Riding Swaptions Curve

Journal of Banking & Finance 59 (2015) 57–75
Contents lists available at ScienceDirect
Journal of Banking & Finance

journal homepage: www.elsevier.com/locate/jbf
Riding the swaption curve

Johan Duyvesteyn a,b,⇑, Gerben de Zwart c,1
a
Robeco, Department of Investment Research, P.O. Box 973, NL-3000 AZ Rotterdam, The Netherlands
b
Erasmus University Rotterdam, Erasmus School of Economics, Burgemeester Oudlaan 50, NL-3062 PA Rotterdam, The Netherlands
c
APG Asset Management, Department of Quantitative Equity Strategies, P.O. Box 75283, NL-1070 AG Amsterdam, The Netherlands
a r t i c l e i n f o a b s t r a c t
Article history: We conduct an empirical analysis of the term structure in the volatility risk premium in the fixed income
Received 4 April 2013 market by constructing long-short combinations of two at-the-money straddles for the four major swap-
Accepted 6 May 2015 tion markets (USD, JPY, EUR and GBP). Our findings are consistent with a concave, upward-sloping matu-
Available online 18 June 2015
rity structure for all markets, with the largest negative premium for the shortest term maturity. The fact
that both delta–vega and delta–gamma neutral straddle combinations earn positive returns that seem
JEL classification: uncorrelated suggests that the term structure is affected by both jump risk and volatility risk. The results
G120
seem robust for macroeconomic announcements and the specific model choice to estimate the risk expo-
G130
G140
sures for hedging.
Ó 2015 Elsevier B.V. All rights reserved.
Keywords:
Volatility risk premium
Term structure
Swaptions
Straddles
Bonds
Interest rate derivatives
1. Introduction the SRVX index as the first interest rate-based volatility index
(Mele and Obayashi, 2012). Only recently, equity variance swaps
Previous research in equity and fixed income strongly supports have been generalized to the fixed income market by Trolle
the market price of volatility risk to be negative for both markets. (2009), Mele and Obayashi (2013), Mueller et al. (2013), Li and
In contrast, investors trade volatility very differently in these mar- Song (2013) and Trolle and Schwartz (2014). This is most likely
kets. The commonly used trading instrument in the equity market because of the ‘non-trivial design issues’ (Li and Song, 2013) and a
is the variance swap (Carr and Wu, 2009), which pays the differ- lack of public data due to the OTC market structure. This might
ence between realized variance and a benchmark variance rate explain why, apart from Mueller et al. (2013), these studies focus
that is set at the start of the contract.2 On the other hand, institu- on studying and replicating variance swap contracts at a single
tional investors in the fixed income market hardly use variance swap maturity and pay little attention to the term structure of the volatil-
contracts, but are very comfortable trading over-the-counter (OTC) ity risk premium. However, swaptions naturally give rise to a matu-
swaptions to get volatility exposure. An important reason behind rity term structure.3
this might be a lack of clear benchmark points for volatility trading
in the fixed income market. This is illustrated by a gap of 20 years 3
A seemingly related, but nonetheless unrelated, line of previous work studies
between the introduction of the VIX in 1993 (Whaley, 1993) as a riding strategies on the yield curve instead of the swaption volatility curve. Yield
benchmark in the equity markets and the recent introduction of curve-riding strategies are popular investment approaches for fixed income managers
to achieve additional returns and have been widely documented; see for example the
study of Dyl and Joehnk (1981). Basically ‘yield curve-riding’ or ‘rolling down’
⇑ Corresponding author at: Robeco, Department of Investment Research, P.O. strategies buy longer-dated bonds and sell before maturity. When these bonds
Box 973, NL-3000 AZ Rotterdam, The Netherlands. Tel.: +31 10 2243405. approach maturity and the yield curve is upward-sloping, they will be valued at a
E-mail addresses: j.duyvesteyn@robeco.nl (J. Duyvesteyn), gerben.de.zwart@ lower yield. A profit will be realized when the bond is sold at the higher price. In
apg-am.nl (G. de Zwart). contrast to these yield curve-riding strategies, this study is the first empirical research
1
Tel.: +31 20 6048182. on the significance of long-short straddle combinations that ‘ride’ the swaption curve.
2
See Carr and Wu (2009) for a detailed discussion on variance swaps in equity Riding the swaption curve and riding the yield curve thus have in common that their
markets. respective forward curves are not realized over time.
http://dx.doi.org/10.1016/j.jbankfin.2015.05.012
0378-4266/Ó 2015 Elsevier B.V. All rights reserved.
58 J. Duyvesteyn, G. de Zwart / Journal of Banking & Finance 59 (2015) 57–75
This paper complements the literature by a comprehensive neutral returns under the SABR model seem, in general, comparable
empirical analysis of the term structure in the volatility risk pre- to the returns under the Black model. For example the 3 vs 6-month
mium for the four major swaption markets (USD, JPY, EUR and strategy has a return (Sharpe ratio) of 0.89% (0.60) under the SABR
GBP).4 We build on Low and Zhang (2005) who relate the volatility model and 0.85% (0.54) under the Black model. Additionally, we do
risk premium to straddle returns by proving that the average return robustness checks of our findings on the 2-year swap rate and the
of a delta-neutral straddle must not be zero if volatility risk is priced. USD swaption smile, we analyze the impact of macroeconomic
We argue that conclusions can be inferred on the term structure of announcements, and we empirically check the exposure of the strat-
the volatility risk premium by studying the average return of a egy returns to the underlying swap rate.
long-short combination of two delta-neutral straddles with different Third, we study the economic importance of our results. For
maturities. In particular, we study long-short straddle combinations example, the average return across the four markets for the 3 vs
which are either delta–gamma or delta–vega neutral. We are the 12-month delta–gamma neutral strategy is 1.89% (t-stat = 4.33)
first to apply these two strategies in the fixed income market. and an annualized Sharpe ratio of 1.35. The delta–vega neutral
Hence, we provide results showing it is plausible that the delta– strategy reports a return of 1.14% (t-stat = 3.69) and an annualized
gamma and delta–vega neutral strategies can be linked to volatility Sharpe ratio of 0.95. However, after calculating break-even costs
risk and jump risk respectively, corroborating the equity market and comparing these with expected trading costs, we conclude
findings of Cremers et al. (2015). Since sellers of volatility risk might that the returns of the strategies are not realizable by investors
also desire a jump risk premium to compensate for sudden and and therefore are not economically significant. This corroborates
extreme losses caused by the unexpected nature of jumps, we use the findings for equity option strategies obtained by Santa-Clara
this link to better understand our empirical results. The presence and Saretto (2009).
of a jump risk premium is not unlikely because there is evidence Our paper relates to several strands of literature. Most impor-
for the presence of jumps in interest rates. Johannes (2004) reports tantly our study is directly related to the literature on the volatility
a significant impact of jumps on the pricing of fixed income deriva- risk premium in fixed income. Earlier studies, such as Goodman
tives on Treasury bills. Dungey et al. (2009) relate jumps in the fixed and Ho (1997) and Duarte et al. (2007), examine the presence
income market to the release of macroeconomic data and show that and sign of the volatility risk premium in the fixed income market
about 2/3 of jumps can be explained by these releases. Using vari- by analyzing the returns of a delta-hedged investment strategy.
ance swaps, Li and Song (2013) show in a recent paper, that jump tail Since then, Almeida and Vicente (2009) have studied the volatility
risk is time varying in the swaption market. risk premium of fixed income Asian options, and Fornari (2010) has
Our research provides a number of new results. We use a large studied the volatility risk premium by calculating the difference
data set of at-the-money implied volatility quotes on the 10-year between the implied volatility and forecast of realized volatility
swap rate and 1 to 12-month swaption maturities between April using a GARCH model. Recently, a growing body of literature which
1996 and December 2011 to calculate the returns of the explores variance swap contracts in fixed income markets is
long-short straddle strategies. Our main finding is that we find sta- emerging. Variance swap contracts provide model-free estimates
tistically significant returns for all markets and for both delta– of the variance risk premium because no assumptions are made
gamma and delta–vega neutral strategies. This finding is consistent about the price process of the underlying swap rate. Trolle
with an upward-sloping term structure in the volatility risk pre- (2009) studies the variance risk premium in the US Treasury mar-
mium implying a less negative premium for longer-term swaption ket by estimating variance swaps under simplifying assumptions
maturities. The strategy returns consistently decrease across matu- and concludes that the variance risk premium is negative.
rities, which suggests that the risk premium curve flattens for Merener (2012) studies a variance strategy on forward swap rates.
longer maturities. The low, although increasing, correlations Mueller et al. (2013) and Mele and Obayashi (2013) both analyze
between the delta–gamma and delta–vega neutral strategies, that variance contracts on Treasury futures. Mele and Obayashi
is 23% for the 3 vs 6-month maturity strategy, 4% for 6 vs (2013) mainly focus on the theoretical derivation of the contract.
9-month and 38% for the 9 vs 12-month, indicate that the two Mueller et al. (2013) introduce a variance contract that is robust
strategies are uncorrelated and probably capture different effects. to jumps and can be replicated in the market at daily frequency.
This suggests that the term structure of the volatility risk premium This approach helps them to empirically analyze the variance pre-
is affected by both jump risk and volatility risk, especially at mium across the maturity and tenor spectrum, and leads them to
short-term maturities. In general, all these empirical findings are conclude that the variance risk premium is negative, but less neg-
consistent across the four individual markets. ative for longer maturities (increasing in maturity), and more neg-
Second, it is important to recognize that our strategy is based on ative for longer-term swap rates (decreasing in tenor). We see our
the Black (1976) model to estimate the risk exposures for hedging work complementing theirs, because our data is on swaptions
and to calculate the returns. To assuage this concern, we re-run our which is a different market, we focus on a straddles trading strat-
strategies on the Vasicek (1977) model for all markets and on the egy and we make a distinction between volatility and jump risk.
stochastic volatility model proposed by (Hagan et al., 2002) for Trolle and Schwartz (2014) and Li and Song (2013) both study vari-
the vega neutral strategy in the USD market.5 The Hagan et al. ance swaps in the swaption market and both have large and pro-
(2002) model is also known as the Stochastic Alpha Beta Rho prietary ‘swaption cube’ data sets from different providers that
(SABR) model. In the Vasicek (1977) framework we find comparable include data along three dimensions: swap tenors, swaption matu-
summary statistics to our main findings for all markets. The vega rities and strike rates. Li and Song (2013) focus on jump risk and
conclude that jump risk is time varying, while Trolle and
4
Schwartz (2014) study variance and skewness risk premiums
Straddles are typically used to speculate on future changes of volatility. A straddle
which are reported to be time varying and negative.
has zero delta exposure at inception. Straddles comprise a combination of a call
option (receiver swaption) and a put option (payer swaption) on a swap with the Our paper is also related to the strand of literature on test
same maturity and the same underlying strike rate. A receiver swaption is a call design for the existence of volatility risk premia. An important con-
option on a receive fixed swap where the swaption holder has the right to receive a tribution in this field includes Branger and Schlag (2008) who pro-
fixed rate on a swap in the future. A payer swaption is a call option on a pay fixed
vide a detailed discussion on the limitations of hedging-based
swap (or a put option on a receive fixed swaption) where the holder has the right to
pay a fixed rate on a swap in the future.
strategies. In particular, discrete trading and model misspecifica-
5
The additional data which is required to estimate the SABR model is not available tion may cause tests to yield unreliable results. Doran (2007)
for other markets. demonstrates that delta–gamma hedged option portfolios are less
J. Duyvesteyn, G. de Zwart / Journal of Banking & Finance 59 (2015) 57–75 59
P
subject to these discrete trading and model misspecification prob- where A ¼ i eri ti . Here the presumption made by Black (1976) is
lems than traditional delta-hedged portfolio tests. Since we con- that the forward swap rate follows a geometric Brownian motion
struct our long-short straddle combinations so that they are (dW) where the volatility is a constant:
either delta–gamma or delta–vega neutral, our strategy returns
dF X ¼ rF X dW; F X ð0Þ ¼ f ð3Þ
are most likely less prone to the limitations raised by Branger
and Schlag (2008). The value of a corresponding receiver swaption giving the right to
Finally, our paper is related to Aït-Sahalia et al. (2012) who receive strike rate F X and pay the floating rate in a swap contract
study the term structure of variance swaps in the equity market is equal to (VR):
and reveal a significant jump component embedded in the variance
swaps, especially at short-term maturities. Their analysis leads LA
VR ¼ ½FNðd1 Þ þ F X Nðd2 Þ ð4Þ
them to the conclusion that the variance risk premium is negative, m
becoming more negative for longer maturities, but we do not find The value of a straddle, S, (both long receiver and long payer
this result in the fixed income market. swaption) with the same strike rate F X is given by the sum of
The remainder of the paper is organized as follows. Section 2 Eqs. (2) and (4):
addresses our methodology and Section 3 describes the data.
Section 4 presents our main empirical findings for the delta– LA
S ¼ VP þ VR ¼ ½Fð2Nðd1 Þ 1Þ F X ð2Nðd2 Þ 1Þ ð5Þ
gamma and delta–vega hedged long-short straddle combinations. m
Section 5 provides a robustness and sensitivity analysis on the effi- Notice that the Black model is a model and therefore does not nec-
ciency of the delta-hedge, scheduled macroeconomic announce- essarily have a perfect fit with empirical data. The basic premise of
ments, choice of the pricing model and additional data sets. a constant volatility is not supported by market dynamics, because
Section 6 discusses the economic importance of our results and swaption implied volatilities tend to vary for different strikes and
we conclude in Section 7. swaption maturities (swaption smile). We discuss our data set later,
but at this stage we emphasize that we only have swaption smile
2. Methodology data for the USD and not for the other markets in our data set. As
such we assume a horizontal smile and rely on the Black model
Straddles are commonly used to speculate on, or to hedge for, (Black, 1976) to convert implied volatility quotes into prices. We will
future volatility changes because they give exposure to volatility use USD swaption smile data for a robustness analysis in Section 5.4.
and have no exposure to the underlying. In this study we analyze We calculate the return of a straddle as follows. Let St be the
long-short combinations of at-the-money swaption straddles with value of a particular straddle position at time t. Then the return
different maturities. The purpose of this section is to describe the of holding the straddle position, including funding costs, during
valuation of swaption straddles, the method to calculate the strad- the period from t 1 to t is:
dle risk parameters and to determine the hedge ratios for
#Days
long-short straddle combinations that are either delta–gamma or St St1 1 þ 360t;t1 ir t1
delta–vega neutral. Rt ¼ ð6Þ
L
where ir t1 is the annualized floating interest rate that is paid to
2.1. Computing straddle returns
borrow money to buy the swaption straddle position for the period
from t 1 to t which comprises #Dayst;t1 number of days. In calcu-
A swaption straddle is a combination of a payer swaption
plus a receiver swaption, both with the same exercise level.6 lating the returns we ignore the bid-offer spread as well as impact
In order to value the straddle we follow market practice and of margin that would be required to back the straddle trades and
use the Black (1976) pricing model to convert quoted implied limit the ability to leverage. Since trading costs and collateral
volatilities into straddle prices (Chaput and Ederington, 2005). requirements can be substantial for short option positions, as
To do so, consider a specific payer swaption giving the right to shown by Santa-Clara and Saretto (2009), we separately analyze
pay the fixed swap strike rate (F X ) and to receive the floating the economic importance in Section 6. In this section we also dis-
rate in a swap contract that will last n years (the tenor), starting cuss how the swaption market changed after the Global Financial
in T years (the maturity), with m coupon payments per year and Crisis (GFC) in 2008.
principal L. Let ti ¼ T þ i=2 be the times of each of the coupon
payments. Then, the contribution of the value of each individual 2.2. Calculating the risk parameters
cash flow (coupon payment) of the underlying swap to the
swaption is: Next to the valuation of the swaption straddles we also want to
estimate the associated risks for hedging purposes and exposure
L ri ti analysis. These risk parameters are collectively known as the
V cf ;i ¼ e ½FNðd1 Þ F X Nðd2 Þ ð1Þ
m ‘Greeks’ and quantify the influence of changes in market factors
on the straddle value. Using the Black model, the analytical delta,
where F is the forward swap rate with the same maturity as the
gamma and vega can be obtained in closed-form formulas
swaption, N is the cumulative normal distribution function,
pffiffiffi (Martellini et al., 2003).
Þþr2 T=2
d1 ¼ lnðF=FrX pffiffi
T
, d2 ¼ d1 r T and r i t i is the spot rate that corre- Delta (D), describes the price change of a straddle with respect
sponds to the maturity of cashflow i at ti. The sum of the values to the underlying swap forward rate. Technically speaking it is the
of the individual cash flows determines the total value of the payer first derivative of the straddle price with respect to the swap for-
swaption (Vp): ward rate:
X LA @S LA
VP ¼ V cf ;i ¼ ½FNðd1 Þ F X Nðd2 Þ ð2Þ D¼ ¼ ½Nðd1 Þ Nðd1 Þ ð7Þ
i
m @F m
At inception an at-the-money delta-neutral straddle position has a
delta equal to zero according to the Black model. This means that
6
At inception, the ATM swaption strike rate is set equal to the swap forward rate. the straddle value is not sensitive to a small change in the
underlying swap forward rate. Straddle exposures to the other forward rate of the long straddle position respectively. The shorter
Greeks are, however, not zero. dated swaptions have larger gammas such that the number of
Gamma (C) is the second derivative of the straddle to the straddles bought to offset the gamma exposure of the short position
underlying swap forward rate and is the rate of change of the delta is more than 1 (Q gamma
ls >1). This results in a positive vega exposure
to the underlying swap forward rate: for the long-short combination because vega is increasing in time to
maturity such that the shorter dated straddle has a smaller vega
@S2 LA
C¼ ¼ pffiffiffi ½N0 ðd1 Þ þ N0 ðd1 Þ ð8Þ than the longer dated straddle. Doran (2007) also emphasizes the
@F 2 mF r T importance of controlling a delta hedged portfolio for gamma expo-
where N 0 is the probability density function of the normal sure to enable a more precise inference on the volatility risk
distribution. premium.
Vega (v) is the sensitivity of the straddle price to the implied Analogously, a long-short straddle strategy constructed to be
volatility. In fact a straddle is quite vulnerable to volatility changes. delta and vega neutral is gamma negative and would be subjected
Technically, vega is the first derivative of the straddle price with to jump risk but almost not impacted by volatility risk. Since vega
respect to the volatility parameter r: is not constant in time to maturity we can construct the delta–vega
pffiffiffi neutral strategy using the Black risk parameters. The strategy con-
@S LAF T 0 sists of two positions (i) a short position in one delta-neutral strad-
v ¼ ¼ ½N ðd1 Þ þ N0 ðd1 Þ ð9Þ
@r m dle contract with maturity T s and (ii) a long position in Q vlsega
Note that under the assumptions of the Black model vega is a com- delta-neutral straddle contracts with maturity T l with T s < T l .
parative statistic and not a sensitivity to a variable that is dynamic Q vlsega is chosen such that the long straddle position creates an over-
such as delta or gamma. all position in the long-short straddle combination that is not only
We use the Black model in this study because of limitations on delta but also vega neutral (Eq. (9)):
data availability for the EUR, GBP and JPY markets. However, just pffiffiffiffiffi
TsFs
because financial markets quote implied volatility in the Black Q vlsega ¼ pffiffiffiffi ð11Þ
framework does not imply that risk parameters should be calcu-
TlFl
lated from the Black model without an adjustment for the pre- Since the shorter dated straddle has a smaller vega, the number of
sumption of constant volatility across strike rates. Hence, as straddles bought to offset the vega exposure of the short position is
discussed by Levin (2004), the return distribution of swap rates less than 1 (Q lsv ega <1). This results in a negative gamma exposure
is not necessarily lognormal as assumed by the Black model. We because shorter dated swaptions have larger gammas.
use USD swaption smile data for robustness analysis in In the remainder of this study we analyze the profitability of
Section 5.3 and 5.4 and show that our results are robust. these long-short straddle combinations for various swaption matu-
rities. The long-short portfolio is formed at month end and rebal-
2.3. Specification of the hedging-based strategy anced monthly in the spirit of Broadie et al. (2009). Intra-month,
we neither change the two straddle positions nor delta-hedge the
Low and Zhang (2005) relate the volatility risk premium to long-short portfolio using the underlying swap forwards. All risk
straddle returns by proving that the average return of a exposures that get into the straddles during the month are sup-
delta-neutral straddle must not be zero if volatility risk is priced. posed to cancel out as a result of offsetting long and short straddle
Following Low and Zhang (2005) we argue that studying the positions. Notwithstanding, we do analyze the risk exposures on a
returns of a long-short combination of delta neutral straddles with daily basis in Section 5.1.
different maturities will enable us to analyze the difference We follow Bakshi and Kapadia (2003) and Low and Zhang
between the volatility risk premium across swaption maturities. (2005) to apply a t-test for testing the null hypothesis that the
Then, the obvious question is: what should be the ratio between long-short profit or loss is zero. We adjust the t-statistics according
the two straddles? In the spirit Cremers et al. (2015) we construct to Newey and West (1987, 1994) to correct for heteroscedasticity
two combinations which are orthogonal and either exposed to and serial correlation. Positive average returns of the strategy
volatility risk (delta–gamma neutral combination) or to jump risk would suggest that the volatility risk premium for the shorter term
(delta–vega neutral combination). This approach does not only swaption maturity is higher than the longer-term maturity. There
enable us to analyze the term structure of the volatility risk pre- would be evidence for a downward-sloping term structure in the
mium but we might also be able to infer conclusions on the drivers volatility risk premium if the long-short returns show positive
of the term structure. returns along a range of subsequent long-short straddle maturities.
A long-short straddle strategy constructed to be delta and
gamma neutral is vega positive and would be subjected to volatil-
3. Data and volatility risk premium
ity risk but almost not impacted by jump risk. Since gamma is not
constant in time to maturity we can construct the delta–gamma
In this section we first describe our data set. We then provide
neutral strategy using the Black risk parameters. The strategy con-
empirical estimates of the volatility risk premium across different
sists of two positions (i) a short position in 1 delta-neutral straddle
swaption maturities.
contract with maturity T s and (ii) a long position in Q gamma ls
delta-neutral straddle contracts with maturity T l with T s < T l . 3.1. Data description
Q gamma
ls is chosen such that the long straddle position creates an
overall position in the long-short straddle combination that is We use an extensive data-set of swap forward rates and swap-
not only delta but also gamma neutral (Eq. (8)): tions that we obtain from Bloomberg and an anonymous major
pffiffiffiffi broker dealer. The advantage of two data sources is that this
T rF
Q gamma
ls ¼ p ffiffiffiffiffil l l ð10Þ enables cross-checking to get a high quality data-set. If available
T s rs F s
we use Bloomberg data, otherwise broker data. The Bloomberg
where T s is the swaption maturity, rs is the implied volatility and F s data are based on the most recent trades and quotes from multiple
the swap forward rate of the short straddle position and T l , rl , and F l pricing sources for each country, among other ICAP which is the
are the swaption maturity, the implied volatility and the swap largest inter-dealer broker in the swap rate derivative market
Trolle and Schwartz (2014). We may therefore expect the quotes to correlation between pairs of monthly changes in the swap yield
be timely and accurate. Given the average swaption trading vol- (below diagonal) and between pairs of monthly changes in the
ume in excess of USD 150 billion notional, we assume that the implied volatility (above diagonal). We can observe a strong and
Bloomberg data represent quotes that can be traded in practice. positive correlation (0.70–0.79) between the monthly changes in
Our sample covers the four largest and most liquid swap mar- the USD, EUR and GBP swap rates, indicating that these markets
kets and spans a period of almost 16 years, from April 1996 to comove strongly. We make a similar observation for the 1-month
December 2011. Daily close prices (midpoint) of the implied change in the implied volatilities. In a similar vein we document
Black volatility are available for at-the-money (ATM) swaptions a weak relationship between these three markets and the JPY mar-
with maturities of one, three, six, nine and twelve months for ket. The 0.15 correlation between the monthly changes of the USD
USD, EUR, JPY and GBP swaptions.7 Before 1999 the EUR data are and JPY implied volatility is a good example of this. This latter
based on Germany, the most liquid fixed income market in the result suggests that the JPY swap rate and implied volatility move
European Monetary Union. Bloomberg only provides ATM data rather independently from the other three markets.
because ATM swaptions are actively quoted in the market. Data on A potential concern about our data set could be that the data
other strike rates is proprietary and broker dealer specific.8 characteristics suggest that the assumptions of the Black model
Furthermore, we focus on the 10 year swap tenor because swaption are violated. It seems that implied volatilities are not
contracts for this maturity are the most liquid.9 log-normally distributed and that jump dynamics are probably in
Fig. 1 plots the time series behavior of the implied volatility of play. The counterfactual on Black does not influence its ability to
3-month swaptions and shows that its’ values have varied greatly follow market practice and convert implied volatility quotes into
over the course of the sample. We make two observations. Firstly, prices but may influence the effectiveness of hedging on basis of
markets seem to show common moves with spikes during major the Black model. To alleviate concerns on the hedge ratios we ana-
stress events in the market in 1998 (unwinding LTCM), 2001 lyze the robustness of our results when the Greeks are based on the
(bursting IT bubble and subsequent recession), 2008 (Lehman Vasicek (1977) model and on a stochastic volatility model (SABR)
Brothers default) and 2011 (FED announced a stimulus policy that in Section 5.3. Moreover, implied volatility from the Black model
was called Operation Twist). Secondly, we note historically high tends to have a negative correlation with the underlying swap rate
values for the Japanese implied volatility in 2003. This jump is (Chan et al., 1992). This might impact the delta-neutrality of our
related to the sharp rise of the 10-year swap rate in Japan. On straddles. We work with the constraints of the data by investigat-
June 12th, 2003 the Japanese 10-year swap rate had an all time ing delta-neutrality in Section 5.1.
low at 0.43%. One month later it had more than doubled to a level
at about 1%, while the rate further increased to 1.5% in September. 3.2. Delta hedged straddle returns
The sharp increase was aggravated by many Japanese banks who
were forced to sell government bonds due the limits on their Using our swap forward and swaption data, we empirically
value-at-risk; see BIS (2003) for more details.10 examine the compensation for volatility risk following the method
Table 1 reports the summary statistics of the swaption data proposed by Bakshi and Kapadia (2003) and Low and Zhang (2005),
across different maturities. We report the mean, standard devia- based on gains/losses by a delta-hedged strategy. We compute
tion, skewness, kurtosis, minimum, and maximum for the month daily delta-hedged returns of swaption straddles for various matu-
end implied volatility quotes in our sample. The number of rities and re-sample the returns at a monthly frequency. After one
monthly observations in each series is 189. We make several month the straddle and swap forward position will be closed and a
observations. The average volatility of Japanese swaptions is about new straddle will be initiated. It is important to note that we have
twice as high as the volatility of the other three markets, as we a holding period of one month for all maturities, whereas most aca-
already have seen in Fig. 1. Furthermore, implied volatilities for demic studies either limit their analysis to a single short maturity
shorter maturities are higher and more volatile than for longer (1-month) or study hold-to-expiration effects. In our view,
maturities in all four markets. This higher standard deviation grad- hold-to-expiration returns for maturities longer than one month
ually decreases for longer maturities. Finally, we also notice that raise issues that complicate the interpretation of straddle returns
the range between the minimum and maximum is the largest for vs the maturity. The main issue is that hold-to-expiration returns
the shortest maturities and narrows when maturity goes up. for maturities larger than one month overlap and might bias any
Together, this illustrates a higher risk embedded in shorter matu- analysis on the relationship between the swaption maturity and
rity swaption contracts. Low and Zhang (2005) argue that this the volatility risk premium.
higher risk warrants a decreasing volatility risk premium in Table 3 gives the summary statistics for daily delta-hedged
maturity. straddle returns maturing in either 1, 3, 6, 9 or 12 months with a
So far, we analyze the data separately and do not address the one month holding period. Not surprisingly the average
issue of commonalities between the four markets. To address this hold-to-expiration returns for the 1-month maturity swaption
question we compute the correlations between the swap yields straddles are negative for all markets and statistically significant
and between the implied volatilities. Table 2 presents the at the 1% level for the USD, JPY and GBP swaptions. This result is
consistent to those in Duarte et al. (2007), Fornari (2010) and
Mueller et al. (2013) who also report average delta-hedged returns
7
At inception, the ATM swaption strike rate is set equal to the swap forward rate. which are negative. Fornari (2010) also finds negative returns for
We use the New York close, which is at EST 17:00, as the closing price for all four Euro swaptions, which are not statistically significant. Second, for
markets. all markets we observe a term structure in the average returns
8
In addition, our broker-dealer provides 5 strikes for USD swaptions (ATM, ATM
which is similar. The term structure is upward-sloping with the
±50, ATM ±100). In Section 5.4 we use this USD data to test the robustness of our
empirical findings. largest negative return for the 1-month maturity. However, unlike
9
This is illustrated by the options on the 10 year U.S. Treasury futures contracts. Mueller et al. (2013) and Fornari (2010), who report negative
Open interests and daily volumes of the options on the 2y, 5y and 30y futures volatility risk premiums across all swaption maturities up to
contracts options are up to ten times lower than for the 10y futures. Source: CME
12 months, we find average returns which are positive for the long-
Group, www.cmegroup.com.
10
A similar but smaller effect can be observed in the U.S. data for the same period.
est maturities. A potential explanation for this is our one month
The USD 10-year swap rate was at a low of 3.46% on June 13th, 2003 and rose back to rebalancing frequency against the hold-to-expiration set-up in
about 5% in only two months due to mortgage hedging activity (see Duarte (2008)). the latter two studies.
90
US EMU
80 Japan UK
70
60
Implied volatility (%)
50
40
30
20
10
0
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
Fig. 1. Time series of swaption implied volatility. This figure shows the time series of the at-the-money Black’s implied volatility of 3-month maturity swaptions with 10-year
swap tenor for the USD, EUR, GBP and JPY markets. The sample period is from April 1996 to December 2011 and the EUR quotes before 1999 are implied volatility quotes for
3-month maturity swaptions on the German (DEM) swap rate.
Table 1
Summary statistics of swaption implied volatilities.
Table 2
Maturity Mean Stdev Skew Kurt Min Max Correlation D10Y swap rate and D implied volatility.
(m) (%) (%) (%) (%)
USD EUR JPY GBP
USD 1 23.5 12.2 1.88 4.87 9.7 94.8
USD – 0.59 0.15 0.57
3 23.2 10.8 1.53 2.97 10.2 80.6
EUR 0.74 – 0.20 0.68
6 22.6 9.6 1.32 1.81 11.0 67.4
JPY 0.27 0.26 – 0.22
9 22.1 8.9 1.25 1.44 11.3 61.4
GBP 0.70 0.79 0.19 –
12 21.6 8.1 1.18 1.07 11.6 55.6
EUR 1 17.2 8.1 2.08 4.53 6.5 54.9 Note: This table reports correlation statistics between the four markets in our
3 16.9 7.5 2.02 4.21 7.6 52.6 sample (USD, EUR, JPY and GBP). Correlations between the monthly changes in the
6 16.5 6.8 1.93 3.84 8.8 50.2 10 years swap rate are presented below the diagonal. Correlations between the
9 16.2 6.4 1.87 3.57 9.4 47.8 monthly changes in the implied volatility are presented above the diagonal and
12 15.8 6.0 1.83 3.36 9.4 45.3 averaged across the 3, 6, 9 and 12 month swaption maturities. The EUR data before
1999 is from Germany. The number of monthly strategy return observations is 189
JPY 1 34.4 11.9 1.54 3.65 15.0 97.0 and each series runs from April 1996 to December 2011.
3 34.3 10.3 1.23 2.68 15.0 84.0
6 33.2 9.1 0.82 1.41 14.0 77.0
9 32.1 8.2 0.64 1.04 14.6 71.0
comprises all four markets. Table 4 shows the summary statistics,
12 31.3 7.7 0.40 0.59 13.0 67.0 including the average annualized return, standard deviation, t-sta-
GBP 1 17.2 7.2 1.98 4.34 9.1 49.5
tistic, skewness, kurtosis, Sharpe ratio, first order autocorrelation
3 16.8 6.4 1.98 4.35 9.3 45.9 and the hit ratio.
6 16.5 5.7 1.90 4.03 9.4 42.3 Most importantly, Table 4 shows consistent positive average
9 16.3 5.2 1.85 3.89 9.5 40.3 returns across all maturities and markets for both delta–gamma
12 15.9 4.7 1.90 4.48 9.7 38.2
and delta–vega neutral strategies. All of the delta–gamma neutral
Note: This table reports summary statistics of annualized, at-the-money swaption returns have statistically significant means, while 13 out of 16
implied volatilities on the 10-year swap forward yield for four markets (USD, EUR, average returns for the delta–vega neutral strategy are statistically
GBP and JPY). For each combination of market and swaption maturity, the table
significant. For example, focusing on the 3 vs 6-month delta–
shows the sample mean (Mean), standard deviation (Stdev), skewness (Skew),
kurtosis (Kurt), minimum (Min), and maximum (Max) of implied volatility mid- gamma neutral strategy average annualized returns are between
quotes for maturities ranging from one to twelve months. The number of obser- 0.64% (JPY) and 0.94% (USD), and Sharpe ratios between 0.72
vations in each series is 189 and runs from April 1996 to December 2011. Data are (JPY) and 0.89 (GBP). For the delta–vega neutral strategies returns
obtained from Bloomberg and an anonymous broker. are slightly wider dispersed with average returns between 0.52%
(EUR) and 1.09% (GBP), and Sharpe ratios between 0.54 (USD)
4. Empirical results riding the swaption curve and 1.21 (GBP). The Sharpe ratios are comparable to those reported
by Cremers et al. (2015) for a similar strategy in the US equity mar-
Following the strategy set-up that was discussed in Section 2, ket.11 In light of our research question, to analyze the term structure
we now calculate the empirical returns of the delta–vega and of the volatility risk premium, the positive returns are intuitively
delta–gamma neutral strategies. We report the outcomes for four
swaption maturity combinations: short 3-month and long 11
Cremers et al. (2015) report a Sharpe ratio of 0.55 for a 1vs2-month delta–
6-month maturity straddles (3 vs 6), 6 vs 9, 9 vs 12 and 3 vs 12. gamma neutral strategy and 0.93 for a 1vs2-month delta–vega neutral strategy.
For each maturity combination we report the return statistics per Note that these strategies are ‘long-short’ while our strategies are ‘short-long’ which
individual market and for an equally weighted portfolio that explains the opposite sign.
Table 3
Pricing of volatility risk.
Maturity (m) Mean (%) t-stat Stdev (%) Skew Kurt AC(1) (%) Sharpe
USD 1 1.49 2.60 2.66 0.16 6.68 13.20 0.56
3 0.43 0.66 2.14 1.87 8.93 19.80 0.20
6 0.53 0.71 2.42 2.51 15.39 22.70 0.22
9 0.99 1.25 2.62 2.88 19.88 20.30 0.38
12 1.56 1.82 2.78 2.90 20.57 22.20 0.56
EUR 1 0.42 0.87 1.68 0.05 3.10 6.00 0.25
3 0.11 0.24 1.32 1.47 5.01 27.70 0.08
6 0.73 1.47 1.52 2.12 9.00 20.70 0.48
9 0.95 1.84 1.65 2.20 9.80 13.90 0.57
12 1.14 2.09 1.75 2.26 9.99 16.20 0.65
JPY 1 1.05 2.53 1.71 0.74 4.08 7.60 0.61
3 0.82 2.43 1.25 1.07 2.73 2.70 0.66
6 0.05 0.14 1.39 1.16 2.94 9.10 0.04
9 0.22 0.60 1.38 1.08 2.48 6.30 0.16
12 0.45 1.24 1.45 0.84 3.22 3.60 0.31
GBP 1 1.39 3.24 1.61 0.16 3.64 4.50 0.86
3 0.70 1.66 1.39 1.17 2.36 25.30 0.50
6 0.21 0.46 1.48 1.35 3.07 27.40 0.14
9 1.23 2.11 1.87 2.18 9.41 29.20 0.66
12 1.01 1.97 1.60 1.44 3.25 26.70 0.63
EQW 1 1.09 3.03 1.35 0.22 3.69 3.50 0.81
3 0.46 1.21 1.14 1.32 2.64 33.40 0.40
6 0.36 0.81 1.29 1.75 4.75 33.70 0.28
9 0.85 1.82 1.40 1.78 5.42 32.80 0.61
12 1.04 2.11 1.43 2.07 7.85 36.80 0.73
Note: This table reports annualized summary statistics of non-overlapping monthly returns on a swaption straddle strategy with daily delta hedging and a one month holding
period. The maturities of the underlying swaptions are 1, 3, 6, 9 and 12 months, respectively. We set the notional value of the straddle position equal to one and compute the
returns from a long straddle perspective. In addition to the USD, EUR, JPY and GBP markets, we compute the summary statistics of an equally-weighted (EQW) portfolio.
Summary statistics include annualized sample mean (Mean), annualized standard deviation (Stdev), skewness (Skew), kurtosis (Kurt), first-order autocorrelation (AC(1)) and
annualized Sharpe ratio. t-statistics are adjusted according to Newey and West (1987) to correct for heteroscedasticity and serial correlation up to four lags. The number of
observations in each series is 189 and runs from April 1996 to December 2011.
consistent with an upward-sloping term structure in the volatility maturities suggest that both strategies are distinctive but get more
risk premium. This is consistent with Fornari (2010) and Mueller in common when maturity increases. The maturity structure in the
et al. (2013) who both report negative risk premiums and a correlations may be linked to how volatility and jump risk is per-
downward-sloping term structure of the volatility risk premium ceived by investors. Aït-Sahalia et al. (2012) for example find that
(in absolute terms) for the fixed income market. a jump component is embedded in the volatility risk premium in
Next, note that looking across the maturity spectrum, we the US equity market and that ‘‘short-term variance risk premia
observe the largest average returns for the shortest maturity com- mainly reflect investors’ fear of a market crash, rather than the
bination (3 vs 6-month) and then a decreasing pattern. This pattern impact of stochastic volatility on the investment opportunity set’’.
is consistent for all markets. For example, the equally weighted Next, we investigate the link between jump risk and the delta–
portfolio that comprises all four markets earns 0.75% per year on vega neutral strategy in more detail. The delta–vega strategy has
average (t-stat of 3.65) with a Sharpe ratio of 1.18 for the 3 vs negative gamma exposure. We expect the strategy to make a large
6-month delta–gamma neutral portfolio. The returns for the 6 vs negative return when a jump occurs and will analyze the strategy
9-month and 9 vs 12-month portfolios are 0.40% (t-stat of 3.89) returns during jumps. In Fig. 2 we show the time-series of the daily
and 0.35% (t-stat of 4.25), respectively. Comparable, we see the changes in the 10-year USD swap rate. The vertical lines in Panel A
equally weighted portfolio returns decrease with the maturity for indicate the 1% largest daily changes (absolute) in the daily swap
the delta–vega neutral strategy with average annualized returns rate. Some of these largest changes could however coincide with
of 0.81% (t-stat of 4.23) for the shortest maturity combination, periods of high volatility, which might misclassify an observation
0.29% (t-stat of 2.50) and 0.23% (t-stat of 2.51). Based on this, we as a jump. For this reason the vertical lines in Panel B indicate real-
conclude that our results suggest that the term structure of the ized jumps according to the non-parametric jump test of Lee and
volatility risk premium is concave. Mykland (2008). If we take out these jump days from our delta–
In the spirit of Cremers et al. (2015) we conjecture that the con- vega strategy returns we expect the returns to be higher. This
sistent positive strategy returns reported in Table 4 suggest that aligns with our results in Panel B in Table 6. In this table we take
both volatility risk (delta–gamma) and jump risk (delta–vega neu- out the 1% days with the largest change in the 10-year swap rate
tral) contribute to the term structure in the volatility risk premium. from the strategy returns. We do the same for the days with real-
In the remainder of this section we investigate the difference ized jumps according to Lee and Mykland (2008). In both analysis
between the delta–gamma and delta–vega neutral strategies in we observe consistent higher returns than the full sample results
more detail. presented in Table 4. From this finding we infer that there seems
In Table 5 we present the pairwise correlations between the to be a link between jump risk and the delta–vega neutral strategy
delta–gamma and delta–vega neutral strategies for each returns. From Panel A in Table 6 we also observe that the returns of
maturity-market combination. The mean correlation across the four the delta–gamma neutral strategy are hardly affected by taking out
markets is 23.2% for the 3 vs 6-month strategy. This indicates that the jumps.
the two strategies are uncorrelated and probably capture a different Finally, we investigate the link between volatility risk and the
effect. The increasing correlations (3.6% and 37.8%) for longer delta–gamma neutral strategy in more detail. The delta–gamma
Table 4
Summary statistics for long-short straddles trading strategies.
Maturity Mean (%) t-stat Stdev (%) Skew Kurt AC(1) (%) Sharpe Hit (%)
Panel A: Delta–gamma neutral combination
USD 3 vs 6 0.94 2.68 1.20 2.33 14.88 13.99 0.78 55
6 vs 9 0.62 3.19 0.76 2.38 19.42 2.71 0.81 62
9 vs 12 0.47 3.35 0.56 0.33 1.06 1.65 0.84 56
3 vs 12 2.57 3.29 2.71 1.86 11.89 9.46 0.95 62
EUR 3 vs 6 0.72 2.99 0.88 2.01 9.06 3.32 0.81 60
6 vs 9 0.29 2.54 0.55 0.92 9.35 20.67 0.53 58
9 vs 12 0.34 2.92 0.49 1.23 5.74 4.14 0.69 61
3 vs 12 1.72 3.41 2.07 2.50 14.66 11.06 0.83 61
JPY 3 vs 6 0.64 2.72 0.89 1.71 8.90 6.79 0.72 59
6 vs 9 0.32 2.13 0.55 0.82 3.64 1.68 0.59 51
9 vs 12 0.35 2.53 0.56 0.01 2.46 7.32 0.62 59
3 vs 12 1.66 3.1 2.06 1.34 6.77 8.70 0.81 62
GBP 3 vs 6 0.69 2.97 0.78 0.88 1.51 13.10 0.89 56
6 vs 9 0.37 2.53 0.55 0.00 1.52 1.40 0.68 57
9 vs 12 0.23 1.71 0.50 0.40 3.52 2.49 0.45 55
3 vs 12 1.60 3.02 1.72 0.95 1.78 15.78 0.93 58
EQW 3 vs 6 0.75 3.65 0.63 1.36 3.26 19.31 1.18 59
6 vs 9 0.40 3.89 0.37 1.34 5.51 4.64 1.10 63
9 vs 12 0.35 4.25 0.29 0.28 1.88 10.98 1.17 66
3 vs 12 1.89 4.33 1.40 1.65 5.37 16.42 1.35 63
Panel B: Delta–vega neutral combination
USD 3 vs 6 0.85 2.51 1.57 2.73 12.81 6.74 0.54 70
6 vs 9 0.36 1.84 0.91 3.05 16.74 1.52 0.40 67
9 vs 12 0.26 1.76 0.67 1.27 5.65 0.62 0.39 62
3 vs 12 1.25 2.34 2.47 2.87 14.05 3.68 0.51 69
EUR 3 vs 6 0.52 2.17 0.90 1.48 5.63 3.83 0.58 65
6 vs 9 0.11 0.73 0.54 1.61 6.32 8.51 0.20 64
9 vs 12 0.14 1.27 0.45 0.13 6.44 3.24 0.32 59
3 vs 12 0.68 1.78 1.39 1.76 7.11 7.59 0.48 64
JPY 3 vs 6 0.79 3.17 1.06 2.65 12.60 8.50 0.75 74
6 vs 9 0.23 1.29 0.74 2.48 16.14 7.03 0.31 64
9 vs 12 0.29 2.09 0.50 1.15 6.30 5.57 0.58 65
3 vs 12 1.10 2.71 1.68 2.96 16.02 5.87 0.65 71
GBP 3 vs 6 1.09 5.05 0.90 2.46 12.00 0.63 1.21 74
6 vs 9 0.44 3.50 0.55 0.77 2.22 3.28 0.80 67
9 vs 12 0.21 1.65 0.45 0.11 2.30 7.04 0.47 59
3 vs 12 1.52 4.46 1.34 2.08 9.37 4.90 1.13 72
EQW 3 vs 6 0.81 4.23 0.76 2.01 8.32 6.00 1.07 68
6 vs 9 0.29 2.50 0.44 1.76 6.44 10.78 0.66 64
9 vs 12 0.23 2.51 0.34 0.91 2.89 6.31 0.66 61
3 vs 12 1.14 3.69 1.19 1.99 7.72 8.94 0.95 67
Note: This table reports annualized summary statistics of non-overlapping monthly returns on long-short swaption straddles strategies. Each month a swaption straddle with
the lowest maturity is sold with a notional of one, and held for one month. At the same time an additional long straddle position, with a higher maturity, is bought at a
notional that either neutralizes the vega or gamma exposure of the combined long-short position. The straddles are initiated on the last business day of the month. In addition
to the USD, EUR, JPY and GBP markets, we compute the summary statistics of an equally-weighted (EQW) portfolio. Summary statistics include annualized sample mean
(Mean), annualized standard deviation (Stdev), skewness (Skew), kurtosis (Kurt), first-order autocorrelation (AC(1)), annualized Sharpe ratio and the percentage of months
with a positive return (Hit). t-statistics are adjusted according to Newey and West (1987, 1994) to correct for heteroscedasticity and serial correlation up to four lags. The
number of observations in each series is 189 and runs from April 1996 to December 2011.
Table 5
returns to be lower if we take out the months with the largest
Pairwise correlations.
increase in volatility. This is what we find in Panel A in Table 6
3 vs 6 (%) 6 vs 9 (%) 9 vs 12 (%) 3 vs 12 (%) where we take out the strategy return for the 5% months with
USD 38.4 30.0 30.7 34.7 the largest increase in realized volatility. We note that all maturity
EUR 19.5 8.2 28.2 32.0 combinations have mean returns lower than the full sample results
JPY 31.1 0.6 40.3 33.7 presented in Table 4. For example, the average return for the 3 vs
GBP 3.8 23.3 51.8 21.4
Average 23.2 3.6 37.8 30.5
6-month USD delta–gamma neutral strategy decreases from
0.94% to 0.47%. This result suggests that there might be a link
Note: This table reports pairwise correlations between the delta–gamma and delta– between the delta–gamma strategy returns and the volatility risk
vega neutral strategy returns. Correlations are computed separately for the com-
binations of market and swaption maturities. The number of monthly strategy
premium. From Panel B in Table 6 we also observe that the returns
return observations is 189 and each series runs from April 1996 to December 2011. of the delta–gamma neutral strategy are positively affected by tak-
ing out the 5% months with the largest increase in realized volatil-
ity. It may be the case that these months (partly) overlap with
strategy has positive vega exposure and is constructed such that it jumps in the underlying swap rate.
makes a positive return if the market expectation of future volatil- The evidence presented in this section, which essentially shows
ity rises. For this reason we expect the delta–gamma strategy that long-short swaption straddles strategies produce positive
50
40
30
1-day yield change 10-year US swap (bps) 20
10
-10
-20
-30
-40
-50
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
(a) Realized jumps defined as 1% largest absolute changes
50
40
30
1-day yield change 10-year US swap (bps)
20
10
-10
-20
-30
-40
-50
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
(b) Realized jumps according to Lee and Mykland (2008)

Fig. 2. Time series of daily swap rate changes and realized jumps. This figure illustrates the difference between detection methods for jumps in the times series of daily
changes of the 10 years swap rate for the USD market. The vertical, dashed lines in Panel A represent realized jumps defined as the 1% largest absolute changes in the daily
swap rate (41 observations). The vertical, dashed lines in Panel B represent realized jumps (22 observations) according to Lee and Mykland (2008). The sample runs from April
1996 to December 2011 and includes 4110 daily observations.
average returns that decrease in maturity, is consistent with a con- 5. Robustness and sensitivity analysis
cave, upward-sloping term structure in the volatility risk premium.
The fact that both delta–vega and delta–gamma neutral strategies In this section we present a variety of additional sensitivity
earn positive returns that seem uncorrelated suggests that the and robustness analysis of our main findings. We first test the
term structure of the volatility risk premium is affected by both delta-neutrality of our strategy and analyze the empirical rela-
jump risk and volatility risk. This finding contributes to recent tionship between the strategy returns and rate changes.
studies on the pricing of jump and volatility risk in the fixed Moreover, we investigate the impact of macroeconomic
income market such as Trolle and Schwartz (2014) and Li and releases. Next, we check if our main results can be confirmed
Song (2013). using the Vasicek (1977) model and a stochastic volatility
Table 6
Effect of jumps on strategy returns.
Excl. Jumps (1%) Excl. Jumps (LM2008) Excl. DVol (5%)

Mean (%) t-stat Sharpe Mean (%) t-stat Sharpe Mean (%) t-stat Sharpe
USD 3 vs 6 0.84 2.54 0.76 0.90 2.51 0.77 0.47 1.74 0.50
6 vs 9 0.52 2.86 0.74 0.59 3.29 0.79 0.36 2.29 0.61
9 vs 12 0.43 3.13 0.75 0.43 3.13 0.76 0.37 2.80 0.68
3 vs 12 2.27 3.31 0.97 2.42 3.23 0.94 1.56 2.50 0.72
EUR 3 vs 6 0.54 2.83 0.70 0.71 3.02 0.81 0.30 1.66 0.46
6 vs 9 0.27 2.62 0.53 0.29 2.50 0.52 0.07 0.79 0.17
9 vs 12 0.30 2.57 0.61 0.35 2.89 0.71 0.22 2.06 0.51
3 vs 12 1.44 3.38 0.79 1.73 3.42 0.84 0.80 2.07 0.53
JPY 3 vs 6 0.34 1.85 0.44 0.56 2.42 0.65 0.28 1.47 0.40
6 vs 9 0.21 1.86 0.39 0.26 1.65 0.48 0.17 1.28 0.34
9 vs 12 0.37 2.36 0.53 0.89 1.50 0.38 0.28 2.04 0.52
3 vs 12 1.26 2.86 0.63 2.47 2.00 0.50 0.95 2.22 0.56
GBP 3 vs 6 0.65 3.03 0.83 0.68 2.94 0.85 0.33 1.87 0.51
6 vs 9 0.26 1.74 0.48 0.38 2.61 0.70 0.23 1.70 0.44
9 vs 12 0.28 1.91 0.53 0.22 1.65 0.43 0.11 0.89 0.22
3 vs 12 1.50 2.93 0.86 1.59 3.02 0.91 0.84 2.11 0.58
EQW 3 vs 6 0.59 3.51 1.04 0.71 3.49 1.13 0.33 1.46 0.77
6 vs 9 0.31 3.38 0.90 0.38 3.89 1.07 0.19 1.80 0.75
9 vs 12 0.34 4.27 1.09 0.47 2.77 0.73 0.23 3.10 0.91
3 vs 12 1.62 4.34 1.26 2.05 3.96 1.16 0.98 2.48 1.06
Panel B: Delta–Vega neutral combination

USD 3 vs 6 1.43 4.29 1.13 1.05 3.24 0.70 1.32 3.80 0.97
6 vs 9 0.64 4.19 0.87 0.45 2.27 0.52 0.68 4.00 0.90
9 vs 12 0.42 3.05 0.67 0.30 2.08 0.46 0.46 2.96 0.75
3 vs 12 2.11 4.56 1.10 1.53 3.01 0.66 2.05 3.86 0.97
EUR 3 vs 6 0.76 3.28 0.91 0.60 2.56 0.70 0.65 2.78 0.79
6 vs 9 0.30 2.38 0.61 0.15 1.00 0.28 0.23 1.61 0.47
9 vs 12 0.24 2.39 0.56 0.17 1.59 0.38 0.27 2.48 0.65
3 vs 12 1.10 3.17 0.87 0.79 2.16 0.60 0.96 2.62 0.76
JPY 3 vs 6 1.11 5.53 1.45 0.83 3.37 0.82 1.09 5.27 1.25
6 vs 9 0.48 3.75 0.90 0.25 1.48 0.35 0.47 2.82 0.78
9 vs 12 0.49 3.61 0.86 0.79 1.62 0.40 0.46 4.31 1.08
3 vs 12 1.70 5.50 1.41 1.43 3.53 0.78 1.66 5.04 1.24
GBP 3 vs 6 1.31 6.21 1.60 1.18 5.62 1.43 1.36 6.59 1.87
6 vs 9 0.49 4.28 0.93 0.49 3.99 0.94 0.64 5.58 1.33
9 vs 12 0.35 2.51 0.75 0.23 1.85 0.51 0.33 2.71 0.79
3 vs 12 1.85 5.66 1.50 1.65 4.99 1.34 1.99 6.36 1.83
EQW 3 vs 6 1.15 6.48 1.81 0.91 5.06 1.29 1.05 5.92 1.71
6 vs 9 0.48 5.66 1.26 0.34 3.08 0.81 0.48 5.15 1.41
9 vs 12 0.38 4.73 1.17 0.38 2.72 0.63 0.36 5.62 1.25
3 vs 12 1.69 6.46 1.71 1.35 4.86 1.17 1.57 5.97 1.67
Note: This table reports annualized summary statistics of non-overlapping monthly returns on long-short swaption straddles strategies for three differently truncated
samples. First, we take out the days on which the absolute daily change in the 10-year swap rate belongs to the 1% largest changes in our sample period (excl. Jumps (1%)).
Secondly, we take out the days on which the 10-year swap rate jumped (excl. Jumps (LM2008)) according to Lee and Mykland (2008). Thirdly, we take out the 5% months with
the largest increase in realized volatility (excl. DVol (5%)). The realized volatility is calculated as the standard deviation of daily changes of 10 years swap rate during the
month. Summary statistics include annualized sample mean (Mean), t-statistics (t-stat) and annualized Sharpe ratio (Sharpe). t-statistics are adjusted according to Newey and
West (1987, 1994) to correct for heteroscedasticity and serial correlation up to four lags. The samples run from April 1996 to December 2011.
model. Finally, we re-run our strategies on two additional data RLSt ¼ b0 þ b1 DSR10Y
t þ et ð12Þ
sets which include the 2-year tenor and swaption smile data
for the USD. where RLSt is the return of the long-short straddles strategy in
month t and DSR10Y
t is the difference between the 10-year swap rate
5.1. Hedging efficiency in month t and t 1.
Most likely, risk exposures will change during the month
A potential concern might be that our straddles are not delta because of the changes in the underlying and in the implied volatil-
neutral but exposed to changes in the underlying swap rate. This ity. This could potentially impact the strategy returns and bias our
exposure could originate from at least two channels: monthly conclusions. To address this discretization error, as highlighted by
rebalancing instead of daily rebalancing and a violation of the Branger and Schlag (2008), Fig. 3 plots the average daily risk expo-
assumptions of the Black model in our data set. To examine the sures for the USD 3 vs 6-month strategies and illustrates how the
impact of monthly rebalancing, we graphically analyze the daily Black delta, gamma and vega develop between two monthly rebal-
intra-month Black risk exposures for the 3- vs 6-month strategies. ances.12 Each panel plots a single risk exposure and shows the 5%
Next, to investigate the exposure to the underlying empirically, we
regress the strategy returns on the rate changes. The specification 12
The results for the other maturities and markets are similar and available upon
of this model is: request.
0.035 0.035
95% perc Median 5% perc 95% perc Median 5% perc
0.030 0.030
0.025 0.025
0.020 0.020
Vega
Vega
0.015 0.015
0.010 0.010
0.005 0.005
0.000 0.000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
#days after initiating the straddles position #days after initiating the straddles position
(a) Vega – vega neutral (b) Vega – gamma neutral

50 50
0 0
Gamma
-50 Gamma -50
-100 -100
-150 -150
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
(c) Gamma – vega neutral (d) Gamma – gamma neutral

0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
Delta
Delta
-0.1 -0.1
-0.2 -0.2
-0.3 -0.3
-0.4 -0.4
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
(e) Delta – vega neutral (f) Delta – gamma neutral

Fig. 3. Strategy risk exposures (‘‘Greeks’’) during the month. This figure plots the average (over time series) risk exposures (‘Greeks’) of the delta–gamma and delta–vega
neutral 3 vs 6 month calendar spread straddles strategy in the USD market. For each working day of the month, the Greeks are calculated using the Black model. Delta is the
sensitivity of the straddle to the underlying swap forward rate, vega is the sensitivity to the implied volatility and gamma is the sensitivity of the delta to the underlying swap
forward rate.
percentile, median and 95% percentile across all months in our data seems unlikely that the outcomes are the result of a large exposure
sample. Most importantly, we note that the median delta exposures to the underlying.
for both the delta–vega hedge (Panel E) and the delta–vega hedge Continuing on the exposure to the underlying, Table 7 reports
(Panel F) stay very close to zero and are 0.04 and 0.06 respectively the results from estimating Eq. (12) and shows some interesting
after one month. Notwithstanding, the 90% bandwidth for the features. For the delta–gamma-neutral strategy we observe nega-
delta–gamma neutral strategy is smaller than for the delta–vega tive estimates for b1 , that are statistically significant for the USD,
neutral strategy. Based on this observation, we conclude that it EUR and GBP markets. The coefficients for JPY are not statistically
significant. This being said, 15 out of 16 estimates for b0 are statis- small day-of-the-month effect in the returns for the delta–vega
tically significant, indicating that the strategy returns do not neutral strategy (Panel B). Average returns are somewhat higher
appear to be explained by exposure to the underlying swap rate. around month end and lower in the middle of the month, but all
These results contrast with the results for the delta–vega neutral returns are statistically significant. Further analysis (not shown)
strategy. Here we see only two positive estimates for b1 , that are indicates that volatilities of the returns are stable over the month
statistically significant; all other coefficients are not statistically implying that the differences between returns are not caused by
significant. The strategy returns after correcting for the change in a higher risk. As Dungey et al. (2009) relates jumps to the release
the swap rate remain statistically significant for 13 out of 16 of macroeconomic data in the fixed income market, the
strategies and are very close to the unadjusted returns reported day-of-the-month results seem to confirm the link between the
in Table 4. delta–vega hedged results and jump risk. Overall, the
Finally, we expand the model in Eq. (12) with three additional day-of-the-month analysis which essentially shows statistically
factors; the monthly change of the slope of the yield curve significant and positive returns for all days of the month, indicates
(D10Y3M t ), the monthly change of the implied volatility (Drt ), that our results are not driven by a specific choice of the inception
and the delta-hedged return of a 1-month short straddle position day.
(Rstraddle
t ). We define the slope of the yield curve as the difference
between the 3-month rate and the 10-year swap rate. That is, 5.3. Beyond Black’s model
the model is given by
To address the concern that the assumptions behind the Black
RLSt ¼ b0 þ b1 DSR10Y
t þ b2 D10Y3M t þ b3 Drt þ b4 Rstraddle
t þ et ð13Þ model can be violated in practice and might affect the effectiveness
of our hedges, we analyze the robustness of our main findings by
The four factor model is estimated and the estimation results are re-running our strategies on the Vasicek (1977) model and the
reported in Table 8. Most importantly, we observe that 15 out of SABR model.
16 and 11 out of 16 estimates for the intercept, b0 , are statistically
significant at a 10% level for the delta–gamma neutral and delta– 5.3.1. Vasicek model
vega neutral strategies respectively. This suggests that the strategy The assumptions behind the Vasicek model are almost equal to
returns do not appear to be explained by the four factors. the Black model. The only difference is that the Vasicek model
Additionally, we see negative sensitivities for the change in volatil- assumes a constant absolute volatility (normal distribution) while
ity and positive sensitivities for the 1-month straddle returns for the Black model assumes a constant relative volatility (lognormal
the delta–vega neutral strategy. Furthermore, the returns for the distribution).13 The consequence is that the gamma of a straddle
delta–gamma neutral strategy are positively correlated with the in the Vasicek model differs from the gamma of the Black model:
change in the implied volatility, as evidenced by the positive and
significant estimates of b3 . Finally, the estimates for the coefficients @S2 LA
CVasicek ¼ ¼ pffiffiffi ½N0 ðd1 Þ þ N0 ðd1 Þ ð14Þ
of the change of the 10-year yield and the slope are insignificant for @F 2 mrabs T
the majority of maturity-market combinations.
where rabs is the absolute volatility that is equal to rF. Numerically
We close this section by concluding that the evidence presented
the delta–gamma neutral hedge on basis of the Vasicek model is
in this section suggest that it seems unlikely that the strategy
exactly the same as the delta–gamma neutral hedge on basis of
returns can fully be explained by exposures to the underlying swap
Black’s gamma (see Section 2.3). The delta–vega neutral hedge ratio
rate, the slope of the yield curve, the implied volatility or 1-month
is different though. The vega risk parameter on basis of the Vasicek
delta-hedged returns.
model is equal to:
pffiffiffi
5.2. Day of the month @S LA T 0
v Vasicek ¼ ¼ ½N ðd1 Þ þ N0 ðd1 Þ ð15Þ
@ rabs m
Several studies have documented that bond markets respond to
To obtain a delta–vega neutral hedge we can now derive that the
scheduled macroeconomic announcements, even more strongly pffiffi
hedge ratio on basis of the Vasicek model should be equal to pffiffiffiffiffiffiffiffi
T ffi
.
than the stock markets (see e.g. Andersen et al. (2007)). In this sec- T hedge
tion we consider a possible day-of-the-month effect because the Table 9 shows the return statistics for the delta–vega neutral
majority of these macroeconomic measures is released around strategies. For all markets we find comparable statistics to our
the turn of the month. Since the scheduled macroeconomic main findings reported in Table 4. This suggests that our findings
announcements impact bond yields but also the volatility risk are robust for assuming a constant absolute volatility.
(Fornari, 2010) and jump risk (Li and Song, 2013) premia, these
announcements might also influence the returns of the delta–vega 5.3.2. SABR model
and delta–gamma neutral strategies. Announcements might have a This section considers a stochastic volatility model as alterna-
larger impact on the long-short returns when the position is not tive to the deterministic volatility models that we have used so
delta, gamma or vega neutral. Therefore, the specific day of the far. We select the SABR model, introduced in Hagan et al. (2002),
month that we have chosen to open and close our positions may because it is widely used in the market (Rebonato et al., 2009).
impact our earlier results. Until now, we used the data of the last The SABR model allows for a relation between the implied
trading day of the month to open new straddle positions and to volatility (r) and the underlying swap forward rate F. For this rea-
close the previous straddle positions. To assess whether our earlier son the delta in the SABR model (DSABR ) differs from the delta in the
results are impacted by macroeconomic announcements we initi- Black model. We use the (Bartlett, 2006) modified SABR delta.
ate the monthly straddle combinations for each individual working
day of the month. For example, we open the long-short straddles
position on the first day of the month instead of the last day, etc.
13
In Fig. 4 we plot the average strategy returns across different In the Vasicek model the volatility is measured as an absolute value and not
relative to the underlying swap forward as in the Black model. For example, a
inception days of the month. For the delta–gamma neutral strategy volatility of 50% for a swap rate of 2% implies an absolute volatility of 100 bp. The
(Panel A) we observe almost no differences between the average same absolute volatility of 100 bp leads to a relative volatility of 20% for a swap rate
returns across different days of the month. In contrast, we see a of 5%.
Table 7
Regression of strategy returns on swap rate changes.
Gamma neutral Vega neutral

2
b0 (%) t-stat b1 t-stat R b0 (%) t-stat b1 Coef R2
USD 3 vs 6 0.83 2.71 0.35 1.74 0.09 0.88 2.52 0.09 0.27 0.00
6 vs 9 0.51 3.38 0.36 3.31 0.22 0.36 1.87 0.00 0.01 0.00
9 vs 12 0.43 3.14 0.15 2.75 0.07 0.28 1.76 0.04 0.30 0.00
3 vs 12 2.23 3.35 1.12 2.89 0.17 1.28 2.35 0.11 0.20 0.00
EUR 3 vs 6 0.57 2.72 0.52 2.92 0.14 0.60 2.48 0.28 1.46 0.04
6 vs 9 0.19 2.05 0.37 3.68 0.18 0.15 0.99 0.14 1.20 0.03
9 vs 12 0.25 2.41 0.31 4.00 0.16 0.16 1.46 0.09 1.02 0.01
3 vs 12 1.29 3.16 1.58 3.59 0.23 0.79 2.08 0.43 1.37 0.04
JPY 3 vs 6 0.68 2.73 0.29 1.04 0.03 0.72 2.77 0.44 1.15 0.05
6 vs 9 0.32 2.06 0.01 0.08 0.00 0.17 0.94 0.38 1.35 0.07
9 vs 12 0.34 2.35 0.03 0.21 0.00 0.26 1.74 0.17 1.07 0.03
3 vs 12 1.70 2.95 0.25 0.40 0.00 0.98 2.29 0.79 1.23 0.06
GBP 3 vs 6 0.62 2.93 0.20 1.93 0.03 1.23 5.95 0.37 1.84 0.08
6 vs 9 0.30 2.28 0.19 2.86 0.06 0.50 4.26 0.15 1.52 0.04
9 vs 12 0.17 1.35 0.13 2.31 0.04 0.25 2.14 0.10 1.51 0.02
3 vs 12 1.34 2.86 0.67 2.94 0.08 1.72 5.52 0.53 1.79 0.08
Note: This table presents estimated results for the contemporaneous regression of the monthly strategy returns on the monthly change of the 10-year swap rate.
t þ et ;
where RLSt is the return of the long-short straddles strategy in month t and DSRt is the difference of the 10-year swap rate (SR10Y
t ) in month t and t 1. For each market and
long-short maturity combination we estimate the model separately. t-statistics are adjusted according to Newey and West (1987, 1994) to correct for heteroscedasticity and
serial correlation up to four lags. The number of observations in each series is 189 and runs from April 1996 to December 2011.
Table 8
Regression of strategy returns on swap rate, yield curve steepness and relative volatility changes and the daily delta-hedged straddle return.
b0 (%) t-stat b1 t-stat b2 t-stat b3 t-stat b4 t-stat R2

USD 3 vs 6 0.80 3.36 0.11 1.06 0.16 2.17 0.06 5.54 0.01 0.31 0.55
6 vs 9 0.51 3.99 0.15 2.64 0.04 0.82 0.04 6.38 0.01 0.61 0.53
9 vs 12 0.39 2.90 0.09 1.64 0.03 0.86 0.02 2.54 0.01 0.78 0.12
3 vs 12 1.99 3.97 0.06 0.28 0.32 1.96 0.29 8.85 0.03 0.56 0.69
EUR 3 vs 6 0.54 3.52 0.11 1.07 0.10 1.49 0.08 10.72 0.04 1.46 0.65
6 vs 9 0.17 2.28 0.09 1.65 0.01 0.25 0.05 9.06 0.01 0.74 0.51
9 vs 12 0.23 2.36 0.15 2.26 0.01 0.39 0.03 4.91 0.03 1.13 0.31
3 vs 12 1.14 4.48 0.02 0.12 0.04 0.33 0.32 8.64 0.08 1.26 0.83
JPY 3 vs 6 0.54 2.76 0.01 0.03 0.22 1.23 0.04 8.22 0.08 2.02 0.52
6 vs 9 0.32 2.46 0.27 1.04 0.35 1.32 0.02 4.23 0.02 0.62 0.24
9 vs 12 0.33 2.11 0.19 0.86 0.14 0.65 0.02 4.08 0.01 0.23 0.12
3 vs 12 1.59 3.13 0.78 1.29 0.59 1.10 0.14 6.43 0.01 0.17 0.55
GBP 3 vs 6 0.52 3.52 0.03 0.32 0.12 1.37 0.08 9.36 0.04 1.40 0.56
6 vs 9 0.24 2.34 0.07 1.13 0.06 1.27 0.04 4.70 0.03 1.18 0.26
9 vs 12 0.19 1.55 0.12 1.70 0.09 2.30 0.02 1.94 0.02 1.15 0.11
3 vs 12 1.11 4.04 0.19 0.85 0.16 0.92 0.32 13.48 0.10 1.85 0.74
Panel B: Delta–vega neutral combination
USD 3 vs 6 0.72 2.44 0.21 0.85 0.12 1.04 0.06 3.69 0.12 2.90 0.35
6 vs 9 0.32 2.15 0.21 1.56 0.14 2.11 0.05 5.19 0.06 3.08 0.46
9 vs 12 0.23 1.77 0.18 1.90 0.04 1.29 0.04 6.94 0.04 2.41 0.37
3 vs 12 0.96 2.04 0.41 1.11 0.32 1.78 0.13 3.57 0.28 5.61 0.32
EUR 3 vs 6 0.51 2.80 0.04 0.28 0.02 0.16 0.03 2.01 0.20 3.28 0.32
6 vs 9 0.11 0.99 0.02 0.19 0.08 1.41 0.03 3.68 0.11 3.15 0.37
9 vs 12 0.14 1.54 0.11 2.03 0.02 0.55 0.03 5.66 0.07 5.83 0.35
3 vs 12 0.64 2.23 0.07 0.28 0.12 0.81 0.07 1.97 0.36 5.45 0.32
JPY 3 vs 6 0.56 2.54 0.33 0.75 0.09 0.28 0.02 2.16 0.19 2.64 0.35
6 vs 9 0.09 0.57 0.49 1.36 0.07 0.27 0.02 2.43 0.09 2.29 0.39
9 vs 12 0.20 1.45 0.01 0.04 0.25 1.40 0.01 1.66 0.06 3.33 0.20
3 vs 12 0.57 1.46 0.72 0.90 0.31 0.59 0.03 1.36 0.40 4.42 0.30
GBP 3 vs 6 0.89 4.52 0.30 1.54 0.11 0.76 0.02 1.38 0.24 5.26 0.34
6 vs 9 0.34 3.49 0.02 0.30 0.03 0.62 0.03 3.92 0.11 5.85 0.37
9 vs 12 0.20 1.97 0.05 0.97 0.01 0.13 0.04 6.14 0.04 2.49 0.32
3 vs 12 1.18 4.33 0.34 1.28 0.14 0.72 0.05 2.32 0.38 7.66 0.39
Note: This table presents estimated results for the following contemporaneous regression model.
t þ b2 D10Y3Mt þ b3 Drt þ b4 Rstraddle
t þ et ;
where RLSt is the return of the long-short straddles strategy in month t, DSR10Y
t is the difference of the 10-year swap rate ðSR10Y Þ in month t and t 1, Drt is the difference of
the relative implied volatility ðrÞ in month t and t 1 and Rstraddle
t is the daily delta-hedged return on a delta neutral straddle in month t. For each market and long-short
maturity combination we estimate the model separately. t-statistics are adjusted according to Newey and West (1987, 1994) to correct for heteroscedasticity and serial
correlation up to four lags. The number of observations in each series is 189 and runs from April 1996 to December 2011.
2.00%
3M vs 6M 6M vs 9M 9M vs 12M 3M vs 12M
1.75%
1.50%
1.25%
Annualized return
1.00%
0.75%
0.50%
0.25%
0.00%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Working day of the month (1=first day of the month)
(a) Delta-gamma neutral strategy
2.00%
3M vs 6M 6M vs 9M 9M vs 12M 3M vs 12M
1.75%
1.50%
1.25%
Annualized return
1.00%
0.75%
0.50%
0.25%
0.00%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Working day of the month (1=first day of the month)
(b) Delta-vega neutral strategy

Fig. 4. Day-of-the-month return effect. This figure shows the portfolio returns of the long-short straddles strategies across inception days at different working days of the
month. Portfolio returns are the equally weighted (EQW) average returns for the USD, EUR, GBP and JPY markets. The number of observations in each series is 189 and runs
from April 1996 to December 2011. t = 0 is the last working day of the previous month.

@ r @ r qm the vega hedge on basis of the SABR model is not necessarily delta
DSABR ¼ D þ v þ ð16Þ
@F @ a F b neutral under the SABR model.
" #
@ r @ r qF b
v SABR ¼v þ ð17Þ
where D and v on the right hand side are the original Black delta (7) @ a @F m
and Black vega (9) respectively.
According to the SABR model, the at-the-money straddle posi- Similarly to our main analysis, we take a short position of one strad-
tion is not necessarily delta neutral. Therefore, we analyze a dle contract in the shorter-term maturity and a long position in the
SABR delta neutral strategy using a long-short combination of longer-term maturity with a hedge ratio that either neutralizes the
two straddles, instead of a delta–gamma neutral strategy. estimated SABR delta or SABR vega. For the empirical analysis we
Likewise, we also analyze a SABR vega neutral strategy using the have gathered swaption smile data for USD swaptions from JP
adjusted vega formula from Bartlett (2006) with the caveat that Morgan because we need smile data to estimate the SABR model.
Table 9 5.4. Swaption smile and 2-year tenor

Summary statistics for trading strategies on the Vasicek model.
Vega neutral As a final robustness check of our main findings we re-run our
Mean (%) t-stat Stdev (%) Sharpe analysis on two additional data sets, (i) USD data for 5 different
strike levels (ATM, ATM 50, ATM 100) and (ii) implied volatil-
USD 3 vs 6 0.86 2.60 1.53 0.56
6 vs 9 0.38 2.10 0.85 0.45
ity data for at-the-money (ATM) swaptions on 2-year swap rates.
9 vs 12 0.25 1.70 0.68 0.37 Table 11 reports USD summary statistics for the delta–vega and
3 vs 12 1.28 2.44 2.41 0.53 delta–gamma neutral strategies where the straddles are valued on
EUR 3 vs 6 0.52 2.21 0.88 0.59 an implied volatility data set with 5 different strike levels. For the
6 vs 9 0.11 0.79 0.52 0.22 delta–gamma neutral strategy we see that the average returns and
9 vs 12 0.15 1.40 0.43 0.35 Sharpe rations are higher than in our main findings (Table 4) for all
3 vs 12 0.69 1.85 1.36 0.51
maturity combinations. A possible explanation for the higher
JPY 3 vs 6 0.78 3.32 1.01 0.78 returns is that the strategy has positive vega exposure in combina-
6 vs 9 0.23 1.41 0.69 0.34
tion typical higher out-of-the-money volatilities (smile or smirk
9 vs 12 0.27 2.04 0.48 0.56
3 vs 12 1.10 2.84 1.61 0.68 pattern). For the delta–vega neutral strategy we observe a similar
return and Sharpe ratio for the 3 vs 6-month strategy and slightly
GBP 3 vs 6 1.09 5.06 0.90 1.21
6 vs 9 0.44 3.57 0.54 0.82 lower returns for the other two maturity combinations. Overall we
9 vs 12 0.22 1.76 0.44 0.50 conclude that our main findings are confirmed after controlling for
3 vs 12 1.52 4.54 1.32 1.16 the existence of an implied volatility smile or smirk.
EQW 3 vs 6 0.81 4.34 0.74 1.09 Table 12 shows summary statistics for our trading strategies on
6 vs 9 0.29 2.73 0.41 0.71 the 2-year swap rate. On this data set, the patterns in the average
9 vs 12 0.22 2.53 0.33 0.67
returns are in general similar to the 10-year swap rate data, essen-
3 vs 12 1.15 3.82 1.16 0.99
tially showing positive returns that decrease in maturity. Yet, we
Note: This table reports annualized summary statistics of non-overlapping monthly note that average returns and Sharpe ratios are lower and in some
returns on long-short swaption straddles strategies. Each month a swaption case slightly negative. For the delta–gamma neutral strategy 10 out
straddle with the lowest maturity is sold with a notional of one, and held for one
month. At the same time an additional long straddle position, with a higher
of 16 strategies have average returns that are statistically signifi-
maturity, is bought at a notional that neutralizes the vega exposure of the combined cant at the 10% level. For the vega-neutral strategy, the average
long-short position according to the Vasicek (1977) model. The straddles are USD and EUR returns are not statistically significant and 5 out of
initiated on the last business day of the month. In addition to the USD, EUR, JPY and 8 strategies for the JPY and GBP. A potential explanation might
GBP markets, we compute the summary statistics of an equally-weighted (EQW)
come from Trolle and Schwartz (2014) who analyze the relation-
portfolio. Summary statistics include annualized sample mean (Mean), annualized
standard deviation (Stdev), skewness (Skew), kurtosis (Kurt), first-order ship between the tenor of the underlying swap rate and the vari-
autocorrelation (AC(1)), annualized Sharpe ratio and the percentage of months ance risk premium. They document a hump-shaped function and
with a positive return (Hit). t-statistics are adjusted according to Newey and West report more negative returns for the 10-year tenor than for the
(1987, 1994) to correct for heteroscedasticity and serial correlation up to four lags. 2-year tenor in both USD and EUR markets. We leave this issue
The number of observations in each series is 189 and runs from April 1996 to
December 2011.
for future work.
We estimate the parameters of the SABR model with a fixed beta 6. Economic importance
parameter of 0.5, which seems to be market practice in the US
according to Rebonato et al. (2009). In this section we explore the economic importance of our find-
Table 10 tabulates the summary statistics for the delta neutral ings. This is motivated by the findings of Santa-Clara and Saretto
and vega neutral long-short strategies using the SABR model for (2009), who show a large disparity between the profitability of
hedging. The average returns are positive across all maturities for option strategies in the equity market before and after taking costs
both delta and vega neutral strategies. Both strategies have statis- into account. Transaction costs can be substantial and collateral
tically significant means for 3 vs 6 and 6 vs 9-month maturity com- requirements, to limit counterparty default risk, may further
binations at a 90% confidence level. In general, the vega neutral reduce profitability. Hence, the tensions in the funding markets
returns and Sharpe ratios under the SABR model seem comparable during the Global Financial Crisis taught swaption dealers that
to the returns and Sharpe ratios under the Black model. For exam- funding costs embedded in derivative operations should not be
ple the 3 vs 6-month strategy has a return (Sharpe ratio) of 0.89% ignored and might result in higher costs for investors.
(0.60) under the SABR model and 0.85% (0.54) under the Black We start from an investor’s perspective. So far, the Sharpe ratios
model. Based on these results we conclude that our main findings of our delta–gamma and delta–vega neutral strategies might look
are robust. appealing to investors. Since the returns of both strategies are
Table 10
Summary statistics for trading strategies on SABR model.
Delta neutral Vega neutral

Mean (%) t-stat Stdev (%) Sharpe Mean (%) t-stat Stdev (%) Sharpe
USD 3 vs 6 0.96 2.36 1.75 0.55 0.89 2.42 1.48 0.60
6 vs 9 1.27 1.79 3.42 0.37 0.31 1.77 0.91 0.34
9 vs 12 0.46 1.19 1.19 0.39 0.19 1.01 0.63 0.30
3 vs 12 1.77 2.14 2.84 0.62 1.21 2.10 2.40 0.50
Note: This table reports annualized summary statistics of non-overlapping monthly returns on long-short swaption straddles strategies using the SABR model for hedging. The
combined long-short straddles position either neutralizes the vega or the delta exposure according to the SABR greeks. The straddles are initiated on the last business day of
the month. Summary statistics include annualized sample mean (Mean), annualized standard deviation (Stdev), skewness (Skew), kurtosis (Kurt), first-order autocorrelation
(AC(1)), annualized Sharpe ratio and the percentage of months with a positive return (Hit). t-statistics are adjusted according to Newey and West (1987, 1994) to correct for
heteroscedasticity and serial correlation up to four lags. The number of observations in each series is 189 and runs from April 1996 to December 2011.
Table 11
Summary statistics for trading strategies on swaption smile.

USD 3 vs 6 1.50 3.31 1.64 0.91 0.89 2.39 1.49 0.60
6 vs 9 0.89 3.40 1.01 0.88 0.29 1.71 0.92 0.31
9 vs 12 0.62 3.43 0.57 1.08 0.18 0.92 0.64 0.28
3 vs 12 3.76 3.48 3.71 1.01 1.19 2.06 2.42 0.49
Note: This table reports annualized summary statistics of non-overlapping monthly returns on long-short swaption straddles strategies for the USD market. Each month a
swaption straddle with the lowest maturity is sold with a notional of one, and held for one month. At the same time an additional long straddle position, with a higher
maturity, is bought at a notional that either neutralizes the vega or gamma exposure of the combined long-short position according to the Black (1976) model. The results
differ from the results in Table 4 because swaption smile data is used to calculate the returns. Swaption data is available at 5 different strike levels (ATM, ATM 50, ATM
100). Summary statistics include annualized sample mean (Mean), t-statistics (t-stat), annualized standard deviation (Stdev) and annualized Sharpe ratio. t-statistics are
adjusted according to Newey and West (1987, 1994) to correct for heteroscedasticity and serial correlation up to four lags. The number of observations in each series is 189
and runs from April 1996 to December 2011.
Table 12
Summary statistics for trading strategies on 2-year tenor.

USD 3 vs 6 0.22 2.12 0.37 0.58 0.06 0.84 0.34 0.19
6 vs 9 0.18 2.96 0.25 0.70 0.03 0.67 0.22 0.15
9 vs 12 0.07 1.39 0.22 0.33 0.04 0.96 0.19 0.23
3 vs 12 0.61 2.48 0.98 0.62 0.06 0.48 0.55 0.11
EUR 3 vs 6 0.17 2.13 0.27 0.63 0.05 0.76 0.25 0.21
6 vs 9 0.07 1.54 0.19 0.38 0.01 0.21 0.16 0.06
9 vs 12 0.09 2.18 0.17 0.52 0.03 0.62 0.17 0.15
3 vs 12 0.44 2.23 0.68 0.64 0.05 0.48 0.40 0.14
JPY 3 vs 6 0.08 1.17 0.25 0.30 0.03 0.89 0.18 0.19
6 vs 9 0.02 0.31 0.21 0.09 0.02 0.80 0.12 0.19
9 vs 12 0.29 3.70 0.22 1.31 0.51 2.27 0.85 0.61
3 vs 12 0.70 3.68 0.60 1.16 0.23 2.15 0.43 0.54
GBP 3 vs 6 0.07 0.82 0.30 0.24 0.15 1.80 0.28 0.53
6 vs 9 0.07 1.23 0.22 0.30 0.07 1.44 0.21 0.34
9 vs 12 0.10 2.19 0.19 0.54 0.08 1.83 0.16 0.48
3 vs 12 0.35 1.68 0.72 0.48 0.23 1.79 0.45 0.51
EQW 3 vs 6 0.13 2.09 0.20 0.65 0.07 1.54 0.17 0.43
6 vs 9 0.07 1.89 0.14 0.53 0.02 0.58 0.11 0.16
9 vs 12 0.14 5.10 0.11 1.24 0.14 2.20 0.24 0.60
3 vs 12 0.52 3.71 0.49 1.06 0.14 1.67 0.30 0.48
Note: This table reports annualized summary statistics of non-overlapping monthly returns on long-short swaption straddles strategies on a 2-year maturity of the underlying
swap rate. Each month a swaption straddle with the lowest maturity is sold with a notional of one, and held for one month. At the same time an additional long straddle
position, with a higher maturity, is bought at a notional that either neutralizes the vega or gamma exposure of the combined long-short position. The straddles are initiated on
the last business day of the month. In addition to the USD, EUR, JPY and GBP markets, we compute the summary statistics of an equally-weighted (EQW) portfolio. Summary
statistics include annualized sample mean (Mean), t-statistics (t-stat), annualized standard deviation (Stdev) and annualized Sharpe ratio. t-statistics are adjusted according
to Newey and West (1987, 1994) to correct for heteroscedasticity and serial correlation up to four lags. The number of observations in each series is 189 and runs from April
1996 to December 2011.
positive (Table 4) and mutual correlations are low (Table 5), inves- Table 13
Return statistics and break-even spread for the 3 vs 12-month mixed strategy.
tors might consider combining the two strategies to benefit from
diversification. Panel A in Table 13 reports the summary statistics Panel A: 50/50 combination Panel B: Break-even spread
for an equally weighted mix of the two strategies. Combining the Mean t- Stdev Sharpe Gamma Vega 50/50
two strategies proves to be successful, in the sense that returns (%) stat (%) neutr. neutr. mix
are strongly significant and Sharpe ratios are higher for the combi- USD 1.91 4.77 1.48 1.29 0.25 0.34 0.3
nations. For example, the return of the 3 vs 12-month maturity EUR 1.20 4.23 1.05 1.14 0.22 0.22 0.22
portfolio has a Sharpe ratio of 2.22 for the combined strategies JPY 1.38 5.15 1.09 1.27 0.34 0.65 0.49
compared to 1.35 and 0.95 for the individual delta–gamma and GBP 1.56 5.39 0.97 1.61 0.18 0.45 0.31
EQW 1.51 7.29 0.68 2.22 0.24 0.39 0.31
delta–vega neutral strategies respectively. In addition, Fig. 5 pro-
vides another way of presenting the combined results by plotting Note: This table shows summary statistics for a 50% delta–vega and 50% delta–
the cumulative wealth curves for the two individual strategies as gamma neutral mixed strategy (Panel A) and the break-even spread for the delta–
gamma neutral, delta–vega neutral and mixed strategies respectively (Panel B) for
well as the 50/50 combination. It is clear that all three curves show
the 3 vs 12-month strategy. The break-even spread is defined as the bid-offer
a positive drift. The combination, however, shows a further implied volatility spread that makes the strategy unprofitable and expressed as
smoothing of the returns. implied volatility points.
To study the left tail risks in the strategies’ returns, we study the
worst loss in any losing period during the historical simulation. cumulative return. We analyze the worst three strategy draw-
This measure is called the maximum drawdown and is defined as downs for the various countries and maturities in our data set.
the percent retrenchment from a peak-to-trough decline in the We focus on the 3 vs 12-month strategy, with the results of the
140
Vega neutral Gamma neutral 50/50 combination
135
130
125
120
115
110
105
100
95
90
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
Fig. 5. Wealth curve swaption riding strategy. This figure shows the wealth curve of a long-short straddles strategies with swaption maturities of 3 months (short) and
12 months (long) respectively. Portfolio returns are the equally weighted (EQW) average returns for the USD, EUR, GBP and JPY markets.
other maturities generally being in line with this strategy. Table 14 dealers make a credit value adjustment (CVA) in the pricing of a
presents the maximum drawdown and the time between the strat- transaction to reflect the counterparty credit risk in uncollateral-
egy’s retrenchment until a new high is reached. For the vega neu- ized transactions. On the other hand, fully collateralized transac-
tral strategy, which is subject to jump risk, we observe large losses tions will rarely be subject to default risk and therefore CVA will
that coincide with the default of Russia in 1998 and the default of be close to zero in these transactions. Second, banks became reluc-
Lehman Brothers in 2008. For the gamma neutral strategy, which is tant to lend to one other after the default of Lehman Brothers, and
subject to volatility risk, we observe the largest drawdowns during the subsequent liquidity squeeze made funding difficult and costly.
the steady decline of worldwide volatilities during 2004–2007. The For quite a while, derivative dealers were faced with a large gap
50/50 mix of the two strategies has much smaller drawdowns. The between the funding costs of their institution and the risk-free rate
largest drawdown of the equally weighted country portfolio is less in the option pricing model for trades that were not collateralized.
than 1%, which is less than a seventh of the largest drawdown that This resulted in derivative dealers charging a funding value adjust-
occurred in the gamma neutral strategy in the US. ment (FVA) to recover their funding costs. From Hull and White
Considering the economic significance we look at the (2014) we know that the inclusion of FVA is a controversial issue
break-even bid-offer spread for the 3 vs 12-month swaption matu- that has resulted in much discussion between practitioners, aca-
rity strategy. The break-even spread is defined as the bid-offer demics and accountants. There are no simple solutions to the use
implied volatility spread that makes the strategy unprofitable of FVA because an FVA violates the law of one price in the market
and is expressed as implied volatility points. Panel B in Table 13 and can lead to conflicts between accountants and traders. Hull
reports the break-even bid-offer spreads for the delta–gamma, and White (2014) conclude that ‘‘an FVA is justifiable only for
delta–vega and combined strategies. The result shows that the the part of a company’s credit spread that does not reflect default
break-even spreads for the equally weighted portfolio are 0.24, risk.’’ Overall, this gives rise to the question whether the profitabil-
0.39 and 0.31 volatility points for the delta–gamma, delta–vega ity of our strategy is impacted by collateralized transactions and
and 50/50 mix strategies, respectively. These break-even spreads potential tensions on funding. We argue that the impact will be
are within the bid-offer spreads in the market, which typically is limited because in our long-short strategy both collateral as well
0.50 volatility points according to two major broker-dealers. This as funding exposures will largely be set-off against one another.
leads us to conclude that, taking into account trading costs, the To conclude, a remark on the economic importance should be
returns of the delta–gamma, delta–vega and 50/50 mix strategies made. Transaction costs have typically not been included in related
are not realizable by investors and therefore are not economically literature on the volatility risk premium in general, nor interest
significant. This corroborates the findings for equity option strate- rate derivatives specifically. For example, Trolle (2009) and
gies obtained by Santa-Clara and Saretto (2009). Doran (2007) both indicate that more research in the direction of
The Global Financial Crisis and subsequent monetary policy including trading and commission costs in the implementation
decisions by central banks have caused derivative dealers to could be done in future.14
change their dealing practices. This might affect the profitability Interestingly, our strategy framework motivates us to consider
of our strategy. In particular, credit and liquidity risks are now rec- the economic importance of our results. Based on the results in this
ognized as having an impact on the economic value of a derivative section we conclude that the returns of the delta–gamma and
security and have changed the manner in which derivative trades delta–vega neutral strategies are not realizable by investors.
are conducted. First, the default of Lehman Brothers showed that
counterparty credit risk cannot be ignored. Counterparty credit risk 14
To our knowledge, the only exception is Duarte et al. (2007) who report returns
in a derivative trade should be carefully managed and either be that are statistically and economically significant for their strategy of selling interest
priced or mitigated with the help of collateral. Today, derivative rate volatility through delta-hedged caps.
Table 14
Maximum drawdowns for the 3 vs 12-month strategy.
Gamma neutral Vega neutral 50/50 combination

Drawdown (%) #Months Start Drawdown (%) #Months Start Drawdown (%) #Months Start
USD Largest 7.08 54 Mar 2004 4.54 12 Oct 2008 2.94 32 Jun 2005
2nd largest 3.61 10 Nov 2009 4.42 21 Apr 2003 2.04 14 Nov 2009
3rd largest 2.06 24 Oct 1996 1.93 6 Jul 1998 1.94 15 Mar 2004
EUR Largest 2.21 40 Jun 2005 3.21 24 Jun 1998 1.52 25 Nov 2006
2nd largest 1.94 15 Oct 2001 2.61 28 Nov 2006 1.20 18 Oct 2001
3rd largest 1.84 11 Aug 2010 2.34 17 Jul 2010 0.78 10 Dec 2005
JPY Largest 2.81 32 May 2000 5.17 24 Apr 1998 1.50 5 May 1999
2nd largest 2.36 28 Jun 2006 2.74 10 Feb 2003 1.40 16 May 2000
3rd largest 2.28 26 Oct 2009 1.78 20 Jul 1996 0.95 20 Jun 2006
GBP Largest 2.51 58 May 2003 3.24 14 Apr 2008 1.06 14 Nov 2006
2nd largest 2.12 19 Mar 1997 1.52 6 Jun 2011 0.90 11 Mar 1997
3rd largest 2.07 26 Aug 1999 1.41 16 Nov 2006 0.68 8 Jul 2000
EQW Largest 2.44 46 Mar 2004 2.05 8 Aug 2008 0.94 13 Oct 2006
2nd largest 1.74 9 Nov 2009 1.63 10 Apr 2003 0.67 5 Jun 2003
3rd largest 0.70 7 Dec 2001 1.59 16 Jul 1998 0.55 6 Nov 2009
Note: This table reports the three largest maximum drawdowns on the delta–gamma neutral, delta–vega neutral and mixed strategies for the 3 vs 12-month strategy. The
maximum drawdown is defined as the worst cumulative loss in any losing period in the historical simulation (April 1996 to December 2011) of the strategy. The length of the
drawdown (#Months) is the number of months between the strategy’s retrenchment until a new high is reached in the cumulative strategy returns.
However, the goal of this paper is to identify the existence of a dif- for equity option strategies obtained by Santa-Clara and Saretto
ference in the volatility risk premium across the term structure of (2009). This limit may prevent markets from taking advantage of
fixed income derivatives. Our results support this for all four swap- ‘riding the swaption curve’, but does not detract from our estab-
tion markets. lished conclusion that there is a statistically significant difference
in the volatility risk premium across the swaption maturity term
7. Conclusion structure.
Since the maturity effect of the volatility risk premium seems to
Existing work has demonstrated that the volatility risk pre- have received limited attention in the literature we encourage
mium is more negative for short-term maturities than for longer future research. A prominent issue that requires attention concerns
maturities. However, while the existence of the volatility risk pre- the changes in the swaption market after the Global Financial
mium in the fixed income market has recently been analyzed by Crisis. Prior to the crisis, swaption dealers relied on a single curve
various papers, the maturity effect was only documented by to forecast rates depending on an underlying index, eg LIBOR or
Fornari (2010) and Mueller et al. (2013) and, to our knowledge, Euribor, and to discount cash flows. The changes in market condi-
has not been analyzed in detail. Our paper contributes to the liter- tions and regulations have resulted in a different method for how
ature by providing a strategy framework to test and analyze the swaptions are valued and risk is managed. In particular, a new
maturity effect in the volatility risk premium in fixed income mar- multi-curve pricing framework that uses separate forecasting and
kets. Specifically, we analyze the returns of two long-short straddle discounting curves became market-standard for new swaption
strategies which both ‘ride’ the swaption curve. The straddle com- trades. Assuming mutual collateral agreements, the market has
binations are either delta–vega neutral and subjected to jump risk, evolved toward discounting future cash flows using Overnight
or delta–gamma neutral and subjected to volatility risk. Index Swap (OIS) rates. The separation between the index rates
Using a large database (April 1996–December 2011) of implied and the OIS rates fundamentally changed the framework for swap-
volatility quotes of the four major bond markets, we find statisti- tion modeling. During the crisis the market slowly transitioned
cally significant returns, which incrementally decrease in swaption from the single curve methodology to the multi-curve methodol-
maturity for all markets. This finding is consistent with a concave, ogy. For example, in September 2010 one of the leading
upward-sloping term structure in the volatility risk premium. The international swaption dealers, ICAP, switched to OIS discounting
fact that both delta–vega and delta–gamma neutral strategies earn and published both LIBOR- and OIS-based swaption implied
positive returns that seem uncorrelated suggests that the term volatilities. In January 2012 ICAP stopped publishing LIBOR-based
structure of the volatility risk premium is affected by both jump volatilities and has since then only published OIS-based volatilities
risk and volatility risk. This effect seems most pronounced at (Bianchetti and Carlicchi, 2013). January 2012 coincides exactly
shorter maturities. Additional robustness analysis indicates that with the end of the empirical data set of this study. Further
the results seem robust when using the Vasicek (1977) and research using OIS-based swaption implied volatilities data post
Hagan et al. (2002) SABR models instead of the Black model. The 2011 could give important complementary evidence on the
results are also robust for macroeconomic announcements, consistency of our strategy. This investigation, however, lies
although we do observe a small effect for the day-of-the-month. beyond the scope of this paper.
Specifically, delta–vega neutral portfolios initiated around the turn
of the month report higher returns than mid-month portfolios. Acknowledgments
Finally, our framework allows more detailed assessment of the
economic importance. Transaction costs and margin requirements We thank Robert Berry for excellent research assistance. We are
are typically not included in related literature on the volatility risk grateful to the participants of the 2008 2nd Investing and Trading
premium. Our break-even cost analysis points to the conclusion in Volatility conference in New York, the 2012 15th Annual
that the returns of the delta–gamma and delta–vega neutral strate- Conference of the Swiss Society for Financial Market Research in
gies are not realizable by investors. This corroborates the findings Zurich, the 2013 EFA Meetings in Cambridge, seminar participants
at Robeco, Guido Baltussen, Kees Bouwman, Nick Firoozye, Duarte, J., Longstaff, F.A., Yu, F., 2007. Risk and return in fixed-income arbitrage:
nickels in front of a steamroller? Review of Financial Studies 20, 769–811.
Winfried Hallerbach, Roy Hoevenaars, Roger Lord, Kasper
Dungey, M., McKenzie, M., Smith, V., 2009. Empirical evidence on jumps in the term
Lorenzen, Martin Martens, Philippe Mueller (discussant), structure of the US Treasury market. Journal of Empirical Finance 16, 430–445.
Sebastian Paik (discussant), Patrick Verwijmeren, Yu-Min Yen Dyl, E.A., Joehnk, M.D., 1981. Riding the yield curve: does it work? Journal of
and three anonymous referees for useful comments and sugges- Portfolio Management 7, 13–17.
Fornari, F., 2010. Assessing the compensation for volatility risk implicit in interest
tions. The views expressed here are those of the authors them- rate derivatives. Journal of Empirical Finance 17, 722–743.
selves and should in no way be attributed to their employers. Goodman, L., Ho, J., 1997. Are investors rewarded for shorting volatility? Journal of
Fixed Income 7, 38–42.
Hagan, P., Kumar, D., Lesniewski, A., Woodward, D., 2002. Managing smile risk.
Witmott Magazine 1, 84–108.
References Hull, J., White, A., 2014. Valuing derivatives: funding value adjustments and fair
value. Financial Analysts Journal 70, 46–56.
Aït-Sahalia, Y., Karaman, M., Mancini, L., 2012. The term structure of variance Johannes, M., 2004. The statistical and economic role of jumps in continuous-time
swaps, risk premia and the expectation hypothesis, Princeton University, interest rate models. Journal of Finance 59, 227–260.
Working paper. Lee, S.S., Mykland, P.A., 2008. Jumps in financial markets: a new nonparametric test
Almeida, C., Vicente, J., 2009. Identifying volatility risk premia from fixed income and jump dynamics. Review of Financial Studies 21, 2536–2563.
Asian options. Journal of Banking and Finance 33, 652–661. Levin, A., 2004. Interest rate model selection. Journal of Portfolio Management 30,
Andersen, T., Bollerslev, T., Diebold, F., Vega, C., 2007. Real price discovery in stock, 74–84.
bond, and foreign exchange markets. Journal of International Economics 73, Li, H., Song, Z., 2013. Jump tail risk in fixed income markets, Working paper.
251–277. Low, B.S., Zhang, S., 2005. The volatility premium embedded in currency options.
Bakshi, G., Kapadia, N., 2003. Delta-hedged gains and the negative market volatility Journal of Financial and Quantitative Analysis 40, 803–832.
risk premium. Review of Financial Studies 16, 527–566. Martellini, L., Priaulet, P., Priaulet, S., 2003. Fixed Income Securities: Valuation, Risk
Bartlett, B., 2006. Hedging under SABR Model. Witmott Magazine 24, 2–4. Management and Portfolio Strategies. John Wiley & Sons Ltd., Chichester (West
Bianchetti, M., Carlicchi, M., 2013. Markets evolution after the credit crunch. In: Sussex).
Bianchetti, M., Morini, M. (Eds.), Interest Rate Modelling after the Financial Mele, A., Obayashi, Y., 2012. CBOE Interest Rate Swap Volatility Index, SRVX White
Crisis. Risk Books, London. Paper. Chicago Board Options Exchange.
BIS, 2003. International banking and financial developments, Quarterly review Mele, A., Obayashi, Y., 2013. Volatility indices and contracts for government bonds
September. and time deposits, University of Lugano, Working paper.
Black, F., 1976. The pricing of commodity contracts. Journal of Financial Economics Merener, 2012. Swap rate variance swaps. Quantitative Finance 12, 249–261.
3, 167–179. Mueller, P., Vedolin, A., Yen, Y.-M., 2013. Bond variance risk premiums, London
Branger, N., Schlag, C., 2008. Can tests based on option hedging errors correctly School of Economics, Working paper.
identify volatility risk premia? Journal of Financial and Quantitative Analysis Newey, W., West, K., 1987. A simple, positive semi-definite, heteroscedasticity, and
43, 1055–1090. autocorrelation consistent covariance matrix. Econometrica 55, 703–708.
Broadie, M., Chernov, M., Johannes, M., 2009. Understanding index option returns. Newey, W., West, K., 1994. Automatic lag selection in covariance matrix estimation.
Review of Financial Studies 22, 4493–4529. Review of Economic Studies 61, 631–653.
Carr, P., Wu, L., 2009. Variance risk premiums. Review of Financial Studies 22, 1311– Rebonato, R., McKay, K., White, R., 2009. The SABR/LIBOR Market Model: Pricing,
1341. Calibration and Hedging for Complex Interest-Rate Derivatives. John Wiley &
Chan, K.C., Karolyi, G.A., Longstaff, F.A., Sanders, A.B., 1992. An empirical comparison Sons Ltd., Chichester (West Sussex).
of alternative models of the short-term interest rate. Journal of Finance 47, Santa-Clara, P., Saretto, A., 2009. Option strategies: good deals and margin calls.
1209–1227. Journal of Financial Markets 12, 391–417.
Chaput, S., Ederington, L., 2005. Volatility trade design. Journal of Futures Markets Trolle, A.B., 2009. The price of interest rate variance risk and optimal investments in
25, 243–279. interest rate derivatives, Swiss Finance Institute, Working paper.
Cremers, M., Halling, M., Weinbaum, D., 2015. Aggregate jump and volatility risk in Trolle, A.B., Schwartz, E.S., 2014. The swaption cube. Review of Financial Studies 27,
the cross-section of stock returns. Journal of Finance 70, 577–614. 2307–2353.
Doran, J., 2007. The influence of tracking error on volatility risk premium Vasicek, O., 1977. An equilibrium characterisation of the term structure. Journal of
estimation. Journal of Risk 9, 1–36. Financial Economics 5, 177–188.
Duarte, J., 2008. The causal effect of mortgage refinancing on interest rate volatility: Whaley, R., 1993. Derivatives on market volatility: hedging tools long overdue.
empirical evidence and theoretical implications. Review of Financial Studies 21, Journal of Derivatives 1, 70–84.
1689–1731.

Riding Swaptions Curve

Uploaded by

Copyright:

Available Formats

Riding Swaptions Curve

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Riding Swaptions Curve

Uploaded by

Copyright:

Available Formats

Journal of Banking & Finance 59 (2015) 57–75

Contents lists available at ScienceDirect

Journal of Banking & Finance

Riding the swaption curve

1-day yield change 10-year US swap (bps) 20

(a) Realized jumps defined as 1% largest absolute changes

(b) Realized jumps according to Lee and Mykland (2008)

Excl. Jumps (1%) Excl. Jumps (LM2008) Excl. DVol (5%)

Panel B: Delta–Vega neutral combination

(a) Vega – vega neutral (b) Vega – gamma neutral

-50 Gamma -50

(c) Gamma – vega neutral (d) Gamma – gamma neutral

(e) Delta – vega neutral (f) Delta – gamma neutral

Gamma neutral Vega neutral

b0 (%) t-stat b1 t-stat b2 t-stat b3 t-stat b4 t-stat R2

(a) Delta-gamma neutral strategy

(b) Delta-vega neutral strategy

Table 9 5.4. Swaption smile and 2-year tenor

Delta neutral Vega neutral

Gamma neutral Vega neutral

Gamma neutral Vega neutral

Gamma neutral Vega neutral 50/50 combination

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.