Gamma, Vanna, and

Higher Greek Exposure:

Compiling the Dealer
Order Book
Garrett DeSimone, Ph.D.
Abhinav Gupta
Joe Falco

August 2021
 Where Volatility Smiles™

Dealer, or market maker, hedging flows have become an extremely hot topic in financial media over
the last two years. An exotic variety of equity market phenomena are attributed to “gamma
imbalances” ranging from the March 2020 coronavirus-induced volatility to various meme stock
manias.

Our research indicates that gamma is only the tip of the iceberg when it comes to explaining market
returns and volatility. A far lesser known greek, vanna, is responsible for a significant portion of
non-fundamental daily market behavior. Vanna is a cross-derivative of volatility with respect to
delta and summarizes hedging exposure to implied volatility changes in the tails. But before we get
too deep into the derivative space, let’s address a superior approach for establishing the dealer
order book.

In the past, dealer models were built using simplifying assumptions around open interest (OI) by
stating that all calls were bought, and all puts were sold, by market makers. These assumptions
provide a nice framework for characterizing dealer hedging, but are suspect at times, as they do not
account for extended periods of out-of-the-money (OTM) put premium harvesting by investors
(such as during the “Volmageddon” blowup of early February 2018).

Unfortunately, dealer originated trades are non-public information. However, we can construct an
aggregated dealer order book using IvyDB Signed Volume. IvyDB Signed Volume uses the quotation
algorithm that identifies buyer- and seller-initiated trades. Market makers are in the business of
providing liquidity and receive the bid/ask spread as compensation. Our estimated order book is
accumulated over time by taking the difference between daily bid trades and ask trades, to form a
running total on each symbol. This represents a cleaner, more accurate picture of dealer
positioning than a simple OI model.

Now that we have created a picture of dealer inventory, we examine their greek exposure.

Gamma - Gamma is the second derivative of the option price with respect to a change in the
underlying security. This is also the rate of change of delta. Long option positions have positive
gamma, and positive gamma is largest for at-the-money (ATM) options with shorter terms to
expiration.

When dealers are positive gamma, this is synchronous with volatility suppression. If dealers are
positioned in long OTM calls (positive gamma), an increase in underlying prices causes a rise in their
deltas, and subsequent selling of the underlying to reel the position back to delta neutral. When
their balance shifts to long OTM puts (negative gamma), a drop in underlying prices accelerates
volatility since this induces selling to maintain hedges. This concept has been popularized in charts
like Figure 1 below:

www.optionmetrics.com
[email protected]
 Where Volatility Smiles™

Figure 1

The readily apparent takeaway is that dealer negative gamma widens the S&P 500 return range,
and positive gamma compresses it. In turn, negative gamma is associated with high volatility
regimes.

A shortcoming of gamma as a risk metric is that it loses its explanatory ability during highly illiquid
markets. This is because the gamma distribution flattens along the strikes by decreasing for ATM
options and increasing for OTM and in-the-money (ITM). During such times, identifying hedging
activity of dealers is not as straightforward.

The “Other” Guys

Gamma receives the most headlines recently, but there are other, lesser known greeks that drive
the tides of volatility. To study the influence of higher order greeks on market variables, we unpack
dollar values using the Black Scholes derived equalities (Carr and Wu 2018). Conveniently, these

www.optionmetrics.com
[email protected]
 Where Volatility Smiles™

exposures are also additive, so we can build a clear picture of dealer position in each one (see
Appendix for calculations). Let’s examine them:

Vega – The first derivative of option price with respect to change in underlying volatility. Long
maturity ATM options have the highest vega.

Vanna - The cross derivative with respect to delta and implied volatility (IV). Vanna is a measure of
how option delta changes with a movement in IV. For OTM calls, vanna is positive indicating that
deltas rise when volatility increases. For OTM puts, vanna is negative signaling a decrease in deltas
for rising volatility. Intuitively, rising volatility increases deltas because it expands the underlying
price distribution, which increases the probability that OTM options land in the money.

Since vanna is additive, the entire dealer book can be summarized into a useful value indicating net
positioning of calls and puts. Dealer positive vanna indicates that the inventory is comprised of a
combination of long calls and short puts, while negative vanna is indicative of short calls and long
puts. Let’s look at how dealer vanna varies throughout time:

Figure 2

www.optionmetrics.com
[email protected]
 Where Volatility Smiles™

The most readily apparent observation is that dealer vanna is negative during periods of low
volatility and exhibits sizable positive increases during the volatility shocks of Volmagaddeon
(February 2018) and the coronavirus sell-off (March 2020). Dealers’ primary flows since 2017 are
long puts and short calls, contrary to the assumption of OI-based gamma models. During periods of
tightened liquidity (volatility shocks), their inventory is dominated by put selling as a result of end-
user demand for hedges.

However, this metric has some ambiguity since it cannot distinguish between an inventory of long
and short positions. The addition of another greek, volga, gives us insight into whether a given
change in vanna can be attributed to buying or selling.

Volga – The second derivative of option price with respect to volatility, or the rate of change of
vega. Since vega is at its maximum for ATM options, volga is zero for these options and positive for
the tails (OTM calls and puts). A positive volga value for a portfolio of options indicates a net long
position. For example, a vanna positive/volga negative inventory indicates that dealers are
primarily short puts.

Quant Evidence
Next, we are going to generate statistical proof for the preceding examples. We run linear
regressions, beginning with daily changes in CBOE Volatility Index (VIX) levels as the dependent
variable. By analyzing contemporaneous changes between VIX and dealer greeks, we establish
which variables provide added explanatory power.

We include several dealer variables in these regressions. $gex_dealer_scaled_chg is the daily

change in dealer gamma exposure where put gamma is scaled by a negative one.
$vega_dealer_chg, $volga_dealer_chg, and $vanna_dealer_chg represent the daily change in
dealer inventory of vega, volga, and vanna, respectively. We utilize control variables for the
previous VIX level (VIX_t-1) and lagged SPX return (SPXlag).

www.optionmetrics.com
[email protected]
 Where Volatility Smiles™

Figure 3

In a baseline univariate regression, dollar gamma exposure (GEX) changes show a significantly
negative relationship with VIX changes with a p-value < .01. This indicates that decreases in dealer
dollar GEX are associated with higher same-day volatility, or reduced liquidity.

We expand these regressions by including calculated dealer exposures as independent variables. In

this updated regression, the only variable to demonstrate strong significance is the vanna change
with a t-stat of 3.9. The higher R^2 value of 0.46 indicates that added inventory variables can explain
approximately four times more of the daily variation than the simple univariate model. This result
also indicates that the explanatory power of vanna dominates any additional predictive power in
gamma exposure.

Intuitively, the coefficient indicates that a positive change in dealer vanna is contemporaneously
related to a rise in volatility. In other words, OTM short puts/long calls added to a dealer’s inventory
result in higher VIX levels at the end of the day. This is likely a result of increased end-user demand
for OTM puts as hedges.

Next, we study the relationship between the dealer variables and Standard & Poor’s 500 Index (SPX)
returns. In a univariate regression, dealer GEX changes show positive significance again, signaling
an increase in dealer gamma on up market days. Changes in dealer GEX explain about 23% of the
daily variation of returns.

www.optionmetrics.com
[email protected]
 Where Volatility Smiles™

Figure 4

In a multivariable regression, the dealer inventory measures show strong significance.

$vega_dealer_chg is significant at a 5% level, with $volga_dealer_chg and $vanna_dealer_chg
showing significance at 1%. The positive sign on $vega_dealer_chg indicates that dealers as buyers
(end users are sellers) of longer dated volatility are concurrent with positive return days. The
negative slope on $volga_dealer_chg signifies short dealer inventory in the wings (long end-users),
which results in down SPX days. This aligns with the concept that option buying by customers pulls
liquidity from the market and is channeled through low SPX returns.

Lastly, the negative coefficient on $vanna_dealer_chg indicates that dealers accumulate short
put/long call inventory on down market days. This is consistent with higher VIX levels during market
selloffs.

Once again, inclusion of the additional dealer terms drastically increases the explanatory power of
the model (R^2 of 66%) and renders the dealer GEX changes insignificant.

The informational power of vanna for understanding the SPX hedging complex cannot be
understated. We have shown this higher order greek provides increased insight into daily volatility
and SPX returns over dollar gamma exposure. Our research highlights the value in modeling this
often-overlooked variable, which is especially pertinent for understanding risk during market
panics.

www.optionmetrics.com
[email protected]
 Where Volatility Smiles™

References
Carr, Peter, and Liuren Wu. "Option profit and loss attribution and pricing: A new framework." The
Journal of Finance 75, no. 4 (2020): 2271-2316.

Lee, Charles MC, and Mark J. Ready. "Inferring trade direction from intraday data." The Journal of
Finance 46, no. 2 (1991): 733-746.

www.optionmetrics.com
[email protected]
 Where Volatility Smiles™

Appendix
These calculations are reproduced from Carr and Wu (2018). The calculations presented below are for the
cash greek position of a single option. Aggregated values are calculated by summing across the total dealer
portfolio daily.

1.
2.
3.
4.

www.optionmetrics.com
[email protected]