Professional Documents
Culture Documents
Diversification Benefits of TIPS
Diversification Benefits of TIPS
Abdullah Mamun
Department of Finance and Management Science
University of Saskatchewan, Saskatoon
25 Campus Drive Saskatoon, Canada SK S7N 5A7.
Email: mamun@commerce.usask.ca
Phone: (306) 966-1862
Fax: (306) 966-2515
Nuttawat Visaltanachoti *
Department of Commerce
Massey University
Private Bag 102 904, NSMC,
Auckland, New Zealand.
E-mail: N.Visaltanachoti@massey.ac.nz
Phone: +64-9-414 0800 ext. 9460
Fax: +64-9-441-8177
Abstract
This paper empirically tests the benefits of Treasury inflation protected securities
(TIPS) for investors. This study examines whether TIPS enhance the risk return
characteristics of an investor’s portfolio. The results of conditional spanning tests
show that adding TIPS to any combined portfolio of stocks, Treasury bonds,
Treasury bills, corporate bonds, and real estate provides investors with
diversification benefits. This paper also shows that United Kingdom (UK) inflation-
linked gilts (ILGs) enhance the risk return characteristics of an investor’s portfolio.
These findings hold in different economic and inflationary environments, and they
confirm the prediction of economic theory that indexed bonds are important for
investors who are vulnerable to inflation.
## Acknowledgements: We are grateful to Charles Corrado, Lawrence Rose, Henk Berkman, Ben
Jacobsen, and Greg Bauer for their valuable comments. We especially wish to thank Raymond Kan for
his helpful comments about spanning test methods. The article has also benefited from participants in
workshops at University of Saskatchewan, Massey University, University of Auckland, Thammasart
University, Bank of Canada, and Reserve Bank of New Zealand and at the 2005 Financial Management
International Annual Meeting.
*Corresponding author: Nuttawat Visaltanachoti. Private Bag 102904, Department of Commerce,
Massey University, Auckland, New Zealand. Phone: +64 9 4140800 ext 9460. Fax: +64 9 4418177.
Email address: N.Visaltanachoti@massey.ac.nz
-1-
Diversification Benefits of Treasury Inflation Protected Securities:
An Empirical Puzzle
Abstract
securities (TIPS) for investors. This study examines whether TIPS enhance
conditional spanning tests show that adding TIPS to any combined portfolio
of stocks, Treasury bonds, Treasury bills, corporate bonds, and real estate
provides investors with diversification benefits. This paper also shows that
United Kingdom (UK) inflation-linked gilts (ILGs) enhance the risk return
economic theory that indexed bonds are important for investors who are
vulnerable to inflation.
diversification
-2-
1. Introduction
investors who are vulnerable to inflation. Fischer (1975) shows that the
inflation, which previously could not be fully hedged. Roll (1996) claims that
Treasury inflation protected securities (TIPS) are the least risky of all assets
because they are immune to both default and inflation risks. Campbell and
nominal bonds are risky and that they are not good substitutes for indexed
bonds. However, empirically indexed bonds (e.g., TIPS) may not deliver the
This study makes two contributions to the literature. First, the results
show that the theoretical prediction about the benefits of indexed bonds to
investors empirically holds for TIPS. In addition, the results show that for
findings based on the United Kingdom (UK) inflation-linked gilts (ILGs) data
from January 1981 to August 2005 show that the diversification benefits hold
for other indexed bonds over different sample periods in different inflationary
-3-
Hunter and Simon (2005) show that TIPS do not enhance the mean-
bonds, and bills do not provide a hedge against inflation. Therefore, indexed
portfolio. The results of the present study suggest that Hunter and Simon’s
The results of this study are in line with the findings of Kothari and
Shanken (2004) and Roll (2004), who investigate asset allocation among
stocks, indexed bonds, Treasury bonds, and a riskless asset using a mean-
Using the hypothetical indexed bond, they investigate asset allocation among
stocks, indexed bonds, Treasury bonds, and a riskless asset. Both studies
particularly ones computed over such short sample periods such as seven
-4-
49). Therefore, Kothari and Shanken’s (2004) and Roll’s (2004) findings could
2. Data
Six classes of assets are used in the present study: equity, Treasury
bills, Treasury bonds, corporate bonds, real estate, and indexed bonds. 2 The
return of the S&P 500 composite index is used as a proxy for equity returns.
The return from the Merrill Lynch U.S. Treasury bond index is used as a
proxy for nominal Treasury bonds returns. The return from the Merrill Lynch
3-month Treasury bill index is used as a proxy for Treasury bill returns. The
return from the Merrill Lynch corporate master bond index is used as a proxy
for corporate bonds returns, and the return from the National Association of
Real Estate Investment Trust index is used as a proxy for real estate
returns. 3 The TIPS return is calculated from an index of all maturity TIPS
available from Barclay Capital. 4 The sample period starts in February 1997
(when the data are first available 5 ) and ends in August 2005.
2On December 31, 2004, the relative weight of the six classes of assets were as follows: real estate was
1.8%, TIPS were 1.7%, Treasury bonds were 14%, Treasury bills were 4.4%, corporate bonds were
11.6%, and equity was 66.6%.
3Although real estate is the smallest in terms of market value compared to other asset classes, it is
included in the present study because of its inflation-hedging capability. Two main sources of historical
real estate data used in the empirical literature are CREFs and REITs. Ibbotson and Siegel (1984)
argue that as a result of appraisal smoothing and imperfect marketability it is necessary to be careful
about directly comparing measured real estate return (i.e., CREFs return) with those of other assets.
Therefore, REITs are used to represent real estate in the present analysis.
4See http://www.barcap.com
5TIPS are designed to offer investors protection against inflation. TIPS pay a semi-annual fixed real
coupon rate on an inflation-adjusted principal. The principal is adjusted on a daily basis using the U.S.
consumer price index of all urban customers (i.e., CPI-U, set at a lag of 3 months). TIPS were first
-5-
2.2 Real Monthly Return and Its Characteristics
for real monthly returns for all assets used in this study. Panel A, Table 1
inflation is calculated from CPI-U data available from the Federal Reserve
website. 6 TIPS returns are exceptionally large during this period compared to
the returns of other asset classes, except for the REIT returns. This
exceptionally high return for TIPS is also noticed by Roll (2004). TIPS have a
higher volatility than Treasury bills, Treasury bonds, and corporate bonds,
Panel B, Table 1 shows the correlation among the real monthly returns
of TIPS, Treasury bills, Treasury bonds, corporate bonds, real estate, and the
S&P 500 equity index. The correlation between TIPS and all other asset
between TIPS and equity indices and a positive correlation between TIPS and
Treasury bills and Treasury bonds in daily returns. As in Roll’s (2004) study,
the results of the present study show a high correlation between TIPS and
Treasury bonds.
issued in 1997. Since then, the U.S. Treasury’s inflation-linked bond market has grown to more than
$281.04 billion as of December 2004. Market capitalization for TIPS is still small compared to nominal
Treasury bonds. The U.S. Treasury has, till the end of December 2004, issued 16 inflation-linked bonds
with different issues sizes. The Treasury issues TIPS with three different maturities (i.e., 5 years, 10
years, and 30 years).
6See http://www.federalreserve.gov
-6-
3. Method
TIPS for investors. Spanning tests investigate whether a set of K assets (i.e.,
(i.e., test assets) plus the K benchmark assets. Let R1t be a K-vector of the
real returns on the K benchmark assets and R2t be an N-vector of the real
returns on the N test assets. When Rt is the N+K risky asset returns at time
t, the expected returns (μ) and the covariance matrix (V) of the N+K risky
assets are
Rt = [ R1t R2t ]
⎡μ ⎤
E [ Rt ] = μ = ⎢ 1 ⎥
⎣ μ2 ⎦
⎡V V12 ⎤
Var [ Rt ] = V = ⎢ 11 ⎥ (1)
⎣V21 V22 ⎦
Huberman and Kandel (1987) provide the necessary, sufficient conditions for
-7-
where δ N = β N × K 1K − 1N . Therefore, a spanning test is a joint test of a constant
that is equal to 0 and the sum of the coefficients (i.e., sum of βs) that is equal
to 1. The null hypothesis determines whether the benchmark assets span the
return space of the benchmark assets and test assets. The rejection of the
null hypothesis indicates that the inclusion of the test assets can shift the
matrix suggested by Newey and West (1987). 7 The null hypothesis is tested
( )
' −1
W = ⎡⎣ Rθˆ − q ⎤⎦ RΩ
ˆ R' ⎡ Rθˆ − q ⎤ ∼ χ 2 (2)
⎣ ⎦ (4)
⎡1 0 ... 0 ⎤ ⎡αˆ ⎤ ⎡0 ⎤
where R = ⎢ ⎥ ; θˆ = ⎢ ⎥ ; q = ⎢ ⎥ ; Ωˆ K +1× K +1 is the covariance
⎣1 1 ... 1 ⎦ 2× K +1 ⎣ βˆ ⎦ K +1×1 ⎣1 ⎦ 2×1
matrix of α̂ and β̂ .
used:
F =W
2
~ F2,T − K −1 (5)
7According to Kan and Zhou’s (1999) simulation, tests based on the regression method have better size
and power properties in small sample tests than tests based on the stochastic discount factor (SDF)
approach. Jaganathan and Wang (2002) argue that Kan and Zhou implicitly assume that the expected
mean of factors is known, and they show that the two approaches deliver similar results when they use
the same moments.
8Kan and Zhou (2001) show a closed form of this test statistics.
-8-
3.2 Conditional Spanning Test
but the reverse is not true. Dybvig and Ross (1985) point out that superior
appropriate.
-9-
There are several ways to incorporate conditioning information into a
spanning test. Bekaert and Urias (1996) and Cochrane (1996) incorporate
return rapidly increases the dimension of the estimation and testing problem.
testing, it is assumed that the expected returns are linear in the conditional
(see Ferson and Schadt, 1996; Shanken, 1990) considers the regression
coefficients that are linear in the conditional variables. DeRoon and Nijman
expected returns are linear in the conditional variables is a special case of the
variables.
α i = ai 0 + z 't ai1
(6)
βi = bi 0 + z 't bi1
- 10 -
Ferson and Schadt (1996) consider equation (6) a first-order Taylor
DeRoon and Nijman (2001) show that spanning under all economic
ai 0 = 0
bi 0ιK = 1
(8)
ai1 = 0
bi1 = 0
F=
(Rθˆ − q )′ [RΩˆ R ′] (Rθˆ − q ) ~ F (J , T − K − J − 1)
−1
(9)
J
where R and q are the coefficient restriction matrix in equation (8), θ is the
F(J,T-K-J-1).
Cochrane (2001) points out that although investors should use all the
observed by investors does not bias the tests, but it does reduce the power.
- 11 -
Four instruments are used in the conditional model presented in this article.
They are the lagged value of yield curve slope between a Treasury bond and
Treasury bill, the lagged yield spread between a Treasury bond and TIPS,
and the lagged return of TIPS, and the lagged return of the S&P 500
composite. As mentioned by Hunter and Simon (2005), the yield curve slope
captures the business cycle and monetary policy expectation, and the yield
spread allows for a risk-adjusted expected inflation rate and periodic trade-off
of returns, inflation risk, and liquidity. In addition, lagged TIPS and S&P 500
returns.
of two to five asset classes (i.e., stocks, nominal Treasury bills, nominal
Treasury bonds, corporate bonds, and real estate). The results of the
investors hold a portfolio with any two, three, or four assets, respectively. In
column 2, Table 2. The results show that for several benchmark portfolios the
- 12 -
spanning hypothesis cannot be rejected. In all these cases, the benchmark
portfolios include Treasury bonds, corporate bonds, or both these assets. This
result is expected given the high correlation between TIPS and Treasury
bonds and TIPS and corporate bonds (see Table 1, panel B).
appropriate method for the present study. Investors and fund managers not
characteristics, they also take into account the condition of the economy and
financial market.
The results of the conditional spanning test conducted in this study are
Therefore, these results imply that during the sample period investors would
diversification from the results in the present study for several reasons. The
addition, the majority of TIPS were auctioned in late 1999 or later. During
this period, the U.S. equity market was performing poorly. 9 TIPS are a new
9The U.S. equity market had a large downturn during this period: Between August 2000 and
September 2002, the S&P 500 lost almost half its value.
- 13 -
class of asset, and there are uncertainties related to investing in TIPS, which
may affect their return. Finally, inflation was low (i.e., 0.20% monthly) and
stable (i.e., standard deviation was only 0.21% a month) during this period,
ILGs were used to test the robustness of the results presented in the
previous section. Before the introduction of TIPS in the United States and
Similar to TIPS, ILGs are rated AAA and Aaa by Standard and Poor’s and
Moody’s, respectively. ILGs are linked to the UK retail price index (RPI). In
addition, data for ILGs is available from January 1981 to the present. During
the sample period, the UK inflation rate was higher and more volatile than
bonds, and government bills. 11 With the UK sample, return on the all-share
composite index (FTA) is used as a proxy for equity return. The return on a 3-
UKbill), and the Barclays Capital sterling bond index is used as a proxy for
- 14 -
Table 3 presents the descriptive statistics for asset returns from June
1981 to August 2005. Mean real return on ILGs is lower than all other asset
classes. ILGs are also less volatile than UKbonds and FTA during the sample
government bills and government bonds; however, unlike TIPS and the S&P
500, the results do not reveal a negative correlation between ILGs and FTA.
presented in Table 4. It is assumed that investors can hold any two of three
asset classes (i.e., stocks, government bills, and government bonds) or all
column 2) show that ILGs do not improve the risk return characteristics of a
portfolio when an investor holds equity and government bills, but they do
findings that show indexed bonds (e.g., TIPS) provide additional benefits to
The results of the present study are different from the results in
Hunter and Simon’s (2005) study. This difference in results may be caused by
- 15 -
a difference in the sample period. Table 5 shows that the asset return
August 2001) and in the rest of the period ending in August 2005.
and Simon (2005) find that Treasury bonds cannot span the mean-variance
frontier generated by Treasury bonds and TIPS, but they obtain the opposite
result with their conditional test. Similarly, in their unconditional test, they
find that Treasury bills cannot span the mean-variance frontier generated by
Treasury bills and TIPS, but they obtain the opposite result with their
conditional test. These results contradict the theory (Hansen and Richard,
efficient, but the opposite is not true. In other words, the rejection of the
bonds, Hunter and Simon (2005) conclude that “Treasury bills may be a
any major inflation shocks” (p. 366). This conclusion contradicts the findings
of Campbell and Shiller (1996) and Campbell and Viceira (2001). They point
out that it is possible to replicate the return of long-term real bonds with
12There are two reasons for a lower p-value in a conditional test: (1) An asymptotic distribution is a
poor approximation of a finite sample distribution when many instruments are used, and/or (2) the use
of conditioning information variables in a conditional test introduces noise and not information.
- 16 -
short-term nominal bonds using a rollover strategy, but it is risky because
5. Concluding Remarks
asset class in the sense that it increases the reward-to-risk ratio when added
(2005), the results of this study show that TIPS provide a diversification
investors can hold any asset class (i.e., stocks, nominal Treasury bills,
nominal Treasury bonds, corporate bonds, and real estate) and in any
consistent with economic theory and in line with the findings of Kothari and
study. The results of the conditional test show that the spanning hypothesis
cannot be accepted in any scenario. These results imply that TIPS constitute
found in this study because of the short history of TIPS data and the
ILG data from January 1981 to August 2005 is used to test the robustness of
the present study’s results. The results of the robustness test show that as
- 17 -
with TIPS adding ILGs to any diversified portfolio enhances its risk return
show that the benefits of indexed bonds hold under different economic and
theories.
- 18 -
References
Chopra, V., and W. Ziemba. 1993. The effect of error in means, variances and
covariances on optimal portfolio choice. Journal of Portfolio
Management Winter: 6–11.
Ferson, W., and A. Siegel. 2001. The efficient use of conditioning information
in portfolios. Journal of Finance 56: 967–982.
Fischer, S. 1975. The demand for indexed bonds. Journal of Political Economy
83: 509–534.
- 19 -
Harvey, C. 1989. Time-varying conditional covariances in tests of asset
pricing models. Journal of Financial Economics 24: 289–317.
Hunter, D., and D. Simon. 2005. Are TIPS the ‘‘real’’ deal? A conditional
assessment of their role in a nominal portfolio. Journal of Banking and
Finance 29: 347–368.
Kan, R., and G. Zhou. 1999. A critique of the stochastic discount factor
methodology. Journal of Finance 54: 1221–1248.
Kan, R., and G. Zhou. 2001. Tests of mean-variance spanning. SSRN Working
paper (http://papers.ssrn.com/sol3/papers.cfm?abstract_id=231522)
Newey, W. K., and K. D. West. 1987. Hypothesis testing with efficient method
of moments estimation. International Economic Review 26: 777–787.
- 20 -
Table 1
Characteristics of TIPS, Treasury Bills, Treasury Bonds, Corporate
Bonds, and S&P 500 Composite Index Real Return
This Table presents the descriptive statistics and a cross-correlation among the
monthly real returns for six major asset classes in the United States. These assets
include Treasury inflation protected securities (TIPS), Treasury bills (Tbills),
Treasury bonds (Tbonds), corporate bonds (Cbonds), real estate equity index (REIT),
and S&P 500 equity index (S&P 500) from March 1997 to August 2005. TIPS is an
index of all maturity TIPS available from Barclays Capital. Tbill is a 3-month
Merrill Lynch U.S. Treasury bill index. Tbond is a Merrill Lynch U.S. Treasury bond
index. Cbond is a Merrill Lynch U.S. corporate bond master index. REIT is a real
estate index from the National Association of Real Estate Investment Trust index.
S&P 500 is the S&P 500 composite index. U.S. inflation is calculated using the
urban consumer price index of U.S. city average (CPI-U).
- 21 -
Table 2
Spanning Tests of U.S. TIPS
This Table presents the results of the unconditional and conditional spanning tests.
Wald test statistics and the p-values from the spanning tests are reported. Under
the null hypothesis, the mean-variance frontier of benchmark portfolios spans the
frontier of benchmark portfolios and TIPS. The benchmark portfolio includes major
asset classes: broad equity portfolio (S&P 500), Treasury bill (Tbill), Treasury bond
(Tbond), corporate bond (Cbond), and real estate (REIT). The instruments include
slope of the yield curve between a Treasury bond and a Treasury bill, the yield
spread between a Treasury bond and TIPS, the lagged TIPS return, and the lagged
S&P return. Panels A, B, and C present the results when a benchmark portfolio
contains any two, three, or four asset classes, respectively. Panel D presents the
extreme case of a benchmark portfolio that contains all five asset classes.
Unconditional Conditional
Benchmark
FWald p(FWald) FWald p(FWald)
Panel A: Benchmark portfolio contains any two asset classes
S&P 500 + Tbill 5.09 0.01 5.33 0.00
S&P 500 + Tbond 1.46 0.24 3.90 0.00
S&P 500 + Cbond 3.11 0.05 4.05 0.00
S&P 500 + REIT 195.47 0.00 91.04 0.00
Tbill + Tbond 12.44 0.00 5.90 0.00
Tbill + Cbond 5.58 0.01 8.69 0.00
Tbill + REIT 2.60 0.08 3.90 0.00
Tbond + Cbond 1.16 0.32 4.64 0.00
Tbond + REIT 0.73 0.49 4.26 0.00
Cbond + REIT 1.66 0.19 2.07 0.02
Panel B: Benchmark portfolio contains any three asset classes
S&P 500 + Tbill + Tbond 12.18 0.00 6.58 0.00
S&P 500 + Tbill + Cbond 5.87 0.00 9.52 0.00
S&P 500 + Tbill + REIT 3.69 0.03 6.54 0.00
S&P 500 + Tbond + Cbond 1.32 0.27 5.06 0.00
S&P 500 + Tbond + REIT 0.84 0.43 3.09 0.00
S&P 500 + Cbond + REIT 2.92 0.06 6.70 0.00
Tbill + Tbond + Cbond 12.09 0.00 7.28 0.00
Tbill + Tbond + REIT 7.43 0.00 6.55 0.00
Tbill + Cbond + REIT 4.49 0.01 10.10 0.00
Tbond + Cbond + REIT 0.65 0.53 3.89 0.00
Panel C: Benchmark portfolio contains any four asset classes
S&P 500 + Tbill + Tbond +Cbond 12.21 0.00 7.74 0.00
S&P 500 + Tbill + Tbond +REIT 7.37 0.00 10.25 0.00
S&P 500 + Tbill + Cbond + REIT 4.41 0.01 23.88 0.00
S&P 500 + Tbond + Cbond + REIT 0.83 0.44 3.90 0.00
Tbill + Tbond + Cbond + REIT 7.41 0.00 8.17 0.00
Panel D: Benchmark portfolio includes all five asset classes
S&P 500 + Tbill + Tbond +Cbond + REIT 7.82 0.00 11.40 0.00
- 22 -
Table 3
Characteristics of Major Asset Classes Real Return in the United
Kingdom
- 23 -
Table 4
Spanning Tests of UK ILGs
This Table presents the result of unconditional and conditional spanning tests. Wald
test statistics and the p-values from the spanning tests are reported. Under the null
hypothesis, the mean-variance frontier of benchmark portfolios spans the frontier of
benchmark portfolios and ILGs. The benchmark portfolio includes major asset
classes: broad equity portfolio (FTA), Treasury bill (UKbill), and Treasury bond
(UKbond). The instruments include slope of the yield curve between a Treasury bond
and a Treasury bill and the yield spread between a Treasury bond and ILG. Panel A
presents the results for a benchmark portfolio that contains any two asset classes,
and panel B presents the results for a benchmark portfolio that contains all three
asset classes.
Unconditional Conditional
Benchmark
FWald p(FWald) FWald p(FWald)
Panel A: Benchmark portfolio contains any two assets
FTSE + UKbill 2.11 0.12 6.17 0.00
FTSE + UKbond 22.28 0.00 8.79 0.00
UKbill + UKbond 6.06 0.00 1.94 0.02
Panel B: Benchmark portfolio includes all three assets
FTA + UKbill + UKbond 7.07 0.00 2.35 0.00
- 24 -
Table 5
US Asset Nominal Returns
This Table shows a monthly mean and standard deviation of nominal asset returns
in the United States in two periods. The first period is from February 1997 to
August 2001, and the second period is from September 2001 to August 2005.
- 25 -