Calibration with Market Tunnel Ordering

Bernard Gourion, Head of R&D Quant Watch

October 2, 2024

The rst main idea is of the budget is to adopt a new criterion to measure the calibration performance of a pricing
model for vanilla options. The most common current criterion is based on the absolute spread between mid-prices of
the market and the prices generated by your calibrated model - aka the 'model' prices. However recent studies from our
R&D quantitative team suggest the mid-prices are not relevant as they implies disordered model-free prices for tunnels -
a tunnel is an European option which pays 1 currency unit -. The leads to the fact the mid price may not be considered
as the target price of a model.
The purpose of this GT is to test pricing models by adopting a new criterion directly based on the real true market
data: the bid and ask prices. Then these calibration will be compared to the classic one to establish a comparison of the
two hierarchy of models. For ease of the work, only pure stochastic volatility models will be studied. Local and Local
stochastic volatility will be set aside given the harder challenge they represent, at least for the beginning of this project.
To ease the realization of the project, we will be required to follow the dierent steps below

1 Collect the market data:

Each student is required to choose a equity index underlying and to retrieve on a daily basis the historic for one year of the
market quotes of vanilla options for this underlying. The Equity Indexes your may select are the CAC40, the Eurostoxx50,
the Dax, the SP500, the Nikkei, the Hang Seng indexes. This list is not limited and you may choose the index you want
as soon as market quotes of European vanilla options are available in BB. I let your organize the retrieval of data from
BB and the organization of data in les like you want. Be ecient is the only criterion.

BB and likewise providers also gives values for the forward Ft,T and discount factor B (t, T ) and at each quoted
maturity. Don't forget to retrieve them!

1
2 Pre-processing in market data

You must understand how you market quotes work. Quotes are for European Options for an underlying impacted by
discounting, repo and dividends. To focus on the calibration, the quotes needs be transformed into cleaned undiscounted
normalized market prices.
A model free formulation of a call is
 
+
Ct (K, T ) = B (t, T ) EQ (ST − K)

with B (t, T ) the discounting factor, T the expiry date, ST the terminal value of the spot and Q the risk neutral probability.
From this quote, the undiscounted normalized market price uct (K, T ) is obtained like:
 + !
Ct (K, T ) F0,T S0
ct (K, T ) = = B (t, T ) EQ XT − M ×
S0 S0 F0,T

The undiscounted normalized price, noted uct (K, T ), is such that:

Ct (K, T ) F0,T
ct (K, T ) = = B (t, T ) uct
S0 S0
Ct (K, T )
⇔ uct =
B (t, T ) F0,T
The Black Scholes is then used to compute the implied volatility σ from the undiscounted normalized market prices
like

uct (K, T, σ) = N (d1 ) − M N (d2 )

√ √
σ (T −t)
with M = K
B(t,T )Ft,T , d1 = −√ln(M ) + 2 and d2 = d1 − σ T − t. Then the implied volatility may be obtained by
σ (T −t)
the inversion of uct seen like a function of σ. The related inverse function taking a quote q gives the implied volatility like

−1
σ = uc (q)
Quoted prices are either bid or ask prices. The mid price is only a simple average of the bid and ask prices

Cbid + Cask
Cmid =
2
Sometimes we also refer to mid-volatility
σmid = σbid + ε (σask − σbid )

3 Computing the tunnel prices from market prices of vanilla options

3.1 The tunnel price

Given a chosen value for the market price of an option of strike K and T, the corresponding implied volatility may be
found. But to obtain the price of corresponding tunnels, the market skew at the quoted strike levels are needed. Indeed
the price of a tunnel which the two strike levels are the two consecutive strike levels are Kj and Kj+1 .
T ut (Kj ; Kj+1 ; T ) = P (XT > Kj ) − P (XT > Kj+1 )
with the probability P (XT > K) given by

∂C BS (K; σI (K)) ∂C BS (K; σI (K)) ∂σI (K)

P (XT < K) = 1 + +
∂K ∂σ ∂K
⇔ P (XT < K) = 1 + DeltaK BS (K; σI (K)) + V egaBS (K; σI (K)) s (K, T )
∂σI (K, T )
s (K, T ) =
∂K
DeltaK BS (K; σ) = −N (d2 (K; σI ))
√
V egaBS (K; σ) = X0 T n (d1 (K; σI ))
√
ln (X0 /K) σI T
d1 (K; σI ) = √ +
σI T 2
√
d2 (K; σI ) = d1 (K; σI ) − σI T .
and s (K, T ) the market skew. But s (K, T ) is not observed. How to choose it ?

2
3.2 Bounds for the skew of implied volatility

In the following reference, the skew is proved to have the following bounds: Given three consecutive quoted strikes Kj−1 ,
Kj and Kj+1 , the skew s (Kj , T ) for the strike level Kj and the quoted maturity T has the following bounds:

C(Kj ,T )−C(Kj−1 ,T ) C(Kj+1 ,T )−C(Kj ,T )

(Kj −Kj−1 ) − δK (Kj , T ) (Kj+1 −Kj ) − δK (Kj , T )
< s (Kj , T ) <
V ega (Kj , T ) V ega (Kj , T )

If you want to know more about how this result is obtained, you may read the two following references:

• Model Free Arbitrability Bounds for the Implied Volatility : https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4

• Calibration Model Scoring with new market induced narrower arbitrability bounds for the implied volatility skew:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4164883

3.3 Impact of the market skew and implied volatility on the tunnel prices

Combining the results from the two subsections, you may see the impact of the skew and of the implied volatilities.

4 Pricing the options with stochastic volatility models

MORE TO COME IN NEXT MEETINGS

5 Calibrating the models

The idea is to compare the calibration performance of the dierent pricing models according the classic calibration cost
function and the new one only based upon the bid and ask prices.
Assuming a set CalT of nT quoted expiry dates {Ti }i=1;...;nT and the existence of nK,Ti quoted strikes for each quoted
expiry date Ti , the calibration process of a pricing model is the way for a model to get the best estimated coecient in
order that the models reproduces market prices of nancial instruments at the current date.

5.1 The classic calibration algorithm

The calibration cost function to minimize is:

nT nX
X K,Ti
2
ζ= wi,j (Cmkt (Kj , Ti ) − Cmod (Kj , Ti ))
i=1 j=1

with a weighting wi,j to be chosen relevantly (The vega of the related option is often chosen. My own experience is to
take the implied volatility σ (Kij ; Ti ) itself ).

5.2 The new calibration algorithm

The calibration cost function to minimize is:

nT nX
X K,Ti h i
+ +
ζ= wi,j (Cbid (Kj , Ti ) − Cmod (Kj , Ti )) + (Cmod (Kj , Ti ) − Cask (Kj , Ti ))
i=1 j=1

with a weighting wi,j to be chosen relevantly (The vega of the related option is often chosen. My own experience is to
take the implied volatility σ (Kij ; Ti ) itself ).