5 days prior. The model-implied yield change is less than 1 basis point. The change in after-tax yields is 39 basis points, so taxes should only 28

account for 18 basis points of the 57 basis point change. Investors seem to be demanding an extra 39 basis points to hold bond subject to income tax. For bonds crossing DM on the same trading day, the increase in after-tax yields is 27 basis points. Similarly, as bonds cross DM to lower tax regions, the decrease in after-tax yields is 44 basis points for bonds where the last and event trade are not on the same day. Thus, investors willingly give up 44 basis points to not subject themselves to income tax. The change in after-tax yields crossing the DM threshold are still large for ask and bid trades in Panels B and C. The smallest after-tax yield change, in absolute magnitude, occurs for ask trades for bonds crossing into the below DM region on the same day. In this case, the after-tax yield increases by 15 basis points. In Panel D, the change in after-tax yields of bonds crossing DM for interdealer trades remains over 17 basis points in absolute magnitude for all trades crossing the de minimis boundary. These are approximately the same as the around 14 basis point changes for bonds involving interdealer trades crossing the RP boundary in Table 8. In summary, bonds crossing into taxable thresholds suddenly trade at higher after-tax yields than what taxes seem to justify. Similarly, investors seem to be willing to give up after-tax yields when bonds cross in regions with lower tax treatments. We now turn to estimating exactly what tax rates are priced by individual investors when the de minimis boundary is crossed.

4.6

Implied Tax Rates

As bonds cross the de minimis boundary downward from the capital gains or no tax regions, bonds suddenly become subject to income tax. We use the rst trade below de minimis to infer the new estimated income tax rate faced by an investor, assuming the bond is held to maturity. This trade is especially important because it is the rst observation where investors are reacting to a new tax treatment. Similarly, as bonds enter the capital gains region from the below de minimis region, their tax burdens are lowered and we can estimate the implied capital gains rate from the rst trade in the capital gains region. In particular, we take all the events where the last trade before the cross happens within ve days, including last trades on the event day. These implied tax rates shed light on how much yield investors are giving up or demanding as tax liabilities lessen or increase. The Implied Income Tax Rate To estimate the implied income tax rate, we consider events where bond prices go down through the de minimis boundary. The rst trade in the below de minimis region has a price of P . The 29

market discount on this is taxed at regular income tax rates if the bond is held to maturity. If P m denotes the model-implied price if the bond were not subject to tax, the difference between P and P m is the present value of the tax liability: (RP P ) I P = Pm (10) (1 + rN /2)N 1+w where I is the estimated income tax rate. This is a very simple expression because all municipal bond coupons are tax exempt. To estimate I , we put a rational expectations error on the RHS of equation (10). This is just an OLS regression to which we add xed year effects and other controls. Panel A of Table 10 reports the estimated implied income tax rates. Using all prices, the implied income tax rates that will equate the actual price and the price implied by the zero curve is 79%. We also consider the effect of only ask (bid) trades, where the last trade within the previous ve days is also an ask (bid). If we take only these ask trades or bid trades, the estimated income tax rates are 73% and 120%, respectively. The implied tax rate from considering only interdealer trades, with the prior trade also being an interdealer trade, is even larger than using all trades, at 101%. These income tax rates are signicantly higher than the highest marginal income tax rate in our sample, at 39.6%. Certainly, the initial jump downwards for bonds rst trading in the below de minimis region are larger than what any reasonable tax rates warrant. We also estimate implied income tax rates in each year. These are always much higher than the statutory income tax rates (which are a maximum of 39.6% in our sample), except during 2002. While the rst trades of bonds entering the below de minimis area are generally at a discount (under-priced) compared to the zero curve, this was not the case in 2002. We estimate an implied income tax rate of 7.4% in 2002, with a standard error of 13.4%. This coincides with a drop in the statutory tax rates from 38.6% in 2002 to 35% in 2003. During the early 2000s, the Economic Growth and Tax Relief Reconciliation Act of 2001 progressively reduced income tax rates from 39.6% in 2000, but The Jobs and Growth Tax Relief Reconciliation Act of 2003 accelerated these and reduced the income tax rate to its present level of 0.35 from 2003 onwards. Our estimated income tax rate shoots up to 163% in 2003, with a standard error of 8.6%. The Implied Capital Gains Tax Rate To estimate implied capital gains tax rates, we take all events where bonds cross upwards across the de minimis boundary and fall into the capital gains tax region (from RP to DM ). Bonds 30

making this transition go from being subject to income tax to being subject only to capital gains tax. To estimate the implied capital gains tax rate, we use the equation: P = Pm (RP P ) C , (1 + rN /2)N 1+w (11)

where the difference between the actual price P and the model-implied price P m is the present value of the tax liability payable at maturity, which is taxed as a capital gain. We estimate C in equation (11) by using an OLS regression with year and other dummies. Panel B of Table 10 shows that for all trades, the estimated capital gains rate is slightly below zero, at -4%. For ask and bid trades, the implied capital gains rates are 1% and 20%, respectively. The implied capital gains rate is negative taking only interdealer trades, at -9%. The implied negative tax rates indicate that prices are moving upward from the below de minimis region into the capital gains region as if the federal government is paying investors to hold these bonds. Alternatively, investors are paying a premium to enjoy relief from the previous income tax regime.

Conclusion

Municipal bond markets are a unique place to study the effects of individual tax rates because individuals are likely to be the marginal pricers in this market. We nd that taxation plays an important role in determining municipal bond prices. Although coupon payments and original issue discount of municipal bonds are exempt from federal income tax, the prots of trading municipal bonds in secondary markets are taxed at capital gains or income tax rates. The number of municipal bonds subject to tax is not small; in some years the proportion of municipal bonds transactions subject to income tax on market discount is above 30%. Furthermore, the tax code mandates discontinuities in the tax treatments of bonds. As the price of a bond originally issued at par falls below par, it rst becomes subject to capital gains tax rates. As it falls further in price, it becomes subject to income tax. The boundary between the capital gains and income tax regions is called the de minimis boundary. We show that the additional yield on municipal bonds carrying income tax liabilities that is required for these bonds to have the same IRR on an after-tax basis as a fully non-taxable municipal bond is theoretically not negligible. Consistent with this, we nd that in data, municipal bonds with income tax liabilities carry higher yields than municipal bonds not subject to market discount taxation. Bonds trading deep below the de minimis threshold have the highest tax lia31

bilities and these bonds have the highest yields. These effects are both highly statistically and economically signicant. Thus, taxes matter for the relative pricing of tax-exempt securities. However, we nd that the yields on municipal bonds subject to tax are higher than what can be explained by valuing the after-tax cashows of the bond using the zero-coupon municipal yield curve. In particular, municipal bonds in A-grade credit classes bearing the highest tax burdens have after-tax yields approximately 30 basis points higher than the tax-exempt yield curve. The high after-tax yields on municipal bonds with market discount persist when we control for default risk and liquidity. The income tax rate implied by transaction prices of bonds rst entering the income tax region is around 80%. These results suggest that taxes play an important role in pricing municipal securities, but the effect of taxes is larger than what standard discounting models can capture.

32

Appendix A Taxation of Market Discount for Bonds Sold Prior to Maturity

The general principle involved in the taxation of market discount for bonds not held to maturity is that there is a second accretion schedule that applies to the purchase price of the bond, which is computed using the yield at the time of purchase. Any resulting capital gain or loss is computed relative to this second accrual of discount, sometimes termed total discount. It is best to illustrate this concept with a few examples. We rst consider the case of original issue premium and par bonds, followed by OID bonds.

A.1

Premium and Par Bonds

Tax Treatment Consider a bond with an original 10-year maturity with a 10% coupon. We refer to this bond as Bond A. Suppose that two years after issue at t = 2, Bond A trades at a price of $95. The market discount on this bond at the time of purchase is 10095 = $5. Suppose that at t = 8, the bond is sold for $99. Thus, from t = 2 to t = 8, the investor has made a gain of 99 95 = $4. A portion of this $4 gain is taxed as ordinary income and a portion is taxed as a capital gain. When Bond A is purchased at t = 2 for $95, the investor purchasing Bond A does so at a yield of 10.9543%. As Bond A approaches maturity, the price of Bond A would deterministically approach 100 under the constant yield method if the yield was constant at 10.9543%. This accretion of discount is taxed as income under IRC 1276. Anything in excess of this accretion is dened as a capital gain. At t = 8, the revised price of Bond A, computed as the present value of the remaining four 5% coupons at a yield of 10.9543%/2 = 5.4772% every six months is $98.3266. The accretion of market discount of 98.3266 95 = $3.3266 is counted as income. Thus, 4 3.3266 = $0.6734 is taxed as income. We illustrate this in the top panel of Figure A-1. The accretion of market discount at the purchase yield of 10.9543% is shown in the dashed red line. This accretion is taxed as income. At t = 8, the difference between the sale price of $99, denoted as the red square, and the accreted purchase price of Bond A of $98.3266 is $0.6734 and it is taxed as a capital gain. The accretion of market discount affects the tax basis of the bond. Any sale of Bond A below the accreted market discount can be deducted as a loss. This also applies to purchases 33

at a premium. While the amortization of municipal bond premiums as a result of secondary market transactions cannot be deducted, the amortization of the premium changes the tax basis of the bond. Figure A-1 also shows the de minimis boundary. Any purchase above the de minimis threshold results in no market discount. At t = 2, Bond As de minimis boundary is 100(1 0.0025 8) = $98. Suppose that at t = 2, Bond A is purchased for $98.50 and sold at t = 8 for $99. Then, the 99 98.50 = $0.50 prot is taxed entirely as a capital gain. Ex-Ante Incentive to Sell Later If the yield on the purchased bond remains constant at y, then a par or premium bond purchased in the secondary market would never be sold early. In this environment, a par or premium bonds value would rise in accordance with the revised price schedule, which changes according to Pn = Pn1 (1 + y/2) 100 C/2, in each subsequent 6-month period, with P1 being the clean purchase price. For a par or premium bond, y > C and the revised purchase price accelerates as the bond approaches maturity. Taxes are not paid each period on Pn Pn1 , which would be the timing appropriate if taxes are levied when the economic income is earned. Instead, taxes are levied on Pn P1 only at the time of sale. Not surprisingly, it is straightforward to show that pushing out the payment of the tax increases the after-tax value of the cashows valued at y, and thus the after-tax yield increases as the sale is deferred. The value to waiting is economically very small. For example, suppose that the income tax rate is 35%. If a par bond with an initial coupon of C = 1% and a remaining 30-year maturity is purchased at a tax-exempt yield of 20%, then the after-tax yield if the bond is sold after 6 months is 19.9771%, compared to 19.9773% if the bond is held to maturity, assuming in both cases that the yield curve remains at at y. Higher tax rates, shorter maturities, or lower coupon rates (initial yields) will cause these differences to be even smaller. There is one counter-argument for an investor to sell a par or premium bond early. In upward-sloping yield curve environments, which has always been the case in the municipal market, Bankman and Klein (1989) show that the constant yield accrual method understates interest income in the early years and there is too much accrual in later years. If yields are held constant, selling early would result in a lower tax payment. However, Sims (1992) numerically shows that such effects are very small, but they would off-set the incentive for investors to not sell early to defer tax. 34

A.2

OID Bonds

Tax Treatment The taxation of market discount for OID bonds depends on the OID accrual schedule relative to the accrual schedule for the market discount. This is best illustrated through an example. Consider Bond B, which is an OID bond originally issued with a 10-year maturity paying a 10% coupon. Bond B was issued at a price of $88.5301 with a par value of 100 with an initial yield of 12%. Suppose that at time t = 2, Bond B is purchased for $84, corresponding to a yield of 13.3105%, and at t = 8, Bond B is sold for $99. The computation of taxable market discount is complicated by the fact that the accreted OID is not taxable. The bottom plot of Figure A-1 illustrates this case. The solid blue line plots the accretion of OID at the original issue yield of 12%. These are the revised prices of Bond B according to the OID accrual schedule. At t = 2, the purchase price is denoted by a red diamond, and this purchase price is accreted at the purchase yield of 13.3105% in the dashed red line. In Figure A-1, the vertical line at t = 2 is the difference between the purchase price of $84 and the revised price of $89.8941. At t = 8, Bond B is sold for $99, which we denote on Figure A-1 by the red square. Normally, the accretion of the purchase price from t = 2 to t = 8 along the dashed red curve would be subject to income tax. However, some of this accretion is exempt from tax because accretion would have happened under the OID accrual schedule. It is only the accretion of the market discount in excess of the OID accretion that is taxable. From t = 2 to t = 8, the revised price of Bond B under the OID schedule increases from $89.8941 to $96.5349. The OID accrual schedule can be computed by valuing the remaining payments of Bond B at the original issue yield of 12%. This difference in OID of 89.8941 96.5349 = $6.6408 is not taxable. When Bond B is purchased at t = 2, we apply the constant accrual yield method to compute the accrual of total discount at the purchase yield of 13.3105%. At t = 8, we can compute this accreted discount by valuing the remaining four payments of $5 every six months plus the nal par value of $100 at a yield of 6.6553% every six months, which totals $94.3495. According to the total discount schedule, there is an increase of 94.349584 = $10.3495. However, according to the OID schedule, we would originally have seen an increase in the bond price of $6.6408. The taxable market discount is the difference in the accrued OID and the total discount accrual, so only 10.3495 6.6408 = $3.7087 is subject to income tax. In Figure A-1, the difference between the two vertical lines represents the accrued market 35

discount at the time of sale. The vertical line at t = 2 is the market discount at purchase, which is 89.8941 84 = $5.8941. The vertical line at t = 8 is the market discount at sale, which is 96.5349 94.3495 = $2.1854. The difference between these two vertical lines is the accrued market discount, which is 5.8941 2.1854 = $3.7087. According to IRC 1276(a)(1), only this portion of the total gain is treated as ordinary income. To summarize, the total gain involved in purchasing Bond B at t = 2 for $84 and selling Bond B at t = 8 for $99 is $15. The OID accrual from t = 2 to t = 8 of $6.6408 is not taxable. Thus, the total taxable income is 99 84 6.6408 = $8.3592 at the time of sale. At t = 8, the investor owes income tax on $3.7087 and the remainder of $4.6505 is taxed as a capital gain. For a second case, suppose that at t = 2, Bond B is trading for $89, which is above the de minimis boundary of Bond B at t = 2, which is DM = 89.89411000.00258 = $87.8941. Bond B is sold at t = 8 for $99. Bond B is not subject to market discount as the purchase price is above the de minimis threshold, but below the revised price of $89.8941 at t = 2. Thus, no income tax is payable but some of the 99 89 = $10 gain is taxable as a capital gain. According to the OID schedule, $6.6408 is not taxable, so the total taxable income is 99 89 6.6408 = $3.3592, which is taxed at the capital gains rate. As a nal case, suppose that at t = 2, Bond B trades above its accreted OID schedule for $91. Again suppose that Bond B is sold at t = 8 for $99. In this case, no income tax on the gain is paid and since some of the OID accretion is tax-free, not all of the 99 91 = $8 is taxed. From t = 2 to t = 8, the change in the OID schedule is $6.6408, so the total taxable income is 99 91 6.6408 = $1.3592 on which capital gains tax is owed. Ex-Ante Incentive to Sell Early With OID bonds, there is a difference between the OID accrual and the purchase price accrual. Because the purchase yield is larger than the initial yield, the accrual of the total discount on the purchase price is faster than the accrual of OID. Since OID is tax-exempt, there is an incentive to bring forward the sale of the bond. This does not occur for a par or premium bond because the revised price is always xed at par value. For realistic cases, the additional advantage of selling early is worth less than 1 basis point. Consider an OID bond with original coupon rate 2.35% (the median coupon in Table 1) with an original issue yield of 5% and a remaining maturity of 10 years. We assume the tax-exempt yield curve is at at 6% and remains constant. If the income tax rate is 0.396, then the after-tax yield assuming a sale in six months is 5.6971% compared to 5.6954% if the investor holds the 36

bond to the full 10-year maturity.

Data Filters

From January 1995 to November 2005, the original MSRB database contains 58,791,805 individual transactions on 1,974,798 unique municipal bonds. Our nal sample consists of 5,372,631 trades on 259,471 unique securities. 1. Tax Status We consider only bonds that are exempt from federal and state income taxes. Some taxexempt municipal bonds issued by state and local governments to nance capital projects are classied as private activity bonds and are subject to the AMT. These bonds comprise approximately 3.4% of all CUSIPS and we exclude these bonds from our analysis. We also limit our bond universe to bonds issued in one of the 50 states, and so we exclude bonds issued in Puerto Rico, the Virgin Islands, other territories of the U.S. such as American Samoa, the Canal Zone, and Guam. Bonds issued in these territories constitute less than 0.4% of all bonds. 2. High Credit Ratings To focus only on the tax implications of municipal bond trades, we focus on bonds of the highest credit classes. We take only bonds rated by S&P in the AAA, AA+, AA, AA-, A+, A, and A- categories. Many A Grade bonds obtain their credit rating because they are insured by a AAA-rated insurer. Slightly over 60% of all bonds are insured in the MSRB sample.14 3. Straight Bonds We further limit our sample to include only bonds paying xed coupon rates (94.1% of all bonds in the MSRB sample). We also take only straight bonds with no embedded option features, so all our bonds are xed maturity paying xed semi-annual coupons. Straight municipal bonds constitute 50.4% of all the bond universe and they generally have shorter maturities than bonds with embedded options. The average maturity at issue of straight bonds is 6.31 years while the average maturity at issue of option-embedded bonds is 15.76
Nanda and Singh (2004) provide a tax-based rationale of why insurance on municipal bonds is valueenhancing for both insurers and municipal bond investors.
14

37

years. The exclusion of option-embedded bonds is to facilitate our computation of yieldto-maturity and market discounts. Including bonds with callable or sinking bond features would entail numerically intensive option-adjusted spread computations involving binomial trees to correctly price the embedded options. After these rst three requirements, we have transactions on 9,950,226 transactions on 464,678 unique municipal bonds. 4. Avoiding Newly Issued Bonds Green, Hollield and Schrhoff (2006) document signicant underpricing in new municipal bond issues and interesting patterns in the aftermarket trading of these bonds between informed and uninformed customers. To avoid the effect of newly issued bonds, we exclude all the transactions that happened within 30 days of issuance. Transaction of newly issued bonds constitute about 31% of the 9.9 million transactions, reecting the fact that municipal bonds transactions are concentrated during the period right after issuance. However, almost all of these transactions are not trades near de minimis because there is little movement in the yield curve over a 30 day period. We obtain nearly identical results when these trades are included in our sample. 5. Maturities Between One and Ten Years Transactions involving straight bonds with maturities longer than 10 years are scarce because most bonds with long maturities are issued with callable or sinking fund provisions. We use only transactions with maturity shorter than 10 years in our analysis. We also take bonds only with maturities greater than one year because long-term capital gains rates apply only to securities held longer than one year and there is no market discount for bonds with a maturity less than one year. 6. Removing Very Small Trades and Outliers To avoid the effect of extremely small trades, we exclude all transactions with par amounts traded less than $10,000. Finally, we take only transactions with prices between $80 and $130, and bonds with coupon rates from 1% to 20%.

38

References
[1] Auerbach, A. J., and M. A. King, 1983, Taxation, Portfolio Choice, and Debt-Equity Ratios: A General Equilibrium Model, Quarterly Journal of Economics, 98, 587-609. [2] Bankman, J., and W. A. Klein, 1989, Accurate Taxation of Long-Term Debt: Taking into Account the Term Structure of Interest, NYU Tax Law Review, 44, 335-348. [3] Barber, B., and T. ODean, 2000, Trading is Hazardous to Your Wealth: The Common Stock Investment Performance of Individual Investors, Journal of Finance, 55, 773-806. [4] Bergstresser, D., and J. Poterba, 2002, Do After-Tax Returns Affect Mutual Fund Inows? Journal of Financial Economics, 63, 381-414. [5] Bernheim, B. D., 2002, Taxation and Saving, in in A. J. Auerbach and M. Feldstein, eds., Handbook of Public Economics Volume 3, Elsevier Science B.V., 1173-1249. [6] Biais, B., and R. C. Green, 2005, The Microstructure of the Bond Market in the 20th Century, working paper, Carnegie Mellon University. [7] Brick, J. R., and H. E. Thompson, 1978, Time Series Analysis of Interest Rates: Some Additional Evidence, Journal of Finance, 33, 93-103. [8] Brooks, R., 2002, The Cost of Tax Policy Uncertainty: Evidence from the Municipal Swap Market, Journal of Fixed Income, Dec, 71-87. [9] Campbell, J. Y., and R. J. Shiller, 1991, Yield Spreads and Interest Rate Movements: A Birds Eye View, Review of Economic Studies, 58, 495-514. [10] Chalmers, J. M. R., 1998, Default Risk Cannot Explain the Muni Puzzle: Evidence from Municipal Bonds that are Secured by U.S. Treasury Obligations, Review of Financial Studies, 11, 281-308. [11] Chalmers, J. M. R., 2000, The Muni Puzzle: Explanations and Implications for Investors, in Jegadeesh, N., and B. Tuckman, ed., Advanced Fixed-Income Valuation Tools, John Wiley and Sons, 302-319. [12] Cole, C. S., P. Liu, and S. D. Smith, 1994, The Capitalization of the State Tax Exemption Benet in Municipal Bond Yields, Journal of Financial and Strategic Decisions, 7, 67-77. [13] Constantinides, G. M., and J. E. Ingersoll, 1984, Optimal Bond Trading with Personal Taxes, Journal of Financial Economics, 13, 299-355. [14] Downing, C. and F. Zhang, 2004, Trading Activity and Price Volatility in the Municipal Bond Market, Journal of Finance, 59, 899-931. [15] Fabozzi, F., and S. Feldstein, 1983, Municipal Bonds, in F. Fabozzi and T. M. Pollack, eds., Handbook of Fixed Income Securities, Dow Jones-Irwin, Homewood, IL. [16] Fortune, P., 1996, Do Municipal Bond Yields Forecast Tax Policy? New England Economic Review, Sep/Oct, 29-48. [17] Green, R. C., 1993, A Simple Model of the Taxable and Tax-Exempt Yield Curves, Review of Financial Studies, 6, 233-264. [18] Green, R. C., and B. A. degaard, 1997, Are There Tax Effects in the Relative Pricing of U.S. Government Bonds? Journal of Finance, 52, 609-633. [19] Green, R. C., B. Hollield, and N. Schrhoff, 2005, Financial Intermediation and Costs of Trading in an Opaque Market, forthcoming Review of Financial Studies [20] Green, R. C., B. Hollield, and N. Schrhoff, 2006, Dealer Intermediation and Price Behavior in the Aftermarket for New Bond Issues, working paper, Carnegie Mellon University.

39

[21] Harris, L. E., and M. S. Piwowar, 2006, Secondary Trading Costs in the Municipal Bond Market, Journal of Finance, 61, 1361-97. [22] Hildreth, W. B., and C. K. Zorn, 2005, The Evolution of the State and Local Government Municipal Debt Market over the Past Quarter Century, Public Budgeting and Finance, 25, 127-153. [23] Hong, G., and A. Warga, 2004, Municipal Marketability, Journal of Fixed Income, September, 86-95. [24] Kochin, L. A., and R. W. Parks, 1988, Was the Tax-Exempt Bond Market Inefcient or Were Future Expected Tax Rates Negative? Journal of Finance, 43, 913-931. [25] Li, D., 2006, Agreeing on a Higher Price, working paper, Carnegie Mellon University. [26] Litvack, D., and F. Rizzo, 2000, Municipal Default Risk, Municipal Finance Journal, 21, 25-42. [27] Litzenberger, R. H., and J. Rolfo, 1984, Arbitrage Pricing, Transactions Costs and Taxation of Capital Gains: A Study of Government Bonds with the Same Maturity Date, Journal of Financial Economics, 13, 337-351. [28] McDonald, R. L., 1983, Government Debt and Private Leverage, Journal of Public Economics, 22, 303325. [29] McDonald, R. L., 2006, Portfolio Choice and Corporate Financial Policy When There are TaxIntermediating Dealers, working paper, Northwestern University. [30] Nanda, V., and R. Singh, 2004, Bond Insurance: What is Special about Munis? Journal of Finance, 59, 2253-79. [31] Nelson, C. R., and A. F. Siegel, 1987, Parsimonious Modeling of Yield Curves, Journal of Business, 60, 473-489. [32] Poterba, J. M., 1986, Explaining the Yield Spread between Taxable and Tax-Exempt Bonds: The Role of Expected Tax Policy, in Rosen, H., ed., Studies in State and Local Public Finance, University of Chicago Press, Chicago, pp5-48. [33] Poterba, J. M., 1989, Tax Reform and the Market for Tax-Exempt Debt, Regional Science and Urban Economics, 19, 537-562. [34] Poterba, J. M., 2002, Taxation, Risk-Taking, and Household Portfolio Behavior, in A. J. Auerbach and M. Feldstein, eds., Handbook of Public Economics Volume 3, Elsevier Science B.V., 1110-71. [35] Poterba, J. M., and A. A. Samwick, 2002, Taxation and Household Portfolio Composition: U.S. Evidence from the 1980s and 1990s, Journal of Public Economics, 87, 5-38. [36] Sialm, C., 2006, Tax Changes and Asset Pricing: Time-Series Evidence, working paper, University of Michigan. [37] Sims, T. S., 1992, Long-Term Debt, the Term Structure of Interest and the Case for Accrual Taxation, NYU Tax Law Review, 47, 313-375. [38] Slemrod, J., and T. Greimel, 1999, Did Steve Forbes Scare the U.S. Municipal Bond Market? Journal of Public Economics, 74, 81-96. [39] Strnad, J., 1995, The Taxation of Bonds: The Tax Trading Dimension, Virginia Law Review, 81, 47-116. [40] Trzcinka, C., 1982, The Pricing of Tax-Exempt Bonds and the Miller Hypothesis, Journal of Finance, 37, 907-923. [41] Woodell, C., W. Montrone, and B. Brady, 2004, U.S. Municipal Rating Transitions and Defaults, 19862003, Municipal Finance Journal, 24, 49-78.

40

Table 1: Sample Summary Statistics

Panel A: Summary Statistics on Unique CUSIPs

Largest States of Issue

CA 10.78

NY 8.92

TX 8.83 Par 24.43

MI 5.38

FL 4.94

NJ 4.67

PA 4.34

OH 3.59

Issue Price

Discount 30.83 General Obligation 48.71 AAA 73.28 Yes 72.36 No 27.64

Premium 44.74

Type of Bond

Revenue 44.04 AA+, AA, AA20.91

Other 7.25 A+,A,A5.80

Credit Rating

Insurance

Panel B: Summary Statistics on Transactions

Largest States of Issues

CA 13.73

NY 12.81

FL 5.96 Par 13.38

TX 5.77

NJ 4.46

MI 3.47

OH 3.64

PA 3.64

Issue Price

Discount 35.67 General Obligation 44.90 AAA 68.84 Yes 63.45 Ask 43.42 5% 1.5% 5% 10,000 1-2 Years 14.25 No 36.55 Bid 24.46 25% 2% 25% 25,000 2-5 Years 41.94

Premium 50.95

Type of Bond

Revenue 49.33 AA+, AA, AA24.59

Other 5.77 A+,A,A6.57

Credit Rating

Insurance

Trade Type

Interdealer 32.12 50% 2.35% 50% 50,000 5-10 Years 43.82 75% 2.62% 75% 100,000 95% 3.35% 95% 1,000,000

Coupon Rate

Par Amount Traded

Maturity at Trade

41

Note to Table 1 The table lists summary statistics of the municipal bonds in our sample from the MSRB database where we take only straight bonds paying xed maturities greater than one year and less than ten years at the time of transaction; bonds with an S&P rating of A- or higher; federal and state-exempt bonds not subject to the AMT; bonds issued in one of the 50 states; bonds trading at least 30 days after original issuance; and transactions of at least $10,000. There are 259,471 unique CUSIPs in our sample with a total of 5,372,631 transactions from January 1995 to December 2005. Panel A lists proportions of the 259,471 unique CUSIPs falling into various categories, while Panel B lists proportions of the 5,372,631 transactions falling into various categories. For the par amount traded, we report the traded par amounts at various percentiles of the distribution.

42

Table 2: Distribution of Original Issue Prices

Issue Price Relative to $100 Par Original Issue Discount 20 20 < and 15 15 < and 10 10 < and 5 5 < and 2.50 2.50 < and 1 1 < and < 0 Par = 0 0 < and 1 1 < and 2.50 2.50 < and 5 5 < and 10 10 < and 15 15 < and 20 20 Number 103 47 57 304 1,223 8,593 69,659 63,400 19,072 34,542 24,928 24,143 10,712 1,987 701 259,471 Percentage 0.04 0.02 0.02 0.12 0.47 3.31 26.85 24.43 7.35 13.31 9.61 9.30 4.13 0.77 0.27 100.00

Par Bonds Premium Bonds

All Bonds

The table lists a breakdown of issue prices (original issue discount, par, and premium bonds) relative to $100 par value, which is denoted by zero. For example, the rst line of the table shows that 0.04% of bonds in our sample were deep discount bonds issued at a price $20 below par.

43

Table 3: Yield Spreads Across Different Price Bins

>1 Panel A: Yield Spreads All Trades # Trades per day Ask Trades # Trades per day Bid Trades # Trades per day Interdealer Trades # Trades per day -3.19 (0.08) 1528 -15.87 (0.12) 1025 13.99 (0.14) 492 1.20 (0.10) 464 -0.15 (0.23) 114 -17.94 (0.23) 87 26.35 (0.41) 40 1.54 (0.25) 39 1.73 (0.33) 129 -15.81 (0.26) 110 30.32 (0.48) 44 1.18 (0.27) 43 3.73 (0.30) 153 -15.18 (0.39) 96 24.81 (0.39) 56 1.77 (0.24) 74 26.52 (0.82) 45 -5.53 (0.93) 25 49.42 (0.96) 24 16.00 (0.62) 19 37.70 (1.05) 32 2.16 (1.11) 20 60.07 (1.25) 18 25.45 (0.86) 13 61.78 (1.58) 108 20.81 (1.46) 70 77.55 (1.62) 56 44.01 (1.13) 46 2101 (0.5 1] [0 0.5] (RP DM ) (-0.5 0] (-1 -0.5] -1 # of Days

989

1372

1758

Panel B: After-Tax Yield Spreads All Trades -3.19 (0.09) -15.87 (0.13) 14.00 (0.14) 1.21 (0.11) -0.15 (0.23) -17.94 (0.24) 26.36 (0.41) 1.55 (0.26) 1.73 (0.34) -15.81 (0.26) 30.32 (0.49) 1.19 (0.27) 1.10 (0.31) -17.50 (0.40) 22.19 (0.40) -1.00 (0.26) 12.48 (0.82) -18.80 (0.93) 35.16 (0.94) 2.08 (0.62) 16.68 (1.01) -16.80 (1.09) 38.00 (1.18) 4.94 (0.82) 27.58 (1.39) -8.87 (1.32) 40.82 (1.45) 11.25 (1.04)

Ask Trades

Bid Trades

Interdealer Trades

The table reports average yield spreads and after-tax yield spreads for bonds partitioned into different price bins. The seven bins are based on the distance between the bond price and the revised price (RP ) or de minimis boundaries (DM ). The three bins containing bond trades with prices higher than RP are dened as: bin 1 (>1) with prices greater than $1.00 above RP ; bin 2 (0.5,1] with prices between $0.50 and $1.00 (including $1) dollar above RP ; bin 3 [0 0.5] with prices between $0 and $0.50 (including $0 and $0.50) above RP ; bin 4 (RP DM ) which includes trades with prices between RP and DM ; bin 5 (-0.5 0] with prices from $0.00 to $0.5 (including $0.00) below DM ; bin 6 (-1 -0.5] with prices from $0.50 to $1.00 below DM ; and the last bin 7 ( 1) contains all the prices more than $1.00 below DM . In Panel A, the yield spread is dened as the actual transaction yield minus the model-implied yield (the yield implied by valuing the bond cashows using the municipal zero-coupon yield curve on that trading day). In Panel B, the aftertax yield spread is computed as the difference between the after-tax transaction yield and the model-implied yield. All spreads are reported in basis points. In computing averages, we only include days for which at least one trade takes place in all the bins. We also report the number of days for each type of trade used to compute the average and the average number of trades per day for each bin. We report standard errors in parentheses.

44

Table 4: Empirical Duration Across Different Price Bins

>1 All Trades Empirical Duration D Model Duration Dm D Dm 7.00 4.88 2.12 ( 0.09) 4.58 4.17 0.41 ( 0.16) 3.47 4.15 -0.68 ( 0.04) 6.31 5.50 0.81 ( 0.13) 6.62 4.16 2.46 ( 0.51) 6.52 4.59 1.93 ( 0.50) 7.31 5.70 1.62 ( 0.19) (0.5 1] [0 0.5] (RP DM ) (-0.5 0] (-1 -0.5] -1

Interdealer Trades Empirical Duration D Model Duration Dm D Dm 6.99 5.13 1.86 ( 0.13) 4.62 4.30 0.31 ( 0.11) 3.87 4.24 -0.37 ( 0.07) 5.57 5.43 0.14 ( 0.12) 6.39 4.20 2.19 ( 0.64) 5.78 4.60 1.18 ( 0.27) 7.34 5.64 1.70 ( 0.30)

In the table we report the empirical duration and model duration of bonds partitioned into different price bins. The seven bins are based on the distance between the bond price and the revised price (RP ) or de minimis boundaries (DM ). The three bins containing bond trades with prices higher than RP are dened as: bin 1 (>1) with prices greater than $1.00 above RP ; bin 2 (0.5,1] with prices between $0.50 and $1.00 (including $1) dollar above RP ; bin 3 [0 0.5] with prices between $0 and $0.50 (including $0 and $0.50) above RP ; bin 4 (RP DM ) which includes trades with prices between RP and DM ; bin 5 (-0.5 0] with prices from $0.00 to $0.5 (including $0.00) below DM ; bin 6 (-1 -0.5] with prices from $0.50 to $1.00 below DM ; and the last bin 7 (< 1) contains all the prices more than or equal to $1.00 below DM . The empirical duration of a trade is computed if the previous trade of the bond occurs within the past 30 days and the price level of the last trade occurs in the same tax region as the current trade following equation (8). Bins 1-3 represent the no-tax region, bin 4 is the capital gains tax region, and bins 5-7 are the income tax regions. Similarly, the model-implied duration is computed using bond prices and yields implied from the zero coupon yield curve using equation (9). We report standard errors of the difference between empirical duration and model implied duration, D Dm , in parentheses.

45

Table 5: Controls for Default Risk

Panel A: Taking Only Short Maturity Bonds >1 Maturity 1-2 Years # Trades per Day -8.54 (0.32) 195 (0.5 1] -1.08 (0.57) 39 [0 0.5] 5.96 (0.75) 43 (RP DM ) 11.98 (0.93) 20 (-0.5 0] 21.24 (1.24) 20 (-1 -0.5] 34.72 (1.57) 12 -1 63.19 (2.08) 13 # of Days 1247

Panel B: Taking Only Insured Bonds >1 Insured # Trades per Day -4.08 (0.10) 968 (0.5 1] -1.79 (0.24) 76 [0 0.5] -0.26 (0.31) 87 (RP DM ) -0.97 (0.32) 106 (-0.5 0] 7.29 (0.73) 31 (-1 -0.5] 11.55 (0.97) 23 -1 17.02 (1.16) 78 # of Days 1900

Panel C: Taking Only Bonds from the Same Series above RP All Trades # Series per Day -5.35 (0.75) 14 (RP DM ) 10.39 (0.99) 12 below DM 24.11 (1.13) 10 # of Days 1658

The table reports average after-tax yield spreads (the difference between the after-tax transaction yield and the model-implied yield) for bonds of different types partitioned into different price bins, representing default risk controls. Panel A takes only bonds of maturity between 1-2 years, while Panel B takes only insured bonds. Panel C considers only bonds of the same issue series, which are issued by the same issuer on the same date. In Panels A and B, the seven bins are based on the distance between the bond price and the revised price (RP ) or de minimis boundaries (DM ). The three bins containing bond trades with prices higher than RP are dened as: bin 1 (>1) with prices greater than $1.00 above RP ; bin 2 (0.5,1] with prices between $0.50 and $1.00 (including $1) dollar above RP ; bin 3 [0 0.5] with prices between $0 and $0.50 (including $0 and $0.50) above RP ; bin 4 (RP DM ) which includes trades with prices between RP and DM ; bin 5 (-0.5 0] with prices from $0.00 to $0.5 (including $0.00) below DM ; bin 6 (-1 -0.5] with prices from $0.50 to $1.00 below DM ; and the last bin 7 (< 1) contains all the prices more than or equal to $1.00 below DM . For the three bins in Panel C, the rst bin contains bond trades with prices greater than RP (bins 1-3); the second bin (RP DM ) includes trades with prices between RP and DM (bin 4); and the last bin contains prices below DM (bins 5-7). In computing averages in each panel, we only include days for which at least one trade takes place in all the bins. We also report the number of days for each type of trade used to compute the average and the average number of series per day in each price bin. We report standard errors in parentheses.

46

Table 6: Controls for Liquidity

Panel A: Taking Only NY and CA Bonds >1 NY and CA # Trades per Day -6.00 (0.26) 451 (0.5 1] -3.03 (0.41) 33 [0 0.5] -1.47 (0.42) 38 (RP DM ) -0.82 (0.39) 45 (-0.5 0] 3.54 (0.79) 14 (-1 -0.5] 6.36 (0.98) 10 -1 12.08 (1.14) 34 # of Days 1463

Panel B: Excluding 1995, 1999-2000 >1 Exclude 1995, 1999, 2000 # Trades per Day -1.68 (0.05) 1774 (0.5 1] 2.03 (0.29) 82 [0 0.5] 3.84 (0.45) 86 (RP DM ) 2.92 (0.42) 103 (-0.5 0] 16.71 (1.12) 27 (-1 -0.5] 22.15 (1.37) 18 -1 37.07 (1.90) 42 # of Days 1409

Panel C: Controlling for Transaction Size >1 All Trades # Trades per Day Average Par Amount Traded per Day 1.59 (0.20) 105 $63,465 -1 27.83 (1.39) 105 $63,465 # of Days 2095

The table reports average after-tax yield spreads (the difference between the after-tax transaction yield and the model-implied yield) for bonds of different types partitioned into different price bins, representing default risk controls. Panel A takes only New York and California bonds. Panel B excludes the years 1995, 1999, and 2000 from our sample. Panel C controls for trade size. In Panels A and B, the seven bins are based on the distance between the bond price and the revised price (RP ) or de minimis boundaries (DM ). The three bins containing bond trades with prices higher than RP are dened as: bin 1 (>1) with prices greater than $1.00 above RP ; bin 2 (0.5,1] with prices between $0.50 and $1.00 (including $1) dollar above RP ; bin 3 [0 0.5] with prices between $0 and $0.50 (including $0 and $0.50) above RP ; bin 4 (RP DM ) which includes trades with prices between RP and DM ; bin 5 (-0.5 0] with prices from $0.00 to $0.5 (including $0.00) below DM ; bin 6 (-1 -0.5] with prices from $0.50 to $1.00 below DM ; and the last bin 7 (< 1) contains all the prices more than or equal to $1.00 below DM . The two bins in Panel C are bins 1 and 7 of Panels A and B. In Panel C, we match a trade in bin 7 with a trade in bin 1 of exactly the same par amount traded. If no matching trade is found, we do not consider that transaction. If there is more than one trade in bin 1 with the same par value traded as the trade in bin 7, we take a random trade of the same transaction size. In computing averages in each panel, we only include days for which at least one trade takes place in all the bins. We also report the number of days for each type of trade used to compute the average and the average number of series per day in each price bin. We report standard errors in parentheses.

47

Table 7: Characterizing the Below De Minimis Yield Spread

Variable Intercept Log Par Amount Maturity in Years Distance to DM Issue Price Selected Fixed Effects Dummy General Obligation Dummy Revenue Dummy Insured Dummy AAA Dummy AA Dummy Bid Dummy Ask -6.17 1.71 1.45 -25.32 -22.61 15.99 -23.42 -28.95 8.34 10.01 -109.64 -103.17 123.80 -173.79 Coefcient 84.66 -8.12 -2.12 1.22 0.27 T-statistic 18.62 -176.23 -107.41 44.62 5.90

In this table we regress the yield spread of bonds, dened as the actual transaction yield less the model-implied yield in basis points, involving all transactions involving below de minimis bonds onto various independent variables. There are 405,871 transactions in total with prices below the de minimis bound. The average yield spread for this sample is 27 basis points. Distance to DM refers to the value of DM P measured in dollars per $100 face value and Issue Price is also measured as per $100 face value. The regression also includes separate dummies for each year (1995 to 2004) and separate dummies for the eight largest states in terms of transactions (CA, NY, FL, TX, NJ, MI, OH, and PA). The Dummy AA is equal to 1 if a bond has a rating of AA+, AA, or AA-.

48

Table 8: Events when Bonds Cross Between No-Tax and Capital Gains Tax Regions
Ylast Panel A: All Trades Crossing Down Crossing Up Crossing Down on the Same Day Crossing Up on the Same Day 4.19% (0.01%) 4.46% (0.00%) 4.24% (0.00%) 4.40% (0.00%) 4.45% (0.01%) 4.09% (0.00%) 4.44% (0.00%) 4.15% (0.00%) 26.76 (0.22) -36.57 (0.13) 20.84 (0.11) -24.47 (0.10) 24.74 (0.21) -34.15 (0.13) 18.99 (0.10) -22.46 (0.10) 0.46 (0.04) -0.32 (0.02) 0.00 0.00 13953 44198 32583 46627 Y0 Y0 Y0, Y0m # of Trades

Panel B: Ask Trades Crossing Down Crossing Up Crossing Down on the Same Day Crossing Up on the Same Day 4.08% (0.01%) 4.28% (0.01%) 4.20% (0.01%) 4.33% (0.01%) 4.24% (0.01%) 4.10% (0.01%) 4.33% (0.01%) 4.19% (0.01%) 16.21 (0.18) -17.52 (0.20) 13.50 (0.14) -13.51 (0.14) 14.66 (0.17) -15.90 (0.20) 12.09 (0.13) -12.09 (0.13) 0.67 (0.06) -0.43 (0.05) 0.00 0.00 5867 6524 7081 7299

Panel C: Bid Trades Crossing Down Crossing Up Crossing Down on the Same Day Crossing Up on the Same Day 4.41% (0.02%) 4.70% (0.02%) 4.43% (0.03%) 4.62% (0.03%) 4.77% (0.02%) 4.33% (0.02%) 4.77% (0.03%) 4.30% (0.03%) 35.51 (0.67) -37.02 (0.65) 33.26 (1.15) -32.04 (1.25) 33.24 (0.66) -34.70 (0.64) 31.04 (1.13) -29.89 (1.22) 0.44 (0.11) -0.44 (0.10) 0.00 0.00 1743 2076 551 522

Panel D: Interdealer Trades Crossing Down Crossing Up Crossing Down on the Same Day Crossing Up on the Same Day 4.45% (0.01%) 4.58% (0.01%) 4.46% (0.01%) 4.64% (0.01%) 4.61% (0.01%) 4.37% (0.01%) 4.64% (0.01%) 4.48% (0.01%) 15.85 (0.31) -20.19 (0.26) 18.34 (0.20) -16.75 (0.20) 14.21 (0.30) -18.36 (0.25) 16.61 (0.19) -15.10 (0.19) 0.97 (0.11) -1.07 (0.06) 0.00 0.00 2625 5675 7547 6802

49

Note to Table 8 The table lists averages of yields and yield changes for events where bonds cross between the no-tax region and the capital gains tax region, with boundaries of revised issue price, RP , and the de minimis boundary, DM . We denote the event time of the transaction crossing the RP boundary as time zero and report the yield in the column labeled Y0 . The yield of the prior trade is denoted as Ylast . We only consider bonds entering or leaving the capital gains tax region to the no-tax region, with no below de minimis trades counted. We report the change in yield Y0 = Y0 Ylast . The change in the after-tax yields between the event trade and the prior trade is reported as Y0, = Y0, Ylast, . We also report the change in the model-implied taxm exempt yields, Y0m = Y0m Ylast , which is zero for intra-day trades. All the columns with yield changes are expressed in basis points. We subdivide by the transaction type in Panels A-D. For Panel B (C), the event trade and the trade prior to the event both occur at the bid (ask), and in Panel D, the event trade and the trade prior are both interdealer trades. For the rows labeled Crossing Down and Crossing Up, we track all events where bonds move down or up, respectively, across the revised price boundary with the last trade happening within the previous ve days (but not the same day as the cross). For the rows labeled Crossing Down on the Same Day and Crossing Up on the Same Day, the last trade occurs on the same trading day as the cross. We report the number of trades in each category in the last column and report standard errors in parentheses.

50

Table 9: Events when Prices Cross Into or Out of the Income Tax Region
Ylast Panel A: All Trades Crossing Down Crossing Up Crossing Down on the Same Day Crossing Up on the Same Day 4.22% (0.01%) 4.78% (0.00%) 4.26% (0.01%) 4.66% (0.01%) 4.79% (0.01%) 4.14% (0.00%) 4.69% (0.01%) 4.17% (0.00%) 57.19 (0.51) -63.57 (0.22) 42.19 (0.30) -49.32 (0.24) 38.86 (0.42) -43.72 (0.18) 26.66 (0.24) -32.46 (0.19) 0.55 (0.05) -0.24 (0.02) 0.00 0.00 8979 42528 20602 33218 Y0 Y0 Y0, Y0m # of Trades

Panel B: Ask Trades Crossing Down Crossing Up Crossing Down on the Same Day Crossing Up on the Same Day 4.02% (0.02%) 4.38% (0.02%) 3.96% (0.02%) 4.18% (0.02%) 4.33% (0.02%) 4.01% (0.02%) 4.23% (0.02%) 3.90% (0.02%) 31.24 (0.66) -36.50 (0.73) 27.36 (0.73) -28.04 (0.86) 17.99 (0.53) -22.11 (0.59) 15.19 (0.59) -15.61 (0.65) 0.81 (0.09) -0.44 (0.08) 0.00 0.00 2339 2784 1981 2100

Panel C: Bid Trades Crossing Down Crossing Up Crossing Down on the Same Day Crossing Up on the Same Day 4.50% (0.01%) 5.05% (0.02%) 4.46% (0.03%) 5.10% (0.04%) 5.14% (0.02%) 4.42% (0.01%) 5.15% (0.04%) 4.45% (0.03%) 63.46 (1.03) -63.45 (0.98) 69.64 (2.68) -64.48 (3.06) 42.98 (0.83) -43.10 (0.80) 47.70 (2.13) -44.13 (2.31) 0.55 (0.11) -0.49 (0.10) 0.00 0.00 2099 2354 548 612

Panel D: Interdealer Trades Crossing Down Crossing Up Crossing Down on the Same Day Crossing Up on the Same Day 4.45% (0.02%) 4.75% (0.01%) 4.53% (0.01%) 4.87% (0.01%) 4.74% (0.02%) 4.38% (0.01%) 4.84% (0.01%) 4.57% (0.01%) 29.46 (1.07) -37.03 (0.54) 30.96 (0.39) -30.54 (0.51) 17.00 (0.82) -22.81 (0.44) 17.82 (0.29) -17.65 (0.40) 0.98 (0.14) -0.94 (0.08) 0.00 0.00 1477 4329 5914 5161

51

Note to Table 9 The table lists averages of yields and yield changes for events where bonds cross the de minimis boundary, DM . We denote the event time of the rst transaction crossing the DM boundary as time zero and report the yield in the column labeled Y0 . The yield of the prior trade is denoted as Ylast . We report the change in yield Y0 = Y0 Ylast . The change in the after-tax yields between the event trade and the prior trade is reported as Y0, = Y0, Ylast, . We also report the change in the model-implied tax-exempt yields, m Y0m = Y0m Ylast , which is zero for intra-day trades. We take only trades where the de minimis boundary does not change across the last trade to the event trade, thus the cross is due to the change in bond prices, not due to a shifting de minimis boundary. All the columns with yield changes are expressed in basis points. We subdivide by the transaction type in Panels A-D. For Panel B (C), the event trade and the trade prior to the event both occur at the bid (ask), and in Panel D, the event trade and the trade prior are both interdealer trades. For the rows labeled Crossing Down and Crossing Up, we track all events where bonds move down or up, respectively, across the de minimis boundary with the last trade happening within the previous ve days (but not the same day as the cross). For the rows labeled Crossing Down on the Same Day and Crossing Up on the Same Day, the last trade occurs on the same trading day as the cross. We report the number of trades in each category in the last column and report standard errors in parentheses.

52

Table 10: Estimating Implied Tax Rates

Panel A: Implied Income Tax Rates Income Tax Rate 0.79 0.73 1.20 1.01

Trade Type All Ask Bid Interdealer

Std Error 0.01 0.03 0.02 0.02

# of Obs 29581 4320 2647 7391

Panel B: Implied Capital Gains Tax Rates Capital Gains Tax Rate -0.04 0.01 0.20 -0.09

Trade Type All Ask Bid Interdealer

Std Error 0.01 0.04 0.07 0.03

# of Obs 39391 3884 1529 7573

In Panel A, we use the rst trades of bonds in the below de minimis region when the bond rst becomes subject to income tax to estimate the implied tax rate I using equation (10). In Panel B, we use the rst trades of bonds in the capital gains region (from RP to DM ) to estimate the implied capital gains tax rate C in equation (11). We only use observations where the prior trade before the crossover happens within 5 days, including when the last trade occurs on the same day as the trade crossing the de minimis or revised price boundaries. In Panels A and B, for the ask, bid, and interdealer trade types, we take only those trades where the previous trade with the previous 5 days is of the same trade type. In estimating I and C , we use xed effects for each year; dummies for different bond types (general obligation or revenue); dummies for different original issue prices (par or premium); and dummies for the eight most traded states (CA, NY, FL, TX, NJ, MI, OH, and PA).

53

Figure 1: Illustration of Market Discount for an OID Bond

100

98

96

94

Bond Price

92

90

88

86

84

5 6 7 Years to Maturity Accreted OID at 12% De Minimis boundary Purchase Price = 84 Accreted Purchase Price at 13.3105% Redemption Price = 100

10

Consider an OID bond originally issued with a 10-year maturity paying a 10% semi-annual coupon. At t = 0, this bond is issued at a price of $88.5301 with a par value of 100. The semi-annual initial yield at issue of this bond is 12%. The solid line plots the accreted OID of this bond, also called the revised price of the bond, which is the value of the remaining payments of the bond discounted at its original 12% yield. At time t = 2, an investor purchases this bond in the secondary market at a price of $84. At t = 2, the revised issue price of the bond is $89.8941, which is equivalent to the original issue price of $88.5301 plus $1.3640 in accreted OID. The market discount at t = 2 is the difference between the revised issue price and the purchase price, which is 89.8941 84.0000 = $5.8941 and graphed as a solid vertical line at t = 2. The plot also shows the accreted purchase price of $84 from t = 2 to the redemption value of $100 at t = 10, representing accretion at a yield of 13.3105%, in the dashed line and the de minimis boundary in black dots.

54

Figure 2: Holdings of Municipal Bonds

0.8 Households Direct and Indirect Banks Insurance Cos

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0 1950

1960

1970

1980

1990

2000

2010

This gure plots the percentage of outstanding municipal bonds held by households, which includes direct ownership and indirect ownership through mutual funds, money market funds, and closed-end funds; banks, which comprise commercial banks and savings institutions; and insurance companies, which are life insurance companies and other insurance companies. The computations are the authors own using raw data from the Flow of Funds compiled by the Federal Reserve.

55

Figure 3: Trades of Municipal Bonds

Number of Bonds Each Month
90

80

70

Number of Bonds in Thousands

60

50

40

30

20

10

0 1994

1996

1998

2000

2002

2004

2006

Proportion of Bonds Trading Below De Minimis

0.5 0.06

0.4

0.05

Percentage of Trades Below de Minimis

0.2

0.03

0.1

0.02

0 1994

1996

1998

2000

2002

2004

0.01 2006

In the top panel, we plot the total number of trades each month in our sample, totalling 5,372,631 over January 1995 to November 2005. The bottom panel plots the proportion of trades below the de minimis boundary each month, as a fraction of the total amount of trades in that month, in the solid line. In the dashed line, we plot the 5-year zero-coupon municipal bond yield, which is computed using the method detailed in Section 3.3.

56

5Yr Zero Rate

0.3

0.04

Figure 4: Municipal Zero-Coupon Yield Curves

1 Year Zero Rate 0.1 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 1995 2000 2005 0.1 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 1995

5 Year Zero Rate

2000

2005

10 Year Zero Rate 0.1 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 1995 2000 2005

Average Zero Curve and Par Yield Curve 0.05 Zero Rate 0.048 Par Yield 0.046 0.044 0.042 0.04 0.038 0.036 0.034 0.032 0.03 0 5 Maturity in Years 10

The plot displays the estimated municipal zero-coupon yield curves for our A-grade municipal bond sample. The construction method is outlined in Section 3.3. In the top two plots, as well as the bottom left plot, we display the estimated zero curves in solid lines for various maturities. For comparison, we overlay Bloombergs AAA yields in the red lines from August 2001. In the bottom right plot, we graph the average zero yield curve and the average par yield curve over the January 1995 to November 2005 sample.

57

Figure 5: Additional Yields Required by Below De Minimis Bonds

180 Maturity = 10 years 160 Maturity = 5 years Maturity = 2 years 140

120

100

80

60

40

20

0.01

0.02

0.03

0.04

0.05

0.06

We consider par bonds of different maturities paying semi-annual coupons of 2.35%. For a given tax-exempt yield (on the x-axis), we compute the additional yield (in basis points on the y-axis) above the tax-exempt yield required to obtain the same IRR on the after-tax cashows as the tax-exempt yield. We assume that I = 0.35 and C = 0.15.

58

Figure 6: Yields across Different Price Bins

All prices 4.8 4.6 4.4 4.2 4 3.8 3.6 >1 (0.5 1] [0 0.5] (RP DM) Ask prices 4.8 4.6 4.4 4.2 4 3.8 3.6 >1 (0.5 1] [0 0.5] (RP DM) Bid prices 4.8 4.6 4.4 4.2 4 3.8 3.6 >1 (0.5 1] [0 0.5] (RP DM) (0.5 0] (1 0.5] <1 Yield Model Yield (0.5 0] (1 0.5] <1 Yield Model Yield (0.5 0] (1 0.5] <1 Yield Model Yield

Interdealer prices 4.6 4.4 4.2 4 3.8 3.6 >1 (0.5 1] [0 0.5] (RP DM) (0.5 0] (1 0.5] <1 Yield Model Yield

59

Note to Figure 6 We plot average yields and average model-implied yields across different price buckets with two standard error bands given by the dotted lines. We divide all the transactions into seven bins based on the distance between the bond price and the revised price (RP ) or de minimis boundaries (DM ). The three bins containing bond trades with prices higher than RP are dened as: bin 1 (>1) with prices greater than $1.00 above RP ; bin 2 (0.5,1] with prices between $0.50 and $1.00 (including $1) dollar above RP ; bin 3 [0 0.5] with prices between $0 and $0.50 (including $0 and $0.50) above RP ; bin 4 (RP DM ) which includes trades with prices between RP and DM ; bin 5 (-0.5 0] contains prices from $0.00 to $0.5 (including $0.00) below DM ; bin 6 (-1 -0.5] has prices from $0.50 to $1.00 below DM ; and the last bin 7 (< 1) contains all the prices more than $1.00 below DM . We plot averages across days where at least one trade takes place in all the bins.

60

Figure A-1: Illustration of Market Discount for Bonds Sold Prior to Maturity
Case of a Par Bond
101

100

99

Bond Price

98

97

96

95

94

5 6 7 Years to Maturity Accreted OID at 10% De Minimis boundary Purchase Price = 95 Accreted Purchase Price at 10.9543% Sale Price = 99

10

Case of an OID Bond

100

98

96

94

Bond Price

92

90

88

86

84

5 6 7 Years to Maturity Accreted OID at 12% De Minimis boundary Purchase Price = 84 Accreted Purchase Price at 13.3105% Sale Price = 99

10

61

Note to Figure A-1 The gure illustrates the taxation of market discount for a bond sold prior to maturity. We consider the case of a par bond in the top panel and an OID bond in the bottom panel. In the top panel, consider a par bond originally issued with a 10-year maturity paying a 10% semi-annual coupon (Bond A). The top panel illustrates the case of a market transaction comprising both income and capital gain components. At t = 2, Bond A is sold at a price of $95 representing a yield of 10.9543%. This point is denoted as a diamond in the gure. The dashed line plots the accreted discount of the $95 purchase price (the revised price) to maturity at a yield of 10.9543%. This accretion is taxed as income. At t = 8, the bond is sold for a price of $99, which is denoted by a square. The revised price of Bond A at t = 8 is $98.3266. The gain in excess of the accretion of market discount is the distance between the red square and the dashed line, which is $0.6734, which is taxed as a capital gain. The remaining $3.3266, which is the accretion of market discount, is taxed as income. In the bottom panel, consider Bond B, which is an OID bond originally issued with a 10-year maturity paying a 10% coupon. Bond B was issued at a price of $88.5301 with a par value of 100 with an initial yield of 12%. The accretion of OID from t = 0 is shown in the solid blue line. Suppose that at time t = 2, Bond B is purchased for $84 at a yield of 13.3105%, which is denoted by the red diamond. The accretion of this purchase price at 13.3105% is shown in the dashed red line. At t = 8, Bond B is sold for $99, denoted in the gure as a red square. The rst vertical line at t = 2 represents the market discount at purchase. The second vertical line at t = 8 represents the market discount at sale. The difference between the two vertical lines represents the accrued market discount at the time of sale, which is taxed as income at the time of sale.

62