Risky Intraday Order Flow and

Option Liquidity∗

Hitesh Doshi† Paola Pederzoli‡ Saim Ayberk Sert§

July 28, 2025

Abstract

This paper provides a novel analysis of marketwide and exchange-specific trading costs
for short- and ultra-short-maturity options. We focus on inventory risk proxies re-
lated to order flow distribution and delta-hedging costs. Intraday order flow volatility
emerges as the primary driver of spreads, while delta-hedging needs play a secondary
role. Leveraging cross-exchange variation, we isolate this effect from broader factors
that may jointly affect order flow volatility and spreads. The findings support models
of active inventory management and suggest that, contrary to standard views, op-
tion liquidity providers rely more on trade matching than on delta-hedging to manage
inventory risk.

Keywords: Bid-ask spread, order flow, order flow volatility, exchanges, big data
∗
We thank Rohit Allena, Pedro Garcia Ares, Vincent Bogousslavsky (discussant), Jean-Philippe
Bouchaud, Kevin Crotty, David De Angelis, Kris Jacobs, Dmitriy Muravyev, Chay Ornthanalai (discussant),
James Upson (discussant), Vipin Kerala Varma, and the participants to the 2024 Lonestar Conference, 2024
UQAM Seminar Series in Canada, 2024 CBOE-FMA Conference on Derivatives and Volatility, 2025 Midwest
Finance Association Conference, and seminar at Capital Fund Management and the University of Houston
for comments and suggestions. We thank SpiderRock Data & Analytics for providing data. This work
was completed in part with resources provided by the Research Computing Data Core at the University of
Houston. All errors are our own.
†
C.T. Bauer College of Business, University of Houston. Email: hdoshi@[Link]
‡
C.T. Bauer College of Business, University of Houston. Email: ppederzoli@[Link]
§
C.T. Bauer College of Business, University of Houston. Email: sasert@[Link]
1 Introduction

Investors are turning into short-maturity options, changing the standard trading dynamics
in both the SPX options market and the market for individual stock options. In 2023, an
impressive 80% of SPX options trading focused on options with expiration less than a month
(Dim, Eraker, and Vilkov 2024; Bandi, Fusari, and Reno 2024). The surge in short-term
options volumes has spurred new research exploring this novel market, its characteristics,
and its implications for market stability. The main focus of these papers is, however, on
the prices and returns of the options (Bandi, Fusari, and Reno 2024; Almeida, Freire, and
Hizmeri 2024; Beckmeyer, Branger, and Gayda 2023), or on the impact of option trading on
the underlying market (Dim, Eraker, and Vilkov 2024; Adams, Fontaine, and Ornthanalai
2024; Brogaard, Han, and Won 2023), with limited analysis on the quality of the market
itself. Yet trading costs in this segment of the option market are exceptionally high. Even
for liquid, at-the-money short-maturity options, bid ask spreads can reach up to 10% of the
option price.
This paper contributes to the literature by analyzing effective trading costs in short-
and ultra-short-maturity U.S. options and their relationship with order flow, with the goal
of identifying trading patterns that may pose liquidity risk in a market characterized by
exceptionally high volumes. We focus on two categories of order flow measures. First, we
examine how the intraday distribution of order flow relates to bid-ask spreads. Intuitively,
one-sided or highly volatile order flow may be risky and costly for liquidity providers, as
it pushes them away from their inventory targets (see, e.g., Stoll 1978; Ho and Stoll 1983;
Bogousslavsky and Collin-Dufresne 2023). Second, drawing on the literature on option
market-making models (Jameson and Wilhelm 1992; Stoikov and Sağlam 2009), we consider
measures that proxy for the delta-hedging costs potentially faced by liquidity providers who

1
manage their option exposure by trading in the underlying equity market.
Our empirical analysis encompasses the sample of options with a maximum maturity of
seven weeks, including S&P500 index options (SPX options) and options on the constituents
of the S&P500 index, spanning the period from 2004 to 2021. While SPX options are
exclusively traded on the Chicago Board Options Exchange (CBOE), individual stock options
are traded across sixteen exchanges, which enables us to conduct an additional analysis across
venues.
We begin by analyzing the order flow of SPX options and document that, although buy
and sell orders are generally balanced throughout the trading day, resulting in small daily
order imbalances in absolute value (as shown by Dim, Eraker, and Vilkov 2024), the intraday
pattern of order flow reveals much greater variation. To capture this, we divide the trading
day into equispaced intraday intervals and compute the order imbalance within each interval.
The resulting series of intraday order imbalances reflects the intraday distribution of order
flow. We find that this distribution has exhibited high volatility since the financial crisis and
in the years thereafter. Comparing days with high trading costs to those with low trading
costs reveals that illiquidity is closely associated with elevated intraday order flow volatility.
This suggests that more dispersed trading activity poses risks for liquidity providers and
contributes to reduced market liquidity. Formal time series and panel regressions confirm
this relationship: higher intraday order flow volatility on a given day is positively associated
with higher trading costs. The effect is present across all option maturities but is particularly
strong for ultra-short maturities, including options with zero days to maturity (0DTE) and
options with one to six days to expiration.1 This underscores the sensitivity of bid-ask spreads
for these ultra-short maturity options to risky intraday trading patterns. Intuitively, for very
short-term options, like 0DTE, liquidity providers do not have time to earn a premium on
1
Formally, 0DTE options are contracts traded on day t that expire at the end of the same day t.

2
their inventory (as in e.g., Fournier and Jacobs 2020) and must quickly adjust to order flow,
incorporating the required premium immediately into the spread.
Overall, our analysis shows that the positive relationship between order flow volatility
and illiquidity is highly robust and statistically significant. It applies to both the SPX
options market and the market for individual stock options (in both the time-series and
cross-sectional dimensions) and survives the inclusion of numerous controls, including daily
measures of volume, order imbalance, volatility, option Greeks, stock characteristics and
past spread levels. Importantly, all our regressions include time-fixed effects, such as day-
of-the-week, month-of-the-year, and year dummies, to account for strong seasonalities in the
spread.
To help isolate the effect of order flow volatility on spreads from that of common macro
variables that may affect both, we leverage a unique feature of the U.S. equity options market:
individual stock options are traded across sixteen exchanges. Trading activity is relatively
well distributed, with no single venue dominating the market. For example, CBOE, the
largest exchange, accounts for only about 20% of total volume. We begin by examining
the cross-sectional dynamics of how liquidity adjusts following trades across the different
exchanges. By tracking quoted spreads at the time of each trade, we find that transactions
typically occur on the exchange offering the most favorable quoted spread, indicating that
order-routing mechanisms help minimize transaction costs. In addition, we observe that after
a trade is executed, the exchange that absorbs the trade tends to widen its quoted spread,
while competing exchanges narrow theirs.2 This behavior suggests an effort to attract order
flow and points to a good degree of competition across U.S. option exchanges. The presence
of multiple exchanges trading the same stock options creates a unique setting to examine
2
In this analysis, we exclude trades whose size is equal to or larger than the quoted size at the best bid
or ask, in order to avoid mechanical effects from trades that sweep the entire top-of-book depth.

3
how the relationship between trading costs and order flow volatility varies across venues
for the same underlying asset on the same day. We exploit this setting and estimate panel
regressions of exchange-specific effective spreads on exchange-specific order flow volatility,
controlling for stock-by-day fixed effects. This approach helps us to isolate the relationship
between intraday order flow volatility and illiquidity by absorbing all common shocks at
the stock-day level. The results consistently confirm a positive and robust relationship,
reinforcing our earlier findings.
Taken together, the results from both the marketwide and exchange-level analyses es-
tablish intraday order flow volatility as a key determinant of trading costs in short-maturity
options. These findings raise a natural question: how do more traditional measures of inven-
tory risk, particularly those associated with delta-hedging, compare in explanatory power?
To explore this, we turn to a second set of inventory risk variables that specifically capture
the hedging needs of liquidity providers. Using data from the CBOE Open-Close database,
we analyze the positions of market-makers in the SPX options market, who are key liquidity
providers and the participants most likely to delta-hedge their inventory. Following Dim,
Eraker, and Vilkov (2024) and Ni, Pearson, Poteshman, and White (2021), we measure, for
each day t in the sample, the intraday gamma of their inventory and the intraday delta of
the new order flow they absorb on day t. These variables capture the extent of delta-hedge
rebalancing for prior positions and the delta-hedging of new positions, respectively. We find
a positive relationship between intraday gamma and trading costs for 0DTE options, and
between the intraday delta of new order flow and trading costs for medium-maturity options,
consistent with options theory (Jameson and Wilhelm 1992). However, when intraday order
flow volatility is included in the regression, the significance of these variables weakens con-
siderably or disappears altogether, suggesting that their explanatory power is overshadowed
by the stronger relationship between volatile order flow and trading costs.

4
 Interpreting our results through the lens of market microstructure theory (Glosten and
Milgrom 1985; Stoll 1978), trading patterns that increase risk and cost for liquidity providers
are naturally reflected in the bid-ask spread.3 A priori, our core variable, the order flow
volatility, could reflect either informed trading or inventory risk. However, our findings
suggest that it is more closely associated with inventory-related liquidity shocks. Specifically,
we observe that order flow volatility is lower on days with a higher likelihood of informed
trading, such as earnings announcements and the preceding day. In contrast, it spikes for
very short-term options on the third Friday of each month, when many contracts expire,
and for medium-term options on the first and last trading days of the month, patterns more
consistent with inventory rebalancing. Supporting this interpretation, we also find that order
flow volatility tends to be lower when stock volatility is high, further suggesting that it is
not driven by market uncertainty either, and it has to be better interepreted as a measure
of inventory risk. Given this, our analysis naturally aligns more closely with inventory risk
models.
Inventory-based theory in the stock market that explicitly account for the stochastic
nature of the order flow (e.g., Bogousslavsky and Collin-Dufresne 2023) emphasize the crit-
ical role of unbalanced order flow distribution and its positive relationship with illiquidity,
consistent with our findings. In this model, liquidity providers actively rebalance inventory
throughout the day, aiming to balance buy and sell orders and maintain a small inventory.
Intraday volatile order flow increases the inventory risk they face while awaiting offsetting
trades, leading to wider spreads for investors.
While this framework is well established in equity markets, inventory models in the op-
tions market typically take a different approach. They emphasize the role of delta-hedging,
3
See Foucault, Pagano, and Röell (2013) for a review of models where bid-ask spreads endogenously
compensate for inventory risk, asymmetric information, or order processing costs.

5
whereby liquidity providers manage inventory risk by offsetting their option positions with
trades in the underlying asset (Jameson and Wilhelm 1992; Stoikov and Sağlam 2009; Cho
and Engle 1999). These models predict that trading costs should reflect the risks and frictions
associated with discrete delta-hedge rebalancing. However, recent empirical evidence from
Hu, Kirilova, Muravyev, and Ryu (2024), based on detailed account-level data from market
makers in the Korean options and futures markets, indicates that liquidity providers rely
primarily on active inventory management and trade matching, with delta-hedging serving
only as a secondary risk management tool. Our findings support this perspective: intraday
order flow volatility emerges as the dominant determinant of trading costs, outweighing the
role of delta-hedging variables. This suggests that option liquidity providers primarily man-
age inventory risk through active inventory management, using delta-hedging as a secondary
tool.
Our paper contributes to different strands of literature. It is primarily related to the
recent literature that studies the novel market of short-maturity options and ultra-short
maturity options (Almeida, Freire, and Hizmeri 2024; Bandi, Fusari, and Reno 2024; Dim,
Eraker, and Vilkov 2024; Beckmeyer, Branger, and Gayda 2023; Adams, Fontaine, and
Ornthanalai 2024). The novel aspect of our investigation is the focus on option liquidity and
its relation to the order flow.
In a closely related paper on option liquidity, Christoffersen, Goyenko, Jacobs, and Karoui
(2018) document a substantial illiquidity premium in the option market for longer maturity
options. They also analyze the determinants of the bid-ask spread in the cross-section of op-
tions and find that the daily absolute value of the order imbalances from non market-makers
are positively related to illiquidity. Goyenko, Ornthanalai, and Tang (2015) also examine the
link between spreads, order imbalance, and delta-hedging costs in the equity option market,
while Cao, Jacobs, and Ke (2024) explore how this relationship varies throughout the trading

6
day in SPX options. Unlike these studies, we focus on the second moment of the order flow
distribution and examine its relationship with illiquidity in short-maturity options, both at
the market level and across exchanges.4
In this sense, our paper is closely related to Bogousslavsky and Collin-Dufresne (2023),
who also examine the second moment of order flow distribution, though in equity markets.
Our contribution is distinct along two key dimensions. First, while their focus is on the
volume-spread puzzle, we take a broader approach to understanding illiquidity in short-
maturity options by comparing the explanatory power of theoretically motivated variables.
This allows us to test competing mechanisms underlying option trading costs. Second, we
exploit unique features of the option market, its maturity structure and multi-exchange
environment, to provide new economic insights not available in the equity setting.

2 Theoretical Framework and Literature Review

According to standard models of inventory management (e.g., Ho and Stoll 1983; Grossman
and Miller 1988), liquidity providers set the bid-ask spread in the market to maximize their
utility based on their final wealth. This wealth is determined by the cash earned from the
bid-ask spread and their inventory position. These models typically assume utility functions
that reflect an aversion to inventory variance, leading to wider bid-ask spreads as inventory
4
Additional research on option bid-ask spread includes early studies such as George and Longstaff (1993)
and Cho and Engle (1999), as well as more recent work on order-routing mechanisms across venues (Huang,
Jorion, and Schwarz 2024) and the role of payment for order flow (Ernst and Spatt 2022). Other related
studies has examined order-flow measures of trading and their impact on option prices, e.g., Bollen and Wha-
ley (2004); Garleanu, Pedersen, and Poteshman (2008); Muravyev (2016); and Fournier and Jacobs (2020).
Unlike these studies, which focus on the first moment of the order flow distribution, our research centers on
the second moment and its impact on liquidity. There is also an extensive literature investigating the impact
of the order flow on stock returns. A non exhaustive list is e.g., Chordia and Subrahmanyam (2004); Kelley
and Tetlock (2013); Brogaard, Hendershott, and Riordan (2014); Chordia, Hu, Subrahmanyam, and Tong
(2019).

7
risk increases. This aversion stems from the preference of liquidity providers to maintain a
minimal and balanced inventory throughout the day, as holding a non-zero position between
offsetting trades is risky.
This is best formalized in the model proposed by Bogousslavsky and Collin-Dufresne
(2023). In their model, liquidity providers actively manage their inventory in the stock
market by matching buy and sell orders to minimize imbalances. The bid-ask spread com-
pensates these risk-averse liquidity providers for the inventory risk they face while awaiting
offsetting order flow. If the arrival rates of buy and sell orders temporarily diverge during
the day, it creates an unbalanced order flow, increasing inventory risk for liquidity providers.
Consequently, holding trade volume constant, the equilibrium bid-ask spread rises with in-
traday volatility in order flow. Intuitively, in this framework, a large volume of shares bought
in one period and sold in another entails more risk than smaller, continuous transactions
spread evenly throughout the day.
In the options market, liquidity providers can manage inventory risk in two primary
(non mutually exclusive) ways: i) by actively managing inventory and matching trades, as
described above, or ii) by delta-hedging their option positions in the underlying stock market.
Hu, Kirilova, Muravyev, and Ryu (2024), who analyze account-level data for market-makers
in options and futures on the Korean Composite Stock Price Index (KOSPI 200), find that
most market-makers do not delta-hedge their option inventory. Instead, they rely on active
inventory reversal strategies as in (i), eliminating undesired positions within minutes. In
such scenarios, intraday order flow volatility is expected to be a primary determinant of
bid-ask spreads.
The first hypothesis tested in our analysis is whether the volatility of order flow is pos-
itively related to the effective spread. Acceptance of this hypothesis would suggest active
inventory management by liquidity providers in the U.S. options market. However, this

8
would not preclude the possibility that liquidity providers also utilize other inventory man-
agement tools, such as delta-hedging, as in (ii). In such cases, additional factors are expected
to influence the bid-ask spread, particularly when perfect inventory hedging is unattainable.5
The risk associated with discrete delta-hedge rebalancing is best represented by the inven-
tory’s gamma, which measures the sensitivity of the delta position to price changes, thereby
reflecting the rebalancing needs of liquidity providers (Jameson and Wilhelm 1992; Ni, Pear-
son, Poteshman, and White 2021; Dim, Eraker, and Vilkov 2024). Gamma also accounts
for errors in discrete rebalancing caused by changes in the delta position. We hypothesize
that if market-makers actively engage in inventory management, as observed in Hu, Kirilova,
Muravyev, and Ryu (2024), inventory gamma will play a secondary role compared to order
flow volatility in driving bid-ask spreads.

3 Data

We obtain options trade data from the CBOE’s LiveVol, including timestamp down to mil-
liseconds, trade price and size in contracts, the prevailing NBBO prices, and the contempora-
neous best bid and offer prices of underlying security for each trade reported by the Options
Price Reporting Authority (OPRA). The dataset spans the intraday trading activity of all
equity and index options from January 01, 2004, to July 16, 2021. We merge the LiveVol
data with the Center for Research in Securities Prices (CRSP), from which we obtain daily
stock returns, trading volumes, prices, and the number of outstanding shares. Additionally,
we combine the intraday trade data with OptionMetrics, allowing us to access daily implied
volatility and Greeks for option series. For each day, option series are required to be present
5
Traditional models of liquidity providers in the option market (e.g., Stoikov and Sağlam 2009; Cho
and Engle 1999) suggest that order flow imbalances and inventory should not impact spreads if perfect
delta-hedging is achievable.

9
in all three data sources.
We focus on S&P500 index options and options on individual stocks which are the con-
stituents of the S&P500 index. We track S&P 500 constituents on a monthly basis following
the historical components file from CRSP. A stock is included in our cross-sectional sample
for a given month if it was part of the S&P500 index in the previous month.
Our focus lies on short-term options with maturities of up to one month, as these have
seen the most significant growth in trading activity over time (Almeida, Freire, and Hizmeri
2024), raising questions about the stability of the option market. Among these options,
at-the-money (ATM) options are of special interest, as they have the highest decline in value
as maturity approaches, and the highest value of gamma, which is particularly relevant for
delta-hedgers liquidity providers (see Ni, Pearson, Poteshman, and White 2021). Moreover,
the prices and spreads of ATM options are less affected by market microstructure noise than
out-of-the-money (OTM) options (Duarte, Jones, and Wang 2024).6 Our main sample is
thus composed by ATM options, defined by an absolute delta between 0.375 and 0.625, with
up to 48 days to maturity. The delta of each option series is assessed at the close of the
preceding business day; for example, an option on day t is considered at-the-money if its
absolute delta, as recorded by OptionMetrics at the close of day t − 1, falls between 0.375
and 0.625.
We examine all options trades recorded by OPRA between 9:30 a.m. and 4:00 p.m.
US Eastern time. The OPRA database encompasses trades occurring across the sixteen
exchanges where investors can trade options. SPX index options are specifically traded only
on the CBOE exchange, with regular trading hours concluding at 4:15 p.m. Additionally,
they are available for trading during global trading hours before the market opens and after
6
In the robustness Section 6, we analyze out-of-the-money options and find results consistent with those
observed for the at-the-money options in the baseline analysis.

10
it closes, with this time frame gradually expanding over time.7 To ensure consistency in
coverage across various securities and over time, we concentrate on the standard trading
hours of 9:30 a.m. to 4:00 p.m. for all underlying stocks and the S&P500 index.
Following the literature, we apply filters to the intraday trade data to clean obvious
errors and outlying records. We filter out the following observations: (1) cancelled trades;
(2) trades with zero or negative price, size, and/or bid-ask spread; (3) trades whose sizes are
higher than 100,000 contracts; (4) trades whose prices are below bid minus spread or above
ask plus spread; and (5) trades whose prices are below $0.10.

4 Empirical Results

This section explores the characteristics of intraday order flow distribution in short-term
at-the-money options and its relation with illiquidity. Sections 4.1, 4.2, 4.3, and 4.4 focus
on the order flow of SPX options, while Section 4.5 examines the cross-section of options on
individual stocks. Section 4.6 performs the analysis at the exchange level.

4.1 Order Flow and Daily Statistics

Our primary focus is on analyzing the distribution of intraday order flow. To achieve this,
we first need to flag every trade as buy (i.e., buyer-initiated) versus sell (i.e. seller-initiated),
since the OPRA data does not explicitly provide this information.
Following the literature on high-frequency data of trades and quotes of stocks (Lee and
Ready 1991; Bogousslavsky and Collin-Dufresne 2023), trades are categorized as buys or sells
based on the quote rule and tick rule. Specifically, if a trade price is closer to the National
7
In 2015, CBOE extended trading hours for SPX options to include 3 a.m. to 9:15 a.m. In 2021, the
start time moved to 8 p.m. of the day before, and in 2022, CBOE added the ‘Curb’ session from 4:15 p.m.
to 5 p.m.

11
Best Offer, it is classified as a buy; otherwise, it is classified as a sell. If a trade price falls
at the NBBO quote midpoint, we follow Bryzgalova, Pavlova, and Sikorskaya (2023), and
apply the quote rule to the Best Bid and Offer (BBO) prices from the exchange where the
trade was executed. In cases where the trade price equals the BBO mid price, the tick rule
is applied: if the current trade price exceeds the price of the last trade in the same option,
the current trade is classified as a buy; conversely, it is classified as a sell.
In the stock market, it is well-known that the quote rule effectively classify trades that
occur without any price improvements, resulting in buyer-initiated (seller-initiated) trade
prices that are very close to the quoted ask (bid) prices. However, when a trade receives
significant price improvement, the trade classification may be prone to misclassification (Ellis,
Michaely, and O’Hara 2000). To validate our quote rule on this critical sample, we obtain a
sample of about one million option trades executed on 2024-02-02 through auctions.8 These
trades are mostly retail orders which have been automatically routed into auctions to receive
the best price improvement. Within the auction database, we have access to the actual trade
direction (buy versus sell) along with the prevailing bid and ask quotes of the exchange where
the trade occurred. Analysis reveals that, in this sample, the quote rule successfully classifies
approximately 85% of the trades.9
We then partition the trading day into equispaced time-intervals, and calculate the option
order flow on day t in each interval d by subtracting the trade size of seller-initiated trades
8
We thank SpiderRock Data & Analytics for providing this auction data.
9
Another potential source of misclassification could occur with trades that are components of multi-leg
strategies. Li et al. (2020) propose an heuristic approach to classify such trades. However, this methodology,
relying on manual trade matching, cannot be verified without a sample containing the actual trade direction.
Additionally, Li et al. (2020) find that in their sample, 70% of vertical spreads and 60% of straddles can be
classified using the quote rule. Therefore, we opt to adhere to the standard quote rule for trade classification.

12
of all options i from that of buyer-initiated trades:

X X
Order Flowt,d = Trade Size of Buysi,t,d − Trade Size of Sellsi,t,d . (1)
i i

Several choices for the length of the time intervals are possible. The optimal choice balances
the need for high frequency data and option liquidity; if the intervals are too short, we risk
having many empty intervals due to insufficient trading activity. While this might not be an
issue for SPX options, it could be problematic for some individual stock tickers. Therefore,
we opt for a 5-minute interval, which provides a suitable balance as an intermediate high
frequency. The first interval spans from 9:30 am to 9:35 am, while the final interval spans
from 3:55 pm to 4:00 pm, and in total we have 78 intervals per day.
To obtain the daily order flow, which we label order imbalance and denote it with the
variable OIt , we sum the order flows across the intra-day intervals:

X
OIt = Order Flowt,d . (2)
d

The order flow measures the buy versus sell pressure in the market. It is positive when
investors are, overall, buying more options than selling them, and negative otherwise.
Finally, we calculate the daily options volume by summing the number of contracts traded
across all option series:
X
Volumet = Trade Sizei,t . (3)
i

[Figure 1 here]

Figure 1 displays the average daily volume and order imbalance for at-the-money put and
call options in each year of the sample period. Panel A1 confirms the well-known upward
trend in SPX option volumes since the years 2012-2013, observed in both call and put

13
options. Panel A2 documents some important characteristics of the daily order imbalances.
On average, the order flow is positive for SPX put options and negative for SPX call options,
displaying some variability across the years; this trend corresponds with findings from Chen,
Joslin, and Ni (2019) and Jacobs, Mai, and Pederzoli (2024), among others. In the aftermath
of the financial crisis, the order flow size surged, reaching an average of 2000 contracts as net
order flow per day in 2010 (positive for put options and negative for call options). Post-crisis,
the daily order flow size remained relatively stable with occasional deviations. For instance,
during the years 2015 or 2018, we observe a modest average daily order flow in both call and
put options. Particularly noteworthy are the last two years of our sample, 2020 and 2021,
where we document an average negative order flow for both call and put options, with a
magnitude around 2000.
Overall, the graph illustrates that, despite the surge in option volumes, buy and sell
orders remain relatively balanced throughout the day, resulting in no significant increase in
the overall size of the daily net order flow, consistent with results documented by exchange
analysts and recent literature.10 The next section will offer a new perspective on order flow
patterns by analyzing the intraday distribution, revealing that even when the daily order
flow is small, there can be substantial intraday variation.

4.2 Intraday Order Flow Distribution

In this section, we start our novel analyzes of the intraday distribution of the order flow.
Every day we calculate mean, standard deviation, skewness, and quartiles (q0.25 , q0.5 , and
q0.75 ) of the seventy-eight 5-minute intervals order flows calculated according to Equation 1.

[Table 1 here]
10
See, for example, [Link]
of-spx-0-dte-options/

14
 Panels A1 and B1 of Table 1 present the average of the daily statistics over the years for
ATM SPX call and put options. Figure 2 complements Table 1 by illustrating the time-series
of the average 5-minute order flow with intraday confidence intervals.11

[Figure 2 here]

The intraday buy and sell orders are largely balanced over the sample period, with the
average 5-minute order flow across years being -6 for ATM call options and 9 for ATM put
options. These averages vary across years, ranging from a minimum of -28.8 (recorded in
2020 for ATM calls) to a maximum of 32.7 (recorded in 2016 for ATM puts). However,
as shown in Figure 2, the mean 5-minute order flow does not exhibit any discernible time
trend. Low skewness estimates across all years further highlight the overall symmetry of the
intraday order flow distribution, which is confirmed by the median and 0.25–0.75 quartiles.
Standard deviations, in contrast, are quite large, ranging from 229.7 (in 2004 for ATM calls)
to 1764.7 (in 2011 for ATM puts). This results in wide confidence intervals for the average
5-minute order flow. For example, in 2011, the average 5-minute order flow for put options
is 6.2 contracts, but with a standard deviation of 1764.7, the confidence interval spans
[−385, 398] contracts, reflecting substantial variability in intraday order flow. Examining
the time-series of the average standard deviation by year, depicted in Figure 2, we find that
the distribution initially exhibited a higher degree of concentration in the early years of the
sample. Subsequently, it became more dispersed during the financial crisis in 2007, and, for
ATM call options, it then stabilizes with some notable spikes around 2018. For ATM puts,
the pattern is similar, with notable spikes in 2011 (concurrent to the European financial
crisis), and 2018 (concurrent with the Volmageddon incident).
11
Specifically, for every day in the sample, we compute the average intraday 5-minute order flow, µt , with
its confidence interval µt ± Z √σtn , where σt is the standard deviation of the intraday 5-minute order flows.
The figure displays the monthly averages of these daily quantities.

15
 In summary, this analysis shows that, beginning with the financial crisis in 2007, the
distribution of intraday order flow has remained stable over the years. It exhibits high
symmetry but also a very high level of standard deviation. Notably, in the ATM put market,
this standard deviation peaks during years marked by significant turbulence in volatility
markets.
To gain a preliminary insight into the relationship between intraday order flow distribu-
tion and option market quality, we compare the distribution of intraday order flow during
days characterized by high transaction costs with those characterized by low transaction
costs. Our goal is to identify the distribution characteristics that are significant for liquidity.
In accordance with Christoffersen, Goyenko, Jacobs, and Karoui (2018) and Bogous-
slavsky and Collin-Dufresne (2023), we measure the cost of trading options with the effec-
tive spread incurred by option traders. Specifically, for each trade i on day t, we define the
percent effective spread as:

Effective Spreadi = 2| ln Pi − ln Mi | (4)

where Pi is the price of the trade i and Mi is the prevailing midpoint of the NBBO. For
each day, the daily effective spread is the volume-weighted average of effective spreads across
trades within the same option category (ATM calls and puts).

[Figure 3 here]

Panel A of Figure 3 displays the time series of the daily effective spread (ESt ) across the
entire sample period for our samples of ATM SPX call and put options. The graph illustrates
a downward trend in the spread throughout the sample period, along with recurrent spikes
that may suggest seasonal patterns in both the spread and the daily changes in the spread.

16
We will account for seasonalities and time-trends in the regression analysis of Sections 4.3,
4.4, and 4.5.
We compare the intraday distribution of order flow on days characterized by low and
high trading costs as follows: for each year in the sample, we identify the days falling in
the bottom 10% and top 10% based on their ESt values.12 We then calculate the summary
statistics (mean, standard deviation, skewness, and quartiles) shown in Table 1 for each of
these subsamples. Panels A2 and B2 of Table 1 present the difference in these statistics
between days with low and high transaction costs, segmented by year.
The results are qualitatively similar across the years for both call and put options markets.
Days with low transaction costs have a distribution of intraday order flow that consistently
shows lower standard deviation and smaller interquartile range compared to days with high
transaction costs. Meanwhile, the distribution remains symmetric and with a small mean
in both subsamples, as evidenced by the minimal change in skewness and mean values. The
table also reports the results of testing whether the differences reported are statistically
significant within each year. We find that for the majority of the years, the differences
in standard deviations and first and third quartiles are statistically significant. None of
the other statistics show the same consistent pattern. The table also reveals no time-trend
in the difference between the standard deviation of order flow on days with low and high
trading costs, indicating that extreme distribution days have not become more pronounced
over time. However, the current high levels of volumes in the option market represent a mass
of traders which could potentially generate a very volatile order flow. This underscores the
importance of understanding the implications of volatile intraday order flow distributions.
In summary, the findings of this section suggest that the distribution of intraday order
12
Similar results are obtained when splitting the sample according to ∆ESt instead of ESt . Results are
provided in Table IA.1 in the Online Appendix.

17
flow holds significant economic implications for market liquidity. Specifically, days in which
the average 5-minute order flow is more volatile, as measured by the standard deviation of
the distribution and the interquantile range, appear to coincide with days with low option
market liquidity. Next section formally tests this pattern through a regression analysis.

4.3 Volatile Order Flow and Option Market Liquidity

In this section, we conduct a formal examination of the relationship between option market
liquidity and order-flow. We estimate separate time-series regressions for SPX call and put
options using the following specification:

ESt = α+β1 log(SDt )+β2 log(Volumet )+β3 |OIt |+Time Controls+Other Controls+t , (5)

where ESt measures the daily effective spread paid by investors for trading options on day
t,13 log(SDt ) denotes the logarithm of the standard deviation of the intraday order flow dis-
tribution on day t, log(Volumet ) is the logarithm of the daily volume calculated according to
Equation 3, and |OIt | is the absolute value of the daily order imbalance calculated according
to Equation 2.14 Time controls include day-of-the-week, month-of-year, and year dummies,
while other controls include the market return and VIX level on day t,15 the absolute value
of the average delta, vega and gamma of the options on day t, and one-day and two-day
13
An alternative measure of trading costs commonly used in the literature is the absolute spread, defined as
the spread in dollar terms rather than as a percentage of the mid-price. The robustness section 6.1 presents
the results using the absolute spread, which are qualitatively similar to those from the baseline analysis.
14
We use the absolute value of the order imbalance, following the findings of Christoffersen, Goyenko,
Jacobs, and Karoui (2018), who demonstrated that this measure is strongly related to illiquidity through a
market-maker inventory channel.
15
Qualitatively similar results are obtained when using maturity-specific implied volatility in place of the
VIX index. Results are available upon request.

18
lags of ESt .16 We further segment call and put option samples into maturity buckets with
one-week intervals, ranging from options expiring on the same day (zero days to maturity
or 0DTE), to options expiring in one week (1-6 days), and up to options expiring in seven
weeks (42-48 days to maturity). All variables are calculated separately for ATM call and
put options in each maturity bucket on day t17 , and standard errors are calculated using
Newey-West with the optimal lag suggested by Andrews and Monahan (1992).

[Table 2 here]

Table 2 presents the summary statistics of the dependent and independent variables
included in the regressions. The average spread is very high, ranging from a maximum of
9% for 0DTE options to a minimum of 2% for options with 21-48 days to maturity. Trading
volume decreases with maturity, while order imbalance increases with maturity, indicating
that the higher trading activity in ultra-short-term options is, on average, less directional
compared to longer-term options.

[Table 3 here]

Panels A1 and B1 of Table 3 present the regression results segmented by option maturity
buckets. The results consistently reveal a positive and statistically significant relationship
between the intraday volatility of order flow log(SDt ) and the effective cost of trading, in-
dicating that days characterized by greater volatility of intraday order flow correspond to
lower liquidity. This result holds across various maturity buckets and put call samples, and
remains robust after accounting for numerous controls. The breakdown of results into ma-
turity buckets reveals a trend in the coefficient of log(SDt ): the coefficient is higher for
16
Table IA.2 of the Online Appendix reports the results using the spread in changes rather than in level
and Table IA.7 reports the results using log(SDt ) scaled by volumes. The findings are qualitatively similar
to those from our baseline specification.
17
For 0DTE options we considered the greeks recorded on day t − 1.

19
short-term options and decreases almost monotonically with option maturity. We formally
test for differences in coefficients between the ultra-short maturity sample, including 0DTE
options, and other maturities, by performing a pooled regression of ESt on log(SDt ), with
dummies identifying each maturity bucket. Specifically, we introduce seven dummies, D1−6 ,
D7−13 , D14−20 , D21−27 , D28−34 , D35−41 , and D42−48 , representing each maturity bucket except
0DTE. The coefficient of log(SDt ) measures the sensitivity of illiquidity to volatile order
flow in 0DTE options, while interactions of log(SDt ) with these dummies assess whether the
coefficient differs in other maturity buckets compared to the 0DTE bucket. Panels A2 and
B2 of Table 3 present the results. The log(SDt ) coefficient is positive and significant, with
a magnitude consistent with the estimate for the 0DTE sample alone. The interaction term
coefficients are all negative and significant, confirming the lower sensitivity to order flow
volatility in options with longer maturities. To further illustrate this pattern, Panels A3 and
B3 display the standardized coefficients of log(SDt ) from Panels A1 and B1. Standardized
coefficients are computed by multiplying the raw coefficients by the standard deviation of
log(SDt ) and dividing by the standard deviation of ESt , each computed within maturity
buckets. The standardized coefficients also decline with maturity, though not strictly mono-
tonically, with the first two buckets (0 and 1–6 days) showing values roughly twice as large
as those for the longest maturities (35–41 and 42–48 days).
The coefficients in Table 3 related to the absolute value of order imbalance also offer
important insights and connection with the literature. The measure has been utilized in the
literature as a measure of demand pressure (Bollen and Whaley 2004; Garleanu, Pedersen,
and Poteshman 2008) or as an indicator of changes in option market-maker positions and
their associated inventory risk (Muravyev 2016; Christoffersen, Goyenko, Jacobs, and Karoui
2018). It may also serve, to some extent, as a proxy for order flow volatility. While it shows
strong significance in univariate regressions, its significance weakens considerably in the full

20
specifications, remaining significant only in a few subsamples.18 Importantly, it does not
overshadow the significance of order flow volatility. These results suggest that order flow
risk is more effectively captured by its intraday volatility, as measured by log(SDt ), rather
than by the aggregate order imbalance, |OIt |.
While the preceding analysis focuses on daily measures of spread, it is well-established
that bid-ask spread follows a U-shaped pattern over the trading day, widening at the open
and close (see, e.g., Cao, Jacobs, and Ke 2024). This raises the question of whether the
previously documented relationship between spreads and order flow volatility is concentrated
during specific periods of the trading day. To investigate this, we estimate separate daily
regressions using only trades executed within each individual trading hour. The regression
specification is as follows:

ESh,t = α+β1 log(SDh,t )+β2 log(Volumeh,t )+β3 |OIh,t |+Time Controls+Other Controls+h,t ,
(6)
where ESh,t denotes the effective spread computed using trades from hour h; log(SDh,t ) is
the volatility of the order flow, calculated from the 5-minute order imbalances within that
hour; log(Volumeh,t ) is the logarithm of the trading volume during hour h; and |OIh,t | is the
absolute value of the order imbalance during that hour.

[Table 4 here]

Table 4 reports the regression results by maturity bucket, using broader maturity group-
ings than in the main specification to ensure sufficient coverage of options with varying
maturities within each trading hour. All regressions include the same set of daily fixed ef-
18
Table IA.3 in the Online Appendix presents the regression results of illiquidity on the absolute value of
order imbalance, both in the univariate regression and together with log(SD).

21
fects as in Table 3, along with control variables computed over hour h.19 The relationship
between log(SDh,t ) and ESh,t remains positive and statistically significant across all trad-
ing hours and maturity groups, with only one exception (put options with 35–48 days to
maturity during the 12 p.m.–1 p.m. interval).
Taken together, the results in this section show that among order-flow variables, order
flow volatility stands out as the key driver, exhibiting a robust positive relationship with
trading costs across maturities and throughout the trading day.
An empirical challenge in this setting is the potential influence of omitted factors that
may simultaneously affect both variables. Although we control for a comprehensive set of
variables, it is not possible to fully eliminate all sources of confounding variation in the
SPX option setting. To address this concern, Section 4.6 leverages the structure of the U.S.
equity options market, where individual stock options are traded across sixteen exchanges
with relatively balanced volume. This structure enables us to estimate panel regressions
with different combinations of exchange, time, and stock fixed effects, thereby effectively
controlling for common shocks at the exchange, time, and stock levels.

4.4 Delta-Hedge Rebalancing Costs and Option Market Liquidity

An important potential determinant of option spreads is the cost of delta-hedging faced by

liquidity providers. To manage inventory risk, they might hedge by trading the underlying
asset in proportion to the option inventory’s delta (∆). This strategy involves transaction
costs and requires frequent rebalancing, as the delta of the position changes over time.
These changes, captured by gamma (Γ), become especially pronounced as options approach
19
Specifically, we control for day-of-the-week, month-of-year, and year fixed effects; one- and two-day lags
of the dependent variable; SPY returns over the interval; and the average implied volatility, delta, gamma,
and vega of the options during the hour.

22
expiration. Inventory models in the options market, which assume that liquidity providers
provide liquidity while maintaining a discretely rebalanced delta-hedged inventory, suggest
that spreads should reflect the associated delta-hedging costs (Jameson and Wilhelm 1992).
In this section, we examine the relationship between two proxies for liquidity providers’
delta-hedging costs in the SPX options market and transaction costs. As a first proxy, we
compute the delta-hedging cost generated by new order flow absorbed by liquidity providers
during each intraday interval d on day t. This is calculated as the absolute value of the
sum of order imbalances across all option series (puts and calls with all moneyness and
maturities up to seven weeks), weighted by the options’ delta and expressed in dollars, as in
Dim, Eraker, and Vilkov (2024).20 The total daily delta-hedging cost from new order flow,
DeltaOIt , is then obtained by summing across all intraday intervals:

X X X
DeltaOIt = DeltaOId,t = | OId,t,j ∆d,t,j Sd,t |
d d j

The second proxy for delta-hedging costs captures the gamma exposure of liquidity
providers’ inventory, measured at each intraday interval d. To construct this measure, we
first approximate the starting inventory on day t using the market-maker inventory at the
end of day t − 1, calculated from the CBOE Open-Close database.21 For each option series
j, we then update this starting value by the cumulative net order flow absorbed by liquidity
providers from the beginning of day t up to the end of interval d, yielding the inventory
position Invd,t,j . We compute the gamma exposure in dollar terms at each interval d by
20
The delta of each option series and the price of the underlying asset are recorded at the end of each
5-minute interval.
21
The CBOE Open-Close database reports the daily number of buy and sell orders by end-users
(non–market-makers) in the SPX options market. The cumulative net order flow from end-users, calcu-
lated from options inception, provides a proxy for (minus) the inventory held by market-makers. We merge
this data with OptionMetrics. See Jacobs, Mai, and Pederzoli (2024) for details on the filtering and merging
procedures.

23
weighting these inventory positions by the option’s gamma and the square of the underly-
ing price. The daily gamma-based hedging cost is obtained by summing across all intraday
intervals:
X X X
2
GammaInvt = GammaInvd,t = | Invd,t,j Γd,t,j Sd,t |.
d d j

This measure reflects the intensity of delta-hedging adjustments required throughout the day.
These costs are higher when gamma is elevated, which typically occurs for near-expiration
options.

[Tables 5(a) and 5(b) here]

Tables 5(a) and 5(b) report regressions of ESt on GammaInvt (Panel A) and DeltaOIt
(Panel B) for call and put options, respectively. For readability, only results for the first four
maturity buckets are shown; results for longer maturities, which are qualitatively similar, are
provided in Tables IA.4 and IA.5 of the Online Appendix. In the univariate specifications,
GammaInvt is positively associated with the spread for 0DTE calls and puts, as well as for
puts with 7–13 days to maturity. In contrast, DeltaOIt shows a positive relationship with
the spread for all maturities except 0DTE. These findings align with inventory-based option
pricing models (Jameson and Wilhelm 1992) and indicate that bid-ask spreads partially re-
flect liquidity providers’ delta-hedging costs, with effects varying across maturities. However,
once we control for order flow volatility, most coefficients on GammaInvt and DeltaOIt lose
significance or even change sign. In the full specification, including log(SD) and all controls
from the baseline analysis, the delta-hedging proxies remain insignificant or negative, while
log(SD) consistently remains positive and significant.
Altogether, these results confirm that the volatility of the order flow is a key determi-
nant of the spread and suggest that liquidity providers actively manage their inventories

24
throughout the day, with delta-hedging costs playing a secondary role in influencing trading
costs.

4.5 Volatile Order Flow and Liquidity in Individual Stock Options

This section analyzes the relationship between trading costs and order-flow in the market for
options on individual stocks. We consider the constituents of the S&P 500, tracking them
monthly from the beginning of our sample. A stock-day is included in our sample if the
stock was part of the S&P 500 index in the preceding month.22 Panel B of Figure 1 displays
the average daily volume and order imbalance of at-the-money equity options with up to 48
days to maturity. Unlike SPX options, we find that investors trade more call options than
put options on individual stocks, with the difference in volumes significantly increasing from
2020 onwards. Panel B2 indicates that the daily order imbalance is, on average, positive for
both call and put options. Our findings are novel but qualitatively align with the summary
statistics provided by Bryzgalova, Pavlova, and Sikorskaya (2023) and Bogousslavsky and
Muravyev (2024) on retail trading, which accounts for a substantial portion of volume in
options on individual stocks in recent years.
We construct order-flow variables separately for each stock s and day t using the same
procedure as for SPX outlined in section 4.1. Specifically, daily volume, order imbalance, and
effective cost of trading are constructed separately for each stock-day option-type following
equations 3, 2, and 4, respectively, and log(SDs,t ), is the logarithm of the daily standard
deviation of the seventy-eight 5-minute order imbalances.23
22
Figure IA.1 in the Online Appendix shows the number of individual stocks in our sample over time. In
the early years, the count ranges from 100 to 200, eventually stabilizing between 300 and 400 from 2009
onward. The sample size aligns with that used by Christoffersen, Goyenko, Jacobs, and Karoui (2018). As
a robustness check, we verified that our findings remain robust even when excluding the early years of the
sample. Results are available upon request.
23
Since options on individual stocks may not be traded as frequently as SPX options, we include a stock-

25
 We perform a panel regression of ESs,t on log(SDs,t ), log(volumes,t ), and |OIs,t | with
stock-fixed effect.24 Other controls include the average implied volatility of options on stock
s and day t, IVs,t , and the average of the options greeks, i.e., gamma, vega, and the absolute
value of delta on stock s and day t.25 We also control for stock characteristics, as stock return,
firm size and stock volume. Time fixed-effects include day-of-the-week, month-of-the-year,
and year controls. Standard errors are double clustered at the day and stock level.

[Table 6 here]

Table 6 presents the results for call options (Panel A) and put options (Panel B), with
samples further divided into two maturity groups: options with up to 24 days to expiration
and those with 25 to 48 days. The results are robust and align with our earlier findings in
SPX options. First, we observe a strong positive relationship between the standard deviation
of order flow and illiquidity for both call and put samples. Moreover, the coefficient is larger
for very short-maturity options (up to 24 days), confirming that trading costs in shorter-
dated contracts are more sensitive to intraday order flow volatility. Finally, the results show
that |OIs,t | is positively related to the spread only in the sample of medium-term options
(25–48 days to expiration).

4.6 Evidence from Exchange-Level Analysis

In this section, we analyze the relationship between spreads and order flow volatility at the
exchange level. We begin by documenting a substantial degree of heterogeneity in volume

day option type (call/put) in the sample if the option group has at least ten non-empty intervals out of the
seventy-eight.
24
Qualitatively similar results are obtained using a cross-sectional Fama-MacBeth regression instead of
the panel regression, and they are presented in Table IA.6 in the Online Appendix.
25
For 0DTE options, we use the greeks recorded on day t − 1.

26
absorption across exchanges, with no single exchange dominating the market. The immediate
cross-exchange liquidity response following trades also suggests a good level of competition
among exchanges. Therefore, this setting allows us to estimate panel regressions of exchange-
specific illiquidity on exchange-specific order flow volatility, while controlling for various sets
of fixed effects.
Individual stock options trade simultaneously across sixteen exchanges, and the OPRA
database reports the exchange identifier for each trade, along with the contemporaneous best
bid and offer quotes across all exchanges. For these analyses at the exchange level we focus
on the constituents of the Dow Jones which have been part of the index since the start of
our sample period, January 2004. Our sample includes the following sixteen tickers: AXP,
BA, CAT, DIS, DOW, HD, IBM, INTC, JNJ, JPM, KO, MMM, MRK, MSFT, PG, and
WMT. These options have been actively traded throughout the sample period, with trading
volumes well distributed across exchanges. Consistent with our main analysis, we examine
one-month (up to 48 days to maturity) at-the-money call and put options, where moneyness
is based on the delta reported by OptionMetrics at the close of the previous day.

[Figure 4 here]

Panel A of Figure 4 shows the percentage of trading volume in our sample absorbed by
each of the sixteen exchanges. The exchange with the highest share is CBOE, which accounts
for 19.8% of the volume, followed by ISE (15%), PHLX (13.5%), ARCA (9.69%), NASDAQ
(8.86%), AMEX (8.08%), BZX (6.87%), and BOX (4.21%). These are the largest exchanges,
and with the exception of BZX, they have operated since 2004. The remaining exchanges
also absorb a non-negligible share of trading activity, although less than the major venues
listed above. Panel B of Figure 4 reports the percentage of trading volume by exchange for
each year in our sample period. It shows that cross-sectional heterogeneity in volume has

27
increased over time, with the leading exchanges, CBOE and ISE, gradually losing market
share to other venues. Overall, the graphs indicate that trading activity in our sample is
relatively well distributed, with no single exchange dominating the market.
To investigate the cross-exchange relationship between spreads and order flow, we begin
with a trade-by-trade analysis aimed at answering two key questions: (i) Are trades, on
average, executed on the exchange quoting the lowest spread? And (ii) Following a trade,
how do exchanges revise their quoted spreads depending on whether they absorbed the trade
or not?
To this end, we track, for each ticker and trade, the level of the quoted spread and its
subsequent change across all exchanges.26 For this analysis only, we exclude trades that
sweep an entire layer of the order book, that is, those with a size greater than or equal to the
corresponding bid or ask quantity on the executing exchange. All other filters are consistent
with those used in the daily baseline analysis. We then estimate the two following pooled
regressions specification, separately for each stock:

Illiqi,j,τ = α + βDummyi,j,τ + τ , (7)

where Illiqi,j,τ is either Spreadi,j,τ , the quoted spread in exchange i for option j at time
τ , or ∆Spreadi,j,τ +1 , the change in the quoted spread in exchange i for option j from the
trade time τ to the next trade time τ + 1. We measure the quoted spread as the difference
between the quoted ask and bid prices on the exchange, divided by the exchange mid price.
Dummyi,j,τ is a dummy variable which equals one for exchange i where the trade in option
j occurred at time τ . It is zero for all other exchanges. The dummy variable thus captures
26
For simultaneous trades occurring on the same option and exchange, we consolidate them into a single
observation. This aggregated observation has a trade size equal to the signed sum of the individual trade
sizes and a trade price that is the average of the individual trade prices.

28
the differential liquidity between the exchange that absorbed the trade versus the others.

[Table 7 here]

The results are presented by stock and option type in Panel A of Table 7, with all coef-
ficients multiplied by 100.27 We find consistent and robust patterns across both individual
stocks and option types (calls and puts). The first specification, which uses the level of the
quoted spread as the dependent variable, shows that trades tend to occur on exchanges quot-
ing spreads approximately 6% lower than the average quoted spread across other exchanges.
This result is in line with the expectation that investors and brokers use routing strategies to
minimize transaction costs. The second specification, which examines the change in spreads
following a trade, reveals that quoted spreads tend to decline by ten to twenty basis points
on exchanges that did not absorb the trade, while increasing on the exchange where the
trade occurred. The net change in the trading exchange’s spread is approximately thirty
basis points. These findings suggest that exchanges respond immediately to order flow:
trading exchanges widen their spreads to incorporate the cost of liquidity provision, while
non-trading exchanges tighten theirs, potentially to attract subsequent volume. Overall, the
results point to a good degree of competition across exchanges.
We next delve deeper into the relationship between spreads and order flow volatility
documented in the main analysis, leveraging this rich cross-sectional structure of trading
across exchanges. Our setting, in which the same stock options trade simultaneously across
multiple venues, offers a unique opportunity to help isolate the impact of order flow volatil-
ity on effective spreads from the influence of other macro-variables that may affect both.
This design helps mitigate concerns about omitted variables that may confound the baseline
27
All regressions include day fixed effect, and the standard errors are clustered at the day and exchange
levels. Spreadi,j,τ and ∆Spreadi,j,τ +1 are also winsorized at the 1% and 99% levels to eliminate instances
of apparently unrealistic quotes reported by OPRA.

29
results. Specifically, we estimate a panel regression of the effective spread for stock s on
exchange i and day t, ESi,s,t , on log(SDi,s,t ), the intraday order flow volatility calculated
exclusively from trades absorbed by exchange i. We begin with a specification that in-
cludes exchange and day fixed effects, which control for structural differences across venues
as well as macro-level variation over time. We then estimate a second specification with
stock-by-day fixed effects, which absorb all variation across both stocks and time, thereby
accounting for any stock-specific or macroeconomic factors. Panel B of Table 7 displays
the results. To further strengthen the analysis, we address the possibility that results are
driven by small versus large exchanges by estimating two separate regressions: one using
data from all exchanges (specification labeled “All Exchanges”) and another using only the
six major venues that have been active throughout the sample period (specification labeled
“Six Major Exchanges”). We also control for exchange-specific volume, log(V olumei,s,t ), and
for exchange-specific order imbalance, |OIi,s,t |. The results document that the coefficient on
log(SDi,s,t ) is positive across all specifications, reinforcing the core finding that greater order
flow volatility is associated with higher transaction costs.
While this cross-exchange approach enhances identification, it is not without limitations.
In an ideal setting, exchanges would differ only in the order flow they absorb. In prac-
tice, however, trading venues may vary in unobservable ways that influence both order flow
volatility and liquidity provision. We have addressed these concerns in two ways. First, we
have restricted part of the analysis to the six largest and most liquid exchanges, which are
more homogeneous in their structure and function. Second, we have included exchange fixed
effects in some specifications to control for time-invariant differences across venues. The
results remain strong and consistent across specifications, alleviating concerns that they are
driven by structural differences across exchanges. We conclude that the cross-sectional ex-
change design meaningfully enhances identification and supports the robustness of our core

30
findings.

5 When is the Order Flow More Volatile?

Order flow can stem from two primary sources: informational trading and liquidity-driven
activity. Disentangling these channels has long been a central challenge in market mi-
crostructure research. In this section, we aim to shed light on the underlying drivers of
our key variable, the order flow volatility. While informed trading may generate order flow
volatility due to asymmetric information, inventory and liquidity needs can create temporary
but substantial intraday imbalances as investors rebalance their positions. To evaluate which
mechanism better explains the behavior of order flow volatility, we estimate panel regressions
of log(SDs,t ) on a set of event-day indicators designed to capture either information-driven
activity or inventory-related rebalancing. The results, presented in Table 8, are reported
separately for ATM calls and puts across two maturity buckets: short-term (0–24 days) and
medium-term (25–48 days).

[Table 8 here]

The table documents that order flow volatility drops significantly on earnings announce-
ment days for all maturities, both on the day of the announcement and slightly on the
preceding day. This pattern is hard to reconcile with the interpretation that order flow
volatility reflects informed trading, as earnings days are typically associated with heightened
informational asymmetry.
In contrast, we observe spikes in order flow volatility on calendar dates more closely linked
to inventory rebalancing. Specifically, log(SDi,s,t ) increases for medium-maturity options at
the end of the month and shows a slight rise at the beginning of the month, consistent with

31
institutional portfolio rebalancing cycles. For short-term options, order flow volatility is
slightly lower on these days, which can be consistent with the idea that end-of-month and
beginning-of-month rebalancing activity is concentrated in longer-dated contracts. For short-
term options, order flow volatility is higher on the third Friday of each month, coinciding
with standard monthly option expirations. In contrast, medium-maturity options exhibit a
slightly lower values of log(SDi,s,t ) on these Fridays, which may reflect the greater flexibility
that investors have in managing longer-dated positions without needing to act precisely
on expiration day. Finally, we find that order flow volatility decreases across all maturity
buckets when underlying stock volatility is high, supporting the view that it is not driven
by market uncertainty either.
Taken together, these results suggest that order flow volatility is more closely tied to
inventory risk than to adverse selection risk.

6 Additional Analysis and Robustness

This section presents the results of several robustness checks. Section 6.1 shows that the
main findings hold when using the dollar spread instead of the relative spread. Section
6.2 incorporates option return volatility as an additional control. Section 6.3 uses order flow
volatility scaled by volume and confirms the robustness of the results in the out-of-the-money
options sample. It also provides additional analyses showing that the relationship between
volatile order flow and illiquidity is not driven by (i) retail trading, (ii) market opening and
closing sessions, and that it remains robust even when time fixed effects are excluded. The
corresponding tables for Section 6.3 are provided in the Online Appendix.

32
6.1 Dollar Spread

The main measure of trading costs used in our analysis is the effective spread, calculated
according to Equation 4. This measure expresses the spread in log terms, providing a relative
measure of trading costs with respect to the option price. It captures the reduction in option
returns that traders incur due to transaction costs. Our choice aligns with the existing
literature (Bogousslavsky and Collin-Dufresne 2023; Christoffersen, Goyenko, Jacobs, and
Karoui 2018) and reflects our goal of understanding how trading costs faced by traders are
influenced by potentially risky patterns in the order flow distribution.
As a robustness check, we test our results using an alternative measure of trading costs:
the spread expressed in dollar terms. This measure quantifies the dollar gain a liquidity
provider earns by supplying liquidity in a trade and immediately reversing the position with
another trade of the opposite sign. Specifically, for each trade i, the dollar spread is defined
as:
Dollar Spreadi = |Pi − Mi |

where Pi is the trade price and Mi is the prevailing midpoint of the NBBO. On each day, the
daily dollar spread is calculated as the volume-weighted average of the dollar spreads across
trades, scaled by the value of the underlying asset on day t, St .

[Table 9 here]

[Table 10 here]

Panel A of Tables 9 and 10 present the regression results for SPX options and equity
options, respectively. In these regressions, we use the dollar spread as the dependent variable
instead of our baseline measure of spread, ESt . All other controls remain consistent with

33
those used in Tables 3 and 6. The results consistently show a positive and robust relationship
between the volatility of the order flow and the dollar spread for both call and put options,
as well as for SPX and equity options. Panel A of Table 9 documents that for SPX options,
the magnitude of the coefficients is particularly high for ultra-short-term options (0DTE
and 1-6 days to maturity), confirming that liquidity is more sensitive to volatile order flow
for these very short maturity categories. For equity options, Panel A of Table 10 shows
that the difference in coefficients between short and medium maturity options is positive
but smaller. Nonetheless, even a small difference in the absolute spread can result in a
substantial difference in the relative spread and trading costs for investors (as documented
in Table 6), given the lower option prices for shorter maturities options.

6.2 Relation with Realized Option Volatility

All market microstructure models of inventory and asymmetric information (see Foucault,
Pagano, and Röell 2013 for a review) predict that transaction costs should be positively
related to asset volatility. This section formally tests this hypothesis and assesses the ro-
bustness of our results to the inclusion of a variable that measures the realized intraday
volatility of options. This addresses the potential concern that order flow volatility might
act as a proxy for the underlying volatility of the options themselves.
We compute the realized option volatility (ORVt ) for day t by summing the squared
average 5-minute option returns across the seventy-eight 5-minute intervals throughout the
day. Panel B of Tables 9 and 10 present the regression results for SPX options and equity
options, respectively. The results confirm a robust positive relationship between ORVt and
illiquidity, consistent with theoretical predictions. However, the significance of the volatility
of the order flow (log(SD)) is not subsumed by ORVt , as shown in the second specification

34
of each subsample. This indicates that, while the two variables are generally correlated
(ranging from a minimum of -10% to a maximum of 30%, depending on the sample), they
convey distinct information about transaction costs and liquidity.

6.3 Additional Robustness

In this section, we perform a series of additional checks to validate the robustness of our main
findings. We begin by scaling both the volatility of order flow and order imbalance by daily
volume, resulting in the variables log(SD/volume)t and |OI/volume|t . Table IA.7 reports
the time-series regression for SPX options using these scaled variables. The findings confirm
a positive relationship between the intraday volatility of order flow log(SD/volume)t and
illiquidity, consistent with our baseline results.
Next, we assess whether the relationship holds when changing the moneyness criterion
of the sample. Specifically, we replace at-the-money (ATM) options with out-of-the-money
(OTM) options, defined as those with absolute delta values between 0.125 and 0.375 (based
on OptionMetrics data at the close of day t − 1). Table IA.8 reports the results. Consistent
with our baseline analysis, we find a positive and statistically significant relationship between
intraday order flow volatility and trading costs. The effect is even more pronounced than in
the ATM sample, with coefficients declining as option maturity increases.
We then examine the sensitivity of our results to the exclusion of time fixed effects,
such as day-of-week, month-of-year, and year dummies. This check addresses the concern
raised by Jennings, Kim, Lee, and Taylor (2024) that controlling for calendar effects may
inadvertently bias regression estimates. Table IA.9 report the results obtained excluding
day-of-week and month-of-year controls. While the adjusted R2 values mildly decrease after
removing these time controls, the primary results and inferences remain consistent.

35
 Finally, we investigate whether retail trading or specific trading sessions contribute dis-
proportionately to the observed effects. To do so, we re-estimate the main regression from
Table 3, first excluding trades identified as retail (using the ‘SLAN’ flag, following Bryz-
galova, Pavlova, and Sikorskaya 2023), and then excluding trades executed during the first
and last 30 minutes of the trading day. Tables IA.10 and IA.11 report the corresponding
results. In both cases, the positive relationship between order flow volatility and illiquidity
remains robust, suggesting that the main effect is not driven by retail activity or by market
open and close dynamics.
Taken together, these robustness checks reinforce our main conclusion: intraday order
flow volatility is a key determinant of trading costs in the options market, and the relationship
is stable across a range of specifications, subsamples, and alternative controls.

7 Conclusion

The recent surge in volumes in option contracts with increasingly shorter expirations has
raised concerns among academics and regulators about the stability of this expanding market.
A key characteristic of the options market is its high level of transaction costs, leaving an
open question as to how effectively liquidity providers can further absorb large, potentially
imbalanced order flows while maintaining an efficient and well-functioning market.
Our analysis documents economically and statistically significant positive relationship
between intraday order flow volatility and illiquidity in options market, particularly for ultra-
short term options. The effect is pervasive: it holds in the time-series and cross-sectional
dimension, and it outweighs the significance of more traditional daily first-moment measures
of order flow dynamics, such as volumes or absolute order imbalances. Furthermore, it
also outweighs the significance of traditional measures capturing the delta-hedging needs

36
of market makers. These findings suggest that liquidity providers rely primarily on active
inventory rebalancing and trade matching throughout the day, with the main source of
inventory risk arising from providing liquidity to unbalanced order flows. An exchange-
level analysis that includes stock-by-day fixed effects further confirms the robustness of this
relationship and helps mitigate concerns related to omitted variable bias.
Our findings underscore the potential risks posed by high volumes in short-term option
contracts, which can amplify intraday order flow volatility and challenge market stability. We
show that as intraday order flow volatility rises, liquidity providers widen bid-ask spreads
to manage the elevated risk, resulting in higher hedging costs for investors increasingly
dependent on short-term rollover strategies over long-term hedges. This spread widening,
in turn, can impair market efficiency by reducing liquidity and price discovery, which may
in turn elevate systemic risk. These dynamics highlight critical aspects that regulators
should consider to maintain stability and market quality in financial markets. An interesting
direction for future research would be to explore the broader implications of unbalanced order
flow in the options market, including its impact on investors’ portfolios, hedging strategies,
and, more generally, on risk premia in financial markets.

37
References
Adams, Greg, Jean-Sebastien Fontaine, and Chayawat Ornthanalai, 2024, The market for
0-days-to-expiration: The role of liquidity providers in volatility attenuation, Working
Paper.

Almeida, Caio, Gustavo Freire, and Rodrigo Hizmeri, 2024, 0DTE Asset Pricing, Working
Paper.

Andrews, Donald WK, and J Christopher Monahan, 1992, An improved heteroskedasticity

and autocorrelation consistent covariance matrix estimator, Econometrica: Journal of the
Econometric Society pp. 953–966.

Bandi, Federico, Nicola Fusari, and Roberto Reno, 2024, 0DTE Option Pricing, Working
Paper.

Beckmeyer, Heiner, Nicole Branger, and Leander Gayda, 2023, Retail Traders Love 0DTE
Options... But Should They?, Working Paper.

Bogousslavsky, Vincent, and Pierre Collin-Dufresne, 2023, Liquidity, volume, and order im-
balance volatility, The Journal of Finance 78, 2189–2232.

Bogousslavsky, Vincent, and Dmitriy Muravyev, 2024, An Anatomy of Retail Option Trad-
ing, Working Paper.

Bollen, Nicolas PB, and Robert E Whaley, 2004, Does net buying pressure affect the shape
of implied volatility functions?, The Journal of Finance 59, 711–753.

Brogaard, Jonathan, Jaehee Han, and Peter Y Won, 2023, How does zero-day-to-expiry
options Trading affect the volatility of underlying assets?, Working Paper.

Brogaard, Jonathan, Terrence Hendershott, and Ryan Riordan, 2014, High-frequency trading
and price discovery, The Review of Financial Studies 27, 2267–2306.

Bryzgalova, Svetlana, Anna Pavlova, and Taisiya Sikorskaya, 2023, Retail trading in options
and the rise of the big three wholesalers, The Journal of Finance 78, 3465–3514.

Cao, Jie, Kris Jacobs, and Sai Ke, 2024, Derivative Spreads: Evidence from SPX Options,
Working Paper.

Chen, Hui, Scott Joslin, and Sophie Xiaoyan Ni, 2019, Demand for crash insurance, interme-
diary constraints, and risk premia in financial markets, The Review of Financial Studies
32, 228–265.

38
Cho, Young-Hye, and Robert F Engle, 1999, Modeling the impacts of market activity on
bid-ask spreads in the option market, National Bureau of Economic Research, Cambridge,
Massachusetts, USA.

Chordia, Tarun, Jianfeng Hu, Avanidhar Subrahmanyam, and Qing Tong, 2019, Order flow
volatility and equity costs of capital, Management Science 65, 1520–1551.

Chordia, Tarun, and Avanidhar Subrahmanyam, 2004, Order imbalance and individual stock
returns: Theory and evidence, Journal of Financial Economics 72, 485–518.

Christoffersen, Peter, Ruslan Goyenko, Kris Jacobs, and Mehdi Karoui, 2018, Illiquidity
premia in the equity options market, The Review of Financial Studies 31, 811–851.

Dim, Chukwuma, Bjorn Eraker, and Grigory Vilkov, 2024, 0DTEs: Trading, Gamma Risk
and Volatility Propagation, Working Paper.

Duarte, Jefferson, Christopher S Jones, and Junbo L Wang, 2024, Very noisy option prices
and inference regarding the volatility risk premium, The Journal of Finance 79, 3581–3621.

Ellis, Katrina, Roni Michaely, and Maureen O’Hara, 2000, The accuracy of trade classifi-
cation rules: Evidence from Nasdaq, Journal of Financial and Quantitative Analysis 35,
529–551.

Ernst, Thomas, and Chester S Spatt, 2022, Payment for order flow and option internalization,
Available at SSRN 4056512.

Foucault, Thierry, Marco Pagano, and Ailsa Röell, 2013, Market liquidity: theory, evidence,
and policy. (Oxford University Press, USA).

Fournier, Mathieu, and Kris Jacobs, 2020, A tractable framework for option pricing with dy-
namic market maker inventory and wealth, Journal of Financial and Quantitative Analysis
55, 1117–1162.

Garleanu, Nicolae, Lasse Heje Pedersen, and Allen M Poteshman, 2008, Demand-based
option pricing, The Review of Financial Studies 22, 4259–4299.

George, Thomas J, and Francis A Longstaff, 1993, Bid-ask spreads and trading activity in
the S&P 100 index options market, Journal of Financial and Quantitative Analysis 28,
381–397.

Glosten, Lawrence R, and Paul R Milgrom, 1985, Bid, ask and transaction prices in a
specialist market with heterogeneously informed traders, Journal of Financial Economics
14, 71–100.

39
Goyenko, Ruslan, Chayawat Ornthanalai, and Shengzhe Tang, 2015, Options illiquidity:
Determinants and implications for stock returns, Rotman School of Management Working
Paper.

Grossman, Sanford J, and Merton H Miller, 1988, Liquidity and market structure, the Jour-
nal of Finance 43, 617–633.

Ho, Thomas SY, and Hans R Stoll, 1983, The dynamics of dealer markets under competition,
The Journal of Finance 38, 1053–1074.

Hu, Jianfeng, Antonia Kirilova, Dmitriy Muravyev, and Doojin Ryu, 2024, Options Market
Makers, Working Paper.

Huang, Xing, Philippe Jorion, and Christopher Schwarz, 2024, Some Anonymous Options
Trades Are More Equal than Others, Available at SSRN 4951825.

Jacobs, Kris, Anh Thu Mai, and Paola Pederzoli, 2024, Identifying Demand Curves in Index
Option Markets, Management Science.

Jameson, Mel, and William Wilhelm, 1992, Market making in the options markets and the
costs of discrete hedge rebalancing, The Journal of Finance 47, 765–779.

Jennings, Jared, Jung Min Kim, Joshua Lee, and Daniel Taylor, 2024, Measurement error,
fixed effects, and false positives in accounting research, Review of Accounting Studies 29,
959–995.

Kelley, Eric K, and Paul C Tetlock, 2013, How wise are crowds? Insights from retail orders
and stock returns, The Journal of Finance 68, 1229–1265.

Lee, Charles MC, and Mark J Ready, 1991, Inferring trade direction from intraday data,
The Journal of Finance 46, 733–746.

Muravyev, Dmitriy, 2016, Order flow and expected option returns, The Journal of Finance
71, 673–708.

Ni, Sophie X, Neil D Pearson, Allen M Poteshman, and Joshua White, 2021, Does option
trading have a pervasive impact on underlying stock prices?, The Review of Financial
Studies 34, 1952–1986.

Stoikov, Sasha, and Mehmet Sağlam, 2009, Option market making under inventory risk,
Review of Derivatives Research 12, 55–79.

Stoll, Hans R, 1978, The supply of dealer services in securities markets, The Journal of
Finance 33, 1133–1151.

40
 Figure 1: Daily Volumes and Order Imbalances
Panel A: SPX options

Panel B: Individual Stock Options

This figure displays the average daily volume and order imbalance for at-the-money (ATM) options
with maturities up to one month (48 days), across each year in our sample period. Daily volume is
the total number of contracts traded, and daily order imbalance is the difference between buy and
sell initiated trades. Panel A displays the average daily volume (A1) and order imbalance (A2) for
SPX call and put options. Panel B plots for call and put options written on the stocks which are
part of the S&P500 index, where we compute average daily volume and order imbalance for each
stock-year, and then we take the cross-sectional averages for each year.

41
 Figure 2: Intraday Order Flow Distribution Over the Years

SPX Call Options SPX Put Options

600 800

600
400
400
200
200

0 0

-200
-200
-400
-400
-600

-600 -800
6

8
0

1
20

20

20

20

20

20

20

20

This figure displays the time-series of the average intraday 5-minute order flow for SPX ATM call
and put options with confidence intervals. The graph is obtained by dividing each trading day into
seventy-eight equal intervals, each covering five minutes, and calculating the order flow (buys minus
sells) of put and call options within each interval. The solid lines display the daily average of these
5-minute order flows, µt , while the dotted lines depict the 95% confidence intervals, calculated
as µt ± Zσ √ t , where σt is the intraday standard deviation of the seventy-eight order flows. For
n
readability, the graph displays the monthly averages of these daily quantities.

42
 Figure 3: Time-Series of ESt
Panel A: SPX Options

Panel B: Individual Stock Options

The figure presents the time-series of the daily effective spread for ATM call and put options.
Panel A presents the graph for SPX options while Panel B presents the graphs for individual stock
options, where a stock-day is included in our sample if the stock was part of the S&P 500 index in
the preceding month.

43
 Figure 4: Percentage of Volume Absorbed by Exchanges
Panel A: Full Sample

AMEX (8.08%)
NASD (8.86%)

EMLD (1.21%)
PEARL (2.47%) CBOE (19.8%)
EDGX (1.35%)

BZX (6.87%)

NASDBX (0.356%)
MIAX (3.35%)

C2 (1.18%)

GEMX (3.74%)
ISE (15%)
MRX (0.312%)
BOX (4.21%)

PHLX (13.5%) ARCA (9.69%)

Panel B: Annual Time-Series

0.9

0.8

0.7

0.6

NASD
0.5 AMEX
CBOE
ISE
0.4 ARCA
PHLX
BOX
0.3 MRX
GEMX
C2
0.2 MIAX
NASDBX
BZX
0.1 EDGX
PEARL
EMLD
0
2004 2006 2008 2010 2012 2014 2016 2018 2020 2022

Panel A reports the percentage of at-the-money call and put option volume (with maturities up
to seven weeks) absorbed by each exchange for the sixteen tickers analyzed in Section 4.6 over the
2004–2021 sample period. Panel B shows the yearly breakdown of these exchange-level volume
shares.

44
 Table 1: Intraday Order Flow Distribution Over the Years
Panel A: SPX Calls
A1: Five-minute Order Flow Summary Statistics A2: Difference in Distribution Between Low and High ES Days
Year Mean Std Skewness Q25 Q50 Q75 ∆Mean ∆Std ∆Skew ∆Q25 ∆Q50 ∆Q75
2004 2.264 229.677 -0.175 -31.429 2.998 41.916 0.068 -161.011∗∗∗ 1.263∗ 27.712∗∗∗ -3.904∗∗∗ -47.394∗∗∗
2005 3.001 350.244 0.166 -48.560 -0.181 48.458 5.256 -192.330∗∗∗ 1.471∗ 40.721∗∗∗ -0.731∗ -44.673∗∗∗
2006 1.217 510.297 -0.172 -68.341 1.978 82.333 -13.141 -194.946∗∗ -1.176 64.615∗∗∗ -8.481 -79.904∗∗∗
2007 5.031 917.111 -0.174 -129.776 3.199 148.560 -11.598 -576.354∗∗∗ -0.546 181.260∗∗∗ -9.250 -209.769∗∗∗
2008 -3.349 969.571 -0.479 -127.992 3.079 153.907 -4.778 -839.847∗∗∗ 0.334 193.269∗∗∗ -11.250 -237.298∗∗∗
2009 -6.875 1047.319 -0.086 -113.103 1.246 111.810 -65.604 -688.481∗∗∗ 0.230 110.087∗∗∗ -3.673 -140.462∗∗∗
2010 -21.329 971.170 -0.593 -119.851 1.587 125.033 -98.646∗ -433.039∗∗ -0.022 41.356 -8.288 -164.808∗∗∗
2011 18.689 922.002 -0.193 -127.552 6.863 151.914 -46.192 -504.000∗∗∗ 0.721 131.096∗∗∗ -19.404∗ -232.990∗∗∗
2012 -6.925 723.389 -0.190 -121.593 3.850 124.954 -20.333 -319.204∗∗∗ 0.924 54.680∗∗ -12.800∗ -98.560∗∗
2013 -23.995 875.890 -0.070 -152.160 -6.550 121.287 -1.147 -235.637 -1.075 111.615∗∗∗ 0.654 -85.404∗∗∗
2014 -19.301 846.771 -0.209 -152.172 -4.938 135.027 -51.440 -331.513∗∗∗ -0.988 111.413∗∗∗ 11.077 -119.740∗∗∗
2015 -3.070 792.471 -0.312 -133.594 3.210 140.449 -42.433 -295.734∗∗ -0.667 78.231∗∗∗ -14.788∗∗∗ -150.106∗∗∗
2016 6.924 852.823 -0.080 -164.524 2.062 179.867 -83.110∗∗ -213.614 -0.300 97.106∗∗∗ -18.154∗ -238.606∗∗∗
2017 -22.156 1157.599 -0.274 -179.111 -4.414 160.257 86.379∗∗ -432.589∗∗ 1.334 142.875∗∗∗ 4.000 -110.856∗∗∗
2018 -1.567 1121.613 -0.042 -196.344 -1.600 189.645 -84.126∗ -247.488 -1.019 170.817∗∗∗ 7.462 -212.990∗∗∗
2019 -2.056 1020.217 0.064 -160.959 -2.187 148.254 10.189 -431.078 0.253 98.981∗∗∗ -1.269 -74.798∗∗∗
2020 -28.763 597.248 -0.354 -155.767 -12.866 124.071 92.753∗∗∗ -407.609∗∗∗ 0.411 199.413∗∗∗ 64.904∗∗∗ -48.010

Panel B: SPX Puts

B1: Five-minute Order Flow Summary Statistics B2: Difference in Distribution Between Low and High ES Days
Year Mean Std Skewness Q25 Q50 Q75 ∆Mean ∆Std ∆Skew ∆Q25 ∆Q50 ∆Q75
2004 10.890 267.683 0.394 -34.179 3.432 50.474 -16.995 -212.596∗∗∗ 0.045 37.644∗∗∗ -2.577 -58.538∗∗∗
2005 0.827 385.758 -0.082 -54.116 2.437 62.690 -10.693 -176.646∗∗ -0.718 48.769∗∗∗ 0.231 -47.260∗∗∗
2006 6.737 486.335 0.097 -76.521 1.319 87.596 20.745 -379.076∗∗∗ 0.687 136.962∗∗∗ -0.212 -119.625∗∗∗
2007 3.936 1011.164 0.007 -163.017 2.255 178.691 28.269 -626.536∗∗ 0.198 221.077∗∗∗ 0.000 -176.558∗∗∗
2008 -12.104 1070.962 -0.249 -186.361 -1.318 180.029 31.167 -812.081∗∗∗ 0.191 240.269∗∗∗ 19.115 -193.548∗∗∗
2009 15.571 980.570 0.074 -90.806 4.847 111.010 28.412 216.043 0.312 94.702∗∗∗ -4.827 -104.769∗∗∗
2010 28.771 1279.395 0.445 -111.428 9.440 143.686 -25.708 -882.587∗∗∗ -0.426 134.298∗∗∗ -24.058 -186.240∗∗∗
2011 6.218 1764.674 -0.096 -143.393 10.411 183.536 -18.898 621.615 -0.691 52.683 -13.577 -99.519
2012 13.453 864.470 0.031 -122.614 3.780 140.172 -25.652 -48.268 0.434 63.590∗ -4.460 -85.960∗∗
2013 19.639 811.893 0.216 -118.472 7.240 145.224 -13.845 -411.241∗∗∗ -0.851 64.644∗∗ -6.885 -106.548∗∗∗
2014 25.477 858.392 0.278 -131.870 11.296 179.445 -64.608 -226.117 -0.796 67.269∗∗∗ -12.635 -117.029∗∗∗
2015 -8.600 803.362 -0.187 -153.960 4.496 161.197 -35.684 -277.738∗ -0.745 113.058∗∗∗ -8.808 -149.885∗∗∗
2016 32.651 756.723 -0.080 -145.812 14.167 214.905 -79.378∗∗∗ -370.184∗∗∗ -0.029 110.067∗∗∗ -18.500∗∗∗ -266.760∗∗∗
2017 22.620 678.144 -0.074 -155.832 8.472 194.332 -39.823 -276.824∗∗∗ -0.405 63.298∗ -13.058 -130.875∗∗∗
2018 9.680 1027.553 0.087 -197.475 7.886 232.516 16.172 -606.494∗ 0.962 120.971∗∗ -17.250 -218.548∗∗
2019 14.007 662.457 0.079 -158.144 2.615 170.965 -20.837 -291.757∗∗∗ -0.975 95.875∗∗∗ -2.846 -105.404∗∗∗
2020 -27.125 563.053 -0.103 -180.096 -22.883 129.011 70.447∗∗∗ -290.051∗∗∗ -0.867 171.837∗∗∗ 66.885∗∗∗ 0.106

This table displays averages of intraday order flow distribution statistics for SPX ATM call (Panel
A) and put options (Panel B). We divide each trading day into seventy-eight equal intervals, each
covering five minutes, and we compute the order flow (buy minus sell orders) within each interval.
Panels A1 and B1 display the daily mean, standard deviation (Std ), skewness, first quartile (Q25 ),
median (Q50 ), and third quartile (Q75 ) of the five-minute order flow distribution. Panels A2 and
B2 display differences in these average statistics (mean, std, skewness, and quantiles) between high
and low liquidity days, classified annually into low liquidity (top 10%) and high liquidity (bottom
10%) days based on ES values. Significance levels are denoted by *, **, and ***, representing the
10%, 5%, and 1% levels, respectively.

45
 Table 2: Descriptive Statistics of ES, log(SD), Volume and Order Imbalance.
SPX Options
Panel A: Calls
ES log(volume)
DTM 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48

Mean 0.087 0.062 0.035 0.028 0.025 0.023 0.021 0.020 9.067 8.623 8.771 8.404 8.206 8.364 7.860 7.620
Std 0.067 0.082 0.031 0.024 0.021 0.020 0.019 0.019 1.144 1.589 1.026 1.255 1.376 1.579 1.921 1.976
Skewness 3.184 14.963 1.894 1.569 1.324 1.635 1.476 1.744 -0.717 -1.175 -0.677 -0.327 -0.146 -0.151 -0.064 -0.135
Kurtosis 26.060 459.054 6.569 3.022 1.676 5.170 2.907 5.931 0.821 1.028 1.614 0.305 -0.351 -0.542 -1.077 -1.093
ρ 0.285 0.257 0.601 0.633 0.671 0.609 0.685 0.651 0.211 0.561 0.424 0.535 0.558 0.585 0.643 0.604
N 1040 2992 2901 2799 2737 2598 2364 2167 1038 2990 2899 2797 2735 2596 2362 2165

log(SD) |OI|
DTM 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48

Mean 4.786 4.895 5.183 5.047 4.996 5.219 4.863 4.703 1.521 1.870 2.239 2.237 2.137 2.929 2.883 2.372
Std 0.855 1.210 0.999 1.143 1.198 1.292 1.630 1.708 2.013 2.947 3.644 3.817 3.798 5.183 5.372 4.352
Skewness -0.425 -0.599 -0.175 -0.189 -0.182 -0.335 -0.224 -0.316 3.016 3.969 5.246 4.860 4.221 3.655 3.429 3.476
Kurtosis 0.508 0.470 0.385 -0.108 -0.319 -0.121 -0.717 -0.743 13.679 24.015 46.830 41.300 24.836 17.151 15.824 16.331
ρ 0.064 0.408 0.315 0.414 0.357 0.414 0.501 0.470 0.026 0.166 0.181 0.231 0.200 0.258 0.272 0.250
N 1038 2990 2899 2797 2735 2596 2362 2165 1038 2990 2899 2797 2735 2596 2362 2165
Panel B: Puts
ES log(volume)
DTM 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48

Mean 0.095 0.063 0.036 0.029 0.025 0.023 0.021 0.019 9.227 8.687 8.782 8.407 8.185 8.269 7.883 7.694
Std 0.060 0.061 0.029 0.024 0.021 0.019 0.019 0.020 1.121 1.539 1.078 1.283 1.398 1.574 1.861 1.920
Skewness 1.970 2.666 1.472 1.577 1.482 1.380 1.491 1.445 -0.818 -1.105 -0.639 -0.202 0.057 0.019 -0.043 -0.042
Kurtosis 8.147 11.526 2.595 4.584 2.982 2.180 3.571 1.880 1.028 1.000 1.622 0.167 -0.447 -0.565 -0.944 -0.976
ρ 0.348 0.439 0.625 0.657 0.657 0.693 0.683 0.712 0.321 0.565 0.415 0.546 0.612 0.656 0.668 0.634
N 1040 2962 2905 2793 2740 2620 2376 2218 1038 2960 2903 2791 2738 2618 2374 2216

log(SD) |OI|
DTM 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48

Mean 4.880 4.941 5.186 5.063 4.965 5.109 4.819 4.665 1.776 1.906 2.224 2.196 2.201 2.772 2.472 2.426
Std 0.843 1.165 1.003 1.124 1.210 1.299 1.553 1.675 2.172 3.152 3.842 4.666 4.589 6.001 5.200 5.795
Skewness -0.728 -0.605 -0.199 -0.123 -0.117 -0.152 -0.308 -0.192 2.336 5.770 5.949 7.604 5.903 5.878 5.352 6.209
Kurtosis 1.228 0.604 0.759 0.129 -0.266 -0.075 -0.577 -0.725 7.352 67.028 53.343 85.581 52.169 50.961 42.339 56.230
ρ 0.118 0.374 0.329 0.422 0.448 0.500 0.554 0.542 0.157 0.162 0.209 0.339 0.243 0.334 0.373 0.257
N 1038 2960 2903 2791 2738 2618 2374 2216 1038 2960 2903 2791 2738 2618 2374 2216

The table reports the time-series mean, standard deviation, skewness, excess kurtosis, AR(1) coef-
ficient (ρ), and total number of observations (N) of the effective spread (ES), logarithm of daily
volume (log(volume)), logarithm of volatility of order-flow (log(SD)), and absolute value of daily
order flow (|OI|) across option maturity buckets. Panel A presents the results for SPX call options
while Panel B presents the results for SPX put options. Absolute value of daily order-flow is divided
by 1000.

46
 Table 3: Time-series Regressions of ESt on log(SDt ) for SPX Options
Panel A: Calls
A1: Subsample Regressions A2: Pooled Regression A3: Standardized Coeff
DTM 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48 with Maturity Dummies DTM

log(SDt ) 0.020∗∗∗ 0.020∗∗∗ 0.007∗∗∗ 0.003∗∗∗ 0.003∗∗∗ 0.004∗∗∗ 0.001∗∗∗ 0.001 log(SDt ) 0.025∗∗∗ 0 0.257
(4.098) (7.302) (7.018) (3.807) (5.484) (6.694) (3.084) (1.336) log(SDt ) D1−6 -0.018∗∗∗ 1-6 0.306
log(volumet ) 0.001 -0.013∗∗∗ 0.002∗∗ 0.003∗∗∗ 0.002∗∗∗ -0.001 0.002∗∗∗ 0.002∗∗∗ log(SDt ) D7−13 -0.016∗∗∗ 7-13 0.216
(0.251) (-4.894) (2.290) (4.954) (4.125) (-0.442) (4.395) (5.644) log(SDt ) D14−20 -0.018∗∗∗ 14-20 0.134
|OIt | -0.016∗∗ 0.002 0.001 0.001 0.002 0.001∗ 0.002∗∗∗ 0.004∗∗∗ log(SDt ) D21−27 -0.020∗∗∗ 21-27 0.165
(-2.067) (0.511) (0.812) (1.144) (1.465) (1.735) (3.490) (4.908) log(SDt ) D28−34 -0.021∗∗∗ 28-34 0.283
RM,t -1.920∗∗∗ -0.007 -0.030 -0.013 -0.002 -0.084 0.022 -0.031 log(SDt ) D35−41 -0.021∗∗∗ 35-41 0.111
(-3.753) (-0.039) (-0.301) (-0.285) (-0.055) (-1.148) (0.749) (-0.946) log(SDt ) D42−48 -0.020∗∗∗ 42-48 0.045
VIXt -0.028 0.039 -0.011 -0.005 -0.001 0.021∗∗ 0.046∗∗∗ 0.026∗∗
(-0.911) (1.621) (-0.858) (-0.695) (-0.051) (1.994) (4.224) (2.157) Maturity Dummies Yes
Time Controls Yes Yes Yes Yes Yes Yes Yes Yes Time Controls Yes
Other Controls Yes Yes Yes Yes Yes Yes Yes Yes Other Controls Yes
N 1035 2990 2899 2797 2735 2596 2362 2165 N 19579
Adj. R2 0.484 0.304 0.640 0.664 0.681 0.629 0.703 0.675 Adj. R2 0.451

Panel B: Puts
B1: Subsample Regressions B2: Pooled Regression B3: Standardized Coeff

47
DTM 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48 with Maturity Dummies DTM

log(SDt ) 0.023∗∗∗ 0.019∗∗∗ 0.008∗∗∗ 0.004∗∗∗ 0.003∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗ log(SDt ) 0.021∗∗∗ 0 0.327
(6.192) (7.004) (7.862) (5.164) (4.339) (3.863) (3.638) (4.292) log(SDt ) D1−6 -0.012∗∗∗ 1-6 0.400
log(volumet ) -0.010∗∗∗ -0.012∗∗∗ 0.001 0.003∗∗∗ 0.003∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.001∗∗∗ log(SDt ) D7−13 -0.011∗∗∗ 7-13 0.283
(-2.588) (-4.686) (0.466) (3.926) (4.647) (4.970) (3.909) (3.514) log(SDt ) D14−20 -0.013∗∗∗ 14-20 0.189
|OIt | -0.004 0.008∗∗ 0.002 0.000 0.004∗∗ 0.001 -0.000 0.001 log(SDt ) D21−27 -0.014∗∗∗ 21-27 0.150
(-0.532) (2.540) (0.980) (0.258) (2.360) (0.925) (-0.241) (1.550) log(SDt ) D28−34 -0.015∗∗∗ 28-34 0.137
RM,t 1.439∗∗∗ -0.070 0.083 0.015 -0.034 0.058 0.120 -0.029 log(SDt ) D35−41 -0.015∗∗∗ 35-41 0.146
(3.670) (-0.615) (1.077) (0.400) (-1.035) (0.773) (1.327) (-0.717) log(SDt ) D42−48 -0.015∗∗∗ 42-48 0.177
VIXt -0.054 -0.015 -0.015 -0.010 0.014 0.015 0.032∗∗ 0.017∗
(-1.641) (-0.740) (-1.237) (-1.478) (1.443) (1.492) (2.551) (1.781) Maturity Dummies Yes
Time Controls Yes Yes Yes Yes Yes Yes Yes Yes Time Controls Yes
Other Controls Yes Yes Yes Yes Yes Yes Yes Yes Other Controls Yes
N 1035 2960 2903 2791 2738 2618 2374 2216 N 19635
Adj. R2 0.513 0.528 0.633 0.666 0.674 0.685 0.685 0.727 Adj. R2 0.611

Panels A1 and B1 present the time series regressions of ESt on log(SDt ) for SPX ATM call (A1) and put options (B1), performed separately for different
maturity buckets. ESt is the effective spread on day t, log(SDt ) is the logarithm of the standard deviation of the intraday order flow distribution on day t,
log(volumet ) is the logarithm of the daily options volume, and |OIt | is the absolute value of the daily order imbalance divided by 10,000. RM,t is daily return
on SPX on day t and V IXt is the level of VIX divided by 100 on day t. Panels A2 and B2 present the results of pooled regressions of ESt on log(SDt ) and the
interaction of log(SDt ) with dummies that identify the different maturity buckets. Time controls include day-of-the-week, month-of-year, and year dummies.
Other controls contain one-day and two-day lags of ESt , absolute value of the average delta, vega and gamma of the options on day t. Standard errors are
computed using Newey-West with the optimal lag suggested by Andrews and Monahan (1992), and corresponding t-statistics are presented in parentheses. *,
**, and *** denote significance at the 10%, 5%, and 1% level. Panels A3 and B3 present the coefficients of log(SD) of Panels A1 and B1 in standardized units.
 Table 4: Time-Series Regressions of ESh,t by Intraday Hour
Panel A: SPX Calls Options
Maturity 0-6 7-34 35-48
log(SDh,t) log(volumeh,t) |OIh,t| log(SDh,t) log(volumeh,t) |OIh,t| log(SDh,t) log(volumeh,t) |OIh,t|
Interval
9:30-10:00 0.011∗∗∗ -0.005∗ 0.059∗∗∗ 0.006∗∗∗ 0.001 0.003 0.002∗∗∗ 0.001 0.011∗∗
(4.669) (-1.934) (3.408) (6.654) (0.368) (0.527) (3.441) (1.446) (2.383)
10:00-11:00 0.012∗∗∗ -0.003 0.026∗∗ 0.003∗∗∗ 0.003∗∗∗ 0.003 0.002∗∗∗ 0.002∗∗∗ 0.004∗∗∗
(4.493) (-1.084) (2.065) (3.418) (3.247) (1.233) (2.937) (3.591) (2.722)
11:00-12:00 0.015∗∗∗ -0.005∗∗∗ 0.002 0.003∗∗∗ 0.003∗∗∗ -0.002 0.001∗ 0.002∗∗∗ 0.006∗∗∗
(7.841) (-2.687) (0.210) (3.394) (3.366) (-1.197) (1.883) (4.191) (4.151)
12:00-13:00 0.020∗∗∗ -0.012 0.005 0.002∗∗ 0.003∗∗∗ 0.003 0.001 0.002∗∗∗ 0.004∗
(3.108) (-1.516) (0.296) (2.430) (3.651) (1.067) (1.478) (4.675) (1.668)
13:00-14:00 0.008∗∗∗ -0.003 0.044∗∗ 0.002∗∗ 0.003∗∗∗ 0.001 0.001∗∗∗ 0.001∗∗∗ 0.004
(4.447) (-1.229) (2.454) (2.218) (4.123) (0.176) (2.822) (2.935) (1.555)
14:00-15:00 0.008∗∗∗ -0.002 0.057∗∗∗ 0.003∗∗∗ 0.003∗∗∗ -0.001 0.001∗ 0.002∗∗∗ 0.002
(3.583) (-0.933) (2.620) (3.061) (2.970) (-0.134) (1.681) (4.300) (1.003)
15:00-16:00 0.005∗∗ 0.006∗∗ -0.001 0.005∗∗∗ 0.003∗∗∗ -0.007∗∗∗ 0.001∗∗ 0.002∗∗∗ 0.003
(2.029) (2.490) (-0.080) (4.526) (3.245) (-2.826) (2.573) (4.465) (1.151)

Panel B: SPX Puts Options

Maturity 0-6 7-34 35-48
log(SDh,t) log(volumeh,t) |OIh,t| log(SDh,t) log(volumeh,t) |OIh,t| log(SDh,t) log(volumeh,t) |OIh,t|
Interval
9:30-10:00 0.013∗∗∗ -0.006∗∗∗ 0.066∗∗∗ 0.006∗∗∗ -0.001 0.005 0.004∗∗∗ -0.001 0.003
(7.913) (-3.514) (4.916) (7.294) (-0.935) (1.404) (3.525) (-0.442) (1.213)
10:00-11:00 0.014∗∗∗ -0.004∗∗ 0.050∗∗∗ 0.004∗∗∗ 0.003∗∗∗ 0.003 0.003∗∗∗ 0.001∗∗ 0.005∗
(6.202) (-2.108) (4.130) (4.372) (2.881) (1.150) (4.435) (2.259) (1.712)
11:00-12:00 0.009∗∗∗ -0.002 0.050∗∗∗ 0.005∗∗∗ 0.001 0.001 0.001∗∗∗ 0.002∗∗∗ -0.000
(4.334) (-0.883) (3.800) (5.003) (0.844) (0.591) (2.932) (5.296) (-0.007)
12:00-13:00 0.007∗∗∗ -0.000 0.011 0.004∗∗∗ 0.002∗∗ 0.001 0.001∗∗∗ 0.002∗∗∗ 0.003
(3.569) (-0.113) (1.144) (4.859) (1.998) (0.776) (3.195) (3.659) (1.015)
13:00-14:00 0.008∗∗∗ 0.002 0.006 0.003∗∗∗ 0.002∗∗∗ -0.000 0.002∗∗∗ 0.001∗∗ 0.001
(3.906) (1.078) (0.601) (3.405) (2.802) (-0.057) (4.800) (2.487) (0.230)
14:00-15:00 0.011∗∗∗ -0.001 0.027∗ 0.002∗∗ 0.0037∗∗∗ 0.001 0.001∗∗∗ 0.002∗∗∗ 0.004
(4.361) (-0.491) (1.836) (2.241) (4.101) (0.346) (2.676) (4.350) (1.187)
15:00-16:00 0.006∗ 0.004 0.032∗ 0.002∗∗ 0.003∗∗∗ -0.001 0.002∗∗∗ 0.002∗∗∗ 0.007∗∗
(1.771) (1.416) (1.809) (2.140) (3.437) (-0.420) (3.230) (3.557) (1.972)

The table reports daily time-series regressions of ESh,t on log(SDt,h ), log(volumeh,t ), and |OIh,t |, where
all variables are computed for specific intraday hour indicated in the first column. The dependent variable,
ESh,t , is the effective spread calculated using only trades executed during hour h. The key independent
variable, log(SDh,t ), measures the volatility of order flow using 5-minute order imbalances within hour h.
The controls include log(volumeh,t ), the logarithm of trading volume during that hour, and |OIh,t |, the
absolute value of the order imbalance, scaled by 10,000. All regressions control for day-of-the-week, month-
of-year, and year fixed effects, as well as the same set of additional controls used in Table 3, computed on
a daily basis over hour h. Standard errors are calculated using the Newey-West estimator with the optimal
lag length selected according to Andrews and Monahan (1992). t-statistics are reported in parentheses.
Statistical significance at the 10%, 5%, and 1% levels is denoted by *, **, and ***, respectively.

48
 Table 5(a): Regressions of ESt on Liquidity Providers Delta-Hedging Variables for Call Options
Panel A: Gamma Inventory
Maturity 0 1-6 7-13 14-20

log(SD)t 0.026∗∗∗ 0.019∗∗∗ 0.010∗∗∗ 0.018∗∗∗ 0.010∗∗∗ 0.006∗∗∗ 0.007∗∗∗ 0.003∗∗∗

(12.390) (3.880) (8.805) (6.412) (20.306) (6.524) (24.020) (3.655)
GammaInvt 0.522∗∗ -0.036 0.040 -0.279 -0.345∗∗ -0.207 0.086 0.057 0.078 -0.032 -0.025 -0.007
(2.240) (-0.157) (0.187) (-1.557) (-2.020) (-1.324) (1.424) (1.023) (1.419) (-0.588) (-0.527) (-0.162)

Time Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Other Controls No No Yes No No Yes No No Yes No No Yes

N 1035 1035 1035 2990 2990 2990 2899 2899 2899 2797 2797 2797
Adj. R2 0.256 0.369 0.472 0.233 0.253 0.283 0.487 0.589 0.629 0.533 0.631 0.660

Panel B: Inventory Change

Maturity 0 1-6 7-13 14-20

49
log(SD)t 0.027∗∗∗ 0.019∗∗∗ 0.010∗∗∗ 0.018∗∗∗ 0.010∗∗∗ 0.006∗∗∗ 0.007∗∗∗ 0.003∗∗∗
(12.528) (3.865) (7.787) (6.347) (20.318) (6.524) (23.281) (3.671)
DeltaOIt -0.009 -0.033 0.009 0.123∗∗∗ 0.050∗∗ 0.049∗∗∗ 0.087∗∗∗ 0.028∗ 0.013 0.035∗∗∗ 0.001 -0.004
(-0.263) (-1.123) (0.317) (6.026) (2.450) (2.578) (5.160) (1.647) (0.934) (3.955) (0.070) (-0.562)

Time Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Other Controls No No Yes No No Yes No No Yes No No Yes

N 1035 1035 1035 2990 2990 2990 2899 2899 2899 2797 2797 2797
Adj. R2 0.254 0.369 0.472 0.236 0.253 0.283 0.499 0.590 0.629 0.536 0.631 0.660

The table reports time-series regressions of ESt on the liquidity providers’ inventory variables GammaInvt (Panel A) and DeltaOIt (Panel B)
for at-the-money SPX call options. Regressions are estimated separately across maturity buckets. The dependent variable, ESt , is the daily
effective spread, while GammaInvt and DeltaOIt are the two proxies introduced in Section 4.4 to capture the delta-hedging costs incurred by
liquidity providers on day t. Specifically, GammaInvt reflects the cost of rebalancing the delta-hedge of pre-existing positions, and DeltaOIt
captures the cost associated with hedging new positions absorbed during the day. The remaining control variables are analogous to those used
in the baseline regressions in Table 3. Standard errors are computed using the Newey-West estimator with the optimal lag length selected
according to Andrews and Monahan (1992). t-statistics are reported in parentheses. *, **, and *** indicate significance at the 10%, 5%, and
1% levels, respectively.
 Table 5(b): Regressions of ESt on Liquidity Providers Delta-Hedging Variables for Put Options
Panel A: Gamma Inventory
Maturity 0 1-6 7-13 14-20

log(SD)t 0.021∗∗∗ 0.024∗∗∗ 0.010∗∗∗ 0.018∗∗∗ 0.009∗∗∗ 0.008∗∗∗ 0.007∗∗∗ 0.004∗∗∗

(12.755) (6.354) (9.869) (6.608) (18.420) (7.781) (19.296) (5.404)
GammaInvt 0.525∗ 0.368 0.197 0.010 -0.033 -0.029 0.125∗ 0.131∗ 0.104 0.015 0.035 0.011
(1.702) (1.196) (0.598) (0.065) (-0.219) (-0.193) (1.765) (1.745) (1.332) (0.236) (0.625) (0.203)

Time Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Other Controls No No Yes No No Yes No No Yes No No Yes

N 1035 1035 1035 2960 2960 2960 2903 2903 2903 2791 2791 2791
Adj. R2 0.357 0.435 0.511 0.433 0.472 0.499 0.511 0.605 0.627 0.535 0.634 0.663

Panel B: Inventory Change

Maturity 0 1-6 7-13 14-20

50
log(SD)t 0.021∗∗∗ 0.024∗∗∗ 0.010∗∗∗ 0.018∗∗∗ 0.009∗∗∗ 0.008∗∗∗ 0.008∗∗∗ 0.004∗∗∗
(12.779) (6.374) (9.118) (6.588) (17.720) (7.786) (18.940) (5.497)
DeltaOIt -0.013 -0.044 -0.052∗ 0.115∗∗∗ 0.040∗∗ 0.054∗∗∗ 0.062∗∗∗ 0.003 0.001 0.030∗∗∗ -0.013∗ -0.015∗∗
(-0.428) (-1.446) (-1.931) (6.618) (2.338) (3.322) (4.966) (0.279) (0.149) (3.497) (-1.762) (-2.195)

Time Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Other Controls No No Yes No No Yes No No Yes No No Yes

N 1035 1035 1035 2960 2960 2960 2903 2903 2903 2791 2791 2791
Adj. R2 0.355 0.435 0.512 0.440 0.473 0.500 0.518 0.604 0.626 0.538 0.634 0.663

The table reports time-series regressions of ESt on the liquidity providers’ inventory variables GammaInvt (Panel A) and DeltaOIt (Panel B)
for at-the-money SPX put options. Regressions are estimated separately across maturity buckets. The dependent variable, ESt , is the daily
effective spread, while GammaInvt and DeltaOIt are the two proxies introduced in Section 4.4 to capture the delta-hedging costs incurred by
liquidity providers on day t. Specifically, GammaInvt reflects the cost of rebalancing the delta-hedge of pre-existing positions, and DeltaOIt
captures the cost associated with hedging new positions absorbed during the day. The remaining control variables are analogous to those used
in the baseline regressions in Table 3. Standard errors are computed using the Newey-West estimator with the optimal lag length selected
according to Andrews and Monahan (1992). t-statistics are reported in parentheses. *, **, and *** indicate significance at the 10%, 5%, and
1% levels, respectively.
 Table 6: Panel Regressions of ESs,t on log(SDs,t ) for Individual Stock Options

Panel A: Calls Panel B: Puts

Maturity 0-24 25-48 0-24 25-48

log(SDs,t ) 0.015∗∗∗ 0.007∗∗∗ 0.014∗∗∗ 0.006∗∗∗

(29.842) (28.965) (24.001) (23.930)
log(volumes,t ) -0.015∗∗∗ -0.007∗∗∗ -0.013∗∗∗ -0.006∗∗∗
(-24.181) (-26.652) (-18.412) (-21.834)
|OIs,t | -0.001 0.003∗∗ -0.001 0.002
(-0.67) (2.014) (-0.349) (1.613)
Returns,t -0.129∗∗∗ -0.006 0.095∗∗∗ 0.021∗∗
(-4.883) (-0.460) (4.548) (2.137)
IVs,t 0.036∗∗∗ 0.010∗∗∗ 0.023∗∗∗ -0.001
(9.856) (4.405) (8.478) (-0.674)

Stock FE Yes Yes Yes Yes

Time Controls Yes Yes Yes Yes
Other Controls Yes Yes Yes Yes

Adj. R2 0.237 0.281 0.190 0.232

This table presents the results of panel regressions of ESs,t on log(SDs,t ) for ATM call options (Panel
A) and put options (Panel B) written on the stocks that are the constituents of the S&P500. The
results are presented for two maturity buckets: 0-24 days to maturity and 25-48 days to maturity.
ESs,t is the daily effective spread on day t for options on stock s. log(SDs,t ) is the logarithm
of the standard deviation of the intraday order flow distribution on day t for options on stock s,
log(volumes,t ) is the logarithm of the daily options volume, and |OIs,t | is the absolute value of the
daily order imbalance (scaled by 10,000). Returns,t is the return of underlying stock on day t, and
IVs,t is the average implied volatility of the options on stock s on day t. Other controls include firm
size, stock volume, one-day and two-day lags of ESs,t , and absolute values of the average delta,
vega and gamma of the options on day t. Time controls include day-of-the-week, month-of-year,
and year dummies. Standard errors are clustered at the day and stock level. The corresponding
t-statistics are presented in parentheses. *, **, and *** denote significance at the 10%, 5%, and
1% level.

51
 Table 7: Exchange-Level Analysis
Panel A: Trade Absoption and Liquidity
Call Options Put Options
Regression of Regression of Regression of Regression of
Spreadi,j,τ ∆Spreadi,j,τ +1 Spreadi,j,τ ∆Spreadi,j,τ +1
α Di,j,τ α Di,j,τ α Di,j,τ α Di,j,τ

AXP 17.10∗∗∗ -5.77∗∗∗ -0.12∗∗∗ 0.28∗∗∗ 17.44∗∗∗ -5.91∗∗∗ -0.15∗∗∗ 0.37∗∗∗

BA 22.05∗∗∗ -10.06∗∗∗ -0.03 0.83∗ 22.65∗∗∗ -10.05∗∗∗ -0.03 0.91∗∗
CAT 13.17∗∗∗ -3.83∗∗ -0.07∗∗ 0.22∗∗∗ 13.19∗∗∗ -3.90∗∗ -0.08∗∗∗ 0.24∗∗∗
DIS 19.00∗∗∗ -9.19∗∗∗ -0.01 0.52∗ 22.20∗∗∗ -11.50∗∗∗ -0.03 0.67∗∗
DOW 16.46∗∗∗ -4.20∗∗∗ -0.20∗∗∗ 0.25∗∗ 16.31∗∗∗ -4.67∗∗∗ -0.32∗∗∗ 0.27∗∗∗
HD 17.71∗∗∗ -6.85∗∗∗ -0.07 0.44∗∗ 17.94∗∗∗ -6.65∗∗∗ -0.09∗∗ 0.44∗∗∗
IBM 14.88∗∗∗ -4.33∗∗∗ -0.04 0.25∗∗∗ 15.02∗∗∗ -4.41∗∗∗ -0.05∗ 0.25∗∗∗
INTC 11.24∗∗∗ -3.39∗∗∗ -0.06∗∗∗ 0.29∗∗∗ 11.06∗∗∗ -3.27∗∗∗ -0.07∗∗∗ 0.30∗∗∗
JNJ 18.45∗∗∗ -6.43∗∗∗ -0.10∗ 0.46∗∗∗ 19.42∗∗∗ -6.51∗∗∗ -0.13∗∗∗ 0.47∗∗∗
JPM 10.19∗∗∗ -3.40∗∗∗ -0.05∗∗ 0.17∗∗∗ 10.48∗∗∗ -3.29∗∗∗ -0.05∗∗∗ 0.17∗∗∗
KO 16.03∗∗∗ -4.95∗∗∗ -0.12∗∗ 0.32∗ 15.97∗∗∗ -4.97∗∗∗ -0.16∗∗∗ 0.34∗∗∗
MMM 18.50∗∗∗ -5.71∗∗∗ -0.20∗∗∗ 0.41∗∗ 18.09∗∗∗ -5.46∗∗∗ -0.25∗∗∗ 0.40∗∗∗
MRK 21.05∗∗∗ -5.72∗∗∗ -0.09∗∗ 0.60∗∗∗ 21.28∗∗∗ -5.88∗∗∗ -0.10∗∗ 0.62∗∗∗
MSFT 18.04∗∗∗ -7.91∗∗ -0.01 0.37∗∗∗ 17.51∗∗∗ -7.50∗∗ 0.00 0.40∗∗∗
PG 16.88∗∗∗ -5.04∗∗∗ -0.10∗∗∗ 0.33∗∗∗ 16.18∗∗∗ -4.99∗∗∗ -0.15∗∗∗ 0.32∗∗∗
WMT 16.18∗∗∗ -6.44∗∗∗ -0.06 0.35∗ 15.81∗∗∗ -5.60∗∗∗ -0.09∗∗∗ 0.32∗∗∗

Panel B: Daily Regressions of Exchange-Specific ESi,s,t

Call Options Put Options
All Six Major All Six Major
Exchanges Exchanges Exchanges Exchanges

log(SDi,s,t ) 0.0079∗∗∗ 0.0038∗∗∗ 0.0081∗∗∗ 0.0036∗∗∗ 0.0072∗∗∗ 0.0041∗∗∗ 0.0076∗∗∗ 0.0044∗∗∗

(9.18) (8.73) (10.21) (7.70) (11.06) (8.44) (10.99) (10.15)
log(Volumei,s,t ) -0.0075∗∗∗ -0.0040∗∗∗ -0.0070∗∗∗ -0.0034∗∗∗ -0.0067∗∗∗ -0.0043∗∗∗ -0.0062∗∗∗ -0.0042∗∗∗
(-7.24) (-7.18) (-7.61) (-5.40) (-7.62) (-7.86) (-7.95) (-7.15)
|OI|i,s,t 0.0033 0.0029∗ 0.0010 0.0013 0.0030 0.0012 -0.002 -0.0002
(1.49) (1.73) (0.03) (1.26) (0.72) (0.66) (-0.61) (-0.20)
Exchange FE Yes No Yes No Yes No Yes No
Day FE Yes No Yes No Yes No Yes No
Stock x Day FE No Yes No Yes No Yes No Yes
Adj. R2 0.22 0.55 0.24 0.63 0.15 0.45 0.20 0.59

Panel A reports the results of panel regressions of the quoted spread (Spreadi,j,τ ) and its change
following each trade (∆Spreadi,j,τ +1 ), estimated across exchanges. The regressions include a con-
stant (coefficient α) and a dummy variable Di,j,τ , which equals one if the trade of option j at time
τ was executed on exchange i. Regressions are run separately by ticker and by option type (calls
and puts), and include day fixed effects. Standard errors are clustered at the day and exchange
level. All coefficients are multiplied by 100. Panel B presents results from panel regressions of
the exchange-specific effective spread, ESi,s,t , on log(SDs,i,t ), the logarithm of option order flow
volatility based on trades executed on exchange i. The regressions also include log(V olumei,s,t ),
the logarithm of daily option volume for stock s on exchange i, and |OIs,i,t |, the absolute value of
daily order imbalance (scaled by 10,000) absorbed by exchange i. Standard errors are clustered at
the day, stock, and exchange level.

52
 Table 8: Log(SD) on Information and Inventory Days

Maturity 0-24 25-48

Calls Puts Calls Puts

EAs,t -0.229∗∗∗ -0.207∗∗∗ -0.189∗∗∗ -0.173∗∗∗

(-20.391) (-16.954) (-20.217) (-16.163)
EAs,t−1 -0.141∗∗∗ -0.110∗∗∗ -0.099∗∗∗ -0.046∗∗∗
(-13.918) (-9.480) (-12.248) (-5.084)

Third Friday 0.007 0.022∗∗∗ -0.054∗∗∗ -0.052∗∗∗

(1.067) (3.305) (-7.582) (-7.221)
BoM -0.025∗∗∗ -0.017∗∗∗ 0.063∗∗∗ 0.047∗∗∗
(-4.972) (-3.338) (7.056) (4.893)
EoM -0.010∗ -0.012∗∗ 0.131∗∗∗ 0.109∗∗∗
(-1.822) (-2.202) (9.845) (8.861)

RVs,t -8.818∗∗∗ -9.619∗∗∗ -6.974∗∗∗ -8.998∗∗∗

(-3.091) (-3.050) (-4.296) (-4.509)

Stock FE Yes Yes Yes Yes

Time Controls Yes Yes Yes Yes
Additional Controls Yes Yes Yes Yes
Adj. R2 0.749 0.753 0.776 0.787

This table presents the results of panel regressions of log(SDs,t ) on a set of event-day dummies
for at-the-money (ATM) call and put options written on S&P 500 constituent stocks. Results
are reported separately for two maturity buckets: 0–24 days and 25–48 days to maturity. The
dependent variable, log(SDs,t ), is the logarithm of the standard deviation of the intraday order
flow distribution for options on stock s on day t. The regression includes several event indicators:
EAs,t (EAs,t−1 ) equals one on the earnings announcement day (or the day before) for stock s; Third
Friday equals one on the third Friday of each month; BoM and EoM are dummies for the first
and last trading days of each month, respectively. RVs,t denotes the realized volatility of stock s
on day t, computed as the sum of squared five-minute intraday returns. Control variables include
day-of-the-week, month-of-year, and year fixed effects, as well as log(V olumes,t ), the logarithm of
the daily option volume, and |OIs,t |, the absolute value of the daily option order imbalance for stock
s. Standard errors are clustered at the day and stock level. t-statistics are shown in parentheses.
*, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively.

53
 Table 9: Robustness: Dollar Spread and Realized Option Volatility for SPX Options
Panel A: Dollar Spreadt and log(SD)t
SPX Calls
Maturity 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48
log(SD)t 0.035∗∗∗ 0.046∗∗∗ 0.020∗∗∗ 0.009∗∗ 0.016∗∗∗ 0.025∗∗∗ 0.006 -0.005
(8.706) (7.508) (4.741) (2.168) (3.994) (5.590) (1.500) (-1.122)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
N 1035 2990 2899 2797 2735 2596 2362 2165
Adj. R2 0.550 0.688 0.743 0.767 0.770 0.766 0.792 0.760
SPX Puts
Maturity 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48
log(SD)t 0.043∗∗∗ 0.066∗∗∗ 0.037 ∗∗∗ 0.024 ∗∗∗ 0.016∗∗∗ 0.023∗∗∗ 0.016∗∗∗ 0.020∗∗∗
(5.720) (7.277) (4.815) (4.350) (2.956) (3.268) (3.334) (3.072)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
N 1035 2960 2903 2791 2738 2618 2374 2216
Adj. R2 0.517 0.652 0.713 0.749 0.746 0.742 0.765 0.760

Panel B: ESt and ORV

SPX Calls
Maturity 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48
log(SD)t 0.027∗∗∗ 0.022∗∗∗ 0.007∗∗∗ 0.003∗∗∗ 0.003∗∗∗ 0.004∗∗∗ 0.001∗∗∗ 0.001
(4.868) (8.187) (6.926) (3.815) (5.557) (6.563) (3.314) (1.623)
ORVt 0.010∗∗∗ 0.007∗∗∗ 0.053∗∗∗ 0.046∗∗∗ 0.029∗∗∗ 0.019∗∗∗ 0.012∗∗∗ 0.010∗∗∗ 0.012∗∗∗ 0.010∗∗∗ 0.016∗∗∗ 0.009∗∗∗ 0.016∗∗∗ 0.011∗∗∗ 0.027∗∗∗ 0.018∗∗∗

54
(13.756) (5.520) (6.050) (5.043) (3.621) (2.845) (5.401) (4.580) (4.446) (4.889) (5.004) (3.300) (4.506) (3.629) (4.927) (4.115)
Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 1035 1035 2990 2990 2899 2899 2797 2797 2735 2735 2596 2596 2362 2362 2165 2165
Adj. R2 0.374 0.514 0.328 0.356 0.510 0.647 0.540 0.666 0.568 0.684 0.533 0.631 0.604 0.705 0.590 0.680
SPX Puts
Maturity 0 1-6 7-13 14-20 21-27 28-34 35-41 42-48
log(SD)t 0.026∗∗∗ 0.020∗∗∗ 0.008∗∗∗ 0.004∗∗∗ 0.003∗∗∗ 0.002∗∗∗ 0.002∗∗∗ 0.002∗∗∗
(5.949) (7.585) (8.002) (5.328) (4.407) (3.849) (3.839) (4.515)
ORVt 0.004∗∗∗ 0.003∗∗∗ 0.038∗∗∗ 0.034∗∗∗ 0.011∗∗∗ 0.017∗∗∗ 0.002 0.007∗∗∗ 0.004∗ 0.007∗∗∗ 0.009∗∗∗ 0.011∗∗∗ 0.011∗∗∗ 0.010∗∗∗ 0.012∗∗∗ 0.007∗∗∗
(5.161) (2.657) (4.340) (4.100) (2.927) (5.406) (1.216) (5.080) (1.649) (4.747) (4.067) (6.667) (3.334) (4.257) (3.948) (3.165)
Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 1035 1035 2960 2960 2903 2903 2791 2791 2738 2738 2618 2618 2374 2374 2216 2216
Adj. R2 0.373 0.518 0.496 0.574 0.515 0.641 0.535 0.668 0.555 0.677 0.595 0.691 0.610 0.687 0.648 0.727

Panel A of the table presents time-series regressions of Dollar Spreadt on the volatility of the order flow log(SD)t for SPX
call and put options. Dollar Spreadt is the daily effective dollar spread. Panel B presents time-series regressions of ESt on
the realized option volatility ORVt , which is calculated as the sum of squared 5-minute average option returns on day t.
The other variables are analogous to those analyzed in the baseline regression in Table 3. Dollar Spreadt is winsorized at
99.5% and 0.5% levels. The coefficients of log(SD)t are multiplied by 1,000 in Panel A. Standard errors are computed using
Newey-West with the optimal lag suggested by Andrews and Monahan (1992), and corresponding t-statistics are presented
in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level.
 Table 10: Robustness: Dollar Spread and Realized Option Volatility for
Individual Stock Options
Panel A: Dollar Spreads,t and log(SD)s,t
Calls Puts
0-24 25-48 0-24 25-48
∗∗∗ ∗∗∗ ∗∗∗
log(SD)s,t 0.114 0.107 0.114 0.110∗∗∗
(28.583) (27.950) (26.861) (22.727)

Stock FE Yes Yes Yes Yes

Time Controls Yes Yes Yes Yes
Other Controls Yes Yes Yes Yes
Adj. R2 0.541 0.564 0.530 0.559

Panel B: ESs,t and ORVs,t

Calls Puts
0-24 25-48 0-24 25-48
∗∗∗ ∗∗∗ ∗∗∗
log(SD)s,t 0.016 0.007 0.015 0.006∗∗∗
(31.704) (29.301) (27.005) (24.448)
ORVs,t 0.018∗∗∗ 0.014∗∗∗ 0.009∗∗∗ 0.006∗∗∗ 0.026∗∗∗ 0.024∗∗∗ 0.011∗∗∗ 0.009∗∗∗
(9.316) (8.932) (5.118) (4.941) (19.724) (19.120) (4.474) (4.525)

Stock FE Yes Yes Yes Yes Yes Yes Yes Yes

Time Controls Yes Yes Yes Yes Yes Yes Yes Yes
Other Controls No Yes No Yes No Yes No Yes
Adj. R2 0.113 0.275 0.087 0.287 0.136 0.258 0.097 0.241

Panel A of the table presents panel regressions of Dollar Spreads,t on the volatility of the order flow log(SD)s,t
for individual stock options. Dollar Spreads,t is the daily effective dollar spread for options on stock s. Panel
B presents panel regressions of ESs,t on the realized option volatility ORVs,t , which is calculated as the sum
of squared 5-minute average option returns for stock s on day t. The other variables are analogous to those
analyzed in the baseline regression in Table 6. The coefficients of log(SD)s,t are multiplied by 1,000. Dollar
Spreads,t is winsorized at 99.5% and 0.5% levels, and standard errors are clustered at the day and stock
level.

55