Quasi-Gaussian Model for Model Validation and Pricing

Analysis
Sebastian Schlenkrich, d-fine GmbH

Frankfurt, April 21, 2017

© d-fine — All—rights
© d-fine All rights
reserved
reserved
| 0
Agenda

» Why is it worth to look at another complex rates model?

» What are the Quasi-Gaussian model dynamics and properties?

» How can the model be calibrated?

» Proof of concept by a callable CMS spread swap and mid-curve option case study

» Summary and References

2017-04-21 | Quasi-Gaussian Model © d-fine — All—rights

© d-fine All rights
reserved
reserved
| 1
Why is it worth to look at another complex rates model?

2017-04-21 | Quasi-Gaussian Model | Why is it worth to look at another complex rates model? © d-fine — All—rights
© d-fine All rights
reserved
reserved
| 2
Model validation and independent price verification exercises benefit from a
flexible model class to assess various product features

ATM Skew/
Smile … Swap partner
Rates Vols
Corr‘s

PV: 12 mm EUR
Structured
swap

How can
various prices
Vendor system
be explained? Pricing service

PV: 9 mm EUR

PV: 10 mm EUR

Quasi-Gaussian models allow to switch on/off effects arising from the number of risk factors, volatility
skew/smile and correlation

2017-04-21 | Quasi-Gaussian Model | Why is it worth to look at another complex rates model? (1/1) © d-fine — All—rights
© d-fine All rights
reserved
reserved
| 3
What are the Quasi-Gaussian model dynamics and
properties?

2017-04-21 | Quasi-Gaussian Model | What are the Quasi-Gaussian model dynamics and properties? © d-fine — All—rights
© d-fine All rights
reserved
reserved
| 4
Quasi-Gaussian models may be described in terms of the scalar short rate
𝑟(𝑡), state variable 𝑥(𝑡) and auxilliary variable 𝑦(𝑡)

Consider short rate 𝑟(𝑡) with dynamics(1)

𝑟 𝑡 = 𝑓 0, 𝑡 + 1⊤ 𝑥 𝑡
𝑑𝑥 𝑡 = 𝑦 𝑡 1 − 𝜒𝑥 𝑡 𝑑𝑡 + 𝜎𝑟 𝑡,⋅ ⊤ 𝑑𝑊 𝑡 , 𝑥 0 =0
𝑑𝑦 𝑡 = 𝜎𝑟 𝑡,⋅ ⊤ 𝜎𝑟 𝑡,⋅ − 𝜒𝑦 𝑡 − 𝑦 𝑡 𝜒 𝑑𝑡, 𝑦 0 =0
Model parameters
𝑑 … number of risk factors
⊤
𝑥 𝑡 = 𝑥1 𝑡 , … , 𝑥𝑑 𝑡 … state variable vector

𝑦11 (𝑡) … 𝑦1𝑑 (𝑡)

𝑦 𝑡 = ⋮ ⋮ … auxilliary variable matrix
𝑦𝑑1 (𝑡) … 𝑦𝑑𝑑 (𝑡)

𝜒1
𝜒 = ⋱ … diagonal matrix of mean reversion speed parameters
𝜒𝑑

𝜎11 (⋅) 𝜎1𝑑 (⋅)

𝜎𝑟 (𝑡,⋅) = … volatility matrix – to be specified in more detail
𝜎𝑑1 (⋅) 𝜎𝑑𝑑 (⋅)

(1) The description follows L. B. G. Andersen and V. V. Piterbarg. Interest Rate Modeling. Volume I-III. Atlantic Financial Press, 2010.

2017-04-21 | Quasi-Gaussian Model | What are the Quasi-Gaussian model dynamics and properties? (1/6) © d-fine — All—rights
© d-fine All rights
reserved
reserved
| 5
It turns out that future yield curves and discount factors may be represented
independent of the choice of volatility

Consider the auxilliary vectors of mean reversion speeds

𝑒 −𝜒1 𝑡 1 − 𝑒 −𝜒1(𝑇−𝑡) /𝜒1

ℎ 𝑡 = ⋮ , 𝐺 𝑡, 𝑇 = ⋮
−𝜒𝑑 (𝑇−𝑡)
𝑒 −𝜒𝑑 𝑡 1−𝑒 /𝜒𝑑

Then future forward rates become

⊤
𝑓 𝑡, 𝑇 = 𝑓 0, 𝑡 + ℎ 𝑇 − 𝑡 𝑥 𝑡 + 𝑦 𝑡 𝐺(𝑡, 𝑇)

Also future zero coupon bonds (i.e. discount factors) become

𝑃(0, 𝑇) ⊤
1 ⊤
𝑃 𝑡, 𝑇 = ⋅ exp −𝐺 𝑡, 𝑇 𝑥 𝑡 − 𝐺 𝑡, 𝑇 𝑦 𝑡 𝐺(𝑡, 𝑇)
𝑃(0, 𝑡) 2

Future forward rates 𝑓 𝑡, 𝑇 are affine functions in terms of the risk factors 𝑥 𝑡

2017-04-21 | Quasi-Gaussian Model | What are the Quasi-Gaussian model dynamics and properties? (2/6) © d-fine — All—rights
© d-fine All rights
reserved
reserved
| 6
Volatility matrix 𝜎𝑟 (⋅) is decomposed into stochastic volatility term 𝑧(⋅) and
local volatility term 𝜎𝑥 (⋅)

Volatility decomposition into stochastic and local volatility part

⊤ ⊤
𝜎𝑟 𝑡,⋅ = 𝑧(𝑡) ⋅ 𝜎𝑥 𝑡, 𝑥, 𝑦

Stochastic volatility is modelled as independent CIR process

𝑑𝑧 𝑡 = 𝜃 ⋅ 𝑧0 − 𝑧(𝑡) ⋅ 𝑑𝑡 + 𝜂(𝑡) ⋅ 𝑧 𝑡 ⋅ 𝑑𝑍 𝑡 , 𝑧 0 = 𝑧0 = 1, 𝑑𝑍 𝑡 ⋅ 𝑑𝑊 𝑡 = 0

For local volatility modelling we choose 𝑑 benchmark forward rates 𝑓𝑖 𝑡 = 𝑓(𝑡, 𝑡 + 𝛿𝑖 ) (𝑖 = 1, … , 𝑑) and

propose the following dynamics

𝑑𝑓𝑖 𝑡 = ⋅ ⋅ 𝑑𝑡 + 𝑧(𝑡) ⋅ 𝜆𝑖 𝑡 ⋅ 𝑎𝑖 𝑡 + 𝑏𝑖 𝑡 ⋅ 𝑓𝑖 𝑡 ⋅ 𝑑𝑈𝑖 (𝑡)

with 𝑑𝑈𝑖 (𝑡) beeing correlated with 𝑑 × 𝑑 correlation matrix Γ

We aim at transfering benchmark forward rate dynamics into our Quasi-Gaussian model

2017-04-21 | Quasi-Gaussian Model | What are the Quasi-Gaussian model dynamics and properties? (3/6) © d-fine — All—rights
© d-fine All rights
reserved
reserved
| 7
Local volatility is specified based on benchmark rate volatility dynamics

Set

𝜆1 (𝑡) 𝑎1 𝑡 + 𝑏1 𝑡 𝑓1 (𝑡)
𝑓
𝜎 (𝑡,⋅) = ⋱
𝜆𝑑 (𝑡) 𝑎𝑑 𝑡 + 𝑏𝑑 𝑡 𝑓𝑑 (𝑡)
and

−1
𝑒 −𝜒1𝛿1 … 𝑒 −𝜒𝑑 𝛿1
−1 −1
𝐻 𝑡 𝐻𝑓 𝑡 −1
= 𝐻𝑓 𝑡 𝐻 𝑡 = ⋮ ⋮
−𝜒1 𝛿𝑑 −𝜒𝑑 𝛿𝑑
𝑒 … 𝑒
and decompose correlation matrix (e.g. by Cholesky decomposition)
Γ = 𝐷⊤ 𝐷
Then Quasi-Gaussian local volatility becomes
⊤ −1 −1
𝜎𝑥 𝑡, 𝑥, 𝑦 = 𝐻𝑓 𝑡 𝐻 𝑡 ⋅ 𝜎 𝑓 𝑡,⋅ ⋅ 𝐷⊤

Note that 𝑥 and 𝑦 enter 𝜎𝑥 implicitely by the future benchmark rates 𝑓1 , … , 𝑓𝑑 in 𝜎 𝑓

2017-04-21 | Quasi-Gaussian Model | What are the Quasi-Gaussian model dynamics and properties? (4/6) © d-fine — All—rights
© d-fine All rights
reserved
reserved
| 8
We may summarize the Quasi-Gaussian dynamics which need to be
implemented e.g. in a Monte Carlo simulation

−1 −1
𝑑𝑥 𝑡 = 𝑦 𝑡 1 − 𝜒𝑥 𝑡 ⋅ 𝑑𝑡 + 𝑧 𝑡 ⋅ 𝐻 𝑓 𝑡 𝐻 𝑡 ⋅ 𝜎 𝑓 𝑡,⋅ ⋅ 𝐷⊤ ⋅ 𝑑𝑊 𝑡 , x 0 =0

−1 ⊤
𝑑𝑦 𝑡 = 𝑧 𝑡 𝐻 𝑡 𝐻 𝑓 𝑡 −1 𝑓
𝜎 𝑡,⋅ Γ 𝜎 𝑓 𝑡,⋅ 𝐻 𝑡 𝐻 𝑓 𝑡 − 𝜒𝑦 𝑡 − 𝑦 𝑡 𝜒 𝑑𝑡, 𝑦 0 =0

𝑑𝑧 𝑡 = 𝜃 ⋅ 𝑧0 − 𝑧 𝑡 ⋅ 𝑑𝑡 + 𝜂(𝑡) ⋅ 𝑧 𝑡 ⋅ 𝑑𝑍 𝑡 , 𝑧 0 = 𝑧0 = 1

Critical piece of a Monte Carlo simulation is the integration of the CIR process for the stochastic volatility
𝑧 𝑡

2017-04-21 | Quasi-Gaussian Model | What are the Quasi-Gaussian model dynamics and properties? (5/6) © d-fine — All—rights
© d-fine All rights
reserved
reserved
| 9
What are properties of the various model parameters?

−1 » 𝛿𝑖 specify explicitely modelled rates;

𝑒 −𝜒1𝛿1 … 𝑒 −𝜒𝑑 𝛿1
−1 −1
𝐻𝑓 𝑡 𝐻 𝑡 = ⋮ ⋮ rates in between are interpolated
−𝜒1 𝛿𝑑 −𝜒𝑑 𝛿𝑑
𝑒 … 𝑒 » 𝜒𝑖 specify fading speed of shocks

𝜎 𝑓 (𝑡,⋅) » 𝜆𝑖 (𝑡) control overall (ATM) volatility

𝜆1 (𝑡) 𝑎1 𝑡 + 𝑏1 𝑡 𝑓1 (𝑡)
» 𝑏𝑖 (𝑡) control volatility skew
= ⋱
𝜆𝑑 (𝑡) 𝑎𝑑 𝑡 + 𝑏𝑑 𝑡 𝑓𝑑 (𝑡) » 𝑎𝑖 𝑡 redundant and set fixed

⊤ ⊤ » Vol-of-vol 𝜂(𝑡) controls volatility smile

𝜎𝑟 𝑡,⋅ = 𝑧 𝑡 ⋅ 𝜎𝑥 𝑡, 𝑥, 𝑦
(i.e. implied vol curvature)
𝑑𝑧 𝑡 = 𝜃 ⋅ 𝑧0 − 𝑧 𝑡 ⋅ 𝑑𝑡 + 𝜂(𝑡) ⋅ 𝑧 𝑡 ⋅ 𝑑𝑍 𝑡 » 𝜃 termstructure of smile

» Correlation matrix Γ controls de-

Γ = 𝐷⊤ 𝐷
correlation of interest rates

Quasi-Gaussian model allows disentangling of the various effects which drivie interest rates

2017-04-21 | Quasi-Gaussian Model | What are the Quasi-Gaussian model dynamics and properties? (6/6) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 10
How can the model be calibrated?

2017-04-21 | Quasi-Gaussian Model | How can the model be calibrated? © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 11
Calibration is based on deriving (approximate) swap rate dynamics in the
Quasi-Gaussian model

Use Ito‘s Lemma and write swap rate dynamics in terms of scalar
Ito‘s Lemma Brownian motion

Apply Markovian projection methods and derive approximate local

Markovian projection volatility function

Apply linearization (and further approximations) to derive time-

Linearization dependent Heston-like dynamics

Use averaging techniques to derive (approximate) time-homogenuous

Parameter averaging Heston-like dynamics

Variable transformation Apply variable transformation to arrive at Heston model

Finally, use semi-analytical methods to price Vanilla option in Heston

Heston model vanilla option model

Given a formula for Vanilla options (i.e. Swaptions) we may calibrate the Quasi-Gaussian model to
observable swaption volatility market data

2017-04-21 | Quasi-Gaussian Model | How can the model be calibrated? (1/3) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 12
How is the fit to market smiles?(1)

0,80% 1,00% 1,00%

0,70% 2Y2Y 5Y2Y 10Y2Y
0,80% 0,80%
0,60%
0,50% 0,60% 0,60%
0,40% Mkt Mkt Mkt
0,30% 0,40% 0,40%
Mdl Mdl Mdl
0,20%
0,20% 0,20%
0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

0,80% 1,00% 1,00%

0,70% 2Y5Y 5Y5Y 10Y5Y
0,80% 0,80%
0,60%
0,50% 0,60% 0,60%
0,40% Mkt Mkt Mkt
0,30% 0,40% 0,40%
Mdl Mdl Mdl
0,20%
0,20% 0,20%
0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

1,00% 1,00% 1,00%

2Y10Y 5Y10Y 10Y10Y
0,80% 0,80% 0,80%

0,60% 0,60% 0,60%

Mkt Mkt Mkt
0,40% 0,40% 0,40%
Mdl Mdl Mdl
0,20% 0,20% 0,20%

0,00% 0,00% 0,00%

-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

(1) Manual fit via analytic formula to market smile

2017-04-21 | Quasi-Gaussian Model | How can the model be calibrated? (2/3) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 13
How accurate are all these approximations?

0,80%

10y x 10y Normal Vol Smile

0,75%

0,70%

QG (Receiver)
0,65%

QG (Payer)

0,60%
QGS (Receiver)

QGS (Payer)
0,55%

Heston Analytic
Strike (BP +/- ATM)
0,50%
-200 -150 -100 -50 0 50 100 150 200
There are manageable variances between Quasi-Gaussian model, approximate Swaption model and
Heston-like model

2017-04-21 | Quasi-Gaussian Model | How can the model be calibrated? (3/3) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 14
Proof of concept by a callable CMS spread swap and mid-
curve option case study

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 15
We consider pricing of a callable CMS spread swap and analyse the impact of
the various model parameters(1)

Legs Receive Pay

Notional 10.000 EUR

Effective Date 2d

Termination Date 10y

Tenor 3m

Payoff Max{ 3 x [CMS10y – CMS2y], 0 } 3m Euribor + 200bp

Conventions mod. following, Act/360

Call Schedule 1y to 9y, annually

Modelling scenarios

2-F Gaussian model 2-F Gaussian model 2-F QG model w/ 2-F QG model w/
1-F Gaussian model
w/ perfect correlation w/ 50% correlation skew skew & smile

» General impact of » Capturing short- » Decoupling short- » Capturing implied » Capturing implied
stochastic rates term and long-term term and long-term volatility skew volatility smile
shocks shocks » Improve vol (curvature
» ATM vol calibration calibration » Improve calibration

It is fairly reasonable that (de-)correlation is of particular importance. But what about skew and smile?
(1) We use market data as of July 2016

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (1/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 16
1-Factor Gaussian model in general may not capture ATM vols for both 2y and
10y swap rates

0,80% 1,00% 1,00%

0,70% 2Y2Y 5Y2Y 10Y2Y
0,80% 0,80%
0,60%
0,50% 0,60% 0,60%
0,40% Mkt Mkt Mkt
0,30% 0,40% 0,40%
Mdl Mdl Mdl
0,20%
0,20% 0,20%
0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

0,80% 0,80% 1,00%

0,70% 2Y5Y 0,70% 5Y5Y 10Y5Y
0,80%
0,60% 0,60%
0,50% 0,50% 0,60%
0,40% Mkt 0,40% Mkt Mkt
0,30% 0,30% 0,40%
Mdl Mdl Mdl
0,20% 0,20%
0,20%
0,10% 0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

1,00% 1,00% 1,00%

2Y10Y 5Y10Y 10Y10Y
0,80% 0,80% 0,80%

0,60% 0,60% 0,60%

Mkt Mkt Mkt
0,40% 0,40% 0,40%
Mdl Mdl Mdl
0,20% 0,20% 0,20%

0,00% 0,00% 0,00%

-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (2/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 17
1-F Gaussian model allows differentiating general stochastic rates impact
from derivative‘s intrinsic value

600

550

500

450

400
NPV

350

300

250

200 Intrinsic
Value
150
1F Gaussian 2F, Full Corr. De-correlation Skew Smile

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (3/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 18
2-F Gaussian model w/ perfect correlation allows improved fit to ATM
volatilities

0,80% 1,00% 1,00%

0,70% 2Y2Y 5Y2Y 10Y2Y
0,80% 0,80%
0,60%
0,50% 0,60% 0,60%
0,40% Mkt Mkt Mkt
0,30% 0,40% 0,40%
Mdl Mdl Mdl
0,20%
0,20% 0,20%
0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

0,80% 0,80% 1,00%

0,70% 2Y5Y 0,70% 5Y5Y 10Y5Y
0,80%
0,60% 0,60%
0,50% 0,50% 0,60%
0,40% Mkt 0,40% Mkt Mkt
0,30% 0,30% 0,40%
Mdl Mdl Mdl
0,20% 0,20%
0,20%
0,10% 0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

1,00% 1,00% 1,00%

2Y10Y 5Y10Y 10Y10Y
0,80% 0,80% 0,80%

0,60% 0,60% 0,60%

Mkt Mkt Mkt
0,40% 0,40% 0,40%
Mdl Mdl Mdl
0,20% 0,20% 0,20%

0,00% 0,00% 0,00%

-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (4/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 19
For 2-F Gaussian model w/ perfect correlation the reduction in callable note
NPV is mainly driven by reduced option value

600

550

500

450

400
NPV

350

300

250

200 Intrinsic
Value
150
1F Gaussian 2F, Full Corr. De-correlation Skew Smile

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (5/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 20
2-F Gaussian model w/ 50% model correlation yields 56% model-implied
correlation between 2y vs. 10y swap rates

0,80% 1,00% 1,00%

0,70% 2Y2Y 5Y2Y 10Y2Y
0,80% 0,80%
0,60%
0,50% 0,60% 0,60%
0,40% Mkt Mkt Mkt
0,30% 0,40% 0,40%
Mdl Mdl Mdl
0,20%
0,20% 0,20%
0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

0,80% 0,80% 1,00%

0,70% 2Y5Y 0,70% 5Y5Y 10Y5Y
0,80%
0,60% 0,60%
0,50% 0,50% 0,60%
0,40% Mkt 0,40% Mkt Mkt
0,30% 0,30% 0,40%
Mdl Mdl Mdl
0,20% 0,20%
0,20%
0,10% 0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

1,00% 1,00% 1,00%

2Y10Y 5Y10Y 10Y10Y
0,80% 0,80% 0,80%

0,60% 0,60% 0,60%

Mkt Mkt Mkt
0,40% 0,40% 0,40%
Mdl Mdl Mdl
0,20% 0,20% 0,20%

0,00% 0,00% 0,00%

-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (6/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 21
De-correlation in 2-F Gaussian model boosts CMS spread leg NPV

600

550

500

450

400
NPV

350

300

250

200 Intrinsic
Value
150
1F Gaussian 2F, Full Corr. De-correlation Skew Smile

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (7/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 22
Incorporating local volatility allows capturing volatility skew

0,80% 1,00% 1,00%

0,70% 2Y2Y 5Y2Y 10Y2Y
0,80% 0,80%
0,60%
0,50% 0,60% 0,60%
0,40% Mkt Mkt Mkt
0,30% 0,40% 0,40%
Mdl Mdl Mdl
0,20%
0,20% 0,20%
0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

0,80% 0,80% 1,00%

0,70% 2Y5Y 0,70% 5Y5Y 10Y5Y
0,80%
0,60% 0,60%
0,50% 0,50% 0,60%
0,40% Mkt 0,40% Mkt Mkt
0,30% 0,30% 0,40%
Mdl Mdl Mdl
0,20% 0,20%
0,20%
0,10% 0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

1,00% 1,00% 1,00%

2Y10Y 5Y10Y 10Y10Y
0,80% 0,80% 0,80%

0,60% 0,60% 0,60%

Mkt Mkt Mkt
0,40% 0,40% 0,40%
Mdl Mdl Mdl
0,20% 0,20% 0,20%

0,00% 0,00% 0,00%

-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (8/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 23
Reduced low-strike volatility reduces CMS spread leg; however effect is
mainly offset by call option (i.e. option on opposite deal)

600

550

500

450

400
NPV

350

300

250

200 Intrinsic
Value
150
1F Gaussian 2F, Full Corr. De-correlation Skew Smile

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (9/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 24
Incorporating stochastic volatility allows capturing volatility smile (i.e.
curvature in implied vols)

0,80% 1,00% 1,00%

0,70% 2Y2Y 5Y2Y 10Y2Y
0,80% 0,80%
0,60%
0,50% 0,60% 0,60%
0,40% Mkt Mkt Mkt
0,30% 0,40% 0,40%
Mdl Mdl Mdl
0,20%
0,20% 0,20%
0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

0,80% 1,00% 1,00%

0,70% 2Y5Y 5Y5Y 10Y5Y
0,80% 0,80%
0,60%
0,50% 0,60% 0,60%
0,40% Mkt Mkt Mkt
0,30% 0,40% 0,40%
Mdl Mdl Mdl
0,20%
0,20% 0,20%
0,10%
0,00% 0,00% 0,00%
-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

1,00% 1,00% 1,00%

2Y10Y 5Y10Y 10Y10Y
0,80% 0,80% 0,80%

0,60% 0,60% 0,60%

Mkt Mkt Mkt
0,40% 0,40% 0,40%
Mdl Mdl Mdl
0,20% 0,20% 0,20%

0,00% 0,00% 0,00%

-200 -100 0 100 200 -200 -100 0 100 200 -200 -100 0 100 200

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (10/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 25
Low-strike vols are increased by stochastic volatility; again with offsetting
effects on CMS spread leg and call option

600

550

500

450

400
NPV

350

300

250

200 Intrinsic
Value
150
1F Gaussian 2F, Full Corr. De-correlation Skew Smile

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (11/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 26
The component prices help for a detailed analysis of pricing results

Scenario Intrinsic 1F Gaussian 2F, Full Corr. De-Corr. Skew Smile

MC Pricing
CMS2y 474 501 498 497 498 500
CMS10y 955 1.055 1.073 1.067 1.072 1.078
Euribor 206 207 207 209 209 209
StructLeg 1.447 1.664 1.747 1.975 1.956 1.974
FundLeg -2.256 -2.257 -2.257 -2.259 -2.259 -2.259
Underlying -809 -593 -510 -284 -303 -286
AMC Pricing
NoteNPV 218 439 373 602 591 590
UnderlyingNPV -809 -593 -512 -286 -305 -288
OptionNPV 1.027 1.032 886 888 895 878

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (12/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 27
Similar analysis may be perfomed for other derivatives. In addition we have a
look at a mid-curve option with 5y exercise into a 5y-5y ATM forward swap

0,90%

0,80%

0,70% 5y x 10y implied vol at 69bp

Implied Normal Volatility

0,60% 5y x 5y implied vol at 65bp

0,50%

0,40%

0,30%

0,20%

0,10%

0,00%
1F Gaussian 2F, Full Corr. De-Corr. Skew Smile

Mid-curve options may be prices by means of the same (semi-)analytical formulas used for Vanilla option
pricing.

2017-04-21 | Quasi-Gaussian Model | Proof of concept by a callable CMS spread swap and mid-curve option case study (13/13) © d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 28
Summary and References

2017-04-21 | Quasi-Gaussian Model | Summary and References © d-fine — All —

© d-fine rights
All rights
reserved
reserved
| 29
Summary and References

Summary

» Quasi-Gaussian model appears to be a powerfull tool for pricing complex rates derivatives

» Model allows clear separation of risk factors and link to model parameters

» Particular parameter choices yield well-established standard modells (e.g. 1F/2F Gaussian

models); which is very practical for model validation

References
» L. Andersen and V. Piterbarg. Interest rate modelling, volume I to III. Atlantic Financial Press,
2010.
» https://github.com/sschlenkrich/QuantLib/tree/master/ql/experimental/templatemodels

2017-04-21 | Quasi-Gaussian Model | Summary and References (1/1) © d-fine — All —

© d-fine rights
All rights
reserved
reserved
| 30
Dr Sebastian Schlenkrich d-fine
Manager Frankfurt
Tel +49 89 7908617-355 München
Mobile +49 162 2631525 London
E-Mail Sebastian.Schlenkrich@d-fine.de Wien
Zürich
Artur Steiner
Partner
Tel +49 89 7908617-288 Zentrale
Mobile +49 151 14819322 d-fine GmbH
E-Mail Artur.Steiner@d-fine.de An der Hauptwache 7
D-60313 Frankfurt/Main

Tel +49 69 90737-0

Fax +49 69 90737-200
www.d-fine.com

© d-fine — All —
© d-fine rights
All rights
reserved
reserved
| 31
 d-fine

(textbox is required to avoid an issue

where this page gets rotated by 90°
if printing (both “physical” and pdf))