0DTEs: Trading, Gamma Risk and Volatility Propagation∗

Chukwuma Dim† Bjørn Eraker‡ Grigory Vilkov§

March 14, 2025

Abstract
We study the recent explosion in trading of same-day expiry (0DTE) options on the S&P500
index and examine if this trading activity is destabilizing for the underlying index. We
find that Market Makers’ inventory, as measured by the net gamma of their positions, is on
average positive and negatively related to future intraday volatility. We also show evidence
suggesting that positive (negative) Market Makers’ inventory gamma strengthens intraday
price reversal (momentum). Our empirical evidence is consistent with delta-hedging but
inconsistent with information-based trading.

Keywords: 0DTE, ultra-short-term options, variance risk premium, volatility trading, gamma
risk, volatility propagation

∗
We appreciate the suggestions from the Editor and three anonymous referees. We also received helpful
comments from Caio Almeida (discussant), Svetlana Bryzgalova, Jefferson Duarte, Amit Goyal, Neil Pearson
(discussant), George Skiadopoulos and Christian Wagner (discussant). We also thank participants of the Cboe
RMC 2023 Academic Roundtable, seminars at Arrowstreet Capital, Deutsche Bank Quant Summit, BlackRock
Global Trading Research, Federal Reserve Board, Frankfurt School, University of Porto, Manchester University,
University of Houston, and Erasmus School of Economics, Frontiers in Quantitative Finance seminar, Derivatives
and Asset Pricing Conference in Cancun, Liverpool Workshop in Option Markets, the 16th Annual SoFiE Meeting
and Western Finance Association Meeting for useful comments.
†
George Washington University; E-mail: [email protected].
‡
Wisconsin School of Business; E-mail: [email protected].
§
Frankfurt School of Finance & Management; E-mail: [email protected].
1 Introduction

The volume of short-dated options trading, especially zero-day-to-expiry (0DTE), has exploded

in recent years. For the S&P500 index alone, 0DTEs accounted for 50% of the index options

volume in August 2023, up from just 5% in 2016.1 The major trading hub for equity options,

Chicago Board Options Exchange (Cboe), has sequentially increased the number of weekly index

option expiration dates to three in 2016 and five in 2022 to eventually have options that expire

every day of the week for the next four weeks. The leading global derivatives exchange CME

Group now offers 0DTEs on the aggregate market (both S&P500 and NASDAQ100) futures.

The surge in 0DTE option trading has raised several concerns among market participants and

stimulated heated discussions in the press.2 The primary concern is that large open positions

and trading flows in 0DTEs may induce delta-hedging needs that can destabilize the underlying

market, even though the underlying instruments are very liquid as in the case of the S&P500

index-based exchange-traded fund (ETF) SPY and S&P500 E-Mini (ES) Futures.3

The rationale behind these concerns, which has been modeled theoretically (e.g., Jarrow 1994,

Sircar and Papanicolaou 1998, Schönbucher and Wilmott 2000), is that if option sellers delta-

hedge, they trade in the direction of realized returns, e.g., sell additional shares of the underlying

asset during a market decline, propagating the return shocks and strengthening momentum. The

intensity with which delta hedgers re-adjust their positions in the underlying is higher for short-

maturity options with strikes close to the current price level because these options’ delta is

more sensitive (measured by “gamma,” which is inversely related to option time to expiry) to

changes in the underlying. Thus, if market makers hold large short (negative gamma) inventory

in 0DTEs and systematically delta-hedge it, their hedging flows can aggravate sudden market
1
E.g., Volatility Insights: Much Ado About 0DTEs-Evaluating the Market Impact of SPX 0DTE Options.
2
See, for example, ”Surge in zero-day options sparks fears over market volatility,” or ”Short-term investors in
SPY and QQQ warned of options risks.”
3
Average daily trading volume in these markets in 2023 was around 400 bn USD.

1
moves. Cboe, who has observed the largest trading flows in S&P500 index 0DTEs, disputes

such scenarios, claiming that these flows and the resulting market makers’ exposure during

the day are well-balanced and do not threaten market stability. Moreover, when the market

makers’ intraday net gamma is positive, delta-hedging generates trades against the realized

return direction, decreasing return persistence and potentially dampening future volatility.

This paper investigates these contrasting arguments to understand the effects of the 0DTEs’

trading activity on the underlying markets. We test two main channels through which 0DTE

trading could impact or be linked to underlying trading and return dynamics. First, we examine

whether market makers’ delta-hedging needs and their potential execution in the underlying

markets affect subsequent volatility and return of the underlying (gamma channel). Next, we

examine an information-based channel in which market participants with superior information

about near-term future volatility trade in 0DTEs prior to changes in volatility. Both channels

can create the same empirically observed patterns, but the gamma channel specifies a direct

cause-and-effect relationship, while the information channel suggests that observed patterns

arise from traders’ anticipatory actions rather than market makers’ hedging flows.

Using intraday data on the underlying variance and the market makers’ 0DTEs net gamma,

we establish a negative and significant association between the market makers’ 0DTEs net

gamma and the underlying index variance, consistent with market makers’ delta-hedging affect-

ing the underlying market. We establish this result in a specification that uses high-dimensional

fixed effects to account for intraday market dynamics and controls for prior realizations of the

underlying variance, returns, and trading volume, thereby accounting for potential 0DTEs trad-

ing based on public information. In magnitude, we document a decline in the underlying log

variance by 7.3% of its standard deviation following a standard deviation (SD) increase in market

makers’ 0DTEs net gamma. This effect is non-negligible, as it equals 32% of the decline in un-

2
derlying log variance following one SD increase in market makers’ net gamma for all longer-term

options with one-day to one-month maturities.

The observed negative association between the market makers’ 0DTEs net gamma and the

underlying variance could result in volatility amplification if, as feared by the media and some

market participants, market makers have negative 0DTEs gamma inventory. It is, however, not

what our analysis suggests. For most of our sample period, market makers are predominantly

long gamma in 0DTEs, implying that their delta-hedging has the opposite effect of dampening

the underlying index volatility rather than amplifying it. Moreover, since there are periods

in our sample where market makers are short gamma in 0DTEs, we can investigate how the

economic magnitude of their potential delta-hedging effect on the underlying variance compares

with that of their long gamma exposure. Indeed, there is potential for market makers’ delta

hedging of their negative 0DTEs inventory to amplify the underlying variance. However, the

economic magnitude of such variance amplification is 65% lower than that of the potential

variance attenuation arising from market makers delta-hedging their long gamma exposure.4

We further quantify the economic effect of the gamma channel on the underlying return

persistence. We construct an intraday momentum strategy and find that its performance is

strongly negatively related to market makers’ 0DTE net gamma, with 10 basis points per hour

difference in average returns between high and low gamma states. We do not find similar results

for market makers’ net gamma in longer-term options, and it indicates that the sizeable hedging

needs engendered by market makers’ 0DTEs exposure tend to significantly affect intraday return

persistence, boosting momentum for low gamma and reversal for high gamma states.

An important consideration is whether the established results are genuinely due to market

makers’ delta-hedging or stem from other market patterns, such as information-based 0DTEs
4
In a related exercise, Vasquez, Amaya, Pearson, and Garcia-Ares (2025) quantifies the maximum negative
return propagation from 0DTEs conditional on the observed data. Their estimates suggest that the maximum
impact of net gamma is around 3.3 percentage points in annual terms for daily volatility, well within the range of
daily changes in annualized realized volatility.

3
trading. We conduct several tests examining this consideration. Besides controlling for the

public signals informative about future underlying volatility in our baseline analysis, which

helps address some endogeneity concerns, we identify a plausibly exogenous instrument for

market makers’ 0DTE net gamma. We then re-examine and confirm our baseline results using

instrumental variable analysis. Furthermore, we directly test whether observable public signals

about future intraday underlying volatility shape the evolution of market makers’ 0DTEs net

gamma and reject this hypothesis.

Finally, we entertain the possibility that 0DTEs trading flow, absorbed by market makers,

is driven not by public but rather by private information. Hence, we assess whether market

makers’ 0DTEs net gamma significantly changes in the trading windows preceding the release of

major unexpected and potentially market-moving news by the media, which privately informed

agents could have anticipated and traded on. Again, we do not find any evidence for such

a pattern. Overall, we rely on a theoretically sound causal mechanism and the rejection of

plausible alternative explanations to support our empirical evidence, consistent with a causal

effect of market makers’ 0DTE exposure on the underlying return volatility.

In the complementary analysis, we study the underlying market dynamics conditional on

abnormal trading activity in 0DTEs. We find that the distribution of intraday returns condi-

tional on large jumps in 0DTEs volume is similar conditional on moderate and small changes in

volume. Quantile regressions show that the magnitude of positive returns on average decreases

after intraday 0DTE volume jumps, whereas the magnitude of negative returns is unaffected.

Importantly, our analysis indicates that 0DTEs volume does not significantly propagate past

underlying returns, which is unsurprising given that a sizeable portion of 0DTEs trading flow is

balanced with no significant effect of the 0DTE flow delta on subsequent volatility in the gamma

channel tests.

4
 To further understand the integration of the underlying and 0DTE markets, we use struc-

tural vector autoregression to analyze joint intraday dynamics of the 0DTE and underlying

assets trading volumes and the underlying variance. 0DTEs and underlying markets are rapidly

becoming more integrated, with a contemporaneous correlation between intraday trading flows

increasing from 0.25-0.30 before 2021 to 0.58 in 2023-24. Positive shocks to 0DTE trading

volume in recent years are associated with and followed by increasingly higher trading volume

in the underlying and vice versa. Despite this increasing market integration, the difference in

the magnitude of the average underlying variance response to 0DTE trading across the early

and later sample periods amounts to an economically negligible 0.15 standard deviations of the

return variance, corroborating the earlier results based on 0DTE volume jumps.

We contribute to several strands of research. We build on the methodology of Ni, Pearson,

Poteshman, and White (2021) who analyze the link between volatility and market makers’

lagged inventory net gamma in stock options. They suggest that more positive inventory gamma

leads to lower volatility, consistent with the delta-hedging effect. Our analysis offers important

additional insights for 0DTE index options markets: delta hedging of 0DTE positions has a

larger potential for market destabilization due to more frequent and extensive rebalancing than

for longer-term options, even in the case of the S&P500 index options, whose underlying markets

are far more liquid than individual stocks. The potential delta hedging due to 0DTEs exposure

has a distinct and non-negligible effect on the underlying index variance compared to longer-term

options analyzed in the earlier studies. A follow-up paper by Adams, Fontaine, and Ornthanalai

(2024) confirms our results by linking intraday net gamma in 0DTEs to subsequent underlying

volatility. In addition, they use variations in 0DTE trading on Tuesdays and Thursdays due

to public holidays to suggest a causal effect of 0DTE trading on the underlying volatility. Our

analysis suggests that this result is sensitive to the empirical design, and 0DTEs trading per

se does not necessarily imply high market markers’ delta-hedging needs, which is the primary

5
mechanism our paper investigates. Another subsequent study that complements ours, Vasquez,

Amaya, Pearson, and Garcia-Ares (2025), estimates models that link market volatility to market

makers’ inventory gamma to quantify the maximum impact of 0DTE positions on volatility.

They find it to be around 3.3 percentage points (pp) per annum for daily volatility, well within the

range of daily changes in annualized volatility, and 6.4 pp for 30-minute intraday volatility. These

results complement our findings, suggesting that when market makers hold negative gamma

inventory, the associated delta-hedging flows can moderately amplify underlying volatility.

Brogaard, Han, and Won (2023) examine the impact of 0DTE trading on daily volatility

and find that 0DTEs’ relative turnover positively relates to future volatility.5 Baltussen, Da,

Lammers, and Martens (2021) link market intraday momentum to the market makers’ gamma

hedging demand. Barbon and Buraschi (2020) show that autocorrelations are significantly re-

lated to the difference between call and put options’ gamma. To our knowledge, our paper is the

first to systematically study the effects of gamma exposures in 0DTEs on autocorrelation and

volatility of intraday returns and analyze joint dynamics of the 0DTE and underlying markets.

A growing number of papers study patterns in 0DTEs trading (Beckmeyer, Branger, and

Gayda 2023), work on specially designed pricing models for short-term options (Bandi, Fusari,

and Renò 2023), and document stylized asset pricing facts about 0DTEs and ultra short-term

options (Almeida, Freire, and Hizmeri 2023, Vilkov 2023, Johannes, Kaeck, Seeger, and Shah

2024). Some papers document similar results for longer-term options. Anderegg, Ulmann, and

Sornette (2022) suggest that exchange rate volatility increases with the aggregate option gamma,

such that delta hedgers’ order flow leads to a 0.7% (0.9%) increase in EUR/USD (USD/JPY)

annualized volatility. Sornette, Ulmann, and Wehrli (2022) study Gamestop’s stock price in

Spring 2021, noting that as the stock price rose, call option sellers were forced to buy, leading
5
We compare the results and reconcile some of the differences in the approaches in Section 5.

6
to a price spiral. Lipson, Tomio, and Zhang (2023) show that shocks to retail option trading

due to Robinhood’s introduction of options increased optionable stocks’ return volatility.

The remainder of the paper is organized as follows. Section 2 conceptualizes the links between

0DTE trading and the underlying process. Section 3 discusses the data, definition of variables

and presents the summary statistics for market composition and dynamics. Section 4 formally

tests the hypotheses related to the channels linking 0DTE trading and underlying volatility.

Section 5 presents robustness tests and additional analysis, and Section 6 concludes. Appendix

and Internet Appendix contain additional materials for the analysis.

2 0DTE Trading and the Underlying Prices

Unlike the underlying stocks or index futures, which offer investors directional exposure, options

expose traders to more types of risks, including time-varying directional risk (delta), realized

(gamma) and implied (vega) volatility risks, time decay (theta), and others. Depending on the

type of trading business and trading objectives (for example, market making with a balanced

inventory and limited directional risk or speculation with both directional and realized volatility

exposure), some options market participants delta hedge their options inventory to limit their

directional risks.

Due to their very short expiration time, some features of 0DTEs are notably different from

those of longer-term options. In particular, 0DTEs are highly sensitive to realized volatil-

ity, measured by gamma, and for close-to-at-the-money options, this sensitivity increases al-

most exponentially as expiration approaches. Hence, while 0DTEs can enable targeted bets

around volatility-linked events, they feature stronger motivations for delta-hedging to limit the

associated risks. In turn, delta-hedging trades can impact the underlying dynamics for non-

informational or non-fundamental reasons.

7
 This non-fundamental-based impact of delta-hedging on the underlying has been derived the-

oretically and extensively tested empirically for stock and foreign exchange long-term options. A

key assumption is that some trader groups, primarily market makers, who absorb other traders’

trading flows, maintain approximately delta-neutral inventory.6 In contrast to information-based

trading, where traders adjust options positions in anticipation of future events, the delta-hedging

channel generates buying or selling pressure in the underlying conditional on the realized re-

turn and inventory characteristics. When the delta-hedging trade direction coincides with the

previous return, it leads to stronger return persistence, magnifying underlying volatility. In the

opposite case, delta-hedging causes lower return autocorrelation, dampening underlying volatil-

ity. We label this pattern the gamma channel because it implies a negative effect of market

makers’ 0DTEs inventory gamma on the underlying variance.

The gamma channel creates a causal effect of 0DTE activity on the underlying, and to test

it, we first illustrate the relevant mechanism and define its main drivers. Assume that the

underlying index price St follows a diffusion process with locally constant volatility so that an
∂V
option’s delta ∆t = ∂S |S=St is a function of stock price and time to expiry ∆t = f (St , T − t).

The delta of an options portfolio is a sum of the deltas of options in the portfolio, and in

the following derivations, we directly work with the delta of an aggregate inventory of market

makers. Assume that at t0 we hold a portfolio with a desired level of market exposure (delta

level). To quantify the rebalancing required to maintain the same level of exposure (ignoring

for now the change in the delta due to trading flow in options) at a later point in time t1 , apply

Ito’s Lemma to ∆t :

2
∂2∆ ∂2∆

∂∆ ∂∆ dS dS ∂∆
d∆ = dS + 0.5 2 (dS)2 + dt = SΓ + 0.5S 2 2 + dt,
∂S ∂S ∂t S ∂S S ∂t
6
Delta-hedging can be expensive, especially in 0DTEs because their gamma and, hence, the required intensity
of rebalancing, are increasing sharply several hours to expiration. Following Ni, Pearson, Poteshman, and White
(2021), we assume that market makers are the major group of traders realistically able to delta hedge 0DTEs.

8
so that a discrete change in delta following small realized return rt0 ,t1 over a short period of

time is approximately

∂2∆ ∂∆
∆t1 − ∆t0 ≈ S × Γ × rt0 ,t1 + 0.5S 2 × 2
× rt20 ,t1 + (t1 − t0 ). (1)
∂S ∂t

The first term quantifies the direct gamma effect, the second term quantifies the effect of gamma

change, called “speed,” and the third term quantifies delta time-decay, called “charm.” It is cus-

tomary to neglect the second-order and time effects for small time intervals, and we concentrate

on the direct gamma effect in the main analysis.7 Assuming that ∆t0 is at an optimal level, the

market maker needs to offset the change in dollar (or cash) delta computed as

c ≈ S 2 × Γt × rt ,t = $Γt × rt ,t ,
$∆ (2)
0 0 1 0 0 1

where $Γt0 = S 2 × Γt0 is the dollar (or cash) gamma of the portfolio. Thus, the direction of a

delta-hedging trade relative to a future realized return direction depends only on the sign of the

portfolio dollar gamma. A delta-hedger with a positive portfolio gamma sells the underlying

asset after positive returns and buys it after negative returns, thus decreasing autocorrelation

in underlying returns and dampening the subsequent volatility. Delta-hedgers with negative

gamma will do the opposite, propagating large jumps, increasing autocorrelation, and potentially

boosting volatility. In addition to delta rebalancing of the existing inventory, market makers

trade periodically in the underlying market to offset the net delta of the order flow. The effect

of such offsetting trades on the underlying depends on the net delta of the aggregated order

flow and the return of the underlying in the period of the trade execution. Anecdotal evidence

suggests that 0DTEs’ order flow is composed mainly of client trades in both directions and is

relatively well-balanced in the short run, but we will formally test the significance of flow delta.
7
In the Robustness and Extensions Section 5, we also discuss the effect of speed and charm and show that
including these higher-order effects does not affect our conclusion.

9
 While the gamma channel generates a non-fundamental effect on the underlying, there are

scenarios where information-based trading in options (e.g., Easley, O’Hara, and Srinivas 1998,

Pan and Poteshman 2006) can generate patterns that resemble the gamma channel. For example,

suppose traders have information that future volatility will be high, and as a result, buy 0DTEs.

If market makers take the opposite side of the trade, their gamma inventory will decline. If

subsequent volatility turns out higher, as informed traders anticipated, it will imply a negative

relationship between market makers’ gamma inventory and subsequent volatility. Although

such a negative relationship is due to informed options trading, it may be mistaken for market

makers’ delta hedging effect. On the other hand, if market makers, rather than simply supplying

liquidity in the market, trade options for information reasons and are better informed than other

traders, there can be a positive relationship between the market makers’ inventory gamma and

future volatility. We label these patterns the information channel and try to distinguish them

from the gamma channel in our tests.8

Besides the gamma and information channels, 0DTEs trading can also affect the underlying

volatility if, as described in French and Roll (1986), trading in the options market substitutes

trading in the underlying. Surge in 0DTEs volume could also dampen underlying volatility if it

increases liquidity in the underlying market by encouraging informed investors to trade options

instead of the underlying, thereby reducing the risk that market makers in the underlying market

are trading against more informed investors, ultimately lowering underlying bid-ask spreads and

volatility in equilibrium (Skinner 1989). In our tests, we investigate and account for these

complementary 0DTEs trading and underlying volatility links.

In sum, the foregoing discussions motivate and inform our subsequent tests, which aim to

document how activity in the 0DTEs market affects the underlying index volatility.
8
It should be noted that because our analysis concentrates on the most liquid broad index, these information-
based considerations, while conceptually valid, are unlikely to obtain in reality. These considerations are much
more likely to apply in the context of individual stocks given their larger scope for obtaining and exploiting private
information, slower pace of incorporating information into prices, and the lower liquidity for some stocks.

10
3 Data Preparation and Summary Statistics

We describe data processing in Section 3.1 and variable construction in Section 3.2, and then

present selected summary statistics in Section 3.3.

3.1 Data Sources

Options. We work with options data for the S&P500 index (SPX) and the SPDR S&P500

ETF Trust (SPY). Index options are European, cash-settled against the S&P500 index, either

at the market close for the weekly options with root SPXW, or at the market open for the

“standard” SPX options. SPX has its expiration on the third Friday of each month, and SPXW

expired once per week before September 2016, then three times per week, and then the fourth

and the fifth days were added by Cboe on April 18, 2022, and May 11, 2022, respectively. SPY

options are American and settled physically on the evening of the expiration date. Currently,

SPY options have a similar expiration schedule to index options. For both assets, there are

options with expiration on each day for the next month and some longer maturities. SPY price

is roughly 1/10th of the index, and the notional of one SPX option is about ten times that of

an SPY contract. Small notional makes SPY options popular among retail investors, and we

include it in parts of our analysis.9

We use three intraday data formats for the options: 30-minute bars, 10-minute Cboe open-

close volume summary, and actual transactions, all from Cboe DataShop. Intraday bars from

2012 to 4/2024 include national best bid and offer, trade volumes, underlying instrument price,

open interest at the beginning of each day, implied volatilities, and selected sensitivities (Greeks,
9
One can also get exposure to S&P500 options using futures options, but due to data availability, we do not
include them in the analysis, though we expect no significant differences in the results. Note also that SPX index
options trading takes place exclusively on the C1 platform, while ETF (SPY) options are traded on multiple
exchanges. To reconstruct open interest by trader types, one needs to observe all transactions; hence, we only
use index options for the analysis of intraday open interest based on the open-close volume data.

11
including delta and gamma).10 Open-close volume summary data for the C1 platform from

2021 to 6/2023 provide trading volumes for option contracts traded on C1, split into aggregate

buying and selling flows classified by trader types: market makers, customers, pro-customers,

firms, and broker-dealers. Transactions data from 2012 to 4/2024 represent enhanced Options

Price Reporting Authority data. It includes trade price and size, the exchange where the trade

printed, the underlying bid-ask, implied volatility, and the trade delta. The provider does not

clean this data, and we parse it using filters and procedures similar to Bryzgalova, Pavlova, and

Sikorskaya (2023).11 After parsing, we merge simultaneous trades with the same conditions and

then aggregate the data to the 1-minute bars.

In different tests, we use options on the S&P500 with both morning (AM) and regular (PM)

expiration schedules (roots SPX, SPXW, and SPY) expiring within the next 30 calendar days.

We compute each option’s days-to-expiry (DTE) on each day, counting only working days, and

concentrate our analysis on options with extremely short maturities. Analysis based on open

interest split by trader types uses only SPXW (for all maturities) and SPX (for non-0DTEs

only) options.

Underlying Markets. We obtain the end-of-the-day (EOD) closing prices and 1-minute

OHLC (open, high, low, close) price bars for SPX index and OHLC with traded volume for
10
We recompute implied volatilities and Greeks, including speed and charm, for deep in- and far out-the-money
options when they are missing in data. We exclude seven dates (24, 26/10, 14/11, and 10, 12, 19, and 24/12 of
2018) from the analysis due to missing open interest numbers.
11
We appreciate having access to the paper replication package at https://tinyurl.com/reppackage, and port the
original R code to Python with slight adjustments. We keep the trades with zero canceled trade condition id,
positive trade size and trade price, non-negative bid-ask spread spread, underlying bid weakly larger than
0.01 (vs. 0.1 in Bryzgalova et al.), and trade price between best bid - spread and best ask + spread.

12
SPY and the continuous front contract of the S&P500 Mini-futures (ES) from DTN IQFeed.12

We add up their volumes to get the underlying volume.

3.2 Construction of the Main Variables

Realized Variances and Intraday Return Distributions. We compute intraday returns

for different time intervals by aggregating 1-minute realized SPX returns. Realized variances

(RV ) for intraday time intervals from a given point in time t0 to a subsequent point t1 on the

same day equal to the sum of squared 1-minute SPX log returns:

1 −1
tX
2
RVt0 ,t1 = rt,t+1 . (3)
t=t0

Trading Activity and Risk Variables. To analyze the dynamics of market activity and

aggregate risks in the options markets, we use intraday open interest by trader types computed

from the Cboe open-close volume summary data, start-of-day aggregate open interest as reported

for each contract in bar data, and trading volume computed from both 30-minute bar data and

actual transactions (for higher frequency). Volumes and open interest are converted to dollar

terms, either dollar notional or cash Greeks.

Start-of-day open interest is aggregated over a set of contracts C (combination of option

roots j ∈ (SP X, SP XW, SP Y ), option type, and days to expiry), and expressed in terms of
12
In the first half of 2023, daily trading volumes in SPY and ES front contract were around 100 million shares
and 1.5-2 million contracts per day, respectively. Futures have a notional of 50 units of the index, and SPY is
approximately 1/10 of the index; hence, turnover in SPY and ES corresponds to 10 and 75-100 million units of the
index, respectively. Thus, SPY is approximately ten times less liquid than ES, and the latter has the advantage
of overnight trading sessions. Anecdotal evidence suggests that delta-hedging of index options happens in both
markets, but large traders prefer Minis.

13
dollar notional, cash (dollar) delta and gamma, respectively:

OId$ = 100 ×
P
c∈C OId,c × Sj,d

OId$∆ = 100 × (4)

P
c∈C OId,c × $∆d,c

OId$Γ = 100 ×
P
c∈C OId,c × $Γd,c ,

2 , with the underlying price

where cash delta $∆d,c = ∆d,c × Sj,d , cash gamma $Γd,c = Γd,c × Sj,d

Sj,d are measured at the close of the first available bar on day d.

Intraday open interest by origin (firm, broker-dealer, market maker, customer, and pro-

customer) is computed from open-close volume data with intraday cumulative trading volumes

by contracts, split into buys and sells for market makers (MM), and buy and sell open (O), and

buy and sell close (C) for the other traders. For each contract c with roots SPX and SPXW,

and trader type tp, at each time t on day d, we compute cumulative order imbalance from the

start of a day. For market makers, it equals to the difference between buys and sells:

CDOId,t,M M,c = Buyd,t,M M,c − Selld,t,M M,c . (5)

For the other trader types tp it also accounts explicitly for opening and closing flows CDOId,t,tp,c =

BuyOd,t,tp,c + BuyCd,t,tp,c − SellOd,t,tp,c − SellCd,t,tp,c . Then we convert cumulative order im-

balance to the intraday order imbalance for all time intervals τ = 30 minutes as the first differ-

ence of CDOI within each day: DOId,t,tp,c = CDOId,t,tp,c − CDOId,t−τ,tp,c , with DOId,0,tp,c =

CDOId,0,tp,c to account for pre-market trading. By default, market makers absorb the net

demand and supply from other market participants when managing their inventory of option

contracts, and their order imbalance is equal to the aggregate order imbalance of other traders.

Accumulating order imbalances DOId,t,tp,c over a long period up to each point d, t gives the

open interest OItp,d,t,c held by tp in contract c at d, t.13 We compute open interest OItp,d,t,c
13
Note that when there is no trading in contract c during day d, order imbalance is missing in the data and open
interest can be computed incorrectly. We track and fill in zero order imbalances for all contracts that exist but
are not traded on a given day. We aggregate order imbalances starting from 180 days before contract expiration.

14
at 30-minute intervals within each day to match bars data, and then we convert it to aggre-

gated notional, delta, and gamma terms following the procedure in (4) and using prices and

cash Greeks for the matching time points. For testing, we are specifically interested in the net

gamma exposure of market makers, denoted netΓd,t , and the delta of their order flow, denoted

f low∆d,t , for a set of contracts C:

P
netΓd,t = 100 × c∈C OId,t,M M,c × $Γd,t,c ,
P (6)
f low∆d,t = 100 × c∈C DOId,t,M M,c × $∆d,t,c .

Volume for a given contract c and a particular time interval t − τ to t on the day d (typically,

one- or 30-minute bar) is converted to absolute cash delta terms using delta value and stock

price at the end of an interval:

V olumed,t,c = V olume#
d,t,c × 100 × |$∆d,t,c |, (7)

where V olume#
d,t,c is the volume for contract c in number of contracts, $∆d,t,c is the cash delta,

and 100 is the contract notional. We also define underlying trading volume for intraday time

intervals t − τ to t by adding up volumes in the nearest futures contract (ES) and SPY:

U ndV olume#
X
U ndV olumed,t = d,t,j × Zj × ×Sj,d,t , (8)
j∈(ES,SP Y )

where U ndV olume#

d,t,j is the volume for underlying j in number of contracts, Zj is the notional

of one contract (50 for ES, and 1 for SPY), and Sj,d,t is the underlying price.

3.3 Market Dynamics: Summary

Trading volumes and aggregate risks. We examine the aggregate market statistics for

options maturing in the next calendar month, splitting maturities into 0, 1, 2-5, 6-10, and 11-22

DTE buckets. To compute the open interest and volume, we add the respective variables for

15
 A: Average Daily Trading Volume (V ol$∆ ) B: Average Daily Open Interest (OI $ )
Average daily trade_volume_delta_usd Average daily open_interest_usd
400 Roots Roots
SPX 3500 SPX
350 SPXW SPXW
3000
300 SPY SPY
2500
Total (bn, USD)

Total (bn, USD)

250
2000
200
150 1500

100 1000
50 500
0 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 0 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
Year Year
$∆
C: Average Daily Open Interest (OI ) D: Average Daily Open Interest (OI $Γ )
Average daily oi_delta_usd Average daily oi_gamma_usd
Roots Roots
SPX 17.5 SPX
400 SPXW SPXW
SPY 15.0 SPY

Total (tln, USD)

Total (bn, USD)

200 12.5
10.0
0 7.5
5.0
200 2.5

2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 0.0 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
Year Year

Figure 1: Aggregate Market Statistics. The figure shows quarterly averages of daily volume and open
interest for SPX, SPXW, and SPY options expiring within the next 22 trading days. We aggregate variables on
each day by summing over contracts up to 22 DTEs and moneyness in [0.5, 1.5] the start-of-day open interest
variables and the sum of intraday volume variables for all available bars, and compute quarterly averages. The
sample period is from 01/2012 to 30/04/2024.

all contracts with up to 50% in- and out-the-money in the required dte range on each day d.

Figure 1 gives the first impression of the market development over the last decade by aggregating

volume and open interest variables for the index and SPY options with maturities within the next

month. SPY options are popular among individual investors due to the low notional value of the

contract. Accordingly, the average trading volume in Panel A for SPY is relatively high—about

30% of trading in index options. At the same time, Panels B to D demonstrate that aggregate

risks are predominantly held in the index options: open interest in terms of dollar notional,

delta, and gamma for them is about six times higher than for the SPY. It indirectly suggests a

more speculative and retail character of SPY options. Figure 2 shows volume and open interest

by DTE buckets and delivers two messages: First, average daily trading volume grows mainly

due to the 0DTEs; second, the open interest dollar delta (i.e., directional risk) in Panel C due

16
 A: Average Daily Trading Volume (V ol$∆ ) B: Average Daily Open Interest (OI $ )
Average daily trade_volume_delta_usd Average daily open_interest_usd
400 DTEs 3500 DTEs
0 DTE 0 DTE
350 1 DTE 1 DTE
3000
300 2-5 DTE 2-5 DTE
6-10 DTE 2500 6-10 DTE
Total (bn, USD)

Total (bn, USD)

250 11-22 DTE 11-22 DTE
2000
200
1500
150
100 1000

50 500
0 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 0 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
Year Year
$∆
C: Average Daily Open Interest Delta (OI ) D: Average Daily Open Interest Gamma (OI $Γ )
Average daily oi_delta_usd Average daily oi_gamma_usd
500 DTEs DTEs
0 DTE 17.5 0 DTE
400 1 DTE 1 DTE
2-5 DTE 15.0 2-5 DTE
300
6-10 DTE 6-10 DTE

Total (tln, USD)

Total (bn, USD)

200 11-22 DTE 12.5 11-22 DTE

100 10.0
0 7.5
100 5.0
200 2.5
300
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 0.0 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
Year Year

Figure 2: Market Statistics by Days to Expiration. The figure shows quarterly averages of daily volume
and open interest variables for SPX, SPXW, and SPY options by expiry buckets (0, 1, 2-5, 6-10, 11-22 DTEs).
We aggregate variables each day by summing over all contracts with moneyness in [0.5, 1.5] in DTE buckets,
including the start-of-day open interest variables and the sum of intraday volume variables for all available bars,
and then compute quarterly averages. The sample period is from 01/2012 to 30/04/2024.

to 0DTEs is an order of magnitude smaller than for longer-term options and has not increased

over time. The open interest in dollar notional and dollar gamma has been relatively stable over

the years and is much smaller for short-term maturities. The pattern fits well with the intuition

that the direction of daily risk in extremely short-term options often changes such that, on

average, the dollar deltas net out. On the other hand, flipping (often) the delta of a longer-term

option portfolio is costly, and its sign generally corresponds to longer-term market sentiment in

a given quarter. 0DTEs’ open interest and dollar gamma are sizeable, indicating a non-trivial

magnitude of daily bets and resulting gamma risk on the market. Interestingly, introducing two

extra weekly expiration dates in August 2016 and two more in April-May 2022 does not seem to

have a pronounced effect on any of the quantities. The trading volume started a relatively sharp

17
increase in 2020, possibly linked to at-home COVID-related trading by individual investors or

some brokers lowering options trading fees for retail customers.

Appendix Table A1 stresses the specialty of 0DTEs and the importance of analyzing ultra-

short-term options separately from traditionally studied longer-term options: on average, the

turnover (in terms of absolute cash delta) in 0DTEs is twice that of 1DTEs and more than four

times that of all options with two to four weeks to expiry. Open interest cash gamma for 0DTEs,

measuring sensitivity to realized variance and required delta-hedging intensity, is half that for

all options with one to four weeks to expiry.

Trading activity by trader types. Market makers face trading flows from other traders and

manage their options inventory to keep risks under control. Exposed to asymmetric information,

they are the most probable trader group systematically delta-hedging their positions, at least

partially (see, Hu 2014, Ni, Pearson, Poteshman, and White 2021). Figure 3 shows the time

A: Daily Net Gamma (netΓ) B: Intraday Net Gamma (netΓ)

0 DTE OI Gamma SPX, SPXW, SPX by Trader Types 0DTE OI Gamma SPX, SPXW, SPX by Time and Trader Types 2021-06/2023
150
400 100

200 50
Total (USD, bn)
Total (USD, bn)

0 0 10:00
11:00
200 Firm 50 12:00
Broker-Dealer 13:00
Market Maker 100 14:00
400 Customer 15:00
Pro-Customer 150 15:30
600
1 4 7 0 1 4 7 0 1 4
2021-0 2021-0 2021-0 2021-1 2022-0 2022-0 2022-0 2022-1 2023-0 2023-0 2023-0
7 Firm Dealer arket Maker Custom
er stomer
quote_date Broker- M Pro-Cu
Trader Types

Figure 3: 0DTE Open Interest Gamma by Trader Types. The figure shows the time series of the average
intraday 0DTE open interest dollar gamma netΓtp by trader types (Panel A) and the average intraday 0DTE
net cash gamma netΓtp by points in time and trader types and their 95% confidence bounds (Panel B). Net
cash gammas are computed only for 0DTE SPXW options. Series in Panel A are smoothed using an exponential
moving average with a half-life of five days. The sample period is from 01/2021 to 14/06/2023.

series of average daily (in Panel A) and intraday (in Panel B) net cash gamma by trader types.

Surprisingly, market makers’ net gamma stays positive for most of the sample period.14 On
14
During our sample period, about one-third of observations have a negative net gamma, with an average
magnitude roughly three times smaller than that of the positive gamma.

18
the opposite side, the major option sellers are non-professional customers, while the others

hold far smaller exposures. During the day, market makers’ gamma is positive and decreases

monotonically towards expiry time, while customers’ gamma stays roughly constant. Figure A1

shows the projected 30-minute delta-hedging flows (as a proportion of underlying volume) based

on the market makers’ intraday cash netΓ values and the subsequent realized index returns:

both average and maximum projected hedging flows are larger for 0DTEs compared to longer-

term buckets containing options from one day (1 DTE) to expiry to a range of expiration dates

(11-22 DTE). Projected hedging flows routinely exceeding 5% of the intraday volume can have

a sizeable impact on the underlying and are therefore worth analyzing.

Overall, similar to the aggregate statistics, we do not observe any connection between the

rapidly increasing 0DTE volume and the market makers’ net gamma throughout the day.15

4 0DTE Trading and Underlying Process: Analysis

This Section specifies and formally tests the links between the 0DTE trading and the underlying

process discussed in Section 2. Section 4.1 investigates the delta-hedging effect, i.e., the gamma

channel. Section 4.2 analyzes alternative explanations, including the potential endogeneity of

market makers’ 0DTE net gamma and the competing information-based 0DTE trading channel.

Section 4.3 looks into trading flows, returns, and the joint dynamics of these variables.

4.1 Shock Propagation through Delta-hedging: the Gamma Channel

We now analyze how the market makers’ intraday open interest gamma relates to the future

intraday realized variance, thereby shedding light on the impact of the 0DTEs market on the

underlying process dynamics.

15
Figure A2 in the Appendix provides additional figures for options in the other DTE buckets, which confirm
that at least in the considered period, market makers are also long gamma in other maturity buckets considered
in our analysis, with customers being on the other side of the open interest.

19
 As discussed in Section 2, a direct causal mechanism through which 0DTEs can affect the

underlying process is the market makers’ delta-hedging activity, resulting in a negative associa-

tion between the market makers’ current inventory gamma and subsequent underlying variance.

It follows from equation (2) that in states when market makers’ gamma is positive, the delta-

hedging flows push prices in the direction opposite to the realized return, thus reducing return

persistence and realized variance; in negative gamma states, the effect is accordingly reversed.

To empirically assess whether the relationship between the intraday market makers’ inventory

in 0DTEs and the subsequently realized variance is consistent with the gamma channel, we

regress the 30-minute realized log-variance following every 30-minute intraday interval up to

15:30 on the levels of market makers’ 0DTEs net gamma (0DTE netΓt ) observed at the end of

such intervals, also accounting for exposures to longer maturities (Non-0DTE netΓt ):

ln RVt+1 = b0 + b1 0DTE netΓt + b2 Non-0DTE netΓt + CX + εt+1 , (9)

where X is a vector of controls including three lags of the dependent variable, intraday returns,

and changes of underlying volume, respectively, as well as fixed effects including date and a com-

bination of year, day-of-week, and time. The dependent and continuous independent variables

are standardized to unit variance for interpretability.

Table 1 shows the estimation results, indicating that market makers’ net cash gamma is

negatively related to future realized variance for both 0DTEs and longer-expiry options, sepa-

rately and jointly. Controls confirm an apparent persistence of realized variance, asymmetric

volatility effect, and high volatility on volume effect. In terms of magnitude, columns (5) and

(6) indicate a decline in the underlying index variance by 8.5% and 7.3%, respectively, relative

to its standard deviation following one standard deviation increase in the market makers’ 0DTE

net gamma (after controlling for longer-term options positions). To put this magnitude in per-

spective, notice that for the market makers’ non-0DTEs exposure, which comprises one-day to

20
 ln RVt+1
(1) (2) (3) (4) (5) (6)
0DTE netΓt -0.097*** -0.074*** -0.085*** -0.073***
(0.011) (0.018) (0.010) (0.018)
Non-0DTE netΓt -0.107*** -0.312*** -0.100*** -0.229***
(0.008) (0.044) (0.009) (0.046)
ln RVt 0.466*** 0.254*** 0.435*** 0.226*** 0.445*** 0.247***
(0.031) (0.024) (0.027) (0.021) (0.030) (0.024)
Rett -0.050*** -0.043*** -0.058*** -0.045*** -0.048*** -0.040***
(0.009) (0.007) (0.007) (0.006) (0.008) (0.007)
∆UndVolumet 0.120*** 0.053*** 0.130*** 0.054*** 0.116*** 0.054***
(0.017) (0.009) (0.014) (0.008) (0.016) (0.009)
Additional Controls Yes Yes Yes Yes Yes Yes
Date FE No Yes No Yes No Yes
Year × Week-Day × Time No Yes No Yes No Yes
R-squared Adj. 0.762 0.829 0.773 0.837 0.768 0.830
Obs. 4,910 4,910 6,150 6,150 4,910 4,910

Table 1: Net Gamma and Underlying Variance. This table reports the results of a regression of the log of
intraday variance over the 30-minute interval, following every 30-minute intraday time point, on the market maker
net cash gamma for 0DTEs (SPXW options) and non-0DTEs (SPX and SPXW options), denoted by netΓt with
a prefix. The dependent and continuous independent variables are standardized to unit variance. The regression
controls for three lags of the dependent variable, past returns and changes in underlying volume, respectively, but
only the first lags are shown for brevity. Standard errors in parentheses use Newey and West (1987) with five
lags. The sample period is from 01/2021 to 14/06/2023.

one-month to expiry, a standard deviation increases in the market makers’ net gamma (i.e.,

Non-0DTE netΓt ) is associated with a decline in underlying variance by 22.9% relative to its

standard deviation. Hence, the economic magnitude of the effect of market makers’ hedging of

their 0DTEs exposures alone is roughly 32% of the magnitude for these longer maturities (from

one day to one month), put together.

Therefore, the potential impact on the underlying variance arising from market makers’

0DTEs delta-hedging can be pretty sizeable. However, the direction of this impact, whether

amplification or attenuation of market volatility, depends on whether the market makers are

predominantly long or short 0DTEs gamma. Because market makers mostly hold positive 0DTEs

net gamma positions in our sample period, as earlier shown in Figure 3, the estimates in Table 1

imply that on average their delta-hedging significantly reduces the underlying index variance

rather than amplifying it.

21
4.1.1 Market Makers’ Net Gamma Components and Conditional Effects

An important observation regarding the market makers’ 0DTEs net gamma is that some of the

positions in the market makers’ 0DTEs inventory were acquired before they became 0DTEs.

Therefore, an interesting insight that is not obvious from the preceding analysis is whether the

negative association between 0DTE netΓt and the future underlying variance is due to hedging

of positions market makers acquired on the expiration day or positions previously accumulated

and held until the expiration day. Such insight is important because it sheds more light on

whether intraday activities on the expiration day matter or the previous result mainly reflects

the effect of older but now expiring options. We examine this by decomposing the total 0DTE

net gamma of market makers in Table 1 into the net gamma evaluated from an open interest at

the start-of-day, denoted Start-of-day 0DTE netΓt , and point-in-time net gamma coming from

the accumulated order imbalance during the day, denoted Within-day 0DTE netΓt .

Table 2 shows the results of regressing the underlying index variance on these components of

market makers’ 0DTEs net gamma. In all the specifications except one, both the prior exposure

and positions acquired during the expiration day matter for the gamma channel. Precisely,

Start-of-day 0DTE netΓt and Within-day 0DTE netΓt are both negatively related to future

intraday variance with similar magnitudes, and both are significant even in the specifications that

include the net gammas jointly in the same regression.16 These results indicate that although

a sizeable chunk of the market makers’ 0DTEs inventory gamma could arise from positions

acquired before they become 0DTEs, hedging exposures accumulated precisely on the expiration

day can have an important effect on the underlying variance.

16
The only exception with respect to the significance is in the specification in column (2), in which the initial net
gamma coefficient loses significance, likely because the high-dimensional fixed effects absorb some of the variation
in Start-of-day 0DTE netΓt , whose variation comes only from the portfolio gamma, given that the market makers’
open interest it fixed throughout the day.

22
 ln RVt+1
(1) (2) (3) (4) (5) (6)
Start-of-day 0DTE netΓt -0.051*** -0.036 -0.070*** -0.056**
(0.010) (0.022) (0.011) (0.025)
Within-day 0DTE netΓt -0.058*** -0.041*** -0.074*** -0.051***
(0.007) (0.011) (0.008) (0.011)
ln RVt 0.484*** 0.261*** 0.479*** 0.257*** 0.466*** 0.254***
(0.032) (0.024) (0.032) (0.024) (0.031) (0.024)
Rett -0.051*** -0.044*** -0.050*** -0.043*** -0.049*** -0.043***
(0.009) (0.007) (0.009) (0.007) (0.009) (0.007)
∆UndVolumet 0.127*** 0.053*** 0.124*** 0.052*** 0.120*** 0.053***
(0.017) (0.009) (0.017) (0.009) (0.017) (0.009)
Additional Controls Yes Yes Yes Yes Yes Yes
Date FE No Yes No Yes No Yes
Year × Day-of-Week FE × Time No Yes No Yes No Yes
R-squared Adj. 0.758 0.828 0.759 0.829 0.762 0.829
Obs. 4,910 4,910 4,910 4,910 4,910 4,910

Table 2: 0DTE Net Gamma and Underlying Variance: Pre-existing vs. New Positions. This
table reports the results of a regression of log of intraday variance over the 30-minute interval, following
every 30-minute time point on the market maker SPXW 0DTE net gamma decomposed into the gamma
at market open (Start-of-day 0DT E netΓt ) and gamma of the cumulative order imbalance during the day
(Within-day 0DT E netΓt ). The dependent and continuous independent variables are standardized to unit vari-
ance. The regression controls for three lags of the dependent variable, past returns and changes in underlying
volume, respectively, but only the first lags are shown for brevity. Standard errors in parentheses use Newey and
West (1987) with five lags. The sample period is from 01/2021 to 14/06/2023.

Figure 4 extends these results by estimating the exact specification as in column (6) of

Table 2 for market makers’ non-0DTE inventory. Precisely, for each non-0DTE expiration

bucket (1-5 DTE and 6-22 DTE ), we obtain the net gamma evaluated from an open interest

at the start-of-day and point-in-time net gamma coming from the accumulated order imbalance

during the day and include them jointly in the regression. Finally, we plot these net gamma

components’ coefficient estimates and confidence intervals. The figure shows that the intraday

order imbalance net gamma plays a significant role for options with up to 5 days to expiry,

hinting at their relatively dynamic portfolio composition. For longer-term options, only the

start-of-day inventory significantly affects the intraday underlying variance.

As discussed in Section 2, delta-hedging of new positions also creates additional trade flow

in underlying enforcing or weakening the return propagation depending on the flow delta and

return signs. We do not find any significant effect of such initial hedging flows, and the results

are provided in Column (4) of Appendix Table IA.1.

23
 Open vs Within Day Gamma Effect on log RV[t+1] by DTE
0.2
Start-of-day
0.1 Within-day

0.0

0.1

0.2

0.3

0.4
0 DTE 1-5 DTE 6-22 DTE
Figure 4: Net Gamma and Underlying Variance: Pre-existing vs. New Positions across DTE
Buckets. The figure reports the coefficient estimates and 90% confidence interval from a regression of log
of intraday variance over the 30-minute interval, following every 30-minute time point on the market maker
net gamma (SPXW for 0DTE, and both SPX and SPXW for longer maturities) decomposed into the gamma
at market open (Start-of-day 0DTE netΓt ) and gamma of the cumulative order imbalance during the day
(Within-day 0DTE netΓt ), jointly for 0 DTE, 1-5 DTE, and 6-22 DTE buckets. All controls and data pro-
cessing are the same as in regression (9). The sample period is from 01/2021 to 14/06/2023.

Because market makers’ net gamma is positive on average across days and intraday, the

gamma channel evidence provided so far implies, on average, a reduction in the underlying

return persistence and variance due to market makers’ delta-hedging of their 0DTEs exposure.

A helpful insight that complements this evidence is whether the magnitude of the effect is

symmetric across scenarios where the market makers’ net gamma implies an amplification vs.

attenuation of the underlying variance. We, therefore, investigate the conditional effects of

the market makers’ 0DTEs net gamma to ascertain how different it is when the market makers’

0DTEs net gamma is positive vs. negative. Furthermore, we investigate whether the documented

unconditional link materially changes when the prevailing underlying volatility is high, there is

a surge in the underlying trading volume, or the underlying return is negative.

Table 3 shows in columns (1) to (3) the estimates of the following regression that includes a

conditioning variable to the previous specification:

ln RVt+1 = b0 + 0DT E netΓt × (b1 + b2 θt ) + CX + εz , (10)

24
where θt is one of the dummy variables, High RV t indicating past realized variance above its

sample median, High ∆U ndV olumet indicating a change in underlying volume above its sample

median, and N eg Rett indicating a negative 30-minute return of the underlying index. Column

(4) of the same table shows results for decomposing net gamma into positive and negative parts

without interaction. In this specification, negative net gamma is denoted 0DTE netΓ−
t and

equals 0DTE netΓt but with the non-negative values set to zero. Positive net gamma is denoted

0DTE netΓ+
t and equals 0DTE netΓt but with the non-positive values set to zero. Hence, by

construction, 0DTE netΓt = 0DTE netΓ− +

t +0DTE netΓt . In all the specifications, the dependent

and continuous independent variables are standardized to unit variance for interpretability.

The interactions of market makers’ 0DTEs net gamma with dummies for high realized vari-

ance, surge in underlying volume, and negative realized return, respectively, are all insignificant

and three to seven times smaller than the base estimate. These estimates indicate that the

gamma channel effects are relatively stable over these market regimes identified using local met-

rics. The most important result, however, is in column (4) of Table 3: the gamma channel is

strong and significant with the coefficient of −0.064 for positive net gamma, while the coefficient

decreases by 65% to −0.022 for negative net gamma. This evidence suggests a much stronger

underlying variance attenuation and return reversal effect coming from market makers’ delta-

hedging of their positive 0DTEs net gamma exposures compared to the variance amplification

and higher autocorrelation generated by delta-hedging negative net gamma exposures. Overall,

the result indicates that there is indeed scope for market makers’ delta-hedging of their 0DTEs

exposure to amplify market shocks, but the attenuation effect dominates in our sample period.

4.1.2 Market Makers’ Net Gamma and the Underlying Return Persistence

To further understand the economic magnitude of the net gamma’s effect on underlying return

autocorrelation, we build a simple momentum trading strategy similar to the one analyzed in

25
 ln RVt+1
(1) (2) (3) (4)
0DTE netΓt -0.067*** -0.082*** -0.076***
(0.017) (0.019) (0.018)
0DTE netΓt × High RVt -0.028
(0.024)
0DTE netΓt × High ∆UndVolumet 0.014
(0.012)
0DTE netΓt × Neg Rett 0.011
(0.013)
0DTE netΓ−
t -0.022**
(0.010)
0DTE netΓ+
t -0.064***
(0.020)
ln RVt 0.260*** 0.252*** 0.250*** 0.254***
(0.026) (0.025) (0.024) (0.024)
Rett -0.043*** -0.042*** -0.031*** -0.043***
(0.007) (0.007) (0.010) (0.007)
∆UndVolumet 0.054*** 0.074*** 0.054*** 0.053***
(0.009) (0.014) (0.009) (0.009)
Additional Controls Yes Yes Yes Yes
Date FE Yes Yes Yes Yes
Year × Day-of-Week FE × Time Yes Yes Yes Yes
R-squared Adj. 0.829 0.830 0.829 0.829
Obs. 4,910 4,910 4,910 4,910

Table 3: 0DTE Net Gamma and Underlying Variance: Conditional Effects. This table reports the
results of a regression of log of intraday variance over the 30-minute interval, following every 30-minute time
point on the market makers net gamma (0DT E netΓt ) for 0DTE SPXW options and its interaction with dummy
variables indicating last 30-minute variance of the underlying index above its sample median (High RV t ), jump
in an underlying volume (High ∆UndVolumet ), or negative last 30-minute index return (N eg Rett ). Column
−
(4) splits the net gamma into positive (0DT E netΓ+ t ) and negative (0DT E netΓt ) values. The dependent and
continuous independent variables before interaction are standardized to unit variance. The regression controls for
three lags of the dependent variable, 30-minute realized returns, and underlying volumes, respectively, but only
the first lags are shown. Dummies and controls’ interactions are not shown. Standard errors in parentheses use
Newey and West (1987) with five lags. The sample period is from 01/2021 to 14/06/2023.

J.P.Morgan (2023) and relate its performance to the market makers’ net gamma for 0DTE and

non-0DTE options. Every 30 minutes during the trading day (from 10:00 to 15:30 ET), we

compute a trading signal equal to the sign of the SPY return from the previous day’s close, take

a constant-size position in the underlying in the signal’s direction, and close that position after

sixty minutes holding period (or at 16:00 for trading at 15:30).17 We regress realized returns of

the strategy on the quartiles of the 0DTE and non-0DTE net gamma at the point of position

taking and date, time, and weekday fixed effects.

17
Strategy based on trading E-Mini futures produces similar results.

26
 Figure 5, Panel A indicates that high 0DTE net gamma significantly decreases and low net

gamma increases the performance of the momentum strategy, and the effects are monotonic.

The evidence is consistent with the gamma channel boosting intraday return reversal for high

positive net gammas and momentum for low (and negative) gammas. The difference between

performance in top and bottom states is economically significant, with 10 basis points per hour

(i.e., more than 0.5% per day). Non-0DTE quartiles are not significantly related to the intraday

momentum; for the most quartiles the point estimates flip sign.

A: 0DTEs
Average Strategy Net
Return Across MM Gamma
NetGamma Quantile (netΓ)
Joint Spec for 0DTEs/Non-0DTEs B: Non-0DTEs Net Gamma (netΓ)
Average Strategy Return Across MM NetGamma Quantile Joint Spec for 0DTEs/Non-0DTEs
20
10
15
5 10
0 5
5 0

10 5

15 10

Q1 Q2 Q3 Q4 Q4 Q1 Q1 Q2 Q3 Q4 Q4 Q1
0DTEs Net Quartile Non-0DTEs Net Quartile

Figure 5: Mean Reversion Strategy and Market Makers’ Net Gamma. The figure shows the performance
of an intraday momentum trading strategy depending on the market makers’ 0DTE and Non-0DTEs net gamma,
respectively. Positions in SPY are taken every 30 minutes corresponding to the sign of the SPY return from the
previous day’s close until now and held for one hour. The resulting returns are regressed on (jointly) dummy
variables for quartiles of market makers’ 0DTEs (shown in Panel A) and Non-0DTEs (shown in Panel B) net
gamma, accounting for date, time, and weekday fixed effects. The coefficient estimates and the 90% confidence
intervals for quartiles 1 (Q1) to 4 (Q4) and the difference of Q4 and Q1 coefficients (Q4−Q1) are plotted. The
confidence intervals are based on the Newey and West (1987) standard errors.

4.2 Accounting for Alternative Explanations

While the observed negative link between market makers’ inventory gamma and future under-

lying variance is consistent with the delta-hedging channel, it could also be explained by some

other market patterns. For example, if market participants trade against market makers on su-

perior information about future market variance and such information turns out to be correct,

market makers’ open interest gamma could be negatively related to future underlying variance

27
even if no delta-hedging effect is in place. Therefore, we conduct tests in this section to rule

out this and related considerations as the primary force driving the results documented in the

previous section. In sum, our instrumental variable analysis and analyses that investigate the

dynamics of market makers’ open interest gamma conditional on information variables indicate

that delta-hedging is the predominant force behind Section 4.1 results.

4.2.1 Instrumental Variable-based Evidence

A concern with the baseline analysis is the potential endogeneity of 0DTE netΓt due to its

possibly being driven by market activities, signals, or expectations related to future underlying

variance but uncontrolled for in our analysis. We address this concern using instrumental variable

(IV) regression that uses as an instrument the average of 0DTE netΓt during the same intraday

window t over the past one-week period ending five trading days (one-week) before the current

day d, denoted 0DTE netΓd−5,t . The instrument must satisfy the validity and exclusion criteria

for the IV regression to help address the potential endogeneity concern.

For the validity criteria, the instrument has to be strongly correlated with 0DTE netΓt , and

there are good reasons to expect this to be the case. The fact that we condition the instrument’s

construction on the intraday interval t, enables 0DTE netΓd−5,t inherit some intraday patterns

in the market makers’ 0DTEs exposure and hence be informative about the latter. More so,

because we average market makers’ open interest gamma over the recent past, we capture recent

market makers’ positioning in 0DTEs, which can indicate their positioning in the near future.

While these intuitive properties of the instrument make it a good candidate, an instrument’s

strength is an empirical question that can be judged using the first stage F -statistic.

For the exclusion criteria, the instrument should not affect future underlying variance directly

but only through its relationship with 0DTE netΓt . The specific structure of the 0DTEs market

makes a strong case for 0DTE netΓd−5,t satisfying the exclusion criteria. To see this, one way

28
to think about the endogeneity concern is that investors might have traded 0DTEs (reflected

in 0DTE netΓt ) due to informative signals about future underlying variance. However, because

0DTE options expire on the day of trading, the horizon of the investors’ expectations or signals

would typically not be much longer than the 0DTEs’ expiration. Otherwise, investors would

have traded longer-dated options with expirations that span the investors’ expectation horizon.18

As such, we assume that the information, if any, that drives 0DTEs trading is short-lived. This

assumption means that trades on 0DTEs that expired several days ago prior to day d − 5,

reflected in 0DTE netΓd−5,t , were unlikely based on signals about the underlying variance that

will be realized at least one week in the future on day d.19

We re-estimate the baseline specification that regresses log of underlying variance at 30

minutes frequency on 0DTE netΓt now using the IV procedure that uses 0DTE netΓd−5,t as the

instrument for 0DTE netΓt , controlling for three lags of underlying variance, return and first

difference of trading volume, respectively. Table 4 summarizes the estimation results, indicating

that the instrument is strong across the different specifications, with the first stage F -statistic

consistently above 20.20 We also consistently observe negative, sizeable, and strongly significant

coefficient estimates for the instrumented 0DTE netΓt , suggesting that market makers delta-

hedging of their 0DTEs exposure affects the underlying variance as hypothesized. Hence, the

earlier results are unlikely due to other mechanisms besides market makers’ delta-hedging.
18
Reports from CBOE and other market participants confirm this view in noting that 0DTEs are suited for
and typically used for targeted bets around events occurring on the 0DTEs expiration days.
19
The exclusion criteria may be violated in the unlikely scenario that the information about the underlying index
variance that drives 0DTEs trades is long-lived and is not captured by lags of the index variables. To address this,
we use data up to one week prior to construct the instrument and include three lags of the underlying variance,
return, and trading volume, respectively, as controls in the IV regression. Furthermore, in subsequent analyses,
we directly investigate if signals about future index variance shape 0DTE netΓt , and we find that they do not.
20
A rule of thumb is that the first-stage F -statistic should be above ten for an instrument to be considered
strong enough to satisfy the relevance criteria.

29
 ln RVt+1
(1) (2) (3) (4) (5)
∗∗∗ ∗∗∗ ∗∗ ∗∗∗
0DTE netΓt -0.259 -0.741 -0.254 -0.128 -0.107∗∗∗
(0.034) (0.142) (0.112) (0.040) (0.040)
ln RVt 0.425∗∗∗ 0.249∗∗∗ 0.274∗∗∗ 0.489∗∗∗ 0.468∗∗∗
(0.031) (0.035) (0.035) (0.027) (0.025)
Rett -0.043∗∗∗ -0.030∗∗∗ -0.037∗∗∗ -0.048∗∗∗ -0.049∗∗∗
(0.009) (0.009) (0.008) (0.009) (0.008)
∆UndVolumet 0.104∗∗∗ 0.059∗∗∗ 0.042∗∗∗ 0.031∗∗∗ 0.043∗∗∗
(0.016) (0.013) (0.010) (0.011) (0.011)
Additional Controls Yes Yes Yes Yes Yes
Date FE No Yes Yes No No
Time FE No No Yes Yes No
Year × Week-day FE No No No Yes No
Year × Week-day × Time FE No No No No Yes
R-squared Adj. 0.746 0.034 0.140 0.664 0.678
Obs. 4,770 4,770 4,770 4,770 4,770
First Stage F -statisitc 82.8 35.8 21.1 31.6 28.7

Table 4: Two-stage Least Squares Estimation of 0DTEs Net Gamma Effect on Underlying Variance.
This table summarizes the results of the instrumental variable regressions of the log of intraday variance over
the 30-minute interval, following every 30-minute intraday time point, on the market maker net cash gamma for
0DTEs (SPXW options), denoted 0DTE netΓt . The average of market maker net cash gamma for 0DTEs during
the same intraday window over the past one-week ending five trading days (one week) before the current day d,
denoted 0DTE netΓd−5,t , is used as the instrument for 0DTE netΓt . The dependent and continuous independent
variables are standardized to unit variance. The regression controls for three lags of the dependent variable, past
returns and changes in underlying volume, respectively, but only the first lags are shown for brevity. Standard
errors in parentheses are clustered by date. The sample period is from 01/2021 to 14/06/2023.

4.2.2 Does Public Volatility Information Drive Market Maker’s Exposure?

While the preceding analysis is re-assuring, we go further and directly examine whether market

makers’ 0DTEs net gamma is driven by observable market variables (public information) that

are likely informative about future underlying variance. We adopt the Ni, Pearson, Poteshman,

and White (2021) approach and regress the future level and first difference of market makers’

0DTEs open interest gamma on three lags of the underlying variance, returns, and volume. If

market makers’ inventory gamma changes in response to public volatility information, we should

observe that these variables, which reflect information about future variance, significantly predict

the subsequent market makers’ 0DTEs open interest gamma.

The results summarized in Table 5 indicate that market makers do not significantly adjust

inventory gamma in response to the public predictors of future volatility, such as past intraday

30
volatility, realized returns, and underlying volume. While this result lends more support for

0DTE netΓt+1 ∆0DTE netΓt+1

(1) (2) (3) (4) (5) (6)
0DTE netΓt 0.754*** 0.755*** 0.755***
(0.077) (0.077) (0.077)
∆0DTE netΓt 0.016 0.017 0.017
(0.029) (0.029) (0.029)
(0.025) (0.025) (0.025)
ln RVt 0.017 0.019 0.024 0.028 0.031 0.040
(0.023) (0.023) (0.031) (0.032) (0.032) (0.043)
ln RVt−1 0.001 0.003 -0.016 0.007 0.010 -0.016
(0.023) (0.023) (0.030) (0.032) (0.033) (0.044)
ln RVt−2 0.009 0.006 0.013 0.020 0.015 0.025
(0.030) (0.030) (0.036) (0.045) (0.045) (0.054)
Rett -0.005 -0.005 -0.008 -0.007
(0.009) (0.009) (0.012) (0.012)
Rett−1 0.012 0.011 0.017 0.016
(0.009) (0.009) (0.014) (0.014)
Rett−2 -0.000 0.001 -0.000 0.001
(0.007) (0.007) (0.010) (0.010)
∆UndVolumet -0.005 -0.008
(0.014) (0.020)
∆UndVolumet−1 0.011 0.013
(0.017) (0.025)
∆UndVolumet−2 0.004 0.003
(0.013) (0.019)
Date FE Yes Yes Yes Yes Yes Yes
Year × Day-of-Week FE × Time Yes Yes Yes Yes Yes Yes
R-squared Adj. 0.564 0.564 0.564 0.088 0.087 0.087
Obs. 4,908 4,908 4,908 4,907 4,907 4,907

Table 5: Market Makers’ 0DTEs Net Gamma Conditional on Public Volatility Information. The
table analyzes whether market makers’ 0DTEs open interest gamma changes in response to information about
future volatility. We regress the t + 1 level of market makers’ 0DTEs net gamma (0DTE netΓt ) and its first
difference (∆0DTE netΓt ) on their respective three lags (only one lag for each is shown) and three lags of log
realized variances, realized returns, and changes in underlying volume, respectively. The dependent variable and
all continuous independent variables are standardized to unit variance. Newey and West (1987) standard errors
with five lags are reported in parentheses. The sample period is 01/2021 to 14/06/2023.

the market maker delta-hedging channel and helps alleviate concerns regarding the exclusion

restriction of the instrument used in Table 4, there is still the possibility that 0DTEs trading

is driven by private rather than public information. The potential concern is that privately

informed agents acquire informative signals about future index volatility and trade 0DTEs before

the news with these signals is publicly released. If market makers take the opposite side of such

trades, their open interest gamma will change in a way that makes it appear as though it is

their hedging trades that drive the relationship between market makers’ 0DTEs net gamma

31
and future volatility, whereas it is the preceding informed trades initiated by privately informed

customers. We address this alternative information-based story next.

4.2.3 Accounting for 0DTEs Trading based on Private Information

We analyze whether market makers’ 0DTEs net gamma changes before major unexpected news

about firms that constitute the index or the aggregate US economy. Our focus on unexpected

or unscheduled news allows us to examine news plausibly unknown to the general public until

it is released, and hence, it could have been uncovered by privately informed traders who seek

information that gives them a trading advantage over other investors.

Suppose the market makers’ 0DTEs net gamma is significantly affected by the options trad-

ing of privately informed investors. In that case, we should see a significant change in the market

makers’ 0DTEs net gamma in the prior intraday trading windows before unexpected news events

that might trigger underlying index volatility are made public by the news media. The idea is

similar to the analysis in earlier works that examine whether informed agents trade individual

stocks ahead of firm-specific news released to the public (Engelberg, Reed, and Ringgenberg

2012, Karpoff and Lou 2010, Christophe, Ferri, and Angel 2004). However, because we are ana-

lyzing the S&P500 index, it is unclear whether the relevant unexpected news privately informed

investors might have uncovered ahead of other market participants pertains to the individual

stocks that make up the index or is primarily about the aggregate economy.

We proceed by collecting two sets of unexpected news events. First, we collect unscheduled

news about the individual stocks that comprise the index from Ravenpack. For each stock, we

count the number of news coverage and separately the number of negative news coverage (i.e.,

coverage with negative sentiment) released within each 30-minute window of the trading day.

We aggregate these numbers to the index level by weighting the counts by the stocks’ market

capitalization as of the previous trading day and summing across stocks. This procedure yields a

32
bottom-up measure of the intensity of news media coverage about the S&P500 index. From this,

we obtain abnormal news coverage the standard way as the difference between the log of time

t index-level news coverage and the log of its average value during the same intraday window

over the past three months ending one month before the current trading day. The second set of

unexpected news events focuses on unscheduled news releases about the aggregate US economy.

We count the number of news coverage and, separately, the number of negative news coverage.

As before, we obtain abnormal news coverage for either count type at each 30-minute window.

For either the bottom-up index-level news coverage or the aggregate news coverage, we

identify episodes of intense media coverage that may be indicative of unexpected market-moving

events that privately informed traders might have anticipated and acted upon before the current

time t. We define such episodes based on whether the abnormal news coverage is above its 75th

percentile and then examine if market makers’ 0DTEs net gamma is significantly different in the

periods leading up to the intense coverage; that is, in the pre-news period during which informed

agents might have traded 0DTEs options, significantly changing market makers’ inventory.21

Table 6 summarizes the results of the analysis. Panel A focuses on the bottom-up media

coverage, and Panel B focuses on the aggregate news coverage. Across the board, we consistently

find no evidence that market makers’ 0DTEs open interest gamma significantly changes in the

intraday periods leading up to major unexpected news releases. This indicates that either

privately informed agents do not systematically trade 0DTEs ahead of unexpected information

releases, as observed in Engelberg, Reed, and Ringgenberg (2012) in the context of individual

stocks, or if such trades do occur, are absorbed by other 0DTEs’ customers and hence do not

materially affect the market makers’ inventory. In any case, the evidence indicates that the

negative relationship between market makers’ open interest gamma and future index volatility

that we document is improbable due to 0DTEs trading based on private information.

21
Results for the abnormal news coverage above its 90th percentile are similar and are available upon request.

33
 Based on All News Based on Negative News
(1) (2) (3) (4) (5) (6)
Panel A: Bottom-Up News Coverage and 75th Percentile Threshold
30-min BeforeBigNews -1.571 -1.587 -0.810 0.034
(1.383) (1.172) (1.206) (1.177)
1-hr ending 30-min BeforeBigNews -0.984 -1.029 -0.987 -0.985
(1.250) (1.260) (1.189) (1.217)
R-squared Adj. 0.812 0.871 0.871 0.812 0.871 0.871
Obs. 6,524 5,518 5,518 6,524 5,518 5,518
Panel B: Aggregate News Coverage and 75th Percentile Threshold
30-min BeforeBigNews -0.455 0.858 -0.883 -1.014
(1.283) (1.226) (1.340) (1.308)
1-hr ending 30-min BeforeBigNews 0.106 0.170 -0.743 -0.806
(1.344) (1.381) (1.438) (1.466)
R-squared Adj. 0.812 0.871 0.871 0.812 0.871 0.871
Obs. 6,524 5,518 5,518 6,524 5,518 5,518

Table 6: Market Maker’s Inventory Gamma and Potential Trading based on Private Information.
This table shows the result of regressing 0DTE netΓt at 30-minutes frequency on dummy variables for trading
windows before major unexpected index-level news releases. Major unexpected news releases are defined as
periods when the index-level abnormal news coverage is above a certain percentile threshold indicated on the
panel labels. 30-min BeforeBigNews is a dummy variable that equals one during the 30-minute intraday window
before a major unexpected news release. 1-hr ending 30-min BeforeBigNews is a dummy variable that equals one
during the one-hour intraday window ending 30-minutes before a major unexpected news release. For expositional
convenience, 0DTE netΓt is divided by its standard deviation and multiplied by 100 before the estimations. In
Panel A, the index-level abnormal news coverage is based on the bottom-up news coverage measure aggregating
news coverage across the time t S&P500 index constituents. In Panel B, the index-level abnormal news coverage
is based on news about the aggregate US economy. In Columns (1)-(3), the news coverage counts are based on
all news (i.e., both positive and negative sentiment news), while in Columns (4)-(6), the counts focus only on the
news with negative sentiment. All regressions include date and year × week-day × time fixed effects. Standard
errors in parentheses are based on the Newey and West (1987) method with five lags.

The analyses in this section indicate that market makers’ delta-hedging of their 0DTEs

exposure is most likely the force behind the evidence of Section 4.1. Neither public nor private

information-based 0DTEs trading can explain the observed relationship, and the IV analyses

suggest a sizeable effect of market makers’ delta-hedging on the underlying variance.

4.3 0DTEs Trading Activity and the Underlying Market

4.3.1 Intraday Return Dynamics

We analyze intraday market returns to assess how sudden spikes in 0DTE trading volume shape

subsequent return realizations. We are interested in the accumulation of realized returns in either

the positive or negative direction following a volume jump, with the negative reaction expected

34
to have stronger implications for market stability. While high volume is often associated with

high return volatility, we want to ascertain whether sharp jumps in 0DTE volume precede

and are potentially related to large underlying returns in the spirit of earlier theoretical and

empirical works that suggest a link between options trading volume and underlying prices and

volatility (e.g., Easley, O’Hara, and Srinivas 1998, Pan and Poteshman 2006). It can be driven

by a preferred trading venue for informed traders or mechanical delta-hedging effects of an

unbalanced order flow—we stay agnostic about the mechanics now and analyze whether we can

document such a temporal link empirically and whether in our extended sample from 2012 to

04/2024 (compared to 01/2021 to 06/2023 used in Sections 4.1 and 4.2) we observe any extreme

return realizations following 0DTE volume spikes.

To obtain 0DTE options volume, we sum the dollar-delta volumes for SPY and SPXW

options at each 1-minute intraday interval and then convert it to log volume vt$∆ = ln V olt$∆ .

We identify jumps in intraday 0DTE trading volume by point in time t with the 1-minute changes

in trading volume (vt$∆ − vt−1

$∆
) larger than three times its standard deviation on a given day. We

use the 5-minute cumulative returns (from t + 1 to t + 5) as a measure of market reaction to a

volume shock at time t and then test whether the market reaction differs conditional on having

experienced a sizeable 0DTE volume jump and whether it depends on the realized return before

and during the 0DTE trading volume jump.

Figure 6 plots the distribution densities of the 5-minute cumulative returns conditional on

jump vs. no jump in 0DTEs volume and separately for four sub-periods. Visually, the distribu-

tion conditional on volume jumps has slightly fatter tails on both sides of the return realizations,

but we do not observe drastic differences in the distributions or out-of-the-ordinary market reac-

tion following 0DTE volume jumps. We further test these observations more formally. First, we

run two non-parametric tests (k -sample Anderson-Darling and two-sample Kolmogorov-Smirnov

tests) to see whether both samples are drawn from the same distribution. Second, because we

35
are especially interested in the distribution’s tails, we run a series of quantile regressions analyz-

ing how relatively infrequent realizations of returns depend on 0DTE volume jumps dummy in

interaction with year fixed effect and cumulative market returns before the volume jump. Such

specification lets us directly evaluate claims that large underlying market moves are propagated

through 0DTE trading, especially in recent years.

A: Before 09/2016 B: From 09/2016 to 01/2020

Before 09/2016 From 09/2016 to 01/2020
After 0DTE volume jump After 0DTE volume jump
Without 0DTE volume jump Without 0DTE volume jump
Density

Density
0.015 0.010 0.005 0.000 0.005 0.010 0.015 0.015 0.010 0.005 0.000 0.005 0.010 0.015
Cumulative 5-minute return Cumulative 5-minute return
C: From 01/2020 to 01/2022 D: After 01/2022
From 01/2020 to 01/2022 After 01/2022
After 0DTE volume jump After 0DTE volume jump
Without 0DTE volume jump Without 0DTE volume jump
Density

Density

0.015 0.010 0.005 0.000 0.005 0.010 0.015 0.015 0.010 0.005 0.000 0.005 0.010 0.015
Cumulative 5-minute return Cumulative 5-minute return

Figure 6: Cumulative Returns Conditional on 0DTE Volume Jump. The figure shows the distribution
density of 5-minute returns (from t + 1 to t + 5) conditional on having a large jump in 0DTE volume at t and not
having such a jump, for four sub-periods. We sample non-overlapping observations, skipping at least five minutes
between each selected t point. The sample period is from 01/2012 to 30/04/2024 and includes only days with
0DTE expiration.

Table 7 shows the results of the non-parametric tests for the conditional samples. We fail

to reject that the samples are drawn from the same distribution for all sub-periods and both

tests. k -sample Anderson-Darling is more suitable for our purpose because, compared to the

Kolmogorov-Smirnov, it puts more weight on the tails of the distributions. Table 8 presents

the results of the quantile regressions for selected percentiles on both sides of the distribution,

using the period before 2020 as the base. We include all double and triple interaction terms

in the regression but concentrate on past cumulative return and 0DTE volume jump, with and

36
 KS Statistic p-val AD Statistic p-val
Full Sample 0.025 0.118 0.705 0.169
Before 09/2016 0.061 0.184 0.721 0.166
From 09/2016 to 01/2020 0.034 0.363 0.422 0.223
From 01/2020 to 01/2022 0.039 0.513 -0.692 0.250
After 01/2022 0.022 0.809 -0.358 0.250

Table 7: Testing Conditional Cumulative Return Samples. This table reports the results of the two-sample
Kolmogorov-Smirnov and k -sample Anderson-Darling tests (Scholz and Stephens 1987) for the distributions of
5-minute cumulative returns (from t + 1 to t + 5) conditional on having a large jump in 0DTE volume at t and
not having such a jump, for four sub-periods. We sample non-overlapping observations in the sample period from
01/2012 to 30/04/2024 and includes only days with 0DTE expiration.

without year dummy interaction. We report only the interaction term coefficients central to our

analysis for brevity. In all cases, we fail to reject the null that there is no propagation of past

returns by 0DTE volume jumps—the interactions of past return and volume jump, with and

without year dummies, are all insignificant. There is limited evidence that the 0DTE volume

jump is inversely associated with positive future returns (Q90 and Q95), but the relationship

is not mediated by the past returns preceding the 0DTE volume jump. In short, there is no

evidence that 0DTE trading propagates market moves.

4.3.2 Underlying and Options Markets Integration

Next, we examine both the underlying and options markets jointly by studying how intraday

trading in both markets is related to realized price movements of the underlying asset. To this

end, we estimate Vector-Auto-Regressions (VARs) on three key time-series: 0DTE options trad-

ing volume, underlying volume, and Realized Variance of the underlying. We seek to understand

the joint dynamics of these variables, their contemporaneous correlation, and how the dynamic

relationship between them has changed over time as 0DTE trading has become more prevalent.

Because trading volumes in the underlying and options markets are interrelated and linked to

realized returns through delta-hedging, we model their dynamics jointly using a structural VAR.

Following Koop, Pesaran, and Potter (1996) and Pesaran and Shin (1998), we analyze responses

to shocks in the system using generalized impulse response functions (gIRF), which account

37
 Q1 Q5 Q10 Q90 Q95 Q99
0DTE Volume Jump -0.006 0.002 0.003 -0.010** -0.019** -0.037*
(0.023) (0.008) (0.005) (0.005) (0.008) (0.022)
Past Return 2.037 1.114 1.029 -6.240*** -6.289** -4.226
(10.933) (3.049) (1.750) (1.500) (2.912) (10.990)
0DTE Volume Jump x Past Return -51.693 -3.653 -1.406 24.149 24.660 17.382
(104.024) (21.889) (11.904) (14.696) (31.023) (110.693)
0DTE Volume Jump x Year 2021 0.016 -0.005 0.009 0.016 0.032 0.017
(0.066) (0.020) (0.014) (0.012) (0.020) (0.065)
0DTE Volume Jump x Year 2022 -0.009 -0.003 -0.002 0.004 0.009 0.157***
(0.050) (0.017) (0.011) (0.010) (0.017) (0.052)
0DTE Volume Jump x Year 2023 -0.052 -0.030* -0.026** 0.029*** 0.079*** 0.075
(0.047) (0.017) (0.011) (0.010) (0.017) (0.049)
0DTE Volume Jump x Year 2024 0.001 -0.006 0.005 0.004 0.029 0.089
(0.088) (0.026) (0.017) (0.015) (0.026) (0.081)
Past Return x 0DTE Volume Jump x Year 2021 83.374 11.315 -2.722 -21.948 -32.356 -25.468
(262.558) (31.844) (25.125) (28.243) (62.968) (327.930)
Past Return x 0DTE Volume Jump x Year 2022 67.084 2.914 14.479 -17.711 -22.025 4.772
(108.011) (24.789) (14.418) (16.309) (32.953) (116.329)
Past Return x 0DTE Volume Jump x Year 2023 73.484 10.704 11.530 -19.975 10.135 11.449
(115.453) (31.885) (16.887) (17.804) (34.878) (121.656)
Past Return x 0DTE Volume Jump x Year 2024 85.573 -34.659 -5.363 -28.763 -21.670 -0.418
(167.320) (34.659) (21.556) (26.513) (54.512) (178.580)
Year Dummies Yes Yes Yes Yes Yes Yes
Obs. 99,707 99,707 99,707 99,707 99,707 99,707

Table 8: Quantile Regressions of Intraday Returns. This table reports the results of the quantile regressions
of the selected percentiles (1/5/10/90/95/99) of the five-minute cumulative returns from t + 1 to t + 5 on the
dummy for the 0DTE Volume Jump at t, past five-minute cumulative return from t − 4 to t (Past Return), year
dummies, and double and triple interactions of the variables. We omit some double interactions from the table
for space reasons. We sample non-overlapping observations, skipping at least five minutes between each selected t
point. The sample period is from 01/2012 to 30/04/2024 and includes only days with 0DTE expiration. Standard
errors are in parentheses.

for the correlation of structural shocks in the system and are invariant to the ordering of the

time-series (Pesaran 2015 and Wiesen and Beaumont 2024).

In order to compute volume for the underlying, we use both intraday 1-minute bar data

for SPY, and the front-month contract of the S&P500 E-mini Futures, ES. At the one-minute

frequency, for each bar within the regular day session (from 9:31 to 16:00), we add up dollar

trading volume Vj$ for both j = SPY and ES to obtain aggregate volume Vd:t
$
and convert it
$ $
to log vd:t = ln Vd:t . For the volume of 0DTE options, we proceed in the same way, adding

dollar delta volumes for each one-minute for SPY and SPXW options and then convert it to
$∆ $∆
log volume vd:t = ln Vd:t . We also compute simple SPY returns Rd:t over the matching time

2 = R2 . To
intervals and use 1-minute squared returns as the instantaneous variance proxy σd:t d:t

38
maintain stationarity in the time-series, we use the first differences of the volume variables for

estimation. We also standardize all variables daily to unit variance.

Because we want to focus on the variables’ intraday association, we estimate their joint dy-

namics each day, compute the generalized impulse response functions, and use the distribution

of these gIRFs over a specified period to get the average impulse responses and their confi-

dence bounds. We split the sample into two sub-periods: 2012-2019, characterized by relatively

less 0DTE volume, and 2020-2024, during which 0DTE trading volumes rapidly increased. As

expected, the trading volume differences for 0DTEs and the underlying are highly correlated,

especially in the latter period: the correlation goes from 0.27 in 2012-2019 to 0.43 in 2020-2024.

The correlations between the variance and volume variables are initially higher for the under-

lying market than for 0DTEs (0.14 vs. 0.08). However, in the latter period, both correlations

almost perfectly align at the level of 0.15–0.16.

We estimate the following structural model daily as a dynamic reduced-form VAR at 1-

minute frequency and using n = 5 lags:

n
X n
X n
X
2 $ $∆ 2
σd:t = c0 + c1,l ∆vd:t−l + c2,l ∆vd:t−l + c3,l σd:t−l + eσ2 ,d:t ,
l=1 l=1 l=1
n
X n
X n
X
$∆ $ $∆ 2
∆vd:t = b0 + b1,l ∆vd:t−l + b2,l ∆vd:t−l + b3,l σd:t−l + e∆v$∆ ,d:t , (11)
l=1 l=1 l=1
n
X n
X n
X
$ $ $∆ 2
∆vd:t = a0 + a1,l ∆vd:t−l + a2,l ∆vd:t−l + a3,l σd:t−l + e∆v$ ,d:t .
l=1 l=1 l=1

We use the estimation output to compute generalized impulse response functions for one-

standard deviation shocks to the selected variables.22 For each sub-period 2012-2019 and 2020-

2024, we obtain the average values and the confidence bounds (5th and 95th percentiles) from

the collection of intraday gIRFs for five time steps. Figure 7 shows three key impulse response
22
Note that one can easily extend the analysis to compute joint impulse response functions as in Wiesen and
Beaumont (2024). However, it requires an assumption about joint shocks, which adds unnecessary complexity.

39
 Response of→retsq
A: Variance to retsq
Variance B:Response of opt_logvol_delta
Variance → 0DTE Volume to retsq Response
C: 0DTEofVolume
retsq to opt_logvol_delta
→ Variance
1.0 1.0
1.50 2012:2019 2012:2019
0.8 2020:2024 0.8 2020:2024
1.25
1.00 0.6 0.6
Response

Response

Response
0.75 0.4 0.4
0.50 0.2
2012:2019 0.2
0.25 2020:2024 0.0 0.0
0.00
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
Lags Lags Lags

Figure 7: 0DTE Volume and Underlying Variance Generalized Cumulative Impulse Response
Functions. The figure shows the average generalized impulse response functions with confidence bounds (5th
and 95th percentiles of empirical daily distribution) for the VAR system in (11) estimated daily with n = 5
lags for one-minute frequency. The averages and percentiles are computed for 2012-2019 (plotted in blue) and
2020-2024 (plotted in red). The response is calculated for one standard deviation shock to a given variable. The
variables are winsorized at 0.01/0.99 levels for the whole sample period and standardized daily to unit variance.
The sample period is from 01/2012 to 30/04/2024.

functions. To interpret the plots, note that a difference between the two time periods will show

as a parallel shift in the gIRFs if the change in the gIRF is driven by a contemporaneous cor-

relation. We see this in the last plot (Panel C), where the average post-2020 gIRF is shifted

upwards by an amount corresponding to the 0-frequency shift, indicating the response of market

variance to shocks in 0DTE volume is entirely driven by an increase in the 0-frequency, con-

temporaneous correlation between these variables. We observe no change between the periods

for the Variance-to-variance IRF (Panel A). Panel B shows how variance shocks impact 0DTE

volume. Here, we see a slight increase at the 0, 1, and 2-minute horizon, followed by a near-zero

difference at longer horizons.

Figure 7 also shows confidence bands for the gIRFs computed from their empirical distribu-

tion. We see that these bands overlap and, thus, suggest that we cannot reject the null that

the dynamic relationship implied by the VARs is unchanged from the pre- to post-2020 periods.

The point estimates, however, imply a small but statistically insignificant increase in the joint

correlation between the three variables at the contemporaneous, zero-frequency horizon. We

interpret this to suggest an increase in market integration over time. For example, following a

one standard deviation shock to 0DTE trading, the underlying index variance rises from about

40
0.18 standard deviation units in the 2012-2019 sub-period, increasing to 0.26 standard deviation

units in 2020-2024. Although these numbers imply a 1.5 times stronger response in the latter

period, the magnitude is economically negligible and statistically insignificant.

5 Robustness and Extensions

We conduct several tests to assess the robustness of our findings, reconcile some of our results

with the existing literature, and collect the additional results in the Online Appendix.

We extend our analysis of market maker open interest gamma effect on underlying variance

to include “speed” and “charm” in Section IA.1.1, and the results indicate that these components

do not propagate past volatility or adverse market shocks. We also find no significant effect of

the delta of intraday 0DTE flow interacted with the realized return sign.

In Section IA.1.2, we revisit the claim in Adams, Fontaine, and Ornthanalai (2024), Table 1,

that regresses intraday volatility on 0DTE Trading dummy (equal to one for days with 0DTE

trading) and uses the variation in the presence of 0DTE trading on Tuesdays and Thursdays due

to public holidays between 01/2019 and 19/05/2022 to identify the treatment effect of 0DTE

trading on the underlying volatility. They note that 0DTE trading has the causal effect of sig-

nificantly lowering the underlying volatility. We replicate the paper’s analysis in our Table IA.2

with two modifications: First, in addition to the 10-minute intraday volatilities, we also use 30-

minute and full-day volatilities, all annualized. Second, we replace the standard errors clustered

at the Y ear × T ime level with the Newey and West (1987) standard errors with five lags, which

is more standard in the context of time-series analysis with potentially serially correlated error

terms. These negligible changes render all coefficients on 0DTE Trading insignificant, indicating

that the suggested treatment effect of 0DTEs trading is not robust.

41
 In Section IA.1.3, we analyze the discrepancies between our results and findings of the positive

effect of 0DTE trading volume on the underlying variance in Brogaard, Han, and Won (2023).

Table IA.3 shows the results of a regression of daily variance on lagged 0DTE volume and open

interest gamma, with and without lagged variance controls. The significant positive coefficient

on lagged 0DTE volume turns insignificant once we control for lags of the underlying variance.

The reported results vary considerably depending on the variables pre-processing and controls

(e.g., applying log transformation to the dependent variable, controlling for lags of overnight

variance and using volume ratio instead of log volume). Given this lack of robustness, we are

cautious in interpreting these numbers and their statistical properties.

Taken together, these two validation exercises show that even though market makers’ 0DTEs

exposure can affect the underlying variance for non-fundamental reasons—in our sample the

effect on average is underlying variance attenuation—0DTEs trading volume by itself does not

necessarily have this same effect. One reason for this is that as Figures 2 and 3 show, high

0DTEs trading volume does not imply high market makers’ exposure to 0DTEs that needs

to be delta-hedged by trading in the underlying market. Instead, a significant portion of the

0DTEs volume could well be trades between other market participants, which does not change

the market makers’ inventory and potential hedging needs.

In Section IA.1.4, we look at the link between aggregate open interest gamma levels and

daily volatility, assuming that aggregate open interest, representing a total outstanding risk

of options inventory, can proxy for the delta-hedging demand. However, we do not find any

significant effect in Table IA.4, confirming the importance of knowing market makers’ inventory

for quantifying the link between option market activity and underlying volatility.

Section IA.1.5 provides additional analysis of sensitivity of impulse response functions to

0DTEs activity in Table IA.5. It also extends the market integration analysis with VAR to

42
longer-term maturities, showing in Figure IA.1 that the higher intraday volume and variance

shocks correlation is driven predominantly by short-term options expiring within one week.

6 Conclusion

Following the introduction of weekly options with daily expiration by Cboe, ultra-short maturity

options have surged in popularity in recent years, with the daily trading volume in zero days to

expiration options (0DTEs) increasing more than tenfold from its 2012 level. As a result, several

large market participants and major media outlets have expressed concerns that the ballooning

0DTE trading could intensify price moves following the re-adjustment of delta hedges in the

same direction as the underlying market.

Following the basic causal mechanism from inventory delta-hedging to trading flows and price

reaction, we document that market makers’ intraday 0DTE open interest gamma is negatively

associated with future short-term volatility and persistence of intraday returns. An intraday

momentum trading strategy earns 10 basis points per hour more in states with low market

makers’ 0DTE gamma than in high gamma states, indicating that the effects are economically

significant. We do not find empirical support for the alternative information channel linking

market makers’ gamma and future volatility, thus confirming the validity of the causal link

between the market makers’ 0DTE inventory and subsequent intraday volatility through the

gamma channel.

In our sample period, market makers carry a positive 0DTE net gamma on average and

most days, meaning that their delta-hedging typically dampens the underlying volatility. While

the gamma-to-volatility link is weaker in negative gamma states, this does not preclude the

possibility of 0DTE market activity strongly propagating adverse market shocks.

43
 We also analyze how spikes in 0DTEs trading volumes shape the subsequent intraday market

index returns. We do not find significant differences between distributions on days with and

without volume jumps. More so, extreme quantiles of the intraday return distributions are not

significantly related to the interaction of 0DTE volume and preceding returns. Nonetheless, we

observe stronger integration of the underlying and 0DTE option markets, with more correlated

trading and simultaneous spikes in volumes and return variance.

In conclusion, while our results suggest that market makers’ 0DTE positions normally do

not propagate large market moves and are linked to a lower, not higher, short-term underlying

volatility, the results are conditioned on a period when market makers, on average, held long

positions in 0DTEs. Thus, while concerns over a potential hedging-based destabilization of

underlying markets are not supported by our evidence and sample period, we do not extrapolate

to a scenario where the market makers’ inventories become very negative. Future studies could

assess the limitations of the existing 0DTEs regulatory frameworks (e.g., intraday exchange

margin and mark-to-market mechanisms) and their potential effect on market stability. Such

analysis could yield insights that can aid the design of optimal risk management and regulatory

practices, which are beyond the scope of the current study.

44
References
Adams, G., J.-S. Fontaine, and C. Ornthanalai, 2024, “The Market for 0-Days-to-Expiration: The Role
of Liquidity Providers in Volatility Attenuation,” Working Paper Government of Canada - Bank of
Canada, University of Toronto - Rotman School of Management.
Almeida, C., G. Freire, and R. Hizmeri, 2023, “0DTE Asset Pricing,” Working paper, Princeton Univer-
sity.
Anderegg, B., F. Ulmann, and D. Sornette, 2022, “The impact of option hedging on the spot market
volatility,” Journal of International Money and Finance, 124, 102627.
Baltussen, G., Z. Da, S. Lammers, and M. Martens, 2021, “Hedging demand and market intraday mo-
mentum,” Journal of Financial Economics, 142(1), 377–403.
Bandi, F. M., N. Fusari, and R. Renò, 2023, “0DTE option pricing,” Working paper.
Barbon, A., and A. Buraschi, 2020, “Gamma fragility,” Working Paper Imperial College Business School
05.
Beckmeyer, H., N. Branger, and L. Gayda, 2023, “Retail Traders Love 0DTE Options... but should
they?,” working paper.
Brogaard, J., J. Han, and P. Y. Won, 2023, “How does Zero-Day-to-Expiry Options Trading Affect the
Volatility of Underlying Assets?,” Working Paper, University of Utah.
Bryzgalova, S., A. Pavlova, and T. Sikorskaya, 2023, “Retail Trading in Options and the Rise of the Big
Three Wholesalers,” The Journal of Finance, 78(6), 3465–3514.
Christophe, S. E., M. G. Ferri, and J. J. Angel, 2004, “Short-selling prior to earnings announcements,”
The Journal of Finance, 59(4), 1845–1876.
Dew-Becker, I., and S. Giglio, 2023, “Risk Preferences Implied by Synthetic Options,” NBER Working
Papers 31833, National Bureau of Economic Research, Inc.
Easley, D., M. O’Hara, and P. Srinivas, 1998, “Option Volume and Stock Prices: Evidence on Where
Informed Traders Trade,” Journal of Fnance, 53, 431–465.
Engelberg, J. E., A. V. Reed, and M. C. Ringgenberg, 2012, “How are shorts informed?: Short sellers,
news, and information processing,” Journal of Financial Economics, 105(2), 260–278.
French, K. R., and R. Roll, 1986, “Stock return variances: The arrival of information and the reaction of
traders,” Journal of Financial Economics, 17(1), 5–26.
Hu, J., 2014, “Does option trading convey stock price information?,” Journal of Financial Economics,
111(3), 625–645.
Jarrow, R. A., 1994, “Derivative Security Markets, Market Manipulation, and Option Pricing Theory,”
The Journal of Financial and Quantitative Analysis, 29(2), 241–261.
Johannes, M. S., A. Kaeck, N. Seeger, and N. Shah, 2024, “Expected 1DTE Option Returns,” Working
paper GSB, Columbia University.
J.P.Morgan, 2023, “0DTE Options: Recent Observations on Intraday Price Patterns,” Technical report,
J.P.Morgan, Global Quantitative & Derivatives Strategy.
Karpoff, J. M., and X. Lou, 2010, “Short sellers and financial misconduct,” The Journal of Finance,
65(5), 1879–1913.
Koop, G., M. Pesaran, and S. M. Potter, 1996, “Impulse response analysis in nonlinear multivariate
models,” Journal of Econometrics, 74(1), 119–147.
Lipson, M., D. Tomio, and J. Zhang, 2023, “A Real Cost of Free Trades: Retail Option Trading Increases
the Volatility of Underlying Securities,” Working paper.
Newey, W. K., and K. D. West, 1987, “A Simple, Positive-semidefinite, Heteroskedasticity and Autocor-
relation Consistent Covariance Matrix,” Econometrica, 55(3), 703–708.
Ni, S. X., N. D. Pearson, A. M. Poteshman, and J. White, 2021, “Does option trading have a pervasive
impact on underlying stock prices?,” The Review of Financial Studies, 34(4), 1952–1986.

45
Pan, J., and A. M. Poteshman, 2006, “The Information in Option Volume for Future Stock Prices,”
Review of Financial Studies, 19(3), 871–908.
Pesaran, H., and Y. Shin, 1998, “Generalized impulse response analysis in linear multivariate models,”
Economics Letters, 58(1), 17–29.
Pesaran, M. H., 2015, Time Series and Panel Data Econometrics. Oxford University Press.
Scholz, F. W., and M. A. Stephens, 1987, “K-Sample Anderson-Darling Tests,” Journal of the American
Statistical Association, 82(399), 918–924.
Schönbucher, P. J., and P. Wilmott, 2000, “The feedback effect of hedging in illiquid markets,” SIAM
Journal on Applied Mathematics, 61(1), 232–272.
Sircar, R. K., and G. Papanicolaou, 1998, “General Black-Scholes models accounting for increased market
volatility from hedging strategies,” Applied Mathematical Finance, 5(1), 45–82.
Skinner, D. J., 1989, “Options Markets, and Stock Return Volatility,” Journal of Financial Economics,
23, 61–78.
Sornette, D., F. Ulmann, and A. Wehrli, 2022, “On the Directional Destabilizing Feedback Effects of
Option Hedging,” Swiss Finance Institute Research Paper Series 22-34, Swiss Finance Institute.
Vasquez, A., D. Amaya, N. D. Pearson, and P. A. Garcia-Ares, 2025, “0DTE Index Options and Market
Volatility: How Large is Their Impact?,” Working Paper Instituto Tecnológico Autónomo de México
(ITAM).
Vilkov, G., 2023, “0DTE Trading Rules,” Working paper, Frankfurt School of Finance and Management.
Wiesen, T. F. P., and P. M. Beaumont, 2024, “A joint impulse response function for vector autoregressive
models,” Empirical Economics, 66(4), 1553–1585.

46
A Additional Tables and Figures

Count Mean StDev Min 25% 50% 75% Max

0 DTE
Trade Volume (USD, bn) 1685 275.3 296.0 4.1 44.1 126.0 485.3 1409.8
Trade Volume Delta (USD, bn) 1685 68.8 75.2 1.6 12.7 29.8 111.3 367.9
Open Interest (USD, bn) 1685 174.2 124.2 11.3 85.5 144.7 223.3 1130.6
Open Interest Delta (USD, bn) 1685 4.2 21.0 -135.4 -3.9 3.2 12.1 198.3
Open Interest Gamma (USD, bn) 1685 1538.5 1008.4 7.0 790.9 1312.2 2036.9 7687.2

1 DTE
Trade Volume (USD, bn) 1689 113.0 89.9 2.9 40.7 84.2 171.6 658.0
Trade Volume Delta (USD, bn) 1689 28.5 24.5 0.8 9.4 20.9 42.9 228.9
Open Interest (USD, bn) 1689 216.8 333.7 7.4 61.9 102.9 228.9 3468.4
Open Interest Delta (USD, bn) 1689 11.3 70.5 -712.3 -2.7 2.5 9.9 881.2
Open Interest Gamma (USD, bn) 1689 1445.3 1876.7 48.7 486.7 872.4 1558.3 20338.6

2-5 DTE
Trade Volume (USD, bn) 2910 94.4 74.0 3.0 40.6 69.3 129.8 438.7
Trade Volume Delta (USD, bn) 2910 21.0 17.1 0.5 8.7 15.1 29.3 121.5
Open Interest (USD, bn) 2910 293.6 408.4 4.2 57.0 131.1 292.3 3325.2
Open Interest Delta (USD, bn) 2910 19.3 88.7 -692.6 -2.6 2.9 14.5 907.4
Open Interest Gamma (USD, bn) 2910 1657.1 2032.9 28.0 421.0 881.4 1954.0 17834.2

6-10 DTE
Trade Volume (USD, bn) 3009 45.9 32.6 0.8 23.2 38.4 60.1 267.4
Trade Volume Delta (USD, bn) 3009 10.7 8.8 0.1 4.7 8.3 14.1 82.6
Open Interest (USD, bn) 3009 277.9 418.3 3.9 29.5 100.3 381.8 3125.9
Open Interest Delta (USD, bn) 3009 18.8 82.3 -655.9 -1.8 1.6 11.4 826.2
Open Interest Gamma (USD, bn) 3009 1368.1 1923.7 10.7 202.0 536.0 1769.4 13466.5

11-22 DTE
Trade Volume (USD, bn) 3053 62.3 44.5 0.6 31.7 52.5 83.9 350.9
Trade Volume Delta (USD, bn) 3053 15.2 12.1 0.1 7.0 12.0 20.4 118.6
Open Interest (USD, bn) 3053 461.4 491.8 0.6 37.7 380.2 688.2 3009.5
Open Interest Delta (USD, bn) 3053 27.7 96.0 -384.2 -1.6 2.1 40.6 742.3
Open Interest Gamma (USD, bn) 3053 1983.0 2631.3 6.7 221.5 1512.5 2882.8 93509.5

Table A1: Volume and Open Interest by DTE Buckets. The table provides statistics for the selected
open interest and volume variables defined in equations (4) to (7) for all options with roots SPX, SPXW and
SPY aggregated by DTE buckets (note that we exclude SPX options from 0DTE statistics, because they expire
at the open of the day). The variables are first aggregated for each day: open interest variables use open interest
in terms of number of contracts at the beginning of the day, and the underlying prices and option deltas and
gammas reported at 10:00; trade volume variables are first computed for each 30-minute bar during the regular
session from 9:30 to 16:00 using volume in contracts during each 30-minute interval, and underlying prices and
deltas at the end of each bar, and then added up for each day. The sample period is from 01/2012 to 30/04/2024.

47
 A: Average Rebalancing Proportion B: Maximum Rebalancing Proportion
Mean 30-min Gamma Channel Rebalancing Percentage Max 30-min Gamma Channel Rebalancing Percentage
17.5
0 DTE 0 DTE
Percentage of underlying volume, %

Percentage of underlying volume, %

1 DTE 15.0 1 DTE
2.0 2-5 DTE 2-5 DTE
6-10 DTE 12.5 6-10 DTE
1.5 11-22 DTE 11-22 DTE
10.0
1.0 7.5
5.0
0.5
2.5
0.0 0.0
1 4 7 0 1 4 7 0 1 4 7 1 4 7 0 1 4 7 0 1 4 7
2021-0 2021-0 2021-0 2021-1 2022-0 2022-0 2022-0 2022-1 2023-0 2023-0 2023-0 2021-0 2021-0 2021-0 2021-1 2022-0 2022-0 2022-0 2022-1 2023-0 2023-0 2023-0
date date

Figure A1: Net Gamma-Projected Rebalancing by Market Makers. The figure reports the required
delta rebalancing by market-makers as a percentage of the total underlying volume (nearest futures, ES, and
SPY combined). We approximate the required delta rebalancing volume as the product of market makers’ cash
netΓt for each DTE bucket at the start of each 30-minute intraday bar and the absolute realized return during
the next 30 minutes. We divide these volumes by the total volume in underlying during these 30 minutes to
compute percentage and take the average (Panel A) or the maximum of this percentage for each day. Series are
smoothed using an exponential moving average with a halflife of five days. The sample period is from 01/2021 to
14/06/2023.

A: Average Daily 1 Open Interest Gamma, 1DTE

DTE OI Gamma SPXW by Trader Types
B: Average Daily 2-5Open Interest Gamma 2-5DTE
DTE OI Gamma SPXW by Trader Types
300
300
200
200
Total (USD, bn)

Total (USD, bn)

100 100

0 0
Firm 100 Firm
100 Broker-Dealer Broker-Dealer
Market Maker 200 Market Maker
200 Customer Customer
Pro-Customer 300 Pro-Customer
1 4 7 0 1 4 7 0 1 4 7 1 4 7 0 1 4 7 0 1 4 7
2021-0 2021-0 2021-0 2021-1 2022-0 2022-0 2022-0 2022-1 2023-0 2023-0 2023-0 2021-0 2021-0 2021-0 2021-1 2022-0 2022-0 2022-0 2022-1 2023-0 2023-0 2023-0
quote_date quote_date

C: Average Daily 6-10

Open Interest Gamma 6-10DTE
DTE OI Gamma SPXW by Trader Types
D: Average Daily Open Interest Gamma 11-22DTE
11-22 DTE OI Gamma SPXW by Trader Types
150 150
100 100
50 50
Total (USD, bn)

Total (USD, bn)

0 0

50 50
Firm Firm
Broker-Dealer 100 Broker-Dealer
100
Market Maker Market Maker
Customer 150 Customer
150 Pro-Customer Pro-Customer
200
1 4 7 0 1 4 7 0 1 4 7 1 4 7 0 1 4 7 0 1 4 7
2021-0 2021-0 2021-0 2021-1 2022-0 2022-0 2022-0 2022-1 2023-0 2023-0 2023-0 2021-0 2021-0 2021-0 2021-1 2022-0 2022-0 2022-0 2022-1 2023-0 2023-0 2023-0
quote_date quote_date

Figure A2: Open Interest Gamma by Trader Types and Days to Expiry. The figure shows the time
series of the average intraday open interest dollar gamma OI $Γ aggregated by trader types for different days to
expiry buckets (from 1DTE in Panel A to 11-22DTE in Panel D). Open interest gammas are computed for SPXW
options expiring on the day of trading. Series are smoothed using exponential moving average with halflife of five
days. The sample period is from 01/2021 to 14/06/2023.

48
 Online Appendix
for
0DTEs: Trading, Gamma Risk and Volatility Propagation

This version: March 14, 2025

IA.1 Robustness and Extensions

IA.1.1 Open Interest Gamma Channel

Intraday Market Maker Open Interest Speed, Charm, and Flow Delta Effects. In

Section 2, we split the delta rebalancing needs into gamma, speed, and charm components:

∂2∆ ∂∆
∆t1 − ∆t0 ≈ S × Γ × rt0 ,t1 + 0.5S 2 × 2
× rt20 ,t1 + (t1 − t0 ), (IA.1.1)
∂S ∂t

where the first term quantifies the direct gamma effect, the second term quantifies the effect of

gamma change, called “speed,” and the third term quantifies delta time-decay, called “charm.”

For small time intervals, it is customary to neglect second-order and time effects, and we also

concentrate on the gamma effect in the main analysis. In dollar terms, by multiplying both sides

of the equation above by the underlying price, the excess portfolio delta to be offset through

rebalancing can be approximated as follows:

c t ≈ S 2 × Γ × rt ,t = $Γ × rt ,t + 0.5$Speed × r2 + $Charm × (t1 − t0 ),

$∆ (IA.1.2)
1 0 1 0 1 t0 ,t1

∂2∆ ∂∆
where we define dollar speed $Speed = S 3 × ∂S 2
and dollar charm $Charm = S × ∂t .

Empirically, for the 01/2021 to 14/6/2023, for which we can compute the open interest of

market makers, both dollar speed and charm are positive on average (and for most days), with

charm being very small in magnitude (we compute it for a 10-minute intraday interval). Being

positive, both speed and charm will increase excess delta; in other words, they would increase

selling pressure on the market, especially after large intraday return realizations (due to the

rt20 ,t1 term accompanying dollar speed). Large dollar speed and negative return realizations

can propagate intraday volatility in case market makers’ open interest gamma is small and

overcompensated by the dollar speed term. We estimate a regression as in (10) in the main text,

but with dollar speed and charm effects:

day $Γ $Speed
ln RVt+1 = b0 +OIM M,t,0DT E × (b1 + b2 θt ) + OIM M,t,0DT E × (b3 + b4 θt )

$Charm
+ OIM M,t,0DT E × (b5 + b6 θt ) + CX + εz , (IA.1.3)

1
 ln RVt+1
(1) (2) (3) (4)
0DTE netΓt -0.075*** -0.074*** -0.062*** -0.070***
(0.018) (0.018) (0.019) (0.018)
0DTE Speedt -0.013*** -0.072 -0.052
(0.004) (0.052) (0.033)
0DTE Charmt 0.010 -0.003 0.010
(0.007) (0.010) (0.009)
0DTE netΓt × High RVt -0.003
(0.024)
0DTE Speedt × High RV t 0.061
(0.053)
0DTE Charmt × High RV t 0.026*
(0.013)
0DTE netΓt × Neg Rett -0.012
(0.018)
0DTE Speedt × N eg Rett 0.042
(0.034)
0DTE Charmt × N eg Rett -0.004
(0.012)
0DTE F low∆t -0.003
(0.007)
0DTE F low∆t × SignRett 0.003
(0.007)
High RV t 0.446***
(0.103)
N eg Rett -0.105*
(0.061)
SignRett 0.054*
(0.030)
Additional Controls Yes Yes Yes Yes
Date FE Yes Yes Yes Yes
Year × Week-day × Time FE Yes Yes Yes Yes
R-squared Adj. 0.832 0.835 0.833 0.833
Obs. 4,899 4,899 4,899 4,899

Table IA.1: Accounting for Speed, Charm, and Initial Hedge. This table reports the results of a
regression of log of intraday variance over 30-minutes, following every 30-minute time point, on the market maker
SPXW 0DTE open interest gamma, speed and charm, denoted 0DTE netΓt , 0DTE Speedt , and 0DTE Charmt .
Each variable interacts with a dummy variable indicating a negative last 30-minute return of the index (Neg Rett )
or high (above its sample median) last 30-minute variance of the index (High RVt ). In the last two specifications,
we include 0DTE F low∆t and its interaction with the sign of the last 30-minute return of the index (SignRett ).
The dependent and continuous independent variables before interaction are standardized to unit variance. The
regression controls for three lags of the dependent variable and of the returns and the first difference of volume
computed over the same intervals as the lags of the dependent variable. All the continuous controls are interacted
with Neg Rett , High RVt , and SignRett . The coefficients of all these controls are suppressed for brevity. Standard
errors are based on Newey and West (1987) with five lags. The sample period is from 01/2021 to 14/06/2023.

where θt is one of the dummy variables—High RV t indicating past realized 30-minute variance

above its sample median, and N eg Rett indicating negative last 30-minute return of the under-

lying index. The results in Table IA.1 indicate that, on average, the dollar speed and charm

2
have mostly an insignificant effect on the future variance. What matters most, however, is that

all interactions with a negative return dummy are insignificant, so neither of the excess open

interest delta drivers contributes to the ”bad” variance propagation. We also find no significant

effect of the delta of intraday 0DTE flow interacted with the realized return sign.

IA.1.2 0DTE Introduction and Underlying Variance

The paper’s main part establishes the relationship between market makers’ net gamma in 0DTE

options and subsequent underlying volatility. By justifying the relationship with a theoretical

mechanism, accounting for potential endogeneity using a plausibly exogenous instrument, and

rejecting alternative explanations, we point to a causal effect due to the market maker’s delta-

hedging. Intraday net gamma can be volatile due to unbalanced trading flows from other

market participants, and we show that positive net gamma reduces subsequent volatility after

controlling for the prevailing volatility regime (using lags of dependent variable) and other

important variables. We do not claim that 0DTE trading reduces volatility for a given day

on which such trading is present. In contrast, Adams, Fontaine, and Ornthanalai (2024) use

variation in the presence of 0DTE trading on Tuesdays and Thursdays due to holidays between

01/2019 and 19/05/2022 to interpret the point estimates for regression of intraday volatility

on 0DTE Trading dummy with Week-day fixed effect as the treatment effect of 0DTE trading.

They claim a causal effect of 0DTE trading that lowers volatility on a given trading day by

60–90 annualized basis points on 0DTE trading days.

Note that for an analysis of the volatility on 0DTE trading vs. non-trading days, it might be

more reasonable to study the trading effect on daily volatility directly as opposed to intraday 10-

minute volatilities for the reason of unknown autocorrelation structure of the intraday volatilities

on a given day (i.e., without accounting explicitly for volatility clustering or reversion, scaling to

daily or annual terms could be inaccurate). In addition, for daily volatility effect using intraday

observations, it is reasonable to cluster standard errors by trading days directly or to use Newey

and West (1987) standard errors to adjust for the serial correlation in the intraday volatility.

We re-estimate the model for realized volatility in Adams, Fontaine, and Ornthanalai (2024),

Table 1, for the sample period from 01/2019 to 19/05/2022, when 0DTE options began trading

3
regularly on every weekday. We regress the realized volatility on 0DTE Trading indicator vari-

able and a set of time fixed effects, controlling for past daily volatility and realized return. The

only significant change in the model specification is that instead of standard errors clustered

at Y ear × T ime level, we use Newey and West (1987) standard errors with five lags, which is

more standard in time-series analysis with serially correlated errors. Following the results in Ta-

10 minutes Frequency 30 minutes Frequency Daily Frequency

(1) (2) (3) (4) (5) (6) (7) (8) (9)
0DTE Tradingd -0.616 -0.518 -0.437 -0.783 -0.681 -0.595 -0.560 -0.475 -0.445
(1.079) (0.657) (0.659) (1.703) (0.971) (0.973) (1.121) (0.698) (0.705)
V olad−1 1.474*** 1.467*** 1.505*** 1.498*** 0.816*** 0.814***
(0.043) (0.043) (0.059) (0.059) (0.043) (0.043)
Retd−1 -0.124 -0.132 -0.048
(0.096) (0.139) (0.095)
Time FE Yes Yes Yes Yes Yes Yes No No No
Week FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year × Week-day FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
R-squared Adj. 0.314 0.612 0.613 0.341 0.675 0.675 0.312 0.771 0.771
Obs. 33,162 33,162 33,162 11,059 11,059 11,059 853 853 853

Table IA.2: S&P500 Index Volatility on Days with and Without 0DTE Expiration. This table reports
the results of a regression model with the annualized 10-minute (columns 1 to 3), 30-minute (columns 4 to 6),
and daily (columns 7 to 9) volatilities of log-returns of the SPX index as the dependent variable, expressed in
percentage points. The indicator variable 0DTE Trading equals one on days when there is an SPXW options
expiration and zero otherwise. The controls include lagged daily realized volatility and return and a set of time
fixed effects. Newey and West (1987) standard errors with five lags are reported in parentheses. The sample
period is from 01/2019 to 19/05/2022.

ble IA.2 we fail to reject the null of realized volatility being the same on days with and without

0DTE trading. The lack of robustness in such analysis stresses the importance of knowing the

composition of market makers’ (i.e., delta-hedgers) options inventory for studying the effects of

0DTE trading on underlying volatility.

IA.1.3 Daily 0DTE Trading Volume Effects

To understand why our finding of no apparent propagation of the underlying index volatility

through the 0DTEs market differs from the positive effect of 0DTE trading volume on the

underlying variance documented in Brogaard, Han, and Won (2023), we note that our approaches

are very different. While Brogaard, Han, and Won directly relate daily variance to the 0DTE

volume, we estimate intraday propagation of shocks in a system with absolute normalized returns

and trading volumes in 0DTEs and underlying, respectively, and only then relate the intensity

4
of shock propagation to 0DTE volume and other variables. Moreover, in other tests, we focus

on whether the potential delta-hedging intensity captured by 0DTE gamma, instead of 0DTE

trading volume, propagates recently realized underlying variance and find that it does not.

We dig deeper into the sources of the somewhat conflicting findings using a specification

similar to Brogaard, Han, and Won’s baseline result. We regress day d intraday variance RVd

of the SPX index on the morning open interest dollar gamma OId$Γ , lagged 0DTE volume, for

which we use either log of 0DTE dollar volume (with and without delta adjustment), denoted

0DT E V olumed−1 , or the proportion of 0DTE dollar trading volume relative to that of all

options for the same underlying maturing within the next month, denoted DT E0%.23 Because

trading volumes and daily variances can be persistent, we estimate the specification with and

without the following controls: lagged values of the dependent variable, which accounts for

the persistence of the outcome, and year-fixed effects to account for common trends. The

results provided in Table IA.3 are consistent with our inferences after accounting for the above-

mentioned controls; importantly, lagged 0DTE volume loses significance in most specifications

with added controls and even turns negative and (borderline) significant in Panel B, columns

(4) and (8).

IA.1.4 Start-of-Day Open Interest and Daily Variance

We also analyze the impact of the total open interest on subsequent volatility. This analysis

is motivated by the possibility that some traders, not necessarily market makers, delta hedge

their inventory. One can hypothesize that short gamma positions have a stronger motive for

maintaining a dynamic hedge because negative convexity leads to losses for short gamma po-

sitions in the case of large market moves. An assumption we make here is that the total open

interest gamma is highly correlated to the net gamma of the delta hedgers. Thus, we project

future variance onto total 0DTE open interest gamma at the market open to establish whether

morning gamma is associated with higher or lower subsequent intraday volatility.

For the test we use log realized variance ln RVdday computed from equation (3) using all

intraday 1-minute intervals for a day, the start-of-day open interest cash gamma OId$Γ computed
23
We use options with roots SPX/SPXW and SPY.

5
 V ol$ V ol$∆
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A. 0DTE Volume as Explanatory Variable
$Γ
OId−1 -0.278*** -0.201*** -0.020 -0.015 -0.283*** -0.210*** -0.021 -0.019
(0.065) (0.047) (0.015) (0.015) (0.066) (0.051) (0.015) (0.016)
0DT E V olumed−1 0.237*** 0.452*** 0.023 0.058 0.232*** 0.410*** 0.024 0.068
(0.032) (0.091) (0.017) (0.063) (0.032) (0.091) (0.017) (0.061)
day
RVd−1 0.489*** 0.486*** 0.489*** 0.485***
(0.150) (0.150) (0.150) (0.150)
day
RVd−2 0.487*** 0.479*** 0.487*** 0.480***
(0.158) (0.157) (0.158) (0.157)
day
RVd−3 0.076 0.078 0.075 0.078
(0.099) (0.099) (0.099) (0.099)
day
RVd−4 -0.029 -0.034 -0.029 -0.033
(0.072) (0.071) (0.071) (0.070)
day
RVd−5 -0.096 -0.098 -0.096 -0.098
(0.069) (0.069) (0.069) (0.069)
Year Dummies No Yes No Yes No Yes No Yes
R-squared Adj. 0.067 0.137 0.726 0.725 0.064 0.137 0.726 0.725
Obs. 1,685 1,685 1,684 1,684 1,685 1,685 1,684 1,684

Panel B. 0DTE% as Explanatory Variable

$Γ
OId−1 -0.244*** -0.131*** -0.013 -0.003 -0.233*** -0.095*** -0.008 0.008
(0.058) (0.031) (0.013) (0.012) (0.049) (0.025) (0.014) (0.013)
0DT E V olumed−1 0.163*** -0.232 0.009 -0.066* 0.101*** -0.259 -0.003 -0.078**
(0.021) (0.185) (0.016) (0.037) (0.028) (0.158) (0.016) (0.032)
day
RVd−1 0.492*** 0.489*** 0.493*** 0.486***
(0.149) (0.148) (0.149) (0.148)
day
RVd−2 0.488*** 0.478*** 0.488*** 0.477***
(0.157) (0.157) (0.157) (0.157)
day
RVd−3 0.074 0.078 0.073 0.082
(0.098) (0.098) (0.097) (0.098)
day
RVd−4 -0.027 -0.033 -0.026 -0.034
(0.072) (0.071) (0.072) (0.071)
day
RVd−5 -0.096 -0.100 -0.096 -0.101
(0.069) (0.069) (0.069) (0.069)
Year Dummies No Yes No Yes No Yes No Yes
R-squared Adj. 0.046 0.132 0.725 0.725 0.034 0.137 0.725 0.726
Obs. 1,685 1,685 1,684 1,684 1,685 1,685 1,684 1,684

Table IA.3: Daily Realized Variance vs. 0DTE Trading Volume. This table reports the results of a daily
time series regression of intraday variance of the SPX index on the lagged values of open interest dollar gamma
$Γ
OId−1 and the lagged 0DTE volume proxy, computed from either dollar or dollar delta volume, as indicated in
the Table headers. In Panel A, 0DT E V olumed−1 is the log of the volume variable indicated in the Table header.
In Panel B, 0DT E%d−1 is the proportion of 0DTE trading volume indicated in the Table header relative to
the total of the corresponding trading volume of all options (with roots SPY, SPX, and SPXW) expiring within
the next month. As additional controls, we use lagged intraday variances and year fixed effects. All continuous
variables are standardized to unit variance. Newey and West (1987) standard errors with five lags are reported
in parentheses. The sample period is from 01/2012 to 30/04/2024.

in equation (4) and aggregated for the pre-defined DTE buckets for SPX (only longer that 0

DTE), SPXW and SPY options, to capture total rebalancing risk due to fluctuations in the

6
underlying market index. We estimate the following regression:

ln RVdday = b0 + OId:sod,dte
$Γ day
× (b1 + b2 ln RVdon + b3 ln RVd−1 ) + CX + εd , (IA.1.4)

where X is a vector of controls including overnight variance RVdon , five lags of intraday variance

RVdday (i.e., lags of the dependent variable), and year dummies.24 We standardize all continuous

variables before an interaction to unit variance. Table IA.4 provides no indication that total risk

ln RVdday
0-DTE 1-DTE 2-5 DTE 6-10 DTE 11-22 DTE
OIΓd × ln RVdon -0.007 -0.012 0.004 -0.015 -0.019
(0.018) (0.017) (0.011) (0.011) (0.014)
day
OIΓd × ln RVd−1 -0.020 -0.009 -0.000 -0.034** -0.009
(0.017) (0.020) (0.018) (0.017) (0.021)
OIΓd -0.264 -0.138 -0.070 -0.456*** -0.155
(0.169) (0.186) (0.188) (0.176) (0.206)
ln RVdon 0.047 0.044 0.035* 0.062*** 0.069***
(0.033) (0.028) (0.019) (0.018) (0.022)
day
ln RVd−1 0.362*** 0.177*** 0.148*** 0.204*** 0.169***
(0.047) (0.042) (0.038) (0.037) (0.040)
Year × Week FE Yes Yes Yes Yes Yes
R-squared Adj. 0.796 0.806 0.787 0.784 0.783
Obs. 1,601 1,605 2,829 2,937 2,969

Table IA.4: Volatility Propagation by Open Interest Gamma. This table reports the results of a
regression of log of intraday variance on the level of overnight variance and lagged intraday variance both interacted
with start-of-day open interest dollar gamma by DTE buckets, using the specification in (IA.1.4). Start-of-day
$Γ
open interest is converted to dollar gamma OId:sod,dte at 10:00 each day and aggregated for SPY, SPXW, and
SPX (only for non-0DTE) options with moneyness levels in [0.5, 1.5] for each DTE bucket. Realized variances
are computed from intraday and overnight returns for the SPX index. Before the interaction, dependent and
continuous independent variables are standardized to unit variance. Five lags of the dependent variable are
included, but only one is shown for brevity. Standard errors in parentheses use Newey and West (1987) with five
lags. The sample period is from 01/2012 to 30/04/2024.

in 0DTEs is associated with the propagation of the overnight and lagged intraday variances to

the subsequent daily variance. The coefficients on the interaction between the gamma levels and

variances are insignificant for 0DTEs. We have weak evidence for the other DTE buckets that

high gamma and overnight variance interaction terms are linked to a lower variance the next

day. Moreover, open interest gamma in the morning of the day d is mostly negatively related
24
Including more lags of overnight variance for interactions has no effect on magnitude and significance of other
coefficients, so we kept only one. Excluding the interaction of gamma with lagged intraday variance does not
materially change the results. In Robustness Section 5 we also estimate a similar regression without interactions,
including lagged daily trading volume in 0DTEs and lagged start-of-day open interest gamma. The volume
coefficient turns insignificant once we control for lags of the dependent variable.

7
to the intraday volatility, and the coefficient is significant for intermediate maturity buckets.

Overall, we do not find evidence that higher open interest gamma levels destabilize prices and

contribute to clustered volatility shocks.

IA.1.5 VAR Analysis Extensions

Impulse Response Functions Dynamics. In the main text, we made an informal statistical

inference about the significance of the impulse response intensity time dynamics based on the

overlap of the confidence bounds. To see whether the propagation of the shocks to volatility and

both 0DTE and underlying trading volumes is related to 0DTE trading activity, we relate each

day’s cumulative generalized impulse response after five time steps (i.e., five minutes for the

1-minute VAR frequency) to year dummies, overnight and intraday variances, and to dummy

variables for considerable (more than one standard deviation) jumps of open interest dollar

gamma and 0DTEs trading volume relative their past averages.25 Table IA.5 shows the results

Impulse Variance 0DTE Vol. Und.Vol.

Propagation Variance 0DTE Vol. Und.Vol. Variance 0DTE Vol. Und.Vol. Variance 0DTE Vol. Und.Vol.
High OId$Γ 0.014 -0.014** -0.010 -0.008 0.005 -0.005 -0.015 -0.006 -0.003
(0.022) (0.006) (0.006) (0.015) (0.006) (0.006) (0.022) (0.007) (0.007)
High ln V old$∆ 0.021 0.012*** 0.013** 0.008 0.012** 0.003 -0.005 0.009 0.010**
(0.019) (0.004) (0.005) (0.014) (0.005) (0.005) (0.018) (0.006) (0.005)
ln RVdon -0.023*** -0.004*** -0.004*** -0.010*** -0.000 -0.004*** -0.020*** -0.004*** -0.003**
(0.005) (0.001) (0.001) (0.003) (0.001) (0.001) (0.004) (0.001) (0.001)
ln RVdday 0.011 0.009*** 0.006*** 0.006 0.002 -0.002 0.006 0.004* -0.003
(0.008) (0.002) (0.002) (0.005) (0.002) (0.002) (0.008) (0.002) (0.002)
Year Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes
Dep. Lags 5 5 5 5 5 5 5 5 5
R-squared Adj. 0.042 0.074 0.039 0.339 0.121 0.444 0.087 0.482 0.097
Obs. 1,679 1,679 1,679 1,679 1,679 1,679 1,679 1,679 1,679

Table IA.5: Conditional Generalized Impulse Response Functions. This table reports the analysis of
the 5-step cumulative responses based on the cumulative gIRFs from the VAR system (11) estimated daily using
1-minute data for Variance, 0DTE Volume (0DTE Vol., in dollar delta terms), and Underlying Volume (Und.Vol.,
in dollar terms). We regress the cumulative gIRFs on its five lags, log of overnight variance (ln RVdon ), log of
intraday variance (ln RVdday ), and dummy variables High OId$Γ and High ln V old$∆ , equal one for large positive
deviations of day d values from their rolling-window averages. Newey and West (1987) standard errors with five
lags are reported in parentheses. The sample period is from 01/2012 to 30/04/2024.

$Γ
of the regressions. Neither high open interest gamma (High OIdte,d ) nor high 0DTE trading

volume (High ln V old$∆ ) is linked to significantly higher responses to shocks in the system,
25
Both standard deviation and averages are computed from the past 21 daily observations.

8
except for one case, where the propagation of 0DTE volume to the underlying market volume

is significantly larger when 0DTE volume is high relative to its recent past average.

VAR Analysis for Other DTE Buckets. To see whether the time trend in impulse response

intensity is specific to the 0DTEs, we re-estimate the VAR system in (11) using option volume

variable for other maturity buckets and also for the volume aggregated over all maturities up to

22 trading days. Figure IA.1 shows the time-series plots of the smoothed cumulative responses

five steps after a shock. In Panel A, we observe a stronger volatility response to short-term

options’ trading volume for all years and its very pronounced upward trend starting around the

second quarter of 2020, led by 0- and 1DTEs. Options with more than a week to expiration

have a stable or even a slightly decreasing (for 11-22DTEs) cumulative response. The responses

are also sizeable, reaching almost 0.4 of the standard deviation. Realized return (volatility)

shocks also propagate stronger to short-term options but with a negligible cumulative effect size.

Option and underlying trading look very similar, with a relatively sizeable shock propagation

in both directions and an increasing trend for short-term options. Notably, the effects (in both

directions) for the aggregate options volume seem to be almost entirely driven by ultra-short-

term buckets. A possible explanation for the observed time effects is an increasing integration

of the underlying and options markets for liquid contracts (e.g., Dew-Becker and Giglio 2023),

which is consistent with the increased correlation between trading volumes (from relatively

uniform annual average levels of 0.25-0.3 before 2021 to 0.38, 0.44, and 0.59 in the next three

years, respectively). At the same time, trading in both 0DTE and underlying markets became

much smoother over the years, with the average daily standard deviation of intraday log volume

differences dropping from 1.49 in 2012 to 0.49 in 2023 for 0DTEs and from 0.80 to 0.47 for the

underlying instruments.

9
 A:Impulse
Option Volume to: response
opt_logvol_delta Variance
retsq B:Impulse
Variance
retsq to Option
: response Volume
opt_logvol_delta
0 DTE 6-10 DTE 0.030 0 DTE 6-10 DTE
0.35 1 DTE 11-22 DTE 1 DTE 11-22 DTE
2-5 DTE All DTE 0.025 2-5 DTE All DTE
0.30
Response

Response 0.020
0.25 0.015

0.20 0.010
0.005
2017 2018 2019 2020 2021 2022 2023 2024 2017 2018 2019 2020 2021 2022 2023 2024
Years Years
C:Impulse
Option Volume to :Underlying
opt_logvol_delta Volume
response und_logvol_delta D:Impulse
Underlying Volume: response
und_logvol_delta to Option Volume
opt_logvol_delta
0 DTE 6-10 DTE 0.18 0 DTE 6-10 DTE
0.18 1 DTE 11-22 DTE 1 DTE 11-22 DTE
0.16
0.16 2-5 DTE All DTE 2-5 DTE All DTE
0.14
Response

Response

0.14
0.12
0.12
0.10
0.10 0.08
0.08 0.06
2017 2018 2019 2020 2021 2022 2023 2024 2017 2018 2019 2020 2021 2022 2023 2024
Years Years

Figure IA.1: Dynamics of Option Volume and Variance Cumulative gIRFs. The figure shows the
smoothed (exponential moving average with a half-life of 252 days) time series of the cumulative generalized im-
pulse response functions after five steps for the VAR system in (11) estimated each day for the 1-minute frequency
with n = 5 lags, separately for options volume in different maturity buckets. The variables are normalized daily
to unit variance, and the response is calculated for one standard deviation shock to a given variable. The sample
period is from 01/2012 to 30/04/2024.

10