Expected 1DTE Option Returns

Michael Johannes, Andreas Kaeck, Norman Seeger and Neel Shah∗

This draft: February 14, 2024

First draft: December 2023

Preliminary and incomplete

Abstract

This paper analyzes short-dated one-day-to-expiry (1DTE) index option re-

turns. Theoretically, we provide new analytical tools to calculate higher-order
moments of option returns for models with known (or easily calculable) charac-
teristic functions and the impact of macroeconomic event risk on option return
distributions for hold-to-maturity option returns. Analytical moments are useful
for understanding observed option returns, for evaluating statistical significance,
and for GMM-based estimation. Empirically, we analyze 1DTE option returns
and document that raw call and put returns are not statistically significant on
most days but are highly significant on macroeconomic announcement days. We
also analyze variance risk premiums and fit models to observed option returns
to understand the underlying risk factors and premia.

∗
Johannes and Shah are at the Graduate School of Business, Columbia University ([email protected]
and [email protected]). Kaeck is with the University of Sussex Business School ([email protected]),
and Seeger is with VU Amsterdam ([email protected]).

Electronic copy available at: https://ssrn.com/abstract=4729757

 In the past few years, trading in very short-dated options, such as those with a day or
less to maturity, has grown dramatically. Although very short-dated options have always
existed (e.g., as time-to-maturity falls for traditional monthly expiries), researchers routinely
discarded these options when analyzing pricing and returns, likely due to a lack of trading
activity and high-quality quote data.1 With daily expirations, these now highly liquid
options are of interest for many reasons as extremely short-dated options provide a unique
view into pricing of short-dated risks as well as investor behavior. In this paper, we study
1DTE option returns, with both theoretical and empirical contributions.
We focus on 1DTE options, in contrast to a growing literature analyzing zero day-to-
expiry (0DTE) options, for a number of reasons. First, many of the most important market
movements occur outside normal trading hours and thus 0DTE samples. For example,
during the financial crisis, the only ’lock limit down’ move in S&P 500 futures occurred pre-
market.2 During COVID, the largest crash since 1987 occurred on 3/16/2020, with most of
the move occurring overnight.3 These large, economically important moves aren’t present
in analyses of 0DTE options.4 Second, short-dated options load heavily on macroeconomic
announcements, and most of these economic announcements also occur outside of normal
trading hours and time frames used to analyze 0DTE options. These include employment,
inflation, GDP and important non-US announcements. The main announcements during
trading hours are the FOMC and ISM services/manufacturing.5
On the theoretical side, we provide two main contributions. First, we provide new
analytical tools to calculate model-based higher moments of option returns. As noted by
Broadie et al. (2009), expected hold-to-maturity option returns can be calculated analytically
for specifications permitting characteristic function based approximate pricing. We extend
these results and show how to calculate option return higher moments for models with
1
For papers either analyzing pricing or option returns, short-dated options are almost always excluded.
Bates (2000) includes options with maturities greater than one week, Pan (2002) uses options with maturities
greater than 15 days, and Bakshi et al. (1997) uses options with more than 6 days to maturity. In studies
of option returns, Broadie et al. (2009) and Hu and Jacobs (2020) use one-month option returns, Coval and
Shumway (2001) analyze weekly returns for options with more than 20 days to maturity, and Constantinides
et al. (2013) analyze daily returns for options with more than 7 calendar days to maturity.
2
https://www.nytimes.com/2008/10/24/business/worldbusiness/24iht-24future.17223870.
html
3
https://www.wsj.com/articles/the-day-coronavirus-nearly-broke-the-financial-markets-
11589982288
4
Close-to-close returns are also more negatively skewed and have higher kurtosis than open-to-close
returns, two characteristics that are particularly important for option prices.
5
See Johannes et al. (2023) for an analysis of FOMC event risk.

Electronic copy available at: https://ssrn.com/abstract=4729757

tractable characteristic functions.6 These analytical higher moments are useful for many
applications, including GMM-based estimation using option returns and for understanding
option return volatility. Option return volatility is crucial for evaluating option trading
strategies and the statistical significance of realized option returns. Both of these can be
difficult to estimate do in observed samples due to the sensitivity to, and importance of,
rare outlier events. This issue is particularly salient for testing option return significance, as
noted in Broadie et al. (2009). As an alternative to Broadie et al. (2009)’s simulation-based
approach, we develop a more efficient analytical approach using model-based moments for
hypothesis testing. We also use these analytical moments for GMM-based estimation of
structural models using short-dated options.
Model-based distributions are useful as sampling biases are likely even more inherent
and acute for short-dated than longer-dated options. While it is clear that events like the
Crash of 1987 are not in either 0DTE or 1DTE samples, many recent market movements
are also not in either of these samples due to the peculiarities of when daily expiries were
introduced. For example, the Flash Crash of 2010 occurred on a Thursday, before the
introduction of Thursday expiries in 2022. Similarly, the market crash on 3/12/2020, which
at the time was the largest single-day drop since 1987, occurred on a Thursday prior to the
introduction of Thursday expiries. These examples highlight biases in observed samples of
very short-dated option returns, as there are important days in broader populations that
aren’t included in typical 0DTE or 1DTE samples. Our model-based population expected
returns, volatilities, and standard errors account for events that occur with low probabilities
in the model but have an important impact on option return moments.
Our second theoretical contribution is to understand the impact of macroeconomic
announcements or event risk on option returns. Macroeconomic event risk introduces two
important effects on underlying index returns: both average stock index returns and volatili-
ties are higher on announcement days. Savor and Wilson (2013) find that stock index returns
are dramatically higher on announcement days, something we confirm in our more recent
sample.7 Macroeconomic announcements generate significant excess volatility at high fre-
quencies, though daily index return volatility on announcement days is only slightly higher
6
Boyer and Vorkink (2014) calculate conditional skewness using properties of truncated log-normal ran-
dom variables. Our approach is more general as it utilizes characteristic functions.
7
For the SPX index post-2012, average SPX index returns are more than 5 times greater on announcement
days.

Electronic copy available at: https://ssrn.com/abstract=4729757

than on non-announcement days. These effects generally have offsetting effects on expected
option returns. Expected option call/put returns are higher/lower with a higher event risk
premium, but option returns are always smaller in absolute value as volatility (event risk)
increases (holding all else constant). This implies that option returns are fundamentally
different for announcement and non-announcement days.
Empirically, we investigate realized 1DTE option returns and their determinants using
high-quality intraday quote data starting in 2012.8 We calculate 1DTE hold-to-maturity
(HTM) returns from 3:55 ET the day prior to expiry, using quote midpoints and report
summaries in log-moneyness buckets.9 As a benchmark for statistical significance, we use the
null of the Black-Scholes model with t-statistics constructed using Black-Scholes expected
returns and volatilities (to construct model-based standard errors). For calls and puts in
the post-2012 sample, while most average returns are statistically significant at the 5% level
(|t| > 1.65) using traditional sample based standard errors, none are significant using t-
statistics constructed using Black-Scholes expected returns and return volatility, with the
exception of deep OTM calls.
We calculate average option returns separately for days with and without macroeco-
nomic announcements and find striking differences: almost all put and call strikes are highly
statistically significant (using Black-Scholes t-statistics) on macroeconomic announcement
days, with the exception of the deepest OTM calls and puts. For example, ATM average
1DTE call returns are 22.15% on days (and highly statistically significant) with macroe-
conomic announcements but only 2.62% on non-announcement days. Average call returns
that are one standard deviation OTM are 60%, also highly significant. Average OTM put
returns on macroeconomic days are quite negative and highly significant. The economically
large (in absolute value) call and put returns are driven by the significant realized macroe-
conomic risk premia on these days and consistent with the model of macroeconomic event
risk described above. These effects of macroeconomic return premia on option returns have
not been previously documented in the option literature and also have important impli-
cations for analyses of longer-dated options using daily returns.10 The striking differences
8
As noted in Rhoads (2014), although weekly Friday SPX options were introduced in 2005, trading
volumes increased dramatically in 2011. See also Bandi et al. (2024), Vilkov (2024), Almeida et al. (2024)
and Dim et al. (2024) for update analyses of trading volumes.
9
The results are similar using fixed strikes.
10
For example, it is common to analyze returns to option positions rebalanced daily, often with delta-

Electronic copy available at: https://ssrn.com/abstract=4729757

between announcement and non-announcement days highlight the first-order importance of
macroeconomic event risk for understanding short-dated options.11
We also analyze variance risk premium (VRP) trades using differences between the
observed realized variance and its risk-neutral expectation, generically variance swaps.12
We analyze variance swaps for various corridors, including left and right tails separately.
We find highly significant 1DTE VRPs for the full sample, with corridor returns indicating
a much larger and more significant VRP in the left tail. VRPs are significant for days
with and without both macroeconomic announcements. Left tail premiums are much higher
on macroeconomic days. Right tail premiums are interestingly insignificant on days with
macroeconomic announcements, highlighting the differential tail properties. These empirical
findings are also in line with our macroeconomic announcement model. An alternative way to
measure risk premiums is via straddle and strangle returns. Straddle returns are insignificant
for days with or without macroeconomic announcements and for variance subsamples. The
variance risk premium is largely driven by OTM put options.
To understand the drivers of option returns, we use our analytical moments in a two-
stage GMM procedure to estimate an extension of Merton’s jump-diffusion model augmented
with macroeconomic risk. The GMM procedure uses multiple sets of moments. The first
set uses short-dated call and put option returns moments for a range of strikes, for days
with and without macroeconomic announcements. These moments provide information on
both the real-world measure P and the risk-neutral measure Q. The second set uses implied
volatility smiles for days with and without macroeconomic events. The final set uses the
first four moments from stock index returns, providing information on the P measure. We
use this model quantify jump and event-risk premia.
hedging. The macroeconomic return premiums will have important implications for understanding these
option returns, as realized and expected returns are systematically different on macroeconomic announce-
ment days. See, e.g., Muravyev and Ni (2020).
11
See Johannes et al. (2023) for an analysis of FOMC announcements.
12
Vilkov (2024) and Almeida et al. (2024) analyze VRPs for 0DTE options, and Londono and Samadi
(2023) analyze variance risk premiums for macroeconomic event risk for longer-dated options, those with
more than 7 days to maturity.

Electronic copy available at: https://ssrn.com/abstract=4729757

1 Option Returns
Option returns combine ex-ante pricing of index returns under the risk-neutral measure Q
and realized P-measure returns. To see this, consider the Black-Scholes model where stock
index values, St , evolve under P via

dSt = µSt dt + σSt dBtP ,

where BtP is a Brownian motion under P, and µ and σ are the drift and volatility coefficients.
Consider a call option on St at time t, struck at K, and maturing at time T with the
price denoted as C = Ct,T (St , K). By combining the option price evolution under P using
Ito’s lemma and the pricing PDE, which captures Q-measure dynamics, the instantaneous
change in the option’s price is

dB P .
 
∂C ∂C
dC = rC + (µ − r) St dt + σSt
∂St ∂St t

These continuous-time representations highlight the determinants of option returns (the

drift, diffusion, risk-free rates and the option’s elasticity) and motivate factor models of
option returns. For example, instantaneous expected excess call returns are

EPt
 
dC 1 ∂C
− rdt = (µ − r) dt,
C C ∂St

which implies that expected call/put returns are always positive/negative, with magnitudes
determined by the equity premium and the option’s elasticity. While these representations
are useful for intuition, the conditional linearity in “dBtP ” and normality of instantaneous
option returns doesn’t hold generally, even in continuous-time models with crash-risk and
certainly doesn’t hold over the non-infinitesimal holding periods, even for 1DTE options.
Because of this, the literature on option returns often focuses on hold-to-maturity
(HTM) returns. For short-dated expiries, this is particularly the case; see, e.g., Beckmeyer
et al. (2023), Vilkov (2024) and Almeida et al. (2024). In the next section, we develop new
theoretical tools to analyze conditional moments of HTM option returns, not only for very
short-dated call or put options but also for HTM option returns for any maturity and also
for common option strategies.

Electronic copy available at: https://ssrn.com/abstract=4729757

 For a call option, the HTM return is given by

c [ST − K]+ [ST /St − K/St ]+ [Rt,T − K/St ]+

rt,T = −1= −1= − 1,
Ct,T (St , K) Ct,T (1, K/St ) Ct,T (1, K/St )

where Rt,T = ST /St is the gross returns of the underlying St and [x]+ = max [x, 0].13 The
second equality holds given the homogeneity of option prices and highlights the nonlinear re-
lationship between realized gross asset returns and option returns. HTM put option returns
p
are defined as rt,T = [K − ST ]+ /Pt,T (St , K) − 1 where Pt,T (St , K) is the put option price.
HTM option returns are generally highly non-normal, with the degree of non-normality de-
termined by the underlying asset return characteristics (e.g., volatility, jump risk, etc.) and
contract characteristics like maturity and moneyness.
As mentioned earlier, characterizing option return distributions is difficult for many
reasons, including the idiosyncrasies of option samples and the impact of extreme outliers.
Compared to historical analyses of stock returns, option samples are always significantly
smaller as stock options were only introduced in the 1970s (e.g., no observations during the
Great Depression), and common databases like OptionMetrics started in 1996. This implies
that index option return analyses omit the largest historical outlier, the Crash of 1987.14
Even accounting for events like the Crash of 1987, Broadie et al. (2009) show that average
1-month HTM option returns are extremely noisy, especially for OTM strikes, due to a small
number of outlier returns combined with the convexity of option payouts.
It is important to note that these sampling issues have persisted and even been ex-
acerbated by events in the past few years. For example, the largest observed SPX index
OTM put returns occurred during the Covid outbreak, at least for longer-dated options,
which have dramatically changed average put returns.15 These issues are particularly true
for short-dated options, which not only have shorter samples but also had a staggered intro-
duction of different daily expiries, with Fridays introduced in 2005, Mondays/Wednesdays
in 2016, and Tuesdays/Thursdays in 2022. As mentioned above, the staggered expiry in-
troduction of daily expiries implies that important dates are missing from 1DTE or 0DTE
option samples, creating important biases. The next section describes new model-based
13
See Nawalkha and Zhuo (2022) for finite holding-period option returns.
14
See, e.g., Dew-Becker and Giglio (2023) for an analysis of historical synthetic option prices.
15
During March 2020, the 5% OTM SPX put had a 1-month HTM return of greater than 4500% and 10%
OTM put had a return of more than 10,000%, both multiples higher than any previously observed.

Electronic copy available at: https://ssrn.com/abstract=4729757

tools for characterizing HTM option return distributions.

2 Option Return Moments – General Framework

It is common to use model-based or population characteristics to understand option return
distributions and to evaluate statistical significance. Monte Carlo simulations are most com-
monly used, with option returns simulated from benchmark models like Black-Scholes or a
stochastic volatility model with crash risk.16 Another approach is simulations by resampling
or bootstrapping existing returns (see, e.g., Kelly et al. (2016) or Almeida et al. (2024)). On
the analytical side, it is also well known that expected HTM option returns (first moments)
can be calculated analytically in models with known characteristic functions (see Broadie
et al., 2009).17 This section extends these analytical results in two directions. First, we
derive put and call HTM option return moments of any order for any model that provides
a known (or easily computable) characteristic function of the log price of the underlying.
Second, we extend these analytical results to any static option portfolio, such as straddles,
strangles, or option spreads.
To relate our results to the literature, consider a general continuous-time model with a
known log price characteristic function, both under the risk-neutral measure Q and under
the real-world measure P. The extended characteristic functions of the log index value are
defined (where finite) as

M eu log ST  ,
t,T (u, St , Lt ) = Et
ΨM

where M ∈ {Q, P}, Lt are latent state variables (e.g., stochastic volatility), u ∈ C (the set
of complex numbers), and EM
t [·] denotes the time-t conditional expectation under measure

M. The characteristic function also depends on structural parameters (which may differ
under the two probability measures) and whose dependence is suppressed for simplicity.
The key to deriving expected option returns is the fact that the numerator and denom-
inator in the expected option return expression are similar. The denominator, the option
16
See, e.g., Broadie et al. (2009), Branger et al. (2010), Hu and Jacobs (2020), Israelov and Kelly (2017),
or Almeida and Freire (2022).
17
For the special case of Black-Scholes, Boyer and Vorkink (2014) calculate second and third-order mo-
ments using the properties of conditional log-normal random variables.

Electronic copy available at: https://ssrn.com/abstract=4729757

price, is the discounted Q-measure expectation of the payoff, whereas the numerator is the
expected payoff under P. Using the derivations in Bates (2006), analytic expressions for
expected payoffs are given by

M M [ST − K]+
t,T (K, St , Lt ) := Et
Λc,
Z∞ " −iu log K M
 # 
= ΨM
e Ψ (iu, S , L )
t,T (1, St , Lt ) − K
1 + 1 ℜ t,T t t
du ,
2 π iu(1 − iu)
0

√
where ℜ [z] denoted the real part of a complex number z and i = −1. Given these
moments, expected call option returns are given by

P
EPt rt,T
Λc,
t,T (K, St , Lt )
Q
c
 
= − 1,
e−r(T −t) Λc,
t,T (K, St , Lt )

where r is the risk-free rate (assumed constant). Similar expressions for expected put option
returns are available.
To analyze higher moments, we follow a slightly different approach, which easily gener-
alizes also to portfolios of options. The key step is the fact that higher-order return moments
can be decomposed using the Binomial formula. The following proposition shows how to
calculate the n-th raw moment of put and call option returns using the mild assumption
that the corresponding moment of the stock price process exists. The result is general and
applies to any pricing model with a known characteristic function under P and Q.

Proposition 1. Suppose the underlying S has finite moments up to n-th order, and the
characteristic function of log S is known in closed form under the risk-neutral measure Q
and under the real-world measure P. Then, the n-th raw moment of the hold-to-maturity
call return is given by

EPt
n n   

n  X X k n
c n
Γct,T (ℓ) Ct,T (St , K)−k (−1)n−k (−K)k−ℓ

rt,T = (−1) + ×
ℓ=0
ℓ k
k=max[ℓ,1]

where

Z∞  −iu log K
Γct,T (ℓ) = ΨP P

1 1 e
t,T (ℓ, St , Lt ) + ℜ Ψt,T (iu + ℓ, St , Lt ) du.
2 π iu
0

Electronic copy available at: https://ssrn.com/abstract=4729757

The n-th raw moment of the hold-to-maturity put option return is given by

EPt
n n   
p

n  X X k n
n
Γpt,T (ℓ) Pt,T (St , K)−k (−1)n−k K k−ℓ

rt,T = (−1) + ×
ℓ=0
ℓ k
k=max[ℓ,1]

where

Γpt,T (ℓ) = (−1)ℓ ΨP ℓ c

t,T (ℓ, St , Lt ) − (−1) Γt,T (ℓ).

Proof. See Appendix A.1.

There are a few notables at this stage. First, calculating the n-th moment requires
n + 2 numerical integrations. For the first moment, the calculation requires three numerical
integrations (one for the option price and one for ℓ = 0 and ℓ = 1 each). This compares
to only two numerical integrations using the results provided above. This approach mirrors
the early derivations of option prices as in Heston (1993) and others, which also lead to
an additional integral. The advantage of this formula is that when calculating higher-order
moments, as Γct,T (ℓ) is not a function of n, these calculations can be used for all the moments
and hence the computational increase for every additional moment is only one additional
integration for every additional moment. Second, the proof of the proposition can be easily
adjusted to other quantities, such as the moments of the P&L distribution, or when option
payoffs are normalized by other quantities, such as the underlying stock price rather than
the option price. Third, as shown in the following paragraph, our approach can be extended
to option portfolios.
To extend these results to returns on option portfolios, consider a portfolio consisting
of a risk-free zero bond investment (with fixed payoff B ∈ R at maturity T ) and call options
with different strikes, allowing for a zero strike price. This assumption is without loss of
generality, as with this extension, we can form portfolios of the bond, stock index (using a
strike zero call) and call options, which allows us to include put options by put-call parity.
We use U to denote the normalization constant in the option portfolio return; this is usually
the price of the portfolio but could be set to any other positive constant, especially in the
case of zero-cost portfolios in which the return is not defined. More specifically, we derive

Electronic copy available at: https://ssrn.com/abstract=4729757

moments of
Pm
o B+ i=1 ci [ST − Ki ]+
rt,T = −1
U

where Ki ≥ 0 are the strikes of call options i = 1, . . . , m, ci denotes the holding of the i-th
option and U > 0 is known at time t.

Proposition 2. Suppose the underlying S has finite moments up to n-th order. The n-th
conditional return moment of a maturity T portfolio consisting of a bond investment (payoff
B ∈ R at time T ) and call options with strikes K = [K1 , . . . , Km ] ≥ 0 and option holdings
c = [c1 , . . . , cm ], with normalization constant U > 0, is given by

EPt (−1)n−q U −q B q−p × EP

q   
n X
 o

n  X n q p
rt,T = t [PC ]
q=0 p=0
q p

where

EPt [PCn ] =
m
!
X X X
α(n, k, j) × Γ K(k),
e ji
k∈K(n) j∈J (k) i=1
 Z∞  −iu log K
1 P P

 1 e
 Ψt,T (ℓ, Vt , Lt ) +
 ℜ × Ψt,T (iu + ℓ, Vt , Lt ) du if K > 0
Γ(K, ℓ) = 2 π iu
0
ΨP (ℓ, V , L )



t,T t t if K = 0
K(k)
e = max Ki
1≤i≤m, ki >0
m
Y m
Y
α(n, k, j) = γ(n, k)β(k, j) cki i (−Ki )ki −ji .
i=1 i=1

The sets K(n) and J (k) and the functions γ(n, k) and β(k, j) are defined in Appendix A.2.

Proof. The proof is provided in Appendix A.2 and follows from two Lemmas, which are also
proofed in Appendix A.2.

We remark that the main computational complexity for our higher-order moment for-
mula for portfolios is similar to option moments of vanilla options. The numerical integra-
tions required are of the same form, although there are now more of these depending on the
number of different strikes in the option portfolio.

10

Electronic copy available at: https://ssrn.com/abstract=4729757

2.1 Option Return Moments – Black-Scholes

To understand these option moments, we start with HTM returns in the Black-Scholes
model and then consider option return moments in a model augmented with event risk.
Expected option returns in the Black-Scholes model can be calculated using variants of the
Black-Scholes formula. For a call option, we have

EPt [ST − K]+

EPt rt,T
 

e−r(T −t) EQ
 c 
=  +
− 1.
t [ST − K]

The numerator can be expressed as

EPt [ST − K]+ = eµ(T −t) EPt e−µ(T −t) [ST − K]+ ,
   

with the expectation given by a modification of the Black-Scholes formula, replacing r with
µ:
EPt e−µ(T −t) [ST − K]+ = St N (d1,µ,σ ) − e−µ(T −t) KN (d2,µ,σ ),
 

where

ln(St /K) + µ + 21 σ 2 (T − t) ln(St /K) + µ − 21 σ 2 (T − t)

d1,µ,σ = √ and d2,µ,σ = √ .
σ T −t σ T −t

Combining, call option expected returns are given by

EPt rt,T
µ(T −t) −µ(T −t)
 
 c
 e St N (d 1,µ,,σ ) − e KN (d2,µ,σ )
= − 1,
St N (d1,r,σ ) − e−r(T −t) KN (d2,r,σ ),

which is a convenient formula. Similar formulas are available for puts as well as static
strategies like straddles, strangles, and spreads, all of which are easy to calculate.
To understand the quantitative implications for 1DTEs, Figure 1 shows expected Black-
Scholes option returns for a range of parameters. Expected call (put) option returns are
increasing in µ and generally decreasing in σ, and both can be quantitatively important.
Call (put) options are leveraged long (short) positions in the underlying, which explains
their dependence on expected returns.
Quantitatively, expected put returns for ATM 1-day options are small, approximately
-2% for the case with µ = 10% and σ = 25%, falling to about −5% when σ = 10%. These

11

Electronic copy available at: https://ssrn.com/abstract=4729757

Figure 1: Expected option returns: Varying underlying expected return and volatility
This figure provides 1-day-to-maturity expected option returns for a range of volatilities and expected stock
index returns. The top panels assume µ = 10% and r = 4%, and the bottom panels use the same interest
rate levels.
Expected call returns, various Expected put returns, various
=10% 1
8 =25%
=40% 2
6 3
4
4 5
6
2 =10%
7 =25%
8 =40%
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness
Expected call returns, various and =15% Expected put returns, various and =15%
16 =5% 0
14 =12.5% 2
12 =20%
4
10 6
8
8
6
4 10
=5%
2 12 =12.5%
14 =20%
0
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness

expected returns are magnified for OTM calls and puts, as well as in low volatility and high
expected parameterizations. While some of the parameters in Figure 1 may seem extreme,
very high values of implied volatility and a range of expected returns are reasonable as both,
in reality, vary significantly with time.
The Black-Scholes model is also useful for thinking about volatility risk premiums,
albeit in an ad-hoc manner. To get a sense of the quantitative implications of a variance risk
premium, consider a Black-Scholes model assuming that the diffusion coefficient is different
under the P and Q measures. Historical comparisons of the VIX index to realized variance
(on a volatility scale) imply that the difference between average Q and P-measure volatility
is approximately 2.5%, which would be generated by, for example, an average historical VIX
around 20% and realized volatility about 17.5%. We provide evidence below for substantial

12

Electronic copy available at: https://ssrn.com/abstract=4729757

1-day variance risk premiums. Johannes et al. (2023) find strong evidence for FOMC event
volatility risk premiums.
To see the implications, via this ad-hoc approach, Figure 2 compares expected option
returns assuming µ = 10% and r = 4%, σ = (17.5%, 35%), and with a variance risk premium
of 2.5%. In the top panels, total Q volatility is always 20%, with two cases: one with a 2.5%
variance risk premium, in which case P volatility is 17.5%, and a second case with no VRP
and P volatility also equal to 20%.18 The bottom panels have the same drift parameters,
but increase Q volatility to 40% and consider a case with a 5% VRP. The results show
a striking impact on especially OTM call and put expected returns. For puts, 5% OTM
expected returns are now close to -100% in the low volatility regime and less than -70% in
the high volatility regime as even a modest VRP results in a dramatic drop in expected put
returns. Expected returns also become similarly strongly negative as the VRP changes the
sign of OTM expected call returns for the same reason as put options.
The results show that very short-dated options are extremely sensitive to variance risk
premiums. This occurs because deep OTM call and put prices are driven by extreme tail
events, and variance risk premiums result in tail probabilities that are dramatically higher
under the Q distribution compared to the P distribution, at least far enough into the tails.
For example, the probability density function of a log-normal at the 95% strike under 20%
and 17.5% volatilities are different by a factor of 35. The results indicate that even modest
variance risk premiums profoundly impact short-dated option returns, far more so than
those for longer-dated option returns.
We also analyze the implications for option return volatility using our analytical ap-
proach in Figure 3. The left (right) panels report put (call) option return volatility. Option
return volatility increases rapidly as strikes move OTM and is also highly dependent on the
underlying asset’s volatility. Option return volatilities are not particularly sensitive to ex-
pected underlying returns, which is not surprising. Call and put return volatilities increase
similarly as the strikes move OTM, though this is not a general property and does not
hold in models with crash risk. Option return volatility and underlying asset volatility are
inversely related, similar to expected option returns, which generally increase (in absolute
value) as underlying asset volatility decreases.
18
As discussed later, this is exactly what happens in a formal model with event risk variance premium.

13

Electronic copy available at: https://ssrn.com/abstract=4729757

 Figure 2: Expected option returns: With and without variance risk premium
This figure provides 1-day-to-maturity expected option returns for high and low volatility ranges with and
without a variance risk premium. All panels assume µ = 10% and r = 4%.
VRP impact on put returns: low volatility VRP impact on call returns: low volatility
0
0
10
10
20
20
30
30
40
40
50
=20%, no VRP 50 =20%, no VRP
60 =17.5%, 2.5% VRP =17.5%, 2.5% VRP
60
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness
VRP impact on put returns: high volatility VRP impact on call returns: high volatility
0
0
10
10
20
20
30
30
40
40
50
=40% no VRP 50 =40% no VRP
60 =35%, 5% VRP =35%, 5% VRP
60
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness

These population moments highlight the high sensitivity of option return volatility to
both moneyness and underlying asset volatility, as option return volatility is driven by the
combination of tail probabilities on index returns and the convexity of OTM option payouts.
As mentioned earlier, measuring option return volatilities in observed samples is challenging,
due to a combination of historical samples with modest lengths, the peculiarities of when
daily expiries were introduced, and the infrequency of outlier events. The impact on outliers
is well known and discussed in Broadie et al. (2009). Suppose the largest one-day index re-
turn is 2.5% in a given sample. This implies that all calls further OTM have -100% returns
(no zero observed volatility), clearly a Peso problem. These model-based return volatilities
provide benchmarks for difficult-to-estimate expected option return volatilities, which, in
turn, can be used for investment purposes (understanding risk/return tradeoffs or for con-
structing optimal portfolios) but also for calculating standard errors under the assumption
of a given model such as the Black-Scholes model or Merton’s jump-diffusion. Given the

14

Electronic copy available at: https://ssrn.com/abstract=4729757

 Figure 3: Expected option return volatility
This figure provides 1-day-to-maturity option return volatilities for various parameters in the Black-Scholes
model for log(moneyness) ranging from -2 to +2. The top panels use µ = 10% while the bottom panels use
σ = 15%. All panels assume r = 4%
Put Volatility for various 16
Call Volatility for various
14 = 15% = 15%
= 25% 14 = 25%
12
= 40% 12 = 40%
10
10
8
8
6 6
4 4
2 2
0 0
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness
Put volatility for various Call volatility for various
14 = 5% 16 = 5%
12
= 12.5% 14 = 12.5%
= 20% = 20%
10 12
10
8
8
6
6
4 4
2 2
0 0
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness

obvious issues with small samples, these analytical moments are also a computationally effi-
cient alternative to Monte Carlo simulations. The advantage of using Black-Scholes is that
the inputs are observed – the realized average index returns and volatilities.

2.2 Option Returns with Event Risk

Short-dated options load heavily on event risk, as the relative importance of diffusive risk,
which scales linearly with time to maturity, is small. We now provide the main intuition
for analyzing option returns with event risk using a simple extension of the Black-Scholes
model. The model builds on Dubinsky et al. (2019) and Johannes et al. (2023), but the
predictions that we derive for option returns require modeling assumptions both under P
and Q since option returns depend crucially on P-measure dynamics. The option pricing
models in Dubinsky et al. (2019) and Johannes et al. (2023) are restricted to Q-measure

15

Electronic copy available at: https://ssrn.com/abstract=4729757

dynamics.
We assume a single macroeconomic announcement event at time τz prior to maturity.
We assume that the underlying asset follows a Geometric Brownian Motion and that the an-
nouncement induces an additional deterministic jump Zτz in the log return that is normally
distributed under P
1 P
2 P
 

2
Zτz ∼ N µz − σ , σz .
2 z
The stock index evolution is, therefore, given by

P P
  
1 2

ST = St × exp µ − σ (T − t) + σ BT − Bt + Zτz ,
2

where BtP is a standard Brownian motion under P and µ and σ are the diffusive drift and
standard deviation, respectively. This evolution implies that

EPt [ST ] = St × exp [µ(T − t) + µz ] .

There is evidence that expected stock returns around macroeconomic announcements

are higher than the expected returns for non-event periods. For example, Savor and Wilson
(2013) find that stock index returns are ten times higher on announcement days than on
non-announcement days. In our 1DTE sample from 2012, stock index returns more than five
times higher on announcement days. In this model, µ captures the ’normal’ expected return
on non-announcement days while µz captures the higher expected return on announcement
days. Note that returns are still normally distributed in this model, with mean µ̃ = µ(T −
e2 (P) = σ 2 (T − t) + σzP

2
t) + µz and variance σ
Under a risk-neutral measure Q, we need to impose the martingale condition on the
stock price, which implies that, if the extra macroeconomic announcement return is assumed
to be normally distributed, then

1 Q
2 Q
 

2
Zτz ∼ N − σz , σz .
2

The stock index evolves via

Q Q
  
1 2  
ST = St × exp r − σ (T − t) + σ BT − Bt + Zτz .
2

16

Electronic copy available at: https://ssrn.com/abstract=4729757

To value options, it is useful to write the model slightly differently, in a form that can be
used to directly apply the Black-Scholes formula. To see this, note that

−r(T −t) Q

! h
var Xt,T i+ 
BS St , K, , r, T − t = e Et St er(T −t)+Xt,T − 21 var(Xt,T )
−K
T −t

where Xt,T is a mean-zero normal variable with variance var(Xt,T ) under Q. Given this,
Q Q
we define Xt,T = σ BT − Bt + σzQ εz and the shocks with εz ∼ N (0, 1) and var (Xt,T ) =
 

σ 2 (T − t) + σzQ . This implies that the Black-Scholes formula applies with a modified

2

(σzQ )
2

e2 (Q) = σ 2 + T −t . Combining, we have that the call option expected

volatility parameter σ
return is

P  c  eµ̃(T −t) St N d1,µ̃,eσ(P) − e−µ̃(T −t) KN (d2,µ̃,,eσ(P) )

Et rt,T = − 1.
St N (d1,r,eσ(Q) ) − e−r(T −t) KN (d2,r,eσ(Q) ),

Figure 4 shows the significant impact of event risk on expected put and call option
returns. Call and put option returns are extremely sensitive to µz , with call (put) returns
increasing (decreasing) in the macroeconomic return premium. The economic magnitudes
are quite striking, and point towards a first-order effect of this premium for explaining option
returns. The impact of macroeconomic announcement volatility and the announcement
volatility premium is also significant, especially for OTM options.

2.3 Option Returns in Merton’s Model

Merton’s model extends Black-Scholes to incorporate crash risk, and in addition, we incor-
porate macroeconomic event risk. We assume that under P stock prices evolve via
 P 
Nt Ntm

dSt = µSt dt + St σdBtP + d  Sτj− (eZj − 1) − λP µP St dt + d

!
X P X
Sτz,j− (eZj − 1) ,
j=1 j=1

where λP is the (constant) intensity of the Poisson process NtP and τj denote the random
jump times of NtP . The jump sizes are given by ZjP ∼ N µP P

2 
y , σy with jump com-
pensator µP = exp µP P 2 − 1. Event risk is modeled by Zj , which represents the
h i
1


y + 2 σy

additional return on the j-th announcement at time τz,j . Note that the counting process
Ntm is deterministic, which is a reasonable assumption for short option maturities we con-

17

Electronic copy available at: https://ssrn.com/abstract=4729757

 Figure 4: Expected option returns with event risk
This figure provides 1-day-to-maturity expected option returns for the Black-Scholes model extended by
event risk. We assume r = 4% and µ = 9% and σ = 15%. In the first row, we fix the event risk volatilities
and vary the event return risk premium. In the second row, we fix the event return premium and vary the
event risk volatilities under P and Q.

Expected Call Return: z = z = 50 bps Expected Put Return: z = z = 50 bps

140 0
z = 0 bps
120 z = 10 bps 10
z = 20 bps
100
z = 30 bps 20
80
30
60
40 40 z = 0 bps
z = 10 bps
20 50
z = 20 bps
0 60 z = 30 bps
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness
Expected Call Return: z = 10 bps Expected Put Return: z = 10 bps
140 0
z = 30 bps z = 30 bps
120 z = 50 bps z = 50 bps 10
100 z = 50 bps z = 60 bps
20
80
30
60
40 40
z = 30 bps z = 30 bps
20 50 = 50 bps = 50 bps
z z
0 60 z = 50 bps z = 60 bps
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness

sider. As before, macroeconomic announcement returns are assumed normally distributed:

Zj ∼ N µz − 12 σzP , σzP . This simple model can be extended in various directions

2
2 

to include stochastic volatility, non-normal crash, or time-varying announcement risk, for

example.
Under the risk-neutral measure Q, stock prices evolve via
 Q 
Ntm
Q
Nt

Sτ Q (eZj − 1) − λQ µQ St dt + d
!
X Q X
dSt = rSt dt + St σdBt + d  Sτz,j− (eZj − 1) ,
j−
j=1 j=1

where λQ is the (constant) intensity of the Poisson process NtQ and τjQ denote the ran-
dom jump times of the risk-neutral Poisson process. The jumps are driven by ZjQ ∼

18

Electronic copy available at: https://ssrn.com/abstract=4729757

 Figure 5: Expected option returns and standard errors: Merton model (in %)
This figure provides expected option returns for the Merton Model with and without macroeconomic
announcement event risk. The left graph presents option returns for calls for a log-moneyness level from
-2.0 to 2.0. The right graph presents option returns for puts. We include model-based standard error
bounds based on 1109 (314 macro and 795 non-macro) observations, and each bound ranges from the
model-based expected return minus 1.96 times its standard error to the expected return plus 1.96 times its
standard error, using the formulae derived in Proposition 1. Parameters used are µ = 0.10, σ = 0.15, λP =
λQ = 2, µQ Q Q P P P P
y = −0.03, σy = 0.09, σz = 0.005, µy = −0.02, σy = 0.04, µz = 0.002, σz = 0.005.

HTM Return in Merton Model HTM Return in Merton Model

20
100 0
20
50 40
60
0
80
50 100
Call (without macro announcement) 120 Put (without macro announcement)
100 Call (with macro announcement) 140 Put (with macro announcement)
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness

N µQ Q Q µQ σyQ

2  h
2 i
1
y , σy with jump compensator µ = exp y + 2
− 1. The macroeco-
nomic announcement returns under the risk-neutral measure are again assumed normally
distributed: Zj ∼ N − 21 σzQ , σzQ , which imposes the martingale restriction as in the

2
2 

Black-Scholes model extension.

As in the case of the Black-Scholes model, it is possible to calculate option prices using a
modification of Merton’s formula (which in turn uses Black-Scholes) or via Fourier transform
methods based on the characteristic function. The characteristic function approach is useful
for more general models with non-normal jump sizes for which prices cannot be calculated
using a simple extension of the Black-Scholes formula. To apply our moment formulas from
Proposition 1, note that he characteristic functions under P and Q are given by

µP u+ 1 P2 2
P(u, St) log(St )+(µ− 21 σ 2 )τ u+ 12 u2 σ 2 τ +λP τ e y 2 (σy ) u −1−µP u +(N m −N m )(µz u+ 1 (σ P )2 (u2 −u))
 

ΨM,
T t 2 z
t,T = e
Q 1 Q2 2
Q(u, St) log(St )+(r− 21 σ 2 )τ u+ 21 u2 σ 2 τ +λQ τ eµy u+ 2 (σy ) u −1−µQ u +(NTm −Ntm )( 12 (σzQ )2 (u2 −u))
 

ΨM,
t,T = e

where τ = T − t.
To provide some intuition on the model assumptions, Figure 5 plots expected option
returns for put and call options of different moneyness levels and their model-based standard

19

Electronic copy available at: https://ssrn.com/abstract=4729757

error bounds, matching the sample sizes of our empirical data below. We use a standard
parameter set that is roughly in line with parameter values reported in the literature. Event
risk has a considerable impact on expected option returns. Call and put option returns differ
quite substantially depending on whether or not an announcement is scheduled before the
maturity of the option. The call option expected returns have a pronounced hump shape
on announcement days.

3 Data and Empirical Results

3.1 Data

We obtain intradaily option quote data on the SPX index from the CBOE.19 The data
include the national best bid/offer (NBBO) at the end of each 5-minute interval during
the regular trading session from 9:30 am ET until 4:15 pm ET, as well as corresponding
S&P 500 index levels. We discard most option maturities in our data set and focus on
contracts with 1-day to maturity. To avoid prices at the close (see Goyenko and Zhang,
2019), we record option mid-quotes at 3:55 pm ET one trading day before maturity. This
initial selection includes option contracts recorded on Fridays (with expiration the following
Monday), as well as options expiring after a public holiday. As weekends have distinct
seasonality patterns (see Jones and Shemesh, 2018), we mainly work with a second sample
for which we restrict the option maturities to exactly one calendar day. Furthermore, we
omit options if trading stops early at 1:15 pm ET on the maturity date (which usually
occurs the day before Independence Day and Thanksgiving and on Christmas Eve) and
options that are not weeklies as they expire in the morning rather than at the close.
In order to maximize the sample size, we focus on the period from January 2, 2012
until December 5, 2023. This is an unbalanced sample, as 1DTE options in the early part of
the sample are only available on Thursdays (for Friday expirations). From February 2016,
the addition of Monday and Wednesday options provides a denser maturity grid, and from
2022, 1DTE options exist for every weekday.
Using the raw data, we calculate a number of option market variables that are required
19
The CBOE does not provide data for three trading days in our sample: January 8, 2013, January 18,
2013, and February 13, 2013. None of these missing data affect our analysis. In addition, the data for March
6, 2020 mainly records unrealistic zero prices and is excluded.

20

Electronic copy available at: https://ssrn.com/abstract=4729757

 Table 1: Macroeconomic announcements
This table provides information about the macroeconomic announcements during our sample period from
January 2012 to December 2023.

Most frequent an-

Announcement Abbreviation Obs Source
nouncement time

Consumer Price Index CPI 143 Wed 08:30 BLS

FOMC Statement FOMC 94 Wed 14:00 Fed
GDP GDP 142 Thu 08:30 BEA
The Employment Situation Monthly Payrolls 143 Fri 08:30 BLS

for our analysis below. First, we use one-month treasury yields from FRED (St. Louis
Fed) as a risk-free rate proxy, and we closely follow the CBOE VIX methodology and
calculate on each day t in the sample the implied forward price Ft,T for 1DTE options
(maturity T ). To do so, we imply the futures price from put-call parity using the put/call
pair whose absolute price difference is the smallest. We then calculate standard moneyness
as M = K/Ft,T and also calculate Black-Scholes implied volatilities. We then linearly
interpolate the implied volatilities for options close to at-the-money to obtain the at-the-
atm
money implied volatility IVt,T for each day. We use this ATM volatility to define log
atm
√

moneyness as m = log(M )/ IVt,T T − t . Log moneyness is more informative in our
option dataset as it adjusts for the volatility of the underlying asset and allows us to more
cleanly compare contracts with different moneyness levels over time.
Our second main dataset consists of macroeconomic announcements and contains the
four main announcement categories: Consumer Price Index (CPI), Gross Domestic Product
(GDP), FOMC announcements (FOMC), and Employment Situation (Monthly Payrolls).
We collect dates and timestamps for these four categories for our sample period from January
2012 to December 2023. Announcement times are important as some announcements are
during the trading day (FOMC) while others usually occur before the market opens at
8:30 am ET. Our data is hand-collected from original sources (e.g., the Bureau of Labor
Statistics). Table 1 summarizes the announcements, including the number of observations
and their most frequent announcement time. There are no announcements on Monday
before the close of the market, which is helpful as weekend option returns behave differently
from the returns during the week.

21

Electronic copy available at: https://ssrn.com/abstract=4729757

3.2 Empirical Results

3.2.1 Put and Call Option Returns

Our first set of empirical results are for call and put options that are held until maturity. For
each option in the sample, we calculate the return of an option using the mid price at 3:55
pm ET one calendar day before maturity (hence we initially omit Monday expirations and
expirations after a public holiday). The payoff at maturity is calculated using the closing
price of the S&P 500 index on the maturity date.
Table 2 provides average returns and higher-order return statistics for puts and calls
for a number of log moneyness targets: -2, -1, 0, 1, and 2. To construct these returns, on
each day in the sample, we assign the option closest to the target that remains within a
tolerance of 0.5. SPX options close to the money usually trade in strike increments of 5,
with larger gaps further out of the money. For robustness, we also provide the same analysis
for the alternative moneyness definition (K/Ft,T ) in the Appendix (see Table A.1). Panel A
summarizes the return distributions for the entire sample period. Average call option returns
are small for in-the-money calls (2.84% for a log moneyness of -2) and monotonically increase
until a moneyness level of 1. The returns turn negative (-52.17%) for a log moneyness level
of 2. We provide standard HAC-corrected t-statistics, which may be severely biased given
the high kurtosis of out-of-the-money options (call returns for log moneyness of -2 have a
kurtosis of 181.78). According to this benchmark, only out-of-the-money calls have returns
that are significantly different from zero. Put option returns also, on average, show a
monotonic increase from -45.43% for out-of-the-money puts (log moneyness -2) to -3.78%
for in-the-money puts (log moneyness 2). Standard t-statistics show that most of the put
returns are significantly negative. Skewness and kurtosis are reasonably similar to those
of the call options, with high excess kurtosis for OTM puts. Panels B and C show that
our results are highly robust across different sample periods, which is important given the
unbalanced nature of our option data set.
In addition to standard t-statistics, Table 2 also contains Black-Scholes-based t-statistics
(labeled t-stat BS). To calculate these, for each return time series, we subtract from the real-
ized average option return the expected option return in the Black-Scholes model and divide
this quantity by the Black-Scholes standard error. This accounts for the sample size and

22

Electronic copy available at: https://ssrn.com/abstract=4729757

centers the t-statistic not around zero but a more realistic model benchmark. For each of
the three samples, we use the sample realized return, average risk-free rate, and the sample
realized variance for the Black-Scholes model estimates. Using this benchmark, we find that
all but one option category has returns that cannot be rejected to be different from a simple
Black-Scholes benchmark. The only category marginally exceeding standard hurdles are
OTM call options with a BS t-statistic of -2.19, which suggests that these returns are lower
than what is expected in the Black-Scholes model. None of the sub-sample returns deviate
significantly from the model benchmark. While 1DTE option returns can be quite large,
our results suggest that they are not different from simple model-based predictions.
Table 3 provides returns for the same put and call options for the same moneyness
targets, but also splits the sample into two subsamples: returns that were realized with
a macroeconomic announcement before the maturity of the options, and returns that oc-
curred without an announcement prior to maturity. Our methodology covers the majority
of announcements during our sample, 314 out of 522. During the later part of the sample,
all announcements have a daily option expiring on the day of the announcement, and we
provide robustness tests in the Appendix that show that our results are remarkably robust
in sub-samples and are not affected by this imbalance.
The All Observation columns in Table 3 are a summary of what was reported in Ta-
ble 2 and are merely reported for convenience. Focusing on With Macro Announcements
columns, we find that call options in Panel A have very different return patterns around
macroeconomic announcements. Call returns for moneyness -2 to 1 are far more positive
and show a strong monotonic pattern, reaching almost 60% average return for moneyness
level 1. For further OTM options (moneyness level 2), returns then turn negative with an
average return of -10.05%. This distinct hump-shaped pattern for the returns on macroe-
conomic announcements is also documented in Figure 6, which shows the same average
returns but also includes further moneyness levels not reported in Table 3. Returns on non-
announcement days are very different, just about positive ITM and only strongly negative for
further OTM options (-68.81% for moneyness 2). The statistical significance of these results
is also interesting: standard t-statistics indicate that almost all option return categories on
non-announcement days are consistent with even the simple BS benchmark, whereas we find
strong deviations from the BS benchmark for the macroeconomic announcement sample.

23

Electronic copy available at: https://ssrn.com/abstract=4729757

 Table 2: SPX hold-to-maturity option returns (in %)
This table provides the average returns for one-day-to-maturity call and put options for different moneyness levels. Moneyness is defined as
atm
√

log(K/Ft,T )/ IVt,T T − t . To calculate the average returns, for each option type in our sample that has a one-day-to-maturity option, we choose the
strike closest to the Moneyness level in column 1 (with a maximum tolerance of 1%) and record the return of the call and put option, respectively. Call
AvgRet (Put AvgRet) is the average call (put) option return, in percentage. t-stat provides the HAC-corrected t-statistic for the mean and t-statistics based
on Black-Scholes moments (t-stat BS), Loss % gives the likelihood of complete capital loss (-100%, i.e. the option is out-of-the-money at maturity). Skew
Electronic copy available at: https://ssrn.com/abstract=4729757

and Kurt are the skewness and the kurtosis of the returns. Obs are the number of observations.

Moneyness Call Option Contract Return Statistics Put Option Contract Return Statistics
t-stat Loss t-stat Loss
mean t-stat Skew Kurt mean t-stat Skew Kurt
BS (%) BS (%)

Panel A: From 2012-01-01 to 2023-12-05 (1109 observations)

-2.000 2.84 (1.90) (0.80) 3 -0.09 -0.36 -45.43 (-3.39) (-1.45) 97 10.96 149.68
-1.000 3.98 (1.68) (0.55) 14 0.38 -0.54 -22.97 (-2.70) (-1.84) 86 6.33 52.68
0.000 8.15 (1.88) (0.86) 46 1.27 0.75 -9.14 (-1.88) (-1.16) 55 3.06 16.02
24

1.000 17.71 (1.79) (1.14) 82 3.60 14.47 -5.38 (-2.03) (-1.18) 18 1.19 2.84
2.000 -52.17 (-3.71) (-2.19) 98 12.89 181.78 -3.78 (-2.37) (-1.49) 2 0.71 2.00

Panel B: From 2016-03-01 to 2023-12-05 (950 observations)

-2.000 3.25 (2.03) (1.06) 3 -0.07 -0.33 -39.40 (-2.58) (-1.13) 97 10.50 136.83
-1.000 4.76 (1.88) (0.87) 13 0.40 -0.53 -20.95 (-2.20) (-1.53) 87 6.25 51.12
0.000 9.64 (2.07) (1.16) 46 1.27 0.71 -8.89 (-1.67) (-1.07) 54 3.21 17.22
1.000 23.61 (2.20) (1.66) 81 3.53 13.90 -5.67 (-1.99) (-1.26) 19 1.26 3.26
2.000 -44.85 (-2.75) (-1.78) 98 11.94 155.57 -4.12 (-2.39) (-1.66) 2 0.75 2.22

Panel C: From 2020-01-01 to 2023-12-05 (562 observations)

-2.000 0.49 (0.23) (-0.11) 3 -0.04 -0.59 -50.74 (-3.53) (-1.28) 97 8.10 68.67
-1.000 0.60 (0.18) (-0.15) 16 0.39 -0.80 -15.58 (-1.41) (-0.99) 84 4.17 19.14
0.000 6.26 (1.07) (0.71) 49 1.24 0.59 0.54 (0.08) (0.38) 51 1.85 3.50
1.000 14.56 (1.14) (0.88) 81 3.53 13.23 -1.40 (-0.39) (-0.09) 19 0.68 -0.02
2.000 -45.78 (-2.22) (-1.30) 98 11.42 147.08 -1.45 (-0.67) (-0.37) 2 0.29 -0.18
 Table 3: SPX hold-to-maturity option returns (in %): Macroeconomic announcement days
This table provides the average returns for one-day-to-maturity call and put options for different moneyness levels. Moneyness is defined as
atm
√

log(K/Ft,T )/ IVt,T T − t . To calculate the average returns, for each option type in our sample that has a one-day-to-maturity option, we choose
Electronic copy available at: https://ssrn.com/abstract=4729757

the strike closest to the Moneyness level in column 1 (with a maximum tolerance of 1%) and record the return of the call and put option, respectively.
Call AvgRet (Put AvgRet) is the average call (put) option return in percentage. t-stat provides the HAC-corrected t-statistic for the mean and and
we also provide t-statistics based on Black-Scholes moments (t-stat BS ). Loss % gives the likelihood of complete capital loss (-100%, i.e. the option is
out-of-the-money at maturity). Skew and Kurt are the skewness and the kurtosis of the returns. Obs are the number of observations.

Moneyness All Observations With Macro Announcements Without Macro Announcements

t-stat t-stat t-stat
mean t-stat Obs mean t-stat Obs mean t-stat Obs
BS BS BS

Panel A: Calls
25

-2.000 2.84 (1.90) (0.80) 1109 8.51 (3.25) (2.45) 314 0.59 (0.36) (-0.60) 795
-1.000 3.98 (1.68) (0.55) 1109 12.79 (2.99) (2.21) 314 0.50 (0.20) (-0.74) 795
0.000 8.15 (1.88) (0.86) 1109 22.15 (2.75) (2.10) 314 2.62 (0.57) (-0.31) 795
1.000 17.71 (1.79) (1.14) 1109 59.72 (2.76) (2.88) 314 1.11 (0.11) (-0.47) 795
2.000 -52.17 (-3.71) (-2.19) 1109 -10.05 (-0.24) (-0.37) 314 -68.81 (-5.68) (-2.36) 795

Panel B: Puts
-2.000 -45.43 (-3.39) (-1.45) 1109 -74.01 (-5.01) (-1.37) 314 -34.15 (-1.93) (-0.85) 795
-1.000 -22.97 (-2.70) (-1.84) 1109 -52.06 (-4.52) (-2.68) 314 -11.48 (-1.04) (-0.49) 795
0.000 -9.14 (-1.88) (-1.16) 1109 -28.22 (-3.60) (-2.97) 314 -1.60 (-0.27) (0.50) 795
1.000 -5.38 (-2.03) (-1.18) 1109 -15.08 (-3.38) (-2.79) 314 -1.55 (-0.51) (0.36) 795
2.000 -3.78 (-2.37) (-1.49) 1109 -10.00 (-3.77) (-3.06) 314 -1.32 (-0.74) (0.17) 795
 Figure 6: Average SPX hold-to-maturity option returns
This figure provides average SPX option 1-day to maturity holding period returns for our sample period
January 2012 to December 2023. We separately show holding period returns for calls and puts, with and
without macroeconomic announcements occurring during the holding period. Additionally, the figure
presents the holding period returns for call options combined and the holding period returns for all
put combined. Shaded areas represent HAC-adjusted standard error bounds. Moneyness is defined as
log-moneyness and ranges from -2 to 2.
Average HTM Returns (Call Options) Average HTM Returns (Put Options)
75 0
50 20
25
40
0
25 60
50 Without Macro Announcement Without Macro Announcement
All Observations 80 All Observations
75 With Macro Announcement With Macro Announcement
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness

Similarly, put option returns also show very different patterns for macro and non-macro
days. Returns on macroeconomic announcement days are far more negative and again only
deviate from the BS benchmark on announcement days where all but one category shows
clear deviations from the model-based predictions.

3.2.2 Model Estimation

We use a two-stage Generalized Method of Moments (GMM) approach to jointly estimate

the parameters of the Merton model with macroeconomic announcement event risk under P
and Q. We impose the restriction that the number of jumps is equal under P and Q resulting
in the structural parameter vector Θ = [µ, σ, λ, µQ Q Q P P P
y , σy , σz , µy , σy , µz , σz ]. The system of

moment conditions E [W ψ(x; Θ)] = 0 with ψ(x;

b Θ) being a vector of the differences between

theoretical and empirical moments, is obtained by stacking a number of different moment

conditions. The different types of moment conditions load on various data sources that are
potentially informative for estimating the parameter vector Θ. In particular, we match the
average returns of SPX 1DTE-options, the implied volatility smile of SPX options, and the
first four moments of the S&P 500 daily returns.
The first set of moments consists of the average option returns as reported in Table

26

Electronic copy available at: https://ssrn.com/abstract=4729757

3 and contributes to the estimation of both the P and Q parameters. The set of moment
conditions is constructed based on average options returns with log-moneyness −2, −1, 0, 1,
and 2 for both call and put options. These are considered for days with and without
announcements, resulting in a total of 20 moment conditions (5 strike prices × 2 option types
× 2 event types). For the implied volatility smile, we use daily smiles from out-of-the-money
options to derive an implied volatility trend line for announcement and non-announcement
days. We then create ten moment conditions by means of a quadratic regression that
contribute to estimating the Q parameters. The remaining moment conditions are derived
from the first four moments of daily S&P 500 index log returns. These moments can be
derived from the characteristic function and are given by:

mP P P
σ2
 
1 = µ− − λµ̄ + λµy ∆,
2
P 
2
h
m 2 = σ + λ σ y + µy P
2 P
2 i
∆,

mP P
2 + µPy
3 ∆,
 
P
3 = λ 3µy σy

mP P P

4
2 P 2 P
4 
4 = λ 3 σ y + 6 σ y (µ y ) + µ y ∆

P P P P
 
2
h
2
2 i2 2
h
2
2 i
+ 3λ σy + µy + 6λσ σ y + µy 4
+ 3σ ∆2 .

where ∆ = 1/252 is the time between observations and µ̄P is as defined above.
Since the system of moment conditions is overidentified, we obtain Θ∗ by minimizing
a quadratic form of moment conditions. The GMM estimator is given by:

b Θ)′ W ψ(x;
ΘGM M = argmin ψ(x; b Θ).
Θ

with W38×38 required to be a positive definite weighing matrix. Appendix A.3 includes
further details on the estimation methodology.
Table 4 provides our parameter estimates. Our estimate for λP = λQ is 0.947, which
translates into slightly less than one jump per year, on average. The estimates in the
literature for this parameter vary, ranging from values lower than ours to substantially higher
values. Our estimate imposes infrequent jump events, and on occurrence, the jump impact
is high. On average, the options price a negative jump of -1.4% with a standard deviation of
8.5%. Again, these values are economically reasonable, especially considering that our jump

27

Electronic copy available at: https://ssrn.com/abstract=4729757

 Table 4: Parameters: Q-measure and P-measure estimation
This table presents the jointly estimated Q-measure and P-measure parameters for the Merton jump-
diffusion model with macroeconomic event risk as using two-stage GMM as described in 3.2.2. Figures in
brackets indicate parameter standard errors.

µ σ λ µQ
y σyQ σzQ µP
y σyP µz σzP

0.089 0.145 0.947 -0.014 0.085 0.005 -0.028 0.026 0.002 0.005
(0.005) (0.002) (0.525) (0.015) (0.067) (0.000) (0.024) (0.019) (0.000) (0.001)

frequency is relatively low. Our estimate for the macro-economic announcement volatility is
roughly 50 bps. This is comparable to findings in Johannes et al. (2023), although marginally
lower than their estimates. One major difference to Johannes et al. (2023) is that their
estimates are time-varying, and this may explain some of the differences we find.20
The P-measure estimates are also economically reasonable. The estimated expected
return is 8.9%, and σ is estimated to be 14.5%. More importantly, as expected and in line
with Broadie et al. (2007), we estimate a substantial jump risk premium for the random
jump component. Average jump sizes under P are -2.8% with a standard deviation of 2.6%,
and therefore, the impact of jumps under P is, on average, more negative but has a much
lower standard deviation than under Q. Of particular interest are the jump parameters of
the macroeconomic jumps. Our estimate for σzP is 50 bps, the same value as under Q, and
implies, contrary to the findings in Londono and Samadi (2023), that there is no significant
wedge between the two measures. But we find a significant drift premium of 20 bps, which
aligns with the findings of Savor and Wilson (2013) that macro-economic announcement
days have higher returns than ordinary trading days.
Figure 7 provides expected option return plots for the estimated model parameters,
using the Merton model with and without macroeconomic announcement events. The shapes
for the option returns are very similar to the empirical results presented earlier in Figure (6).
In particular, our estimated parameters imply a large difference between expected option
returns on announcement vs. non-announcement days. The shape of the expected return
distribution for calls on announcement days has the distinctive hump shape reported earlier,
20
It is also highly likely that jump distributions change over time. We do not explicitly allow for this, as
the exact variation of the jump distribution is not of interest to our purposes. Our estimated parameters
have to be interpreted as average values of a potentially changing jump distribution.

28

Electronic copy available at: https://ssrn.com/abstract=4729757

 Figure 7: Expected option returns for the Merton model (in %)
This figure provides expected option returns for the Merton Model with and without macroeconomic
announcement event risk based on the estimated parameters that are provided in Table (4). The graph
presents option returns for calls and puts for a log-moneyness level from -2.0 to 2.0 along with analytical
one standard error bounds based on 1109 (314 macro and 795 non-macro) observations and the formulae
derived in Proposition 1. We add the corresponding realized average holding period returns in dotted lines
for comparison.

80 Call Returns (%) Put Returns (%)

60 0
40
20
20
0 40
20 60
40 80
Expected Returns (Without Macro) Expected Returns (Without Macro)
60 Expected Returns (With Macro) Expected Returns (With Macro)
80
Realized Returns (Without Macro) 100 Realized Returns (Without Macro)
Realized Returns (With Macro) Realized Returns (With Macro)
2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.0 1.5 1.0 0.5 0.0 0.5 1.0 1.5 2.0
Log Moneyness Log Moneyness

and there is a substantial difference to the shape observed on non-announcement days. Our
model achieves this mainly by imposing a return premium for announcement days. Average
put option returns are also similar to the empirical values, with lower option returns on
macroeconomic announcement days.

3.2.3 Portfolio Returns

Next, we turn to a number of other popular trading strategies. In Figure 8, we summarize

average hold-to-maturity returns for ATM straddles and strangles for different moneyness
levels. Straddle returns are insignificant, both on non-announcement and announcement
days. In general, macroeconomic announcements push straddle and strangle returns down-
ward, in line with the predictions of the simple Merton model parameterization. The model
parameters presented above are in line with the empirical returns for the straddle and far
out-of-the-money strangles but slightly overestimate the returns for other strangles.

3.2.4 Further Empirical Results

In this section, we provide further empirical results for a number of other popular option
trading strategies. We start with variance risk premium strategies, as these have been consid-

29

Electronic copy available at: https://ssrn.com/abstract=4729757

 Figure 8: SPX option portfolio returns (in %)
These figures provide realized and expected strangle returns at various moneyness levels with and without
macroeconomic announcement event risk. For instance, 1% represents a portfolio long one percent out of
the money call and put options. 0% corresponds to ATM straddles. The figure on the left gives realized
portfolio returns, while the one on the right shows expected returns using Merton model parameters
calibrated above. We add the corresponding realized average hold-to-maturity returns in dotted lines for
comparison. Shaded areas represent HAC-adjusted standard error bounds for the realized returns and
analytical one standard error bounds based on 1109 (314 macro and 795 non-macro) observations and the
formulae derived in Proposition A.2 for expected returns.

Realized Returns Expected Returns

10
10
0
0
10
10
20
20
30
30
40
40
50 50
60 Without Macro Announcement 60 Without Macro Announcement
With Macro Announcement With Macro Announcement
70
0.0% 0.5% 1.0% 1.5% 0.0% 0.5% 1.0% 1.5%
% Out of the Money % Out of the Money

ered in the literature. Londono and Samadi (2023) find a significant variance risk premium
for macroeconomic announcements, whereas our model estimation in the last section does
not imply an economically meaningful difference between macroeconomic volatility under P
and Q.
We define the variance risk premium as the difference between the observed realized
variance and its expectation under the risk-neutral measure Q. Realized variance between
time-t and time-T is defined as

n    
X Fti ,T Fti ,T
RVt,T =2× − 1 − log
i=1
Fti−1 ,T Fti−1 ,T

where Ft,T is the time-t forward contract with maturity T and t = t0 < t1 < · · · < tn = T (see
Neuberger, 2012, and Bondarenko, 2014). With this definition of variance, the expectation
of realized variance under the risk-neutral measure is (due to the martingale property of
Ft,T and the telescoping sum)

EQ Q  
t [RVt,T ] = −2 × Et [log (ST )] − log (Ft,T ) ,

30

Electronic copy available at: https://ssrn.com/abstract=4729757

which is proportional to the squared VIX index. This construction is model-free and does,
in particular, not depend on the monitoring frequency of the realized variance. We use a
standard approach to calculate the Q-measure expectation from option prices.21
To calculate the realized variance, we rely on the 5-minute spot price return data pro-
vided by the CBOE. The empirical option pricing literature often relies on daily realized
volatilities calculated from the option-implied forward price as a substitute for the unobserv-
able forward contract. While this is sensible for daily return calculations, this procedure for
intradaily observations would mean we need to calculate these implied forwards for options
with only a few minutes to maturity. To avoid this, we use the spot index returns as for our
short-term option contracts, the spot and futures will be extremely highly correlated. Our
realized variance calculation includes the overnight period and every 5-minute interval dur-
ing the trading day. The variance risk premium theory above works for any chosen partition
of the trading day, lower frequencies are also possible but increase the sampling error.
Table 5 provides our empirical results. We define the realized variance risk premium
as follows:

V RPt,T = RVt,T /EQ

t [RVt,T ] − 1.

Overall, our results show that in the entire sample the variance risk premium was -26.22%
(t-statistic -17.66, Panel A), a value that is similar in magnitude to the monthly variance
risk premium reported in the literature (see Carr and Wu, 2006, Neuberger, 2012, Bon-
darenko, 2014). The VRP is slightly larger in the sample with macroeconomic announce-
ments (-28.48%, t-statistic -9.28, Panel B) compared to the sample without macroeconomic
announcements (-25.33%, t-statistic -15.09, Panel C). But overall, unlike our results for
the raw put and call returns, these risk premiums are economically similar across the two
samples with and without macroeconomic announcements.22
21
To do so, we first fill in option prices for strikes that are not observed by using values from an interpolated
implied volatility curve, and extend option prices beyond the traded range by fixing the implied volatility
to the last observed value. The option price integral required for the calculation is then approximated using
the trapezoidal rule.
22
As a robustness check, we calculate the Variance Risk Premium using an alternative methodology.
Specifically, the risk-neutral variance is calculated from option prices following Bakshi et al. (2003), and
realized variance is calculated using intraday and overnight returns. We then augment Corsi (2009)’s HAR
model with an announcement indicator to project the variance forward. The results using this approach are
qualitatively and quantitatively very similar to what we report above.

31

Electronic copy available at: https://ssrn.com/abstract=4729757

 To shed light on the finer structure of the variance risk premium, we also consider
corridor variance risk premia. We define the realized corridor variance by

n
X
a,b
RVt,T = h(Fti ,T ) − h(Fti−1 ,T ) − h′ (Fti−1 ,T ) × (Fti ,T − Fti−1 ,T ),
i=1

where

2 × − log b − xb + 1 if x > b





h(x) = −2 log x if x ∈ [a, b]


 2 × − log a − x + 1 if x < a


a

and h′ (x) denotes the partial derivative of h(x). As before, the expectation of the realized
volatility under the risk-neutral measure is an option portfolio related to the VIX index,
with options included only if their strikes are between a and b. This corridor realized
variance is a natural extension of the realized variance defined above and has the same
theoretical advantages (the estimation of the risk premium is model-free and independent of
the monitoring frequency).23 To be able to compare realized risk premia over time, rather
than fixing a and b, we calculate the bounds in log moneyness levels α and β. For instance,
if α = −2 and β = 2, on each trading day, we calculate the strikes a and b such that they
coincide with a two and minus two log moneyness level. This allows us to cleanly compare
our estimates over time. The corridor variance premium is a generalization of our earlier
Q h i
definition: V RPt,T = RVt,T /Et RVt,T − 1.
a,b a,b a,b

23
Setting a = 0 and b = ∞ yields the definition of realized variance used above.

32

Electronic copy available at: https://ssrn.com/abstract=4729757

 Table 5: SPX option risk premium returns (in %)
This table provides the return statistics for one-day-to-maturity variance and corridor variance risk
premium trades. For the variance risk premium (labeled Variance in column 1), the hold-to-maturity
return from t to T is defined as V RPt,T = RVt,T /EQ t [RVt,T ] − 1, where RVt,T is the realized variance,
calculated using 5-minute S&P 500 index returns, and EQ t [RVt,T ] is its risk-neutral counterpart, calculated
from option prices. Columns 4 to 8 (mean HPR, t-stat, Skew, Kurtosis and Obs) provide statistics of
the return distribution over the sample from January 2012 to December 2023. For hthe corridor variance
/EQ
i
a,b a,b a,b
premia (labeled Corridor Variance in column 1), defined as V RPt,T = RVt,T t RV t,T − 1, we also
specify a corridor in columns 2 and 3 and only variance accumulated in the corridor is used for the realized
variance calculation and its risk-neutral counterpart. We obtain index-level corridors by choosing, at the
inception of the trade at time t, corridors that correspond to a left and right moneyness bound of α and β,
atm
√

respectively. Moneyness is defined as log(K/Ft,T )/ IVt,T T −t .

mean
Risk Premium α β t-stat Skew Kurtosis Obs
HPR

Panel A: All Observations

Variance – – -26.22 (-17.66) 2.50 10.40 1107
Corridor Variance -2 2 -19.40 (-13.53) 2.04 6.50 1107
Corridor Variance -4 4 -23.49 (-15.64) 2.43 9.79 1107
Corridor Variance -4 0 -30.36 (-11.95) 2.46 9.36 1107
Corridor Variance 0 4 -13.40 (-5.01) 2.16 7.55 1107
Corridor Variance -3 -1 -53.54 (-13.39) 3.83 16.61 1107

Panel B: Observations with Macro Announcement

Variance – – -28.48 (-9.28) 2.27 7.68 313
Corridor Variance -2 2 -21.95 (-7.11) 2.05 5.71 313
Corridor Variance -4 4 -26.17 (-8.43) 2.32 7.96 313
Corridor Variance -4 0 -43.32 (-9.79) 2.61 10.27 313
Corridor Variance 0 4 -1.76 (-0.29) 2.29 8.28 313
Corridor Variance -3 -1 -67.68 (-11.00) 4.79 25.80 313

Panel C: Observations without Macro Announcement

Variance – – -25.33 (-15.09) 2.57 11.26 794
Corridor Variance -2 2 -18.40 (-11.34) 2.04 6.83 794
Corridor Variance -4 4 -22.43 (-13.18) 2.47 10.40 794
Corridor Variance -4 0 -25.26 (-8.35) 2.38 8.82 794
Corridor Variance 0 4 -17.99 (-6.21) 2.02 6.42 794
Corridor Variance -3 -1 -47.96 (-10.17) 3.57 14.40 794

33

Electronic copy available at: https://ssrn.com/abstract=4729757

 Our results in Table 5 provide much more striking differences between these realized
corridor risk-premia for different sub-samples. For the entire sample, we find evidence that
down variance (α = −4, β = 0) has a more negative risk premium (-30.36%, t-statistic
-11.95) than up variance (α = 0, β = 4) (-13.40%, t-statistic -5.01). This difference arises
almost entirely due to macroeconomic announcements in our sample. For the sub-sample of
macroeconomic announcements, the difference between up and down variance is extremely
strong (-43.32% versus -1.76%), whereas it is much more modest for observations without an
announcement (-25.26% versus -17.99%). This suggests that while the overall risk premium
on announcement days is not too different from other trading days, the structure of the risk
premia is very different, with most of the premium arising from priced downside risk and
no evidence of priced upside risk.
The structure of these empirical results is consistent with our macroeconomic announce-
ment model. The restriction that the risk-neutral announcement drift is equal to zero, com-
bined with a positive drift under the real-world measure, increases the relative likelihood
of price increases on announcement days. Therefore, under P, volatility is more likely to
accumulate in corridors on the right rather than the left side of the current index level. With
higher P-realized variances in right-hand-side corridors and no flexibility for the risk-neutral
drift to adjust, the higher realized variance under P pushes the returns of these corridor
swaps downwards. The opposite effect can be observed for left corridors where the likeli-
hood of variance accumulating in these drops on announcement days. Another interesting
feature is that not only does the size of the drift µz matter for the difference between left
and right corridors, but a second important feature is the timing of the announcement. If
announcements occur earlier, they push prices (on average) toward the right corridors early
during the monitoring interval, increasing the likelihood of variance accumulating in right-
hand-side corridors. Most of the announcements in our sample occur before the start of the
regular trading session and, hence, at the beginning of the monitoring interval.

4 Conclusions
This paper provides theoretical and empirical contributions to understanding short-dated
SPX options, in particular those with one day to maturity. Our theoretical contributions

34

Electronic copy available at: https://ssrn.com/abstract=4729757

focus on the impact of event risk on 1DTE options and on deriving model-based higher-order
moments for option portfolios. These analytical results are useful for many applications,
especially for significance tests of observed option returns. Empirically, we find evidence for
significant macroeconomic event risk. We provide evidence that option returns are highly
sensitive to a (statistically and economically) significant macroeconomic return premium.
We also investigate the evidence for a volatility event risk premium but find much weaker
evidence (if any at all).
We are currently working on a number of extensions and robustness checks. We are
considering other return indices to provide further robustness checks and are working on
presenting results for individual macroeconomic announcement categories. We are also
extending the announcement sample to earnings announcements of large corporations.

35

Electronic copy available at: https://ssrn.com/abstract=4729757

References
Almeida, C. and Freire, G. (2022). Pricing of index options in incomplete markets. Journal
of Financial Economics, 144(1):174–205.

Almeida, C., Freire, G., and Hizmeri, R. (2024). 0DTE asset pricing. SSRN working paper.

Bakshi, G., Cao, C., and Chen, Z. (1997). Empirical performance of alternative option
pricing models. Journal of Finance, 52(5):2003–2049.

Bakshi, G., Kapadia, N., and Madan, D. (2003). Stock return characteristics, skew laws,
and the differential pricing of individual equity options. Review of Financial Studies,
16(1):101–143.

Bandi, F. M., Fusari, N., and Renò, R. (2024). 0DTE option pricing. SSRN working paper.

Bates, D. S. (2000). Post-’87 crash fears in the S&P 500 futures option market. Journal of
Econometrics, 94:181–238.

Bates, D. S. (2006). Maximum likelihood estimation of latent affine processes. Review of

Financial Studies, 19(3):909–965.

Beckmeyer, H., Branger, N., and Gayda, L. (2023). Retail traders love 0DTE options... but
should they? SSRN working paper.

Bondarenko, O. (2014). Variance trading and market price of variance risk. Journal of
Econometrics, 180(1):81–97.

Boyer, B. H. and Vorkink, K. (2014). Stock options as lotteries. Journal of Finance,

69(4):1485–1527.

Branger, N., Hansis, A., and Schlag, C. (2010). Expected option returns and the structure
of jump risk premia. SSRN working paper.

Broadie, M., Chernov, M., and Johannes, M. (2007). Model specification and risk premia:
Evidence from futures options. Journal of Finance, 62(3):1453–1490.

Broadie, M., Chernov, M., and Johannes, M. (2009). Understanding index option returns.
Review of Financial Studies, 22(11):4493–4529.

36

Electronic copy available at: https://ssrn.com/abstract=4729757

Carr, P. and Wu, L. (2006). A tale of two indices. The Journal of Derivatives, 13(3):13–29.

Constantinides, G. M., Jackwerth, J. C., and Savov, A. (2013). The puzzle of index option
returns. Review of Asset Pricing Studies, 3(2):229–257.

Corsi, F. (2009). A simple approximate long-memory model of realized volatility. Journal

of Financial Econometrics, 7(2):174–196.

Coval, J. D. and Shumway, T. (2001). Expected option returns. Journal of Finance,

56(3):983–1009.

Dew-Becker, I. and Giglio, S. (2023). Risk preferences implied by synthetic options. Tech-
nical report, NBER.

Dim, C., Eraker, B., and Vilkov, G. (2024). 0DTEs: Trading, gamma risk and volatility
propagation. SSRN working paper.

Dubinsky, A., Johannes, M., Kaeck, A., and Seeger, N. J. (2019). Option pricing of earnings
announcement risks. Review of Financial Studies, 32(2):646–687.

Goyenko, R. and Zhang, C. (2019). Option returns: Closing prices are not what you pay.
Working paper.

Hansen, L. P. (1982). Large Sample properties of generalized method of moments estimators.

Econometrica, 50(4):1029–1054.

Heston, S. L. (1993). A closed-form solution for options with stochastic volatility with
applications to bond and currency options. Review of Financial Studies, 6(2):327–343.

Hu, G. and Jacobs, K. (2020). Volatility and expected option returns. Journal of Financial
and Quantitative Analysis, 55(3):1025–1060.

Israelov, R. and Kelly, B. T. (2017). Forecasting the distribution of option returns. SSRN
working paper.

Johannes, M. S., Kaeck, A., and Seeger, N. (2023). FOMC announcement event risk. SSRN
working paper.

37

Electronic copy available at: https://ssrn.com/abstract=4729757

Jones, C. S. and Shemesh, J. (2018). Option mispricing around nontrading periods. Journal
of Finance, 73(2):861–900.

Kelly, B., Pastor, L., and Veronesi, P. (2016). The price of political uncertainty: Theory
and evidence from the option market. Journal of Finance, 71(5):2417–2480.

Londono, J. M. and Samadi, M. (2023). The price of macroeconomic uncertainty: Evidence

from daily option expirations. SSRN working paper.

Muravyev, D. and Ni, X. C. (2020). Why do option returns change sign from day to night?
Journal of Financial Economics, 136(1):219–238.

Nawalkha, S. K. and Zhuo, X. (2022). A Theory of equivalent expectation measures for

contingent claim returns. Journal of Finance, 77(5):2853–2906.

Neuberger, A. (2012). Realized skewness. Review of Financial Studies, 25(11):3423–3455.

Newey, W. K. (1985). Generalized method of moments spcification testing. Journal of

Econometrics, 29(3):229–256.

Pan, J. (2002). The jump-risk premia implicit in options: Evidence from an integrated
time-series study. Journal of Financial Economics, 63(1):3–50.

Rhoads, R. (2014). Trading weekly Options: Pricing characteristics and short-term trading
Strategies. John Wiley & Sons.

Savor, P. and Wilson, M. (2013). How much do investors care about macroeconomic risk?
Evidence from scheduled economic announcements. Journal of Financial and Quantitative
Analysis, 48(02):343–375.

Storn, R. and Price, K. (1997). Differential evolution - A simple and efficient heuristic for
global optimization over continuous spaces. Journal of Global Optimization, 11(4):341–
359.

Vilkov, G. (2024). 0DTE trading rules. SSRN working paper.

38

Electronic copy available at: https://ssrn.com/abstract=4729757

A Appendix

A.1 Proof of Proposition 1

Proof. Assume the processes are defined on a probability space (Ω, F, P) where ω ∈ Ω are
the elementary events. F = σ (∪t≤T Ft ) where Ft = σ (Ss , Ls : s ≤ t) (conditional expecta-
tions with respect to Ft are indicated by a subscript t). For the call, we can write

n
[ST − K]+

c

n
rt,T = −1
Ct,T (St , K)

where Ct,T (St , K) is the call price. Applying the binomial formula twice, this can be written
as

n X k   
k n
Ct,T (St , K)−k (−1)n−k (−K)k−ℓ (1ST ≥K )k STℓ .
c

n X
rt,T =
k=0 ℓ=0
ℓ k

Therefore, to derive higher-order option return moments, we need to derive moments of

the form EP
h i
t (1 ST ≥K )k ℓ
ST (0 ≤ ℓ ≤ k ≤ n). Suppose k = 0. Then ℓ = 0 and we have
EPt (1ST ≥K )k STℓ = 1. Next, suppose k > 0. Then 1ST ≥K k = 1ST ≥K . Define the Radon-
h i

Nikodym density

ST (ω)ℓ
ΨP
L(ω) = .
0,T (ℓ, V0 , L0 )

L is equal to one in P-expectation, strictly positive (since stock prices are strictly positive)
and well-defined by assumption, hence a valid density.24 The new measure Sℓ is defined by
Z
S (A) =
ℓ
L(ω)P(dω) ∀A ∈ F.
A

For an integrable random variable Y the conditional expectations under P and Sℓ are related
by Bayes formula for conditional expectations (EP S P
t [Y L] = Et [Y ] × Et [L]). Here choose
ℓ

24
Note that for ℓ = 0 the density is equal to one.

A.1

Electronic copy available at: https://ssrn.com/abstract=4729757

Y (ω) = ΨP
0,T (ℓ, V0 , L0 )1ST (ω)≥K , then Y (ω)L(ω) = 1ST (ω)≥K ST (ω) . Hence
ℓ

P = ΨP S P
" #
STℓ
Et 1ST ≥K ST 0,T (ℓ, S0 , L0 )Et [1ST ≥K ]Et
ΨP
ℓ
 ℓ

0,T (ℓ, S0 , L0 )

= ΨP S
t,T (ℓ, St , Lt )Et [1ST ≥K ] .
ℓ

To derive the expectation under the new measure, we use the characteristic function
of log(ST ) under Sℓ :

S (u, St, Lt) = ESt eu log ST EPt Leu log ST ΨP

 
t,T (u + ℓ, St , Lt )
P ΨP
ℓ ℓ  
Ψt,T = = .
Et [L] t,T (ℓ, St , Lt )

Therefore, putting these together and using the standard inversion formula for characteristic
functions yields

Z∞ " −iu log K P

 # 
P
Et 1ST ≥K ST = ΨP 1 + 1 ℜ e
Ψ (iu + ℓ, St , Lt )
P
ℓ
 t,T
t,T (ℓ, St , Lt ) × × du .
2 π iu Ψt,T (ℓ, St , Lt )
0

This concludes the proof for the call (after changing the order of summation). The proof
for the put returns follows the same logic.

A.2 Proof of portfolio return moments

We define the payoff of a call option portfolio consisting of m call options, where ci denotes
the holding of the i-th call option, as PC = m +
P
i=1 ci [ST − Ki ] . We allow Ki ≥ 0, which

implies that one of the call options may be the underlying asset, in which case Ki = 0.
We define the following set
( m
)
(k1 , . . . , km ) ∈ N0 × · · · × N0 :
X
K(n) := ki = n
i=1

which is the set of all non-negative integer-valued m-tuples for which the elements sum to
n. We suppress the dependence on m in our notation. Also define, for each k ∈ K(n),
( m
)
(j1 , . . . , jm ) ∈ N0 × · · · × N0 :
X
J (k) := ji ≤ n, ∀ 1 ≤ i ≤ m : ji ≤ ki
i=1

A.2

Electronic copy available at: https://ssrn.com/abstract=4729757

which collects all non-negative integer-valued m-tuples smaller than the m-tuple k. Fur-
thermore, for k ∈ K(n) and j ∈ J (k) define

n!
γ(n, k) :=
k ! × · · · × km !
1   
k1 km
β(k, j) := × ··· × .
j1 jm

Lemma 1. The n-th power of the payoff of a call option portfolio can be written as

Pm
i=1 ji
1ST ≥K(k)
X X
PCn = α(n, k, j) ST e
k∈K(n) j∈J (k)

where

K(k)
e := max Ki
1≤i≤m, ki >0
m
Y m
Y
α(n, k, j) := γ(n, k)β(k, j) cki i (−Ki )ki −ji .
i=1 i=1

Proof. Writing the i-th call payoff as ci ST 1ST ≥Ki + ci (−Ki )1ST ≥Ki and applying the multi-
nomial theorem to the n-th power of the call option payoff yields:

m
cji i STji 1jSiT ≥Ki cki i −ji (−Ki )ki −ji 1SkiT−j
X X Y
PCn = γ(n, k) β(k, j) i
≥Ki .
k∈K(n) j∈J (k) i=1

1kSiT ≥Ki = 1ST ≥max1≤i≤m, ki >0 Ki concludes the proof.

Qm
Collecting terms and observing that i=1

Lemma 2. Suppose the underlying S has finite moments up to n-th order. Then the n-th
conditional moment of a call option portfolio payoff is given by

EPt [PCn ] =
m
!
X X X
α(n, k, j) × Γ K(k),
e ji
k∈K(n) j∈J (k) i=1

where

Z∞  −iu log K
1 P P

1 e
Γ(K, ℓ) = Ψt,T (ℓ, St , Lt ) + ℜ × Ψt,T (iu + ℓ, St , Lt ) du
2 π iu
0

A.3

Electronic copy available at: https://ssrn.com/abstract=4729757

for K > 0 and Γ(0, ℓ) = ΨP
t,T (ℓ, St , Lt ).

Proof. The results follow from the previous lemma, the linearity of the conditional expecta-
tion and the same line of argument as in Proposition 1, since Γ(K, ℓ) = EP 1
 ℓ

t ST ≥K S T .

To extend these results to general option portfolios, we add a constant (bond invest-
ment) to the call portfolio. With this addition, we can form portfolios of the bond, stock
and call options, which provides full flexibility to include put options by put-call parity. We
define this extended payoff by PC,B = B + m +
P
i=1 ci [ST − Ki ] . This case may be of interest

in its own right, especially when the option portfolio is a zero-cost strategy. Finally, we can
derive the n-th conditional moment of the option portfolio return, provided its price is pos-
itive. Allowing for normalizations with quantities other than the option portfolio value, we
o PC,B
write rt,T = U
− 1 where U > 0 is either the option portfolio value or another normalizing
quantity such as the current stock price (Ft measurable). Putting these two steps together
yields Proposition 2 stated in the main text:

Proposition. Suppose the underlying S has finite moments up to n-th order. The n-th
conditional return moment of a maturity T portfolio consisting of a bond investment (payoff
B at time T ) and call options with strikes K = [K1 , . . . , Km ] ≥ 0 and option holdings
c = [c1 , . . . , cm ], with normalization constant U > 0, is given by

EPt (−1)n−q U −q B q−p × EP

q   
n X
 o

n  X n q p
rt,T = t [PC ] (A.1)
q=0 p=0
q p

where

EPt [PCn ]
m
!
X X X
= α(n, k, j) × Γ K(k),
e ji
k∈K(n) j∈J (k) i=1
 Z∞  −iu log K
P P

 1 1 e
 Ψt,T (ℓ, Vt , Lt ) + ℜ × Ψt,T (iu + ℓ, Vt , Lt ) du if K > 0


Γ(K, ℓ) = 2 π iu
0
ΨP (ℓ, V , L )



t,T t t if K = 0
K(k)
e = max Ki
1≤i≤m, ki >0
m
Y m
Y
α(n, k, j) = γ(n, k)β(k, j) cki i (−Ki )ki −ji .
i=1 i=1

A.4

Electronic copy available at: https://ssrn.com/abstract=4729757

The sets K(n) and J (k) and the functions γ(n, k) and β(k, j) are defined in the text above.

Proof. This again follows directly from the Binomial theorem, and an application of the
previous Lemma.

A.3 GMM Estimation Details

To jointly estimate model parameters of the Merton model with macroeconomic announce-
ment event risk under P and Q we use a GMM approach as laid out in Section (3.2.2). To
do so, we minimize the quadratic form using differential evolution as proposed in Storn and
Price (1997). Following Hansen (1982), the optimal weighing matrix for the second step
is obtained by inverting the variance-covariance matrix of ψ(x;
b Θ1 ) where Θ1 refers to the

consistent estimates obtained from the first step of GMM. Since our theoretical moment con-
ditions are essentially point estimates (for announcement and non-announcement days), the
variance-covariance matrix essentially reflects the variation in empirical moments. Further,
the moment conditions on announcement days are independent of the moment conditions
on non-announcement days. The second stage optimal moment vector Θ2 = Θ∗ is used to
calculate standard errors using
r  
b′ Ω
diag (G b −1 G)
b −1

h√ √ i  ∗)

where Ω
b = E
b T ψ (x; Θ∗ ) T ψ (x; Θ∗ )′ and the gradient G = E ∂ψ(x;Θ
∂Θ ∗ is approxi-
mated numerically using simultaneous perturbation. Finally, since the parameters are over-
identified we test for the validity of the moment specification as proposed by Newey (1985).
b Θ∗ ) = 0.003 and hence fail to reject the null hypothesis that all moment
We obtain T ψ(x;
conditions have zero expectation at Θ∗ .

A.4 Appendix: Further Results

A.5

Electronic copy available at: https://ssrn.com/abstract=4729757

 Table A.1: SPX Option Hold-to-Maturity Returns (in %): Standard Moneyness
This table provides the average returns for one-day-to-maturity call and put options for different moneyness levels. Moneyness is defined as K/F . To
calculate the average returns, for each option type in our sample that has a one-day-to-maturity option, we choose the strike closest to the Moneyness level
in column 1 (with a maximum tolerance of 1%) and record the return of the call and put option, respectively. Call AvgRet (Put AvgRet) is the average call
(put) option return, in percentage. t-stat provides the HAC-corrected (probably highly unreliable) t-statistic for the mean, Loss % gives the likelihood of
complete capital loss (-100%, i.e. the option is out-of-the-money at maturity). Skew and Kurt are the skewness and the kurtosis of the returns. Obs are
Electronic copy available at: https://ssrn.com/abstract=4729757

the number of observations.

Moneyness Call Option Contract Return Statistics Put Option Contract Return Statistics
t-stat Loss t-stat Loss
mean t-stat Skew Kurt mean t-stat Skew Kurt
BS (%) BS (%)

Panel A: From 2012-01-01 to 2023-12-05 (1109 observations)

0.970 1.34 (1.28) (0.33) 1 -0.19 1.61 -79.98 (-8.19) (-0.38) 99 23.76 637.03
0.980 1.94 (1.36) (0.30) 4 0.08 0.72 -46.89 (-2.37) (-1.16) 96 26.38 788.21
0.990 3.27 (1.51) (0.32) 12 0.44 -0.04 -24.55 (-2.26) (-1.84) 88 9.05 103.96
A.6

1.000 8.12 (1.87) (0.85) 46 1.27 0.75 -9.14 (-1.88) (-1.16) 55 3.06 16.02
1.010 -3.57 (-0.30) (-0.99) 85 10.82 188.95 -3.97 (-1.67) (-0.65) 15 1.00 1.36
1.020 -68.94 (-7.58) (-2.32) 97 15.83 300.07 -2.89 (-1.96) (-1.01) 3 0.55 1.19
1.030 -87.62 (-11.46) (-0.59) 99 28.89 893.43 -1.93 (-1.87) (-0.98) 1 0.39 1.61

Panel B: From 2016-03-01 to 2023-12-05 (950 observations)

0.970 1.30 (1.13) (0.29) 2 -0.20 1.64 -76.63 (-6.74) (-0.40) 98 21.98 545.13
0.980 2.01 (1.30) (0.35) 4 0.09 0.75 -38.70 (-1.68) (-0.94) 96 24.45 676.24
0.990 3.58 (1.54) (0.46) 12 0.47 -0.00 -21.63 (-1.77) (-1.51) 88 8.93 100.29
1.000 9.60 (2.06) (1.15) 46 1.27 0.72 -8.89 (-1.67) (-1.07) 54 3.21 17.22
1.010 -1.76 (-0.14) (-0.73) 85 10.96 186.14 -4.16 (-1.62) (-0.71) 15 1.05 1.50
1.020 -65.24 (-6.19) (-2.21) 97 14.81 260.89 -2.85 (-1.78) (-0.92) 3 0.59 1.24
1.030 -85.54 (-9.59) (-0.63) 99 26.73 764.59 -1.92 (-1.69) (-0.90) 1 0.41 1.63

Panel C: From 2020-01-01 to 2023-12-05 (562 observations)

0.970 0.29 (0.17) (-0.16) 2 -0.01 0.74 -78.36 (-8.33) (-0.72) 98 13.08 181.22
0.980 0.50 (0.22) (-0.12) 6 0.23 0.11 -54.95 (-4.69) (-1.73) 94 7.62 66.87
0.990 0.61 (0.18) (-0.16) 17 0.55 -0.35 -17.30 (-1.40) (-1.22) 83 6.06 46.35
1.000 6.20 (1.05) (0.70) 49 1.24 0.59 0.54 (0.08) (0.38) 51 1.85 3.50
1.010 11.37 (0.84) (0.71) 81 4.88 30.57 -1.08 (-0.29) (0.02) 19 0.82 0.43
1.020 -56.66 (-4.57) (-2.03) 95 9.97 111.60 -1.71 (-0.72) (-0.42) 5 0.40 0.30
1.030 -77.90 (-5.24) (-0.87) 99 21.20 472.28 -1.17 (-0.69) (-0.41) 1 0.25 0.67
 Table A.2: SPX hold-to-maturity returns (in %): Macroeconomic announcements (standard moneyness)
This table provides the average returns for one-day-to-maturity call and put options for different moneyness levels. Moneyness is defined as K/F . To
calculate the average returns, for each option type in our sample that has a one-day-to-maturity option, we choose the strike closest to the Moneyness level
in column 1 (with a maximum tolerance of 1%) and record the return of the call and put option, respectively. Call AvgRet (Put AvgRet) is the average
call (put) option return, in percentage. t-stat provides the HAC-corrected t-statistic for the mean and we also provide t-statistics based on Black-Scholes
moments (t-stat BS ). Loss % gives the likelihood of complete capital loss (-100%, i.e. the option is out-of-the-money at maturity). Skew and Kurt are the
Electronic copy available at: https://ssrn.com/abstract=4729757

skewness and the kurtosis of the returns. Obs are the number of observations.

Moneyness All Observations With Macro Announcements Without Macro Announcements

t-stat t-stat t-stat
mean t-stat Obs mean t-stat Obs mean t-stat Obs
BS BS BS

Panel A: Calls
0.970 1.34 (1.28) (0.33) 1109 4.43 (2.19) (1.95) 314 0.12 (0.11) (-0.84) 795
0.980 1.94 (1.36) (0.30) 1109 6.14 (2.23) (1.79) 314 0.27 (0.18) (-0.77) 795
0.990 3.27 (1.51) (0.32) 1109 10.34 (2.56) (1.79) 314 0.48 (0.21) (-0.74) 795
A.7

1.000 8.12 (1.87) (0.85) 1109 22.15 (2.75) (2.10) 314 2.57 (0.56) (-0.32) 795
1.010 -3.57 (-0.30) (-0.99) 1109 18.48 (0.92) (0.58) 314 -12.28 (-0.87) (-1.53) 795
1.020 -68.94 (-7.58) (-2.32) 1109 -31.07 (-1.03) (-0.64) 314 -83.90 (-14.39) (-2.33) 795
1.030 -87.62 (-11.46) (-0.59) 1108 -64.50 (-2.12) (-0.24) 314 -96.76 (-45.54) (-0.54) 794

Panel B: Puts
0.970 -79.98 (-8.19) (-0.38) 1109 -88.42 (-9.97) (-0.23) 314 -76.64 (-5.80) (-0.31) 795
0.980 -46.89 (-2.37) (-1.16) 1109 -71.56 (-6.81) (-1.02) 314 -37.15 (-1.36) (-0.73) 795
0.990 -24.55 (-2.26) (-1.84) 1109 -49.10 (-3.88) (-2.30) 314 -14.85 (-1.04) (-0.73) 795
1.000 -9.14 (-1.88) (-1.16) 1109 -28.22 (-3.60) (-2.97) 314 -1.60 (-0.27) (0.50) 795
1.010 -3.97 (-1.67) (-0.65) 1109 -10.60 (-2.39) (-1.89) 314 -1.36 (-0.51) (0.41) 795
1.020 -2.89 (-1.96) (-1.01) 1109 -6.97 (-2.49) (-2.12) 314 -1.28 (-0.80) (0.14) 795
1.030 -1.93 (-1.87) (-0.98) 1108 -4.84 (-2.40) (-2.19) 314 -0.79 (-0.71) (0.22) 794
 Figure A.1: Average hold-to-maturity returns: Standard moneyness
This figure provides average SPX option 1-day to maturity holding period returns for our sample period
January 2012 to December 2023. We separately show holding period returns for calls and puts, with and
without macroeconomic announcements occurring during the holding period. We show returns for up to 3%
out of the money, sampled at intervals of 0.5%. Shaded areas represent HAC-adjusted standard error bounds.
Average HTM Returns (Call Options) Average HTM Returns (Put Options)
40 0
20
20
0
20 40
40 60
60
Without Macro Announcement 80 Without Macro Announcement
80 All Observations All Observations
100 With Macro Announcement With Macro Announcement
100
0.97 0.98 0.99 1.00 1.01 1.02 1.03 0.97 0.98 0.99 1.00 1.01 1.02 1.03
Moneyness Moneyness

A.8

Electronic copy available at: https://ssrn.com/abstract=4729757