Visit https://ebookfinal.

com to download the full version and

explore more ebooks

Pricing models for Bermudan style interest rate

derivatives Raoul Pietersz

_____ Click the link below to download _____

https://ebookfinal.com/download/pricing-models-for-
bermudan-style-interest-rate-derivatives-raoul-
pietersz/

Explore and download more ebooks at ebookfinal.com

Here are some suggested products you might be interested in.
Click the link to download

The SABR LIBOR Market Model Pricing Calibration and

Hedging for Complex Interest Rate Derivatives 1st Edition
Riccardo Rebonato
https://ebookfinal.com/download/the-sabr-libor-market-model-pricing-
calibration-and-hedging-for-complex-interest-rate-derivatives-1st-
edition-riccardo-rebonato/

Interest Rate Modeling Theory and Practice 3rd Edition

Lixin Wu

https://ebookfinal.com/download/interest-rate-modeling-theory-and-
practice-3rd-edition-lixin-wu/

An Elementary Introduction To Stochastic Interest Rate

Modeling 2nd Edition Edition Nicolas Privault

https://ebookfinal.com/download/an-elementary-introduction-to-
stochastic-interest-rate-modeling-2nd-edition-edition-nicolas-
privault/

American style derivatives valuation and computation 1st

Edition Jerome Detemple

https://ebookfinal.com/download/american-style-derivatives-valuation-
and-computation-1st-edition-jerome-detemple/
Property Derivatives Pricing Hedging and Applications 1st
Edition Juerg M. Syz

https://ebookfinal.com/download/property-derivatives-pricing-hedging-
and-applications-1st-edition-juerg-m-syz/

Commodities and Commodity Derivatives Modelling and

Pricing for Agriculturals Metals and Energy 1st Edition
Helyette Geman
https://ebookfinal.com/download/commodities-and-commodity-derivatives-
modelling-and-pricing-for-agriculturals-metals-and-energy-1st-edition-
helyette-geman/

Credit derivatives instruments applications and pricing

1st Edition Mark Jonathan Paul Anson

https://ebookfinal.com/download/credit-derivatives-instruments-
applications-and-pricing-1st-edition-mark-jonathan-paul-anson/

Le vy processes in finance pricing financial derivatives

1st Edition Wim Schoutens

https://ebookfinal.com/download/le-vy-processes-in-finance-pricing-
financial-derivatives-1st-edition-wim-schoutens/

Basic Option Volatility Strategies Understanding Popular

Pricing Models 1st Edition Sheldon Natenberg

https://ebookfinal.com/download/basic-option-volatility-strategies-
understanding-popular-pricing-models-1st-edition-sheldon-natenberg/
Pricing models for Bermudan style interest rate
derivatives Raoul Pietersz Digital Instant Download
Author(s): Raoul Pietersz
ISBN(s): 9789058920997, 9058920992
Edition: Kindle
File Details: PDF, 4.22 MB
Year: 2005
Language: english
Erim - 05 omslag Pietersz 9/23/05 1:41 PM Pagina 1

Pricing models for Bermudan-style interest rate

Design: B&T Ontwerp en advies www.b-en-t.nl

71 RAOUL PIETERSZ
derivatives
Bermudan-style interest rate derivatives are an important class of

RAOUL PIETERSZ
options. Many banking and insurance products, such as mortgages,
cancellable bonds, and life insurance products, contain Bermudan
interest rate options associated with early redemption or cancella-
tion of the contract. The abundance of these options makes evident
Pricing models for
that their proper valuation and risk measurement are important to
banks and insurance companies. Risk measurement allows for off-
setting market risk by hedging with underlying liquidly traded assets
Bermudan-style

Pricing models for Bermudan-style interest rate derivatives

and options. Pricing models must be arbitrage-free, and consistent
interest rate derivatives

Print: Haveka www.haveka.nl

with (calibrated to) prices of actively traded underlying options.
Model dynamics need be consistent with the observed dynamics of
the term structure of interest rates, e.g., correlation between inte-
rest rates. Moreover, valuation algorithms need be efficient: Finan-
cial decisions based on derivatives pricing calculations often need to
be made in seconds, rather than hours or days. In recent years, a
successful class of models has appeared in the literature known as
market models. This thesis extends the theory of market models, in
the following ways: (i) it introduces a new, efficient, and more
accurate approximate pricing technique, (ii) it presents two new and
fast algorithms for correlation-calibration, (iii) it develops new models
that enable efficient calibration for a whole new range of deriva-
tives, such as fixed-maturity Bermudan swaptions, and (iv) it presents
novel empirical comparisons of the performance of existing calibra-
tion techniques and models, in terms of reduction of risk.

ERIM
The Erasmus Research Institute of Management (ERIM) is the Research
School (Onderzoekschool) in the field of management of the
Erasmus University Rotterdam. The founding participants of ERIM
are RSM Erasmus University and the Erasmus School of Economics.
ERIM was founded in 1999 and is officially accredited by the Royal
Netherlands Academy of Arts and Sciences (KNAW). The research
undertaken by ERIM is focussed on the management of the firm in its
environment, its intra- and inter-firm relations, and its business
processes in their interdependent connections.
The objective of ERIM is to carry out first rate research in manage-
ment, and to offer an advanced graduate program in Research in
Management. Within ERIM, over two hundred senior researchers and
Ph.D. candidates are active in the different research programs. From
a variety of academic backgrounds and expertises, the ERIM commu-
nity is united in striving for excellence and working at the forefront
of creating new business knowledge.

www.erim.eur.nl ISBN 90-5892-099-2

 Pricing Models for Bermudan-Style
Interest Rate Derivatives

1
2
 Pricing Models for Bermudan-Style
Interest Rate Derivatives
Waarderingsmodellen voor Bermuda-stijl rente derivaten

Proefschrift

ter verkrijging van de graad van doctor aan de

Erasmus Universiteit Rotterdam
op gezag van de
rector magnificus

Prof.dr. S.W.J. Lamberts

en volgens besluit van het College voor Promoties.

De openbare verdediging zal plaatsvinden op

donderdag 8 december 2005 om 16.00 uur

door

Raoul Pietersz
geboren te Rotterdam

3
 Promotiecommissie

Promotoren:
Prof.dr. A.A.J. Pelsser
Prof.dr. A.C.F. Vorst

Overige leden:
Prof.dr. P.J.F. Groenen
Prof.dr. F.C.J.M. de Jong
Dr.ir. M.P.E. Martens

Erasmus Research Institute of Management (ERIM)

Erasmus University Rotterdam
Internet: http://www.erim.eur.nl

ERIM Ph.D. Series Research in Management 71

ISBN 90-5892-099-2

c 2005, Raoul Pietersz

°

All rights reserved. No part of this publication may be reproduced or transmitted in

any form or by any means electronic or mechanical, including photocopying, recording,
or by any information storage and retrieval system, without permission in writing from
the author.

4
 Voor Beata, Karsten en Daniël

5
6
 Acknowledgements

First and foremost, I would like to thank my promotor Antoon Pelsser. His guidance
throughout the Ph.D. period has been excellent. Chapters 2, 5 and 6 were written in
cooperation with Antoon. The research benefitted greatly from his invaluable suggestions,
and he truly is an inspirator. Thank you Antoon.
Second, I thank my promotor and former employer Ton Vorst. He is the one who
suggested me to start the Ph.D. track with Antoon Pelsser, and who part-time employed
me at ABN AMRO Bank, Quantitative Risk Analytics (at the time “Market Risk –
Modelling and Product Analysis”), for the period September 2001 till June 2004. For all
this, I am very grateful.
Third, many thanks go to the other members of the small committee; Patrick Groenen,
Frank de Jong, and, Martin Martens. Also, I express my gratitude to the other members
of the committee; Lane Hughston, Farshid Jamshidian, and, Thierry Post.
Fourth, special thanks go to Marcel van Regenmortel, for teaching me many technical
and exciting aspects of interest rate derivatives pricing, and for part-time employing me
at Product Development Group, Quantitative Analytics, ABN AMRO Bank, from July
2004 onwards. Chapters 5 and 7 were written in cooperation with Marcel.
Fifth, I am grateful to Patrick Groenen, for co-authoring Chapter 3.
Sixth, I thank Igor Grubišić for co-authoring Chapter 4.
I am much obliged to my colleagues at Erasmus University Rotterdam: Jaap Spronk
for support through the Erasmus Center for Financial Research (ECFR); Martin Martens
for his help while I was co-teaching one of his example classes and for suggesting me
as a lecturer at the Rotterdam School of Management (RSM); Winfried Hallerbach for
help with preparation of RSM lectures; Wilfred Mijnhardt for guiding me through the
publication process; Tineke Kurtz, Tineke van der Vhee, Elli Hoek van Dijke, and Ella
Boniuk, for efficiently aiding me in administrative matters; and, Mariëlle Sonnenberg, for
being a pleasant room-mate.
During the Ph.D. period I have been able to present my work at leading international
conferences. I am very grateful for the financial support that made this possible, received
from Erasmus Research Institute of Management (ERIM), from the Econometric Institute
(EI), and from ECFR.

7
 viii ACKNOWLEDGEMENTS

A special thank you to past and present managers at ABN AMRO Bank: Ton Vorst,
Dick Boswinkel, Nam Kyoo Boots, Marcel van Regenmortel, Bernt van Linder and Geert
Ceuppens. Thank you to past and present colleagues at Product Development Group:
Nicolas Carré in Amsterdam, and Thilo Roßberg and Russell Barker in London. Thanks to
members of CAL: Nancy Appels, Reinier Bosman, Danny Wester, Frank Putman, André
Roukema, Jelper Striet, and Willem van der Zwart. Thank you to past and present
colleagues at ‘Product Analysis’: Steffen Lukas, Benjamin Schiesslé, Martijn van der
Voort, Lukas Phaf, Alice Gee, Rutger Pijls, Drona Kandhai, and Alex Zilber. A thank
you to colleagues that have become friends: Dion Hautvast, Bram Warmenhoven, and
Glyn Baker.
I am also grateful to a number of anonymous referees and to Riccardo Rebonato and
Mark Joshi who provided valuable comments and suggestions to earlier versions of the
papers that form the basis for this thesis. Many thanks to Frank de Jong and Joanne
Kennedy for providing much appreciated feedback to my research proposal.
Finally, I would like to thank my parents for always being there for me. I thank my
son Karsten for bringing so much joy to my life. I thank Beata for her love, support, and
kindness.

Raoul Pietersz
February 7th 2005, Amsterdam

8
 Contents

Acknowledgements vii

Notation xix

Outline xxiii

1 Introduction 1
1.1 Arbitrage-free pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Use of models in practice . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Interest rate markets and options . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Linear products: Deposits, bonds, and swaps . . . . . . . . . . . . . 7
1.2.2 Interest rate options: Caps, floors, and swaptions . . . . . . . . . . 8
1.3 Interest rate derivatives pricing models . . . . . . . . . . . . . . . . . . . . 11
1.3.1 Short rate models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.2 Market models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.3 Markov-functional models . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 American option pricing with Monte Carlo simulation . . . . . . . . . . . . 17

2 Risk-managing Bermudan swaptions in a LIBOR model 19

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Recalibration approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Explanation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4 Swap vega and the swap market model . . . . . . . . . . . . . . . . . . . . 27
2.5 Alternative method for calculating swap vega . . . . . . . . . . . . . . . . 29
2.6 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.7 Comparison with the swap market model . . . . . . . . . . . . . . . . . . . 30
2.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.A Appendix: Negative vega two-stock Bermudan options . . . . . . . . . . . 34

9
 x CONTENTS

3 Rank reduction of correlation matrices by majorization 39

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2.1 Modified PCA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2.2 Majorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2.3 Geometric programming . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2.4 Alternating projections without normal correction . . . . . . . . . . 45
3.2.5 Lagrange multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.6 Parametrization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.7 Alternating projections with normal correction (d = n) . . . . . . . 47
3.3 Majorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4 The algorithm and convergence analysis . . . . . . . . . . . . . . . . . . . 50
3.4.1 Global convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4.2 Local rate of convergence . . . . . . . . . . . . . . . . . . . . . . . . 52
3.5 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.5.1 Numerical comparison with other methods . . . . . . . . . . . . . . 54
3.5.2 Non-constant weights . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.5.3 The order effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.5.4 Majorization equipped with the power method . . . . . . . . . . . . 62
3.5.5 Using an estimate for the largest eigenvalue . . . . . . . . . . . . . 62
3.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.A Appendix: Proof of Equation (3.11) . . . . . . . . . . . . . . . . . . . . . . 64

4 Rank reduction of correlation matrices by geometric programming 67

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.1.1 Weighted norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.2 Solution methodology with geometric optimisation . . . . . . . . . . . . . . 71
4.2.1 Basic idea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2.2 Topological structure . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2.3 A dense part of Mn,d equipped with a differentiable structure . . . . 74
4.2.4 The Cholesky manifold . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2.5 Choice of representation . . . . . . . . . . . . . . . . . . . . . . . . 76
4.3 Optimisation over the Cholesky manifold . . . . . . . . . . . . . . . . . . . 76
4.3.1 Riemannian structure . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.3.2 Normal and tangent spaces . . . . . . . . . . . . . . . . . . . . . . . 78
4.3.3 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.3.4 Parallel transport along a geodesic . . . . . . . . . . . . . . . . . . 80
4.3.5 The gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.3.6 Hessian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

10
 CONTENTS xi

4.3.7 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.4 Discussion of convergence properties . . . . . . . . . . . . . . . . . . . . . 81
4.4.1 Global convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.4.2 Local rate of convergence . . . . . . . . . . . . . . . . . . . . . . . . 83
4.5 A special case: Distance minimization . . . . . . . . . . . . . . . . . . . . . 85
4.5.1 The case of d = n . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.5.2 The case of d = 2, n = 3 . . . . . . . . . . . . . . . . . . . . . . . . 85
4.5.3 Formula for the differential of ϕ . . . . . . . . . . . . . . . . . . . . 85
4.5.4 Connection normal with Lagrange multipliers . . . . . . . . . . . . 86
4.5.5 Initial feasible point . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.6 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.6.1 Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.6.2 Numerical comparison . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.A Appendix: Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.A.1 Proof of Theorem 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.A.2 Proof of Proposition 2 . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.A.3 Proof of Proposition 3 . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.A.4 Proof of Theorem 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.A.5 Proof of Theorem 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.A.6 Proof of Lemma 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.A.7 Proof of Theorem 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

5 Fast drift-approximated pricing in the BGM model 97

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.2 Notation for BGM model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.3 Single time step method for pricing on a grid . . . . . . . . . . . . . . . . . 100
5.3.1 Justification of the above assumptions . . . . . . . . . . . . . . . . 100
5.3.2 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.3.3 Separability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.3.4 Single time step method . . . . . . . . . . . . . . . . . . . . . . . . 101
5.3.5 Valuation of interest rate derivatives with the single time step method103
5.4 Discretizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.4.1 Euler discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.4.2 Predictor-corrector discretization . . . . . . . . . . . . . . . . . . . 104
5.4.3 Milstein discretization . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.4.4 Brownian bridge discretization . . . . . . . . . . . . . . . . . . . . . 105
5.5 The Brownian bridge scheme for single time steps . . . . . . . . . . . . . . 107
5.5.1 Theoretical result . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

11
 xii CONTENTS

5.5.2 LIBOR-in-arrears case . . . . . . . . . . . . . . . . . . . . . . . . . 108

5.6 The Brownian bridge scheme for multi-time steps . . . . . . . . . . . . . . 110
5.6.1 Weak convergence of the Brownian bridge scheme . . . . . . . . . . 110
5.6.2 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.7 Example: one-factor drift-approximated BGM . . . . . . . . . . . . . . . . 114
5.7.1 A simple numerical example . . . . . . . . . . . . . . . . . . . . . . 115
5.8 Example: Bermudan swaption . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.8.1 Two-factor model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.9 Test of accuracy of drift approximation . . . . . . . . . . . . . . . . . . . . 124
5.9.1 Drift-approximation accuracy test based on no-arbitrage . . . . . . 125
5.9.2 Numerical results for single time step test . . . . . . . . . . . . . . 125
5.10 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.A Appendix: Mean of geometric Brownian bridge . . . . . . . . . . . . . . . 127
5.B Appendix: Approximation of substituting the mean . . . . . . . . . . . . . 128
5.C Appendix: MATLAB code for Brownian bridge scheme . . . . . . . . . . . 129

6 A comparison of single factor Markov-functional and multi factor market

models 133
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.2.1 The LIBOR and swap market models . . . . . . . . . . . . . . . . . 139
6.2.2 The Markov-functional model . . . . . . . . . . . . . . . . . . . . . 141
6.2.3 Estimating Greeks for callable products in market models . . . . . . 143
6.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
6.4 Accuracy of the terminal correlation formula . . . . . . . . . . . . . . . . . 146
6.5 Empirical comparison results . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6.5.1 Delta hedging versus delta and vega hedging . . . . . . . . . . . . . 150
6.5.2 ‘Large’ perturbation sizes versus constant exercise decision method
with ‘small’ perturbation sizes . . . . . . . . . . . . . . . . . . . . . 151
6.5.3 Delta-vega hedge results . . . . . . . . . . . . . . . . . . . . . . . . 152
6.6 The impact of smile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

7 Generic market models 161

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
7.2.1 Absence of arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . . 166
7.3 Necessary and sufficient conditions for no-arbitrage . . . . . . . . . . . . . 168
7.3.1 Main result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172

12
 CONTENTS xiii

7.4 Generic expressions for no-arbitrage drift terms . . . . . . . . . . . . . . . 175

7.4.1 Terminal measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
7.4.2 Spot measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
7.4.3 An example: The LIBOR market model . . . . . . . . . . . . . . . 179
7.5 Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
7.6 Generic calibration to correlation . . . . . . . . . . . . . . . . . . . . . . . 185
7.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
7.A Appendix: Rationale for Approximation 1 . . . . . . . . . . . . . . . . . . 186
7.B Appendix: Proof of Lemma 3 . . . . . . . . . . . . . . . . . . . . . . . . . 187

8 Conclusions 189

Nederlandse samenvatting (Summary in Dutch) 193

Bibliography 195

Author index 207

13
14
 List of Figures

1 Outline of the thesis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiv

1.1 Payoffs of caplets and floorlets versus realized LIBOR. . . . . . . . . . . . 9

2.1 Recalibration swap vega results for 10,000 simulation paths. . . . . . . . . 24

2.2 Empirical standard errors of vega for 10,000 simulation paths. . . . . . . . 25
2.3 Recalibration THFRV vega results for 1 million simulation paths. . . . . . 25
2.4 Observed change in swap rate instantaneous variance. . . . . . . . . . . . . 26
2.5 Natural increment of Black implied swaption volatility. . . . . . . . . . . . 28
2.6 Swap vega results for 10,000 simulation paths. . . . . . . . . . . . . . . . . 31
2.7 Comparison of LMM and SMM for swap vega per bucket, 5% strike. . . . . 31
2.8 Comparison of LMM and SMM for total swap vega against strike. . . . . . 32

3.1 The idea of majorization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3.2 Performance profile for n = 10, d = 2, t = 0.05s. . . . . . . . . . . . . . . . 56
3.3 Performance profile for n = 20, d = 4, t = 0.1s. . . . . . . . . . . . . . . . . 57
3.4 Performance profile for n = 80, d = 20, t = 2s. . . . . . . . . . . . . . . . . 58
3.5 Convergence run of the power method versus p lambda=max(eig(B)). . . . . 63
3.6 The equality kPy(∞) (y(k) − y(∞) )k = δ (k) 1 − (δ (k) )2 /4. . . . . . . . . . . . 65

4.1 Shell representing the set of 3 × 3 correlation matrices of rank 2 or less. . . 69

4.2 Convergence runs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.3 Performance profile for n = 30, d = 3, t = 2s, equal weights. . . . . . . . . 89
4.4 Performance profile for n = 50, d = 4, t = 1s, equal weights. . . . . . . . . 90
4.5 Performance profile for n = 60, d = 5, t = 3s, equal weights. . . . . . . . . 91
4.6 Performance profile for n = 15, d = 3, t = 1s, non-equal weights. . . . . . . 92

5.1 LIBOR-in-arrears test. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

5.2 Monte Carlo convergence. . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.3 Exercise boundaries for the eight-year deal. . . . . . . . . . . . . . . . . . . 121
5.4 Risk sensitivities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.5 Timing inconsistency in the single time step framework for BGM. . . . . . 124

15
 xvi LIST OF FIGURES

5.6 Set-up for inconsistency test. . . . . . . . . . . . . . . . . . . . . . . . . . . 125

6.1 Fitted a-parameter versus β-parameter. . . . . . . . . . . . . . . . . . . . . 141

6.2 Bermudan swaption values per trade date. . . . . . . . . . . . . . . . . . . 149
6.3 Comparison of delta versus delta and vega hedging. . . . . . . . . . . . . . 150
6.4 ‘Large’ versus ‘small’ perturbation sizes and constant exercise method. . . 151
6.5 Delta-vega hedge results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

7.1 Swaptions from swaption matrix to which various models are calibrated. . 169
7.2 An overview of the forward swap agreements for various market models. . . 170
7.3 Test results of exact versus approximate drift terms in CMS(q) models. . . 183

16
 List of Tables

1.1 Some short rate models and their specification of short rate dynamics. . . . 13

2.1 Market European swaption volatilities. . . . . . . . . . . . . . . . . . . . . 23

2.2 Swap vega per bucket test results for varying strikes. . . . . . . . . . . . . 33
2.3 Deal description. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4 Results for negative vega per bucket for two-stock Bermudan option. . . . 36

3.1 Excerpt of Table 3 in De Jong et al. (2004). . . . . . . . . . . . . . . . . . 55

3.2 Comparative results of the parametrization and majorization algorithms. . 59
3.3 Results for the ratchet cap and trigger swap. . . . . . . . . . . . . . . . . . 60
3.4 The order effect. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

5.1 A simple numerical example. . . . . . . . . . . . . . . . . . . . . . . . . . . 116

5.2 Specification of the Bermudan swaption comparison deal. . . . . . . . . . . 118
5.3 Results of the Bermudan swaption comparison deal. . . . . . . . . . . . . . 119
5.4 Computational times for the Bermudan swaption comparison deal. . . . . . 120
5.5 Simulation re-run using pre-computed exercise boundaries. . . . . . . . . . 122
5.6 Two-factor model comparison. . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.7 Quality of drift approximations: volatility/mean-reversion 15%/10%. . . . 126
5.8 Quality of drift approximations: various scenarios. . . . . . . . . . . . . . . 126

6.1 Statistical description of the swaption volatility data. . . . . . . . . . . . . 145

6.2 Discount factors for the USD data of 21 February 2003. . . . . . . . . . . . 146
6.3 Smile swaption volatility USD data of 21 February 2003. . . . . . . . . . . 146
6.4 Error analysis of the terminal correlation. . . . . . . . . . . . . . . . . . . . 147
6.5 The Bermudan swaption deal used in the comparison. . . . . . . . . . . . . 148
6.6 Fitted displaced diffusion parameters. . . . . . . . . . . . . . . . . . . . . . 155
6.7 Fitted swaption volatility and fit errors with the displaced diffusion model. 155
6.8 Benchmark results for the displaced diffusion Markov-functional model. . . 157
6.9 The Bermudan swaption deal used in the test of impact of smile. . . . . . . 158
6.10 Prices of Bermudan swaptions in smile versus non-smile models. . . . . . . 159

17
 xviii LIST OF TABLES

7.1 Example of a hybrid coupon swap payment structure for the floating side. . 163
7.2 Deal description for test of exact versus approximate drift in CMS models. 184

18
 Notation

s A lower case italic denotes a scalar. %(p,s) Performance ratio for algorithm s on
test problem p.
v A lower case bold denotes a vector.
σ Volatility.
M An upper case bold denotes a matrix.
τi Discretization time point. Time dis-
α Day count fraction.
cretization: τ1 < · · · < τm .
β Correlation parameter, see (2.4).
φ Cumulative normal distribution func-
γ Weights; correlation parameters. tion; performance profile.

Γ Lagrange multipliers. ϕ Objective function for rank reduction

of correlation matrices.
δ Delta: risk sensitivity with respect to
underlying rate or asset price. χ Auxiliary majorization function.

∆ Tangent vector for geometric program- Ψ Error per-entry in correlation matrix:

ming. Ψ = YYT − P.

ε Perturbation size; convergence criterium. ω Outcome of probability space, ω ∈ Ω.

η Pay (+1)/receive (−1) fixed index; caplet Ω Probability space.

(+1)/floorlet (−1) index.
Sn Sphere in Rn+1 .
θ Drift in short rate models.
E Expectation.
λ Largest eigenvalue of B.
P Real-world probability measure.
Λ Diagonal eigenvalues matrix.
Q Arbitrage-free pricing measure.
µ Drift.
R Set of real numbers.
ν Vega: risk sensitivity with respect to
Sn Set of real n × n symmetric matrices.
volatility.
F Filtration.
π Realized numeraire relative payoff.
N Normal distribution.
ρ Target correlation, matrix form: P.

19
 xx NOTATION

O Order. r Short rate; instantaneous continuously

compounded interest rate.
S Swap market model.
s Swap rate; start index of a forward
a Model parameter. rate; asset price.
b Discount bond price; model parame- t Time.
ter.
ti Tenor time. Tenor structure: 0 = t0 <
bi Price of a discount bond maturing at · · · < tn .
time ti .
v Value of a security, asset or derivative.
B Helper matrix for majorization.
w Brownian motion; weight coefficient.
c Model parameter.
Y Decomposition matrix: YYT = P,
C Correlation matrix. with P a correlation matrix.
d Number of stochastic factors; the usual z Normally distributed random variable.
d1 , d2 in Black-type formulas.
i (Subscript i) Associated with the pe-
e End index of a forward rate. riod [ti , ti+1 ] (e.g., fi , σi , αi ); associ-
f Forward rate. ated with ti (e.g, bi ).

i(t) Spot LIBOR index at time t, see (2.3). s:e (Subscript s : e) Associated with the
period [ts , te ] (e.g., fs:e , σs:e ).
I Identity matrix.
(i+1)
(Superscript (i + 1)) Associated with
k Strike rate. the forward measure for which bi+1 is
the numeraire.
m Number of discretization time points.
d Infinitesimal differential.
n Number of forward rates; numeraire
value. bp Basis point, 0.01%.

o Option price. P&L Profit and loss.

p PVBP, present value of a basis point, xNCy x non-call y option, exercisable on an

see (2.1). underlying with a maturity of x years
from today but callable only after y
P Target correlation matrix, P = (ρij )ij .
years.
Q Orthogonal matrix, QQT = I, with I
· Scalar product of vectors.
the identity matrix.

20
 NOTATION xxi

h·, ·i Scalar product of vectors; quadratic

cross-variation.

h·i Quadratic variation.

Var Variance.

Cov Covariance.

k.k Vector length: Square root of sum of

squares of vector entries.

k · kF Frobenius norm, kYk2F := tr(YYT )

for matrices Y.

¿ Much smaller than.

T
(Superscript T ) Matrix transpose.

21
22
 Outline

The purpose of this thesis is to further knowledge of efficient valuation and risk man-
agement of interest rate derivatives (mainly of Bermudan-style but other types are also
included) by extending the theory on market models. Here, we provide an outline of the
thesis. Readers that are non-experts in the field of interest rate derivatives pricing could
skip the outline at first reading and return here after reading the introductory Chapter 1.
A schematic outline of the thesis is given in Figure 1.
Chapter 2 investigates various popular calibration choices for the LIBOR market model
and their effect on the quality of risk sensitivities of Bermudan swaptions. The results
show that care should be taken when selecting a calibration method: Certain choices,
e.g., so-called time-homogeneous volatility, may lead to non-efficient estimates of risk
sensitivities. Poor and unstable estimates of risk, in turn, lead to fluctuations in a hedge
portfolio that are spurious and have no economic meaning, and the risk associated with
the derivative is not adequately reduced. The results however also show that so-called
constant volatility leads to efficient and stable estimates of risk sensitivities. The combined
results are important and valuable to financial institutions that need to select a calibration
method for market models, with the aim to risk manage Bermudan swaptions and other
interest rate derivatives.
Chapter 2 has been published in the Journal of Derivatives, see Pietersz & Pelsser
(2004a). An extended abstract of the chapter has been published in Risk Magazine, see
Pietersz & Pelsser (2004b). The Risk article has been republished as part of a Risk book,
see Pietersz & Pelsser (2005b).
Chapters 3 and 4 solve the same problem in two completely different ways. The
problem is the so-called rank reduction of correlation matrices, and occurs as a key part
of calibrating multi-factor market models to correlation. Mathematically formulated, rank
reductions of correlation matrices are non-convex optimization problems, which are known
to be difficult to solve: The problem is to minimize, over low-rank correlation matrices
attainable by the model, an objective value, which is the error with the original given
correlation matrix. We present two elegant solution algorithms. The benefit over existing
algorithms is the enhanced efficiency: In terms of computational speed, the algorithms of
Chapters 3 and 4 outperform existing algorithms, in the numerical tests we considered.

23
 xxiv OUTLINE

Ch. 2: Model A Model B New model

Comparison:
performance of Calibration method A Ch. 7:
calibration Develop-
methods Calibration method B ment of a
new
Ch.s 3&4 New calibration method model

Calculation method A

Ch. 5 New calculation method

Performance in practice: Perfor-

Reduction of risk, mance in
efficiency, etc. practice

Ch. 6: Comparison: hedge performance of models

Figure 1: Outline of the thesis. Here, models A and B denote market models and Markov-
functional models, respectively. The new model denotes CMS and generic market models.

24
 OUTLINE xxv

Chapter 3 presents a solution for rank reduction of correlation matrices, based on

majorization, which is a general technique from optimization. We perform the task of
showing that majorization can be applied to rank reduction of correlation matrices. The
resulting algorithm is globally convergent (i.e., from any starting point) to a local mini-
mum. The algorithm is shown to be straightforward to implement, which makes its use
accessible to non-experts. The majorization algorithm is extremely efficient, because of
its low cost per iterate.
Chapter 3 has been published in Quantitative Finance, see Pietersz & Groenen (2004b).
An extended abstract of the chapter has been published in Risk Magazine, see Pietersz
& Groenen (2004a).
Chapter 4 develops a solution for rank reduction of correlation matrices based on geo-
metric programming, which is optimization over curved space (manifolds). The manifold-
equivalents of Newton and conjugate gradient optimization algorithms are presented for
the problem of rank reduction of correlation matrices. By carefully selecting the man-
ifold, we are able to bring the gradient and Hessian to natural forms, enabling an effi-
cient implementation. The geometric curved algorithms enjoy the same super-linear local
convergence properties as their Euclidean flat counterparts: Quadratic convergence for
Newton and m-steps quadratic convergence for conjugate gradient, where m denotes the
dimension of the manifold. Additionally, we develop a novel method to immediately check
whether a stationary point is a global minimum, by extending the Lagrange multiplier
results of Zhang & Wu (2003) and Wu (2003). This feature is very rare for non-convex
optimisation problems, and makes the problem of rank reduction of correlation matrices
all the more interesting. Extensive numerical tests show that geometric programming
compares favourably with other existing algorithms, in terms of computational speed.
Chapter 4 has been submitted. For the working paper version, see Grubišić & Pietersz
(2005).
Chapter 5 introduces a new discretization for the LIBOR market model, the Brow-
nian bridge discretization. Discretizations are required for implementation of a pricing
algorithm. The benefit of Brownian bridge is its accuracy when single or large time steps
are used. For single time steps, we show that it is least-squares optimal to use Brow-
nian bridge (in a to be defined natural sense). This is also confirmed in the numerical
LIBOR-in-arrears test extended from Hunter, Jäckel & Joshi (2001). As a multi step dis-
cretization, we show that Brownian bridge converges weakly with order one. The multi
step convergence is illustrated by numerical tests. Finally, we show that a single time
step discretization combined with a separability assumption on the volatility, allows for
an even more efficient implementation via pricing on a grid or on a recombining lattice,
instead of Monte Carlo.

25
 xxvi OUTLINE

Chapter 5 has been published in the Journal of Computational Finance, see Pietersz,
Pelsser & van Regenmortel (2004). An extended abstract has been published in Wilmott
Magazine, see Pietersz, Pelsser & van Regenmortel (2005).
Chapter 6 presents novel empirical comparisons on the performance of models in terms
of reduction of risk. The profit and loss (P&L) of hedge portfolios of Bermudan swaptions
are recorded for USD swap rates and swaptions data over a one year period. We compare
LIBOR and swap market models, and the Markov-functional model of Hunt, Kennedy
& Pelsser (2000). The Markov-functional model is representative of single-factor models,
such as short rate models. The market models are representative of multi-factor interest
rate pricing models. Both market models and Markov-functional models can be calibrated
to relevant interest rate correlations. Therefore, correlation pricing effects can be captured
in both model types. The three main conclusions of the hedge tests are quite remarkable:
First, delta hedging is compared to delta and vega hedging. Delta hedging is the-
oretically justified by the replication argument of Black & Scholes (1973) and Merton
(1973), of continuous trading in the underlying asset. Vega hedging is the offsetting of
volatility risk by trading in underlying options. Vega hedging is not based on a replica-
tion argument, and it is considered a financial engineering trick, widely applied by traders
and practitioners. We show that delta and vega hedging significantly outperforms delta
hedging, in terms of reduction of variance of P&L.
Second, the algorithm of Longstaff & Schwartz (2001) for estimating the optimal
exercise decision of American options in Monte Carlo is investigated. This algorithm is
required for market models, but not for the Markov-functional model. We show that the
algorithm contains a discontinuity, which renders convergence of finite difference estimates
of risk sensitivities to be slower, see, for example, Glasserman (2004, Section 7.1). The
hedge tests show that the less effective estimation of risk sensitivities adversely affects
reduction of variance of P&L. Moreover, we propose a novel adjustment of the Longstaff &
Schwartz (2001) algorithm, termed constant exercise decision method. With our proposed
modification, a far greater reduction of variance of P&L is attained than with the original
algorithm. The reduction is comparable to the reduction in the Markov-functional model.
Our proposal thus enables market models to function properly as risk management tools
of callable derivatives.
Third, the effect of the number of stochastic factors and correlation specification is
investigated. The hedge tests show no significant differences in terms of reduction of
variance of P&L, across: models, number of factors and correlation specification.
Finally, the effect of smile on pricing is investigated. Volatility smile is the phenomenon
that different Black (1976) implied volatilities are quoted for different strikes of otherwise
equal options. The results show that the impact of smile can be much larger than the
impact of correlation. Also, the impact of smile is similar in both market models and
Markov-functional models.

26
 OUTLINE xxvii

Chapter 6 has been submitted. For a working paper version, see Pietersz & Pelsser
(2005a).
In Chapter 7, new CMS and generic market models are developed, which allow for
ease of volatility calibration for a whole new range of derivatives, such as fixed-maturity
Bermudan swaptions and Bermudan CMS swaptions. CMS and generic market mod-
els allow for a choice of forward rates other than the classical LIBOR and swap rates
in the LIBOR and swap market models, respectively. We present a theoretical result
with necessary and sufficient conditions for an arbitrary structure of forward rates to be
arbitrage-free at all possible states of the model. CMS and generic drift terms for the
forward rates are derived, by use of matrix notation, for both terminal and spot measures.
A fast algorithm is presented that approximately, but accurately, calculates forward rates
over time steps for CMS market models.
Chapter 7 has been submitted. For a working paper version, see Pietersz & van
Regenmortel (2005).

27
28
 Chapter 1

Introduction

Bermudan-style interest rate derivatives are an important class of options. Many banking
and insurance products, such as mortgages, cancellable bonds, and life insurance products,
contain Bermudan interest rate options associated with early redemption or cancellation
of the contract. The abundance of these options makes evident that their proper valuation
and risk measurement are important to banks and insurance companies. Risk measure-
ment allows for offsetting market risk by hedging with underlying liquidly traded assets
and options.
The purpose of this thesis is to further knowledge of efficient valuation and risk man-
agement of Bermudan-style interest rate derivatives. In this chapter, we provide a historic
background and comprehensive framework for the chapters that are to follow.
The outline of this chapter is as follows. First, we introduce the use of models for
arbitrage-free pricing. Second, we briefly describe interest rate markets and options.
Third, we provide an overview of interest rate derivatives pricing models relevant to the
thesis. Fourth, American option pricing with Monte Carlo simulation is discussed.

1.1 Arbitrage-free pricing

In this thesis, pricing models produce relative valuations. The relative valuation of an
asset (most often a derivative) is in terms of other asset prices. Pricing models are thus
viewed as an ‘extrapolation tool’ that aim to extrapolate derivative prices from underlying
assets.
Key to relative valuation models is the exclusion of arbitrage. An arbitrage is an
opportunity to make a risk-less profit with positive probability, with no costs at time
of execution. If an arbitrage opportunity occurs, then many investors buy the arbitrage
opportunity, driving up the arbitrage price. Eventually, this causes the arbitrage to

29
 2 CHAPTER 1. INTRODUCTION

disappear. Arbitrage opportunities are therefore not likely to occur in an efficient and
competitive economy.
When we construct a relative valuation model, then we usually do so by specifying
dynamics of certain base asset prices in a frictionless arbitrage-free market. We then
consider, for example, derivatives whose values derive from these base assets. A self-
financing portfolio is a portfolio without injection or withdrawal of funds. A derivative
that we added to the model is said to be attainable if its payoff can be exactly replicated
by a dynamically managed self-financing portfolio of the base assets. We call a model
complete if all the derivatives that we added to the model are attainable. In a complete
model, any added derivative is effectively redundant (in this theoretical world), since it
is merely a particular dynamic portfolio of the base assets. The added derivatives are
said to be spanned by the underlying assets. In a complete model, any derivative price
is already known when the current underlying prices and their dynamics are known: The
current derivative price is simply equal to the current value of a replicating portfolio.
Next to absence of arbitrage, we also assume absence of transaction costs: there is
no difference in price when buying or selling an asset. This assumption obviously does
not reflect reality. In the presence of transaction costs, arbitrageurs cannot exploit all
theoretical arbitrage opportunities, since transaction costs make some of these no longer
profitable. However, for large market participants, transaction costs are sufficiently low
relative to transaction sizes. In effect, the assumption of zero transaction costs is quite
accurate for the market as a whole. Moreover, the assumption of absence of transaction
costs leads to a theory that is still sufficiently accurate, but much more tractable.
Model dynamics of returns on asset prices, in the real-world measure, consist of two
parts: An average part (drift) and a random part (diffusion). The diffusion part can be
modelled as either continuous (e.g., Brownian motion) or discontinuous (e.g., jumps as in
Merton (1976)). The use of continuous diffusion models is widespread, foremost because
continuous models provide a more than sufficiently accurate description of reality, and
also because of analytical tractability. This thesis therefore considers only continuous
diffusion models. If it is then assumed that asset returns over disjoint time periods are
independent, then it follows from the Lévy-Khinchin theorem that the diffusion term is a
(time- and state-dependent) volatility coefficient times a Brownian motion.
We summarize these concepts in terms of stochastic differential equations (SDEs). We
consider a model with a filtered probability space (Ω, P, F) with filtration F = (F(t))t≥0 ,
on which is defined a F-adapted Brownian motion w. Here, P denotes the real-world
measure. The asset price is denoted by s, its F-adapted drift by µ and its F-adapted
volatility by σ. A realization in Ω is denoted by ω. We have:

ds(t)
= µ(t, ω)dt + σ(t, ω)dw(t).
s(t)

30
 1.1. ARBITRAGE-FREE PRICING 3

Also, we assume the existence of a money market account with price b and return r(t, ω):

db(t)
= r(t, ω)dt.
b(t)

The market price of risk (or Sharpe ratio, see Sharpe (1964)) is defined to be the
excess average return over the risk free rate divided by the volatility of the asset. In other
words, it is (drift-[risk free rate])/volatility and (µ − r)/σ, see, e.g., Baxter & Rennie
(1996, page 119). A sufficient condition for no-arbitrage is equality of market prices of
risk for all assets, e.g., Hull (2000, Equation (19.6)). The actual levels of market prices of
risk then turn out not to matter for valuation.
We introduce some key concepts of arbitrage-free pricing. A numeraire is an asset
with a strictly positive value at all times. Asset prices may be denominated in terms
of amounts of the numeraire. A martingale is a process with zero drift. Suppose we
can construct a new measure (the so-called risk neutral measure) such that all numeraire-
expressed asset prices become martingales. It can be shown (e.g., Hunt & Kennedy (2000,
Theorem 7.32)) that the assumption of existence of such a measure automatically implies
that the model is arbitrage-free. Moreover, if there exists a single unique risk neutral
measure, then it can be shown (e.g., Hunt & Kennedy (2000, Theorem 7.41)) that the
model is complete, i.e., every derivative security is attainable by a replicating portfolio in
the underlying assets.
Given the assumption of equality of all market prices of risk, we may apply Girsanov’s
theorem (e.g., Øksendal (1998, Theorem 8.6.4)) to construct the risk-neutral measure.
Under the risk neutral measure, market prices of risk then turn out to disappear. Therefore
these do not affect arbitrage-free pricing.
The martingale property of numeraire-relative asset prices implies that their future
expectations take today’s value. If s and n denote prices of an asset and a numeraire,
respectively, then · ¸
s(0) s(t)
=E , for t ≥ 0.
n(0) n(t)
where the expectation is with respect to the risk-neutral measure. The price v of a
derivative (also an asset), necessarily sharing the same market price of risk, then satisfies
· ¸
v(t)
v(0) = n(0)E , for t ≥ 0, (1.1)
n(t)
which is the fundamental arbitrage-free pricing formula, see, e.g., Björk (2004, Theorem
10.18). We note that the market price of risk does not occur in this formula. If we calculate
(1.1) for a call option on a stock that is modelled as geometric Brownian motion, then we
obtain the famous formula of Black & Scholes (1973).

31
 4 CHAPTER 1. INTRODUCTION

In practice, we directly model under a risk-neutral measure, by specifying the part

that stems from the real-world measure, the diffusion part. No-arbitrage requirements
then fully fix the drift term. In fact, for European call and put options, traders quote
the diffusion part: so-called implied volatility. It is the volatility to be used in the Black-
Scholes formula to obtain the European option price. A European option is an option
that is exercisable only at a single point in time (usually during a single trading day).
The reason that the derivative price is fully fixed by (1.1) is replication by a self-
financing portfolio. Suppose amounts δs of s and δb of b are held, then the value v is given
by:
v = sδs + bδb , (1.2)
The change in value v of a self-financing portfolio thus satisfies (e.g, Joshi (2003a, Equa-
tions (5.58) and (5.59))):
dv = δs ds + δb db. (1.3)
In other words, value changes due to trading in the asset s and risk-free asset b cancel
exactly:
sdδs + bdδb = 0.
A replicating portfolio is a dynamically managed self-financing portfolio of underlying
assets s and b, which has a value equal to the payoff of derivative v, in all possible future
states of the economy.
The replicating portfolio holds a dynamic amount of δs of the underlying asset s. The
amount δs can be calculated from (1.2), (1.3), and from Itô’s formula: If s is a stochastic
process, and v : R → R is a function, then for the process v(s) we have:

∂v 1 ∂2v
dv = ds + dhsi, (1.4)
∂s 2 ∂s2
see, for example, Karatzas & Shreve (1991, Theorem 3.3.3). Here, angle brackets h·i
denote quadratic variation, see, e.g., Øksendal (1998, Exercise 2.17). By rewriting (1.4)
in the form of (1.3), we can show that

∂v
δs = . (1.5)
∂s
The quantity δs is called the delta, and it is an example of a risk sensitivity: the risk
sensitivity with respect to the underlying asset price.

1.1.1 Use of models in practice

A dynamic hedge of the derivative v consists of taking an opposite position in the repli-
cating portfolio. In practice, the hedge is not re-balanced on a continuous basis, rather at

32
 1.1. ARBITRAGE-FREE PRICING 5

discrete points in time. Re-balancing usually takes place when delta risk exceeds a cer-
tain threshold level. The adaption from continuous-time to discrete-time hedging works
extremely well in practice, and forms the basis for the success of arbitrage-free pricing
models.
While we use risk-neutral pricing models as relative valuation and hedging tools, it is
interesting to note that these models also make assertions on real-world price dynamics,
through the connection with the real-world measure. Risk-neutral models can thus also
be viewed as economic models, attempting to model economic reality. Though modelling
of real-world price dynamics is a vital aspect of risk-neutral models, it is not the most
important aspect. More important to risk-neutral models are:

1. To produce sensible prices of derivatives, and to reproduce prices of underlying

assets and options exactly.

2. To adequately reduce variance of profit and loss (P&L), when a hedge is set up.

3. To efficiently produce prices, i.e., within a limited amount of computational time.

The difference between the use of a model as a hedging tool or economic model has
important implications. For example, consider modelling the term structure of interest
rates. Extensive empirical research has shown that the term structure is driven by more
than one stochastic factors, see the review article of Dai & Singleton (2003). For an
economic model, we should thus use at least two factors, and a single-factor model is
simply not acceptable. However, for a model used as hedging tool, it is perfectly sensible
to consider a single-factor model, as long as it satisfies the above three properties.
The necessity that pricing models need reproduce prices of underlying assets and
options has two further implications:
First, option price data (implied volatility) determines the diffusion part (or, equiv-
alently, volatility) of the model, rather than time-series estimates from historic data on
asset returns. The reason is straightforward: Certain features of a derivative may become
redundant during the life of the derivative, which may render the derivative equivalent to
(or almost equivalent to) a market traded option. In that case, the model should produce
a derivative price equal to the market traded option price, otherwise the financial institu-
tion holding the derivative can incur arbitrage, which should be avoided. The only way
to avoid arbitrage is to make the model consistent with prices of underlying options, i.e.,
to use implied volatility for the diffusion term in the model. The process of making the
model consistent with market prices is called calibration.
Second, a pricing model is re-calibrated to the most recent implied volatility data,
whenever the derivative needs be valued. The important reason is again that of no-
arbitrage, and is the same as above for the use of implied volatility in models. Implied

33
 6 CHAPTER 1. INTRODUCTION

volatility quotes change over time. The practice of re-calibration to unpredictably chang-
ing volatility is not consistent with most pricing models (excluding stochastic volatility
models), since most models assume volatility to be known over the model time horizon.
When implied volatility changes, then a derivative value may change too, due to the
practice of re-calibration. As a result, derivative traders face volatility risk (vega). The
risk may be offset by vega hedging. If σ denotes the volatility, and o the price of an
underlying option, then the following portfolio has zero volatility risk (for small changes
in σ):
∂v/∂σ
one derivative (v) and − options (o). (1.6)
∂o/∂σ
A delta hedge with the underlying asset as in (1.5) can then be applied to the vega-neutral
portfolio in (1.6).
Vega hedging is out-of-model hedging, since we hedge parameters that are input to
the model. Delta hedging with the underlying asset in (1.5) is in-model hedging, since
the underlying asset price is a state variable of the model. Nonetheless, vega hedging
is not inconsistent with arbitrage-free pricing models: We are only holding a different
portfolio of derivatives and options that needs to be delta hedged. From an arbitrage-free
pricing perspective though, there is simply no need to add the additional options (though
such addition is allowed): the original derivative is already perfectly delta-replicable in
the theoretical model world. In practice however, vega hedging enables a significant
additional reduction of variance of P&L, and it thus contributes to wealth preservation.
Another practice for arbitrage-free pricing models is that of customizing a model to a
certain product. We construct the model in such way that all parts of economic reality,
relevant to the product, are incorporated into the model. The benefit is that the product
is priced correctly, while not having to fully model all parts of the market, thereby often
attaining a more efficient implementation.

1.2 Interest rate markets and options

In interest rate markets, participants trade primarily in interest rate agreements. An
interest rate agreement is an agreement to borrow or lend money, over an agreed period
of time, against agreed periodical payments (interest rate payments) that are in some
form denoted as a percentage (interest rate) of the underlying borrowed or lent amount
(notional amount).
We refer to the length of an interest rate agreement as its tenor. Different tenors may
attract different interest rates, which gives rise to the so-called term structure of interest
rates.

34
 1.2. INTEREST RATE MARKETS AND OPTIONS 7

The above description of an interest rate agreement includes money market deposits,
bonds, forward interest rate agreements, and swaps. Money market deposits, bonds, and
swaps are of particular relevance to this thesis, therefore we explain their workings in
some detail.

1.2.1 Linear products: Deposits, bonds, and swaps

Money market deposits usually have a maturity of one year or less. The two parties
agree on an interest rate and one party deposits the notional amount. At the end of
the agreement (at maturity), the other party returns the notional and makes the agreed
interest rate payments.
A bond is an agreement between two parties, the borrower and the lender, on a
designated notional amount. At initiation, the borrower receives a pre-negotiated amount
for the bond (not necessarily equal to the notional amount). During the life of the bond,
the borrower makes coupon payments on the notional amount, usually on the basis of a
fixed contractually agreed rate, but the coupon payments could also be based on a floating
interest rate. By a floating interest rate, we mean the prevailing market interest rate for
the tenor spacing between the floating interest rate payments. We discuss the method for
determining this floating interest rate below. If there are no coupon payments during the
life of the bond, then we call such a bond a zero coupon bond. At maturity of a bond, the
borrower returns the notional amount to the lender.
Interest rate swaps typically have a maturity of two years or more. Interest rate swaps
involve only exchanges of interest rate payments, but normally do not involve exchanges
of notional. The two parties agree on an interest rate. Periodically, one party pays this
agreed interest rate (the fixed rate), while the other party pays a floating interest rate. We
remark that the frequency of fixed and floating payments may differ. Typical frequencies
are annually, semi-annually, quarterly, and monthly.
The fixed rate at which market participants can enter into a swap agreement at other-
wise zero cost is called the swap rate. Swap rates can be seen as long term borrowing and
lending rates. In fact, the swap rate for a swap with a particular tenor is more or less the
interest rate for that particular tenor. The reason is that a swap can be used to create a
synthetic borrowing or lending agreement at a single interest rate over the tenor period
of the swap: Suppose we borrow money through deposits with floating interest rates, and
we enter into a swap in which we pay fixed, on a notional equal to the amount borrowed
from the deposit. At the end of each deposit, we borrow the same amount again in the
deposit market, in order to pay back the notional from the previous deposit agreement.
Rolling over money market deposits in such way, the resulting deposit interest payments
cancel against the floating interest we receive from the swap; the fixed swap payments
remain. Effectively, we then pay a fixed interest rate on our loan over the life of the swap.

35
 8 CHAPTER 1. INTRODUCTION

The two parties in a swap determine the floating interest rate usually via a reference
interest rate. Reference rates are used to calculate payments not only of swaps, but also
of other securities, such as interest rate derivatives. A reference interest rate is a rate
that is set by a financial authority or calculation agent. Examples are:

LIBOR: London inter-bank offered rate, published by the British Bankers’ Association
(BBA), each trading day at noon (12.00am) London time.

EURIBOR: Euro inter-bank offered rate, published by the European Banking Federation
(FBE) and by the Financial Markets Association (ACI), each trading day at around
11.00am central European time.

These reference rates are published for several tenors and currencies. Upon publication
of the reference rate, practitioners say that the rate then fixes.
Financial authorities determine reference rates normally along the following lines: A
number of panel banks are consulted. Each panel bank provides rates at which it conceives
it possible to borrow money in the inter-bank market, for various tenors and currencies.
For each tenor and currency, some percentile of the top and bottom of the quotes are
discarded. The remaining quotes are averaged to form the reference rate for that tenor
and currency. It is interesting to note there is now an interest rate derivatives pricing
model that bears the name of a reference rate: the LIBOR market model, see Section 1.3.
An interesting note is that the first major swap took place only in 1981, between IBM
and the World Bank, see Valdez (1997, pages 269–270).

1.2.2 Interest rate options: Caps, floors, and swaptions

The plain-vanilla European interest rate options most relevant to this thesis are (i) caps
and floors, and, (ii) swaptions. A cap consists of a sequence of consecutive caplets, and,
likewise, a floor consists of a sequence of consecutive floorlets. Caplets and floorlets are
call and put options, respectively, on LIBOR rates. Swaptions are options on swap rates.
A caplet (respectively, a floorlet) gives its holder at expiry the right, but not the
obligation, to enter into a borrowing deposit (lending deposit) at a pre-arranged strike
rate. If an option holder claims the option right, then we say that he or she exercises
the option. If LIBOR fixes below (above) the respective strike rate, then it is cheaper to
borrow (more rewarding to lend) in the market; whereby it is sensible not to exercise the
caplet (floorlet) and it ends worthless. If LIBOR fixes above (below) the respective strike
rate, then it is sensible to exercise the caplet (floorlet), since we then receive the positive
difference LIBOR minus strike (strike minus LIBOR) at the deposit payment date. The
option gains at the deposit payment date as dependent on realized LIBOR are displayed
in Figure 1.1.

36
 1.2. INTEREST RATE MARKETS AND OPTIONS 9

Caplet Floorlet

Strike Strike
rate LIBOR→ rate LIBOR→
Figure 1.1: Payoffs of caplets and floorlets versus realized LIBOR.

37
 10 CHAPTER 1. INTRODUCTION

A European swaption gives its holder at expiry the right, but not the obligation, to
enter into a swap with a fixed rate equal to a pre-arranged strike rate. Market participants
invariably indicate the direction of swap cash flows from the point of view of whether fixed
payments are payed or received. Thus, if we hold a swaption that gives us the right to
enter into a swap for which we pay or receive fixed, then such a swaption is a payer or
receiver swaption, respectively. A payer (respectively, a receiver) swaption corresponds
to a call option (put option). The payoff structures for payer and receiver swaptions are
similar to those of caplets and floorlets in Figure 1.1: instead of ‘LIBOR’ read ‘swap rate’,
and instead of ‘caplet’ or ‘floorlet’ read ‘payer swaption’ or ‘receiver swaption’.
Cash settled contracts differ from normal option contracts, in that they pay the relevant
difference between realized rate and strike, if this is positive. For a cash settled swaption
at expiry, both parties need to agree on ‘the’ swap rate manifest in the market. Usually,
again a reference rate is used. An example of a reference swap rate is ISDAFIX, published
for various tenors and currencies by the International Swaps and Derivatives Association
(ISDA). Financial authorities calculate swap reference rates more or less in the same way
as deposit reference rates are set; via consultation of a group of panel banks.
From Figure 1.1, we find that an option always provides a nonnegative cash flow, with a
positive probability to provide a positive cash flow, therefore we require a positive premium
for the option. To calculate cap and floor or swaption premiums, market participants
initially used a Black-type formula that is based on assuming a log-normal distribution for
the LIBOR or swap rate at expiry. This approach however lacked a theoretical justification
for long, but the approach later turned out to be valid. Moreover, an assumption of
jointly log-normal LIBOR and swap rates is inconsistent. Many researchers therefore
considered the use of the Black formula for caps and floors or swaptions to be unsound
for a considerable period: While the Black swaptions approach had already been justified
in 1990, articles establishing its validity kept appearing at least until 1997, see Rebonato
(2004a, Section 4(d)).
We present the Black approach for caps and floors; the approach is similar for swap-
tions. Prior to that, we introduce some terminology: We set out by specifying a tenor
structure, 0 = t0 < t1 < · · · < tn+1 . Let αi denote the day count fraction for the period
[ti , ti+1 ]. A discount bond is a hypothetical security that pays one unit of currency at its
maturity, and has no other cash flows. The time-t price of a discount bond with maturity
ti is denoted by bi (t).
To present the Black approach for caps and floors, we consider two tenor points 0 <
t1 < t2 . LIBOR fixes at time t1 and interest is paid at time t2 . We define the forward
LIBOR rate f1 (t) by
b1 (t) − b2 (t)
f1 (t) = . (1.7)
α1 b2 (t)

38
 1.3. INTEREST RATE DERIVATIVES PRICING MODELS 11

We consider the forward measure, which is the measure associated with the discount bond
b2 maturing at the payment date of the LIBOR deposit. By the assumption of absence
of arbitrage, it follows that f1 is a martingale under its forward measure. To see this,
note that f1 (t) in (1.7) is the value of a portfolio ((b1 (t) − b2 (t))/α1 ) expressed in terms
of amounts of the numeraire b2 (t).
Continuing, we assume that the forward LIBOR rate is a log-normal martingale under
its forward measure:
µ ¶
df1 (t) 1
= σ1 dw(t), equivalently, f1 (t) = f1 (0) exp − σ12 t + σ1 w(t) , (1.8)
f1 (t) 2

where σ1 is a scalar constant. The payoff v(t2 ) at time t2 of a caplet with strike rate k is
then given by α1 max(f1 (t1 ) − k, 0). From (1.1), we find for the caplet value v(0):
· ¸ · ¸
v(t2 ) α1 max(f1 (t1 ) − k, 0)
v(0) = n(0)E = b2 (0)E
n(t2 ) b2 (t2 )
(∗) £ ¤
= α1 b2 (0)E max(f1 (t1 ) − k, 0) . (1.9)

Equality (∗) holds since b2 (t2 ) = 1. If we calculate (1.9) in full, we obtain the formula of
Black (1976):
n ¡ ¢ ¡ ¢o
v(0) = α1 b2 (0)η f1 (0)φ ηd1 − kφ ηd2 ,
³ ´
ln f1k(0) ± 12 σ12 t1
d1,2 = √ .
σ1 t1
Here, η denotes +1 for a caplet, and −1 for a floorlet; φ(·) denotes the cumulative normal
distribution function.

1.3 Interest rate derivatives pricing models

Up to here we have examined options on a single interest rate, such as caps, floors and
European swaptions. There are however many interesting interest rate derivatives that
depend not only on a single interest rate, but on multiple interest rates. Examples in-
clude Bermudan-style interest rate derivatives. Bermudan means that the derivative is
exercisable (equivalently: callable) at multiple discrete time points, usually separated by,
e.g., annual or semi-annual periods.
Exercise of a Bermudan derivative is a trade-off between taking the option gains now
or holding onto the option at possibly more favourable option gains later. Inherently,
values of Bermudan interest rate derivatives therefore depend on multiple interest rates.
There are of course also many non-Bermudan interest rate derivatives that dependent on

39
Exploring the Variety of Random
Documents with Different Content
“I don’t see why. To me it seems delightfully restful.”
From an ancient horse-hair trunk he brought forth a box, and seating
himself at her feet, emptied its contents upon the floor.
“This is why,” and he arranged in parallel lines the little leaden
soldiers, diminutive cannons, some with wheels and some without,
and a quantity of dominos, two by two. “These are troops, and if you
care to know how I passed the rainy days of boyhood this will show
you.”
“But, what are the dominos?”
“They are the enemy. These lead soldiers are mine, and they are all
veterans, and all brave. This is myself,” and he held up a bent and
battered relic on a three-legged horse.
“And who are you in these fights, Goosey?”
“Napoleon, generally; often Cæsar and Frederick, and sometimes
George Washington and General Lee.”
“But you have no head. Isn’t that a drawback for a commander?”
“Not with troops like these. I lost that head at Quebec, as Montcalm.”
She looked down upon him with a wish that she also might have
been one of those absurd little soldiers and shared his victories.
“The cracks between the floor-boards,” he continued, “are railroads,
rivers, canals, stone walls, or mountain ranges, according to the
campaign.”
“They must have been a nuisance, though. Could not a soldier
disappear and not return?”
“I should say he could! Why, those ravines are gorged with heroes,
and that recalls the most humiliating event of my career. I was
leading the charge of the Light Brigade, six of these cavalrymen,
each representing a hundred men. I of course was in front, and it
was a supreme moment. As we dashed across the open field—the
cracks, mind you, didn’t count this time—I, the leader, suddenly
disappeared, head downward, feet up, in an open field! Of course
the charge could not stop, and the others rushed on to a magnificent
death.”
With a sigh he gathered the motley company together again, and laid
them away in their box. She got up and moved about. “I should like
to live in an attic. It is mysterious and poetic, and so crammed with
history. Each of these things has its little story for somebody,” and
she stopped before a curious feminine garment in India silk, of a
long-ago fashion.
Pointing to a quaint old cap with ear-laps, she exclaimed, “What a
funny rig that is! Put it on.” And she took it from its peg and placed it
upon his head, then laughed and led him to a broken mirror that was
hanging from a rafter. “Unless you wear it in New York next winter, I
shall never marry you!”
“Then I promise, but at present it is a trifle warm.”
As he removed it a letter slipped from the lining and fell to the floor.
She picked it up and turned it over in her fingers. “Why, it has never
been opened! It is directed to Mr. Josiah Judd.”
Amos examined it, studied the date, then looked at the old cap. “He
wore this at the time of his death, when he had just come from the
post-office, and the Daleford postmark says December fifth, the very
day before. That is very curious.” And he stood looking down at the
letter, deep in thought.
“Why don’t you open it? You are the one who should do it, I
suppose.”
“Yes, I suppose so.”
“Where is it from?”
“India. From Mr. Morton Judd, his brother, the one who sent me
here.”
“Oh, yes! I remember. Is Mr. Morton Judd alive?”
“No, he died ten years ago.”
“Well, please open it, for it may be interesting. Come over near the
light.”
As they stood by the open window, leaning against the sill, he tore
open the envelope and began reading aloud, she looking idly out
upon some haymakers in a neighboring field. Their voices came
faintly to her ears, and they made a pleasant picture in the afternoon
sunlight with the village spires, the tall elms, and the purple hills for a
background. She wondered if India was at all like New England.
DEAR Josiah: The case ought to reach you about a
fortnight after this letter, and if you will write to Mr.
Wharton, or better still, visit him, he will see that there is
no trouble at the Custom House. Give my love to Sarah,
but don’t show her the shawl and the silks before her
birthday, in January. What you say about the boy Amos
does not surprise me, and I was only waiting for you to
make your own discoveries. He gave clear indications
when a very small child of this same faculty in which his
mother and the rest of his family had great faith. In the box
you will receive I send a book giving an account of the
Rajah Sirdar Sing, his ancestor, a hero of prophetic
powers who died ninety-eight years ago, so this boy,
according to tradition, should inherit the same
supernatural faculties. Be careful that he does not see this
book before coming of age, as it might put dangerous
ideas into his head, and if he should suspect what he
really is great mischief might ensue. I am glad he is
turning out such a sensible boy. But if he should ever
come over here and make himself known it would cause a
great disturbance, and might result fatally to himself. Am
sorry to hear about Phil Bates’s wife. She was a fool to
marry him. Your affectionate brother,
Morton Judd.
Amos stood looking down at the letter and remained silent. She laid
a hand upon his arm and said, “What does it mean, Amos, about not
letting you know who you are? Who are you?”
He looked up with a smile. “I don’t know; I can only guess.”
“Well, what do you guess?”
“I guess that I am the rajah of that province.”
“Really? Why, you don’t mean it! And have you always known it?”
“I don’t know it now, but I have always suspected it.”
“You funny old thing! Why, this is awfully exciting! And you never told
me!”
“Why should I? Your father would only have hastened my departure
if I had tried to pass myself off as a fairy prince; and you would have
laughed in my face.”
“No. I am not so sure. But that was long ago, and to-day I should
believe anything you told me.”
“Well, I believe you would,” and there, at the open window, he put his
arm about her waist and did that unnecessary thing true lovers seem
unable to resist. She jumped away to turn with an anxious face and
look cautiously through the window. But the distant haymakers gave
no signs of having received a shock.
“Could they have seen?” she demanded.
He looked over upon the sunlit field. “No, poor things, they missed
it!”
But Molly moved away and seated herself upon a venerable little
horse-hair trunk whose bald spots were numerous and of
considerable extent. Brass-headed nails, now black with age,
studded all its edges and formed at each end the initials of Josiah
Judd.
“Tell me, little Amos, what happened to you as a child, that you
should consider yourself a fairy prince.”
The trunk was short for two, but Amos, by a little pushing and
crowding, managed to sit beside her.
“Well, in the first place, I was always too wise and too amiable for an
ordinary mor—”
“No, no! Be serious.”
“Well, almost everything I remember seems to point in that direction.
For instance, there was a separate seat for me on swell occasions; a
sort of throne, I should say, and all the other people stood up. In the
big hall I told you about where the fight took place, I used to sit in an
ivory chair with gold ornaments on it, cocked up on a platform apart
from other people. And that afternoon I was walking across the hall
toward it when the fierce-looking chap with the beard caught me up
and passed me along.”
“Gracious! This is very exciting! Go on.”
“I could give you this sort of stuff by the yard if the conditions were
favorable. The conditions now are unfavorable.”
Their eyes met, but experience had taught her caution. “Go on.
There are no rajahs in America, and you will do as I tell you.”
“That is very true, but we are too far apart.”
“And all the while you are crowding me off this trunk!”
“Yes, but at the same time I am holding you on. Do you see that old
rocking-chair over there with one arm that is beckoning to us?”
There followed a brief, illogical discussion, then finally a gentle force
was used by the stronger party, and a moment later the old chair
groaned beneath a heavier burden than it had borne for thirty years.
After persistent urging the reminiscences were continued. “They
always helped me first at table, no matter how old the other guests
were, or how many or how swell. The bowing and saluting was much
more elaborate toward me than toward anyone else, and in
processions they always stuck me in front. Shortly after my father
died there was a grand ceremony in a sort of courtyard with awnings
over us, and I remember what an everlasting affair it was, and how
my uncle and an old general stood behind my chair, while all the
swells and panjandrums came up and saluted me, then passed
along. I should say there might have been a million. I know I went to
sleep and my uncle kept tapping me on the shoulder to keep me
awake.”
“You poor little thing! But you must really have been something
tremendously important, mustn’t you?”
“It seems so.”
“Well, go on.”
“After that there were some big reviews, and I sat on a white pony
with officers in a semicircle behind me, while the troops marched by,
and the generals and colonels all saluted. That was great fun. And I
shall never forget my saddle of crimson leather with the gold
trimmings.”
“How romantic! Why, it seems impossible!”
“Do you remember the head-dress in my mother’s miniature?”
“Yes.”
“Well, I find that sort of thing is only worn by royalty.”
There was a pause, during which the old chair rocked gently to and
fro, but noisily, as if in protest against its double burden, while the
voices from the neighboring field came drifting in the window and
with them the occasional tinkling of a cow-bell.
“And to think of your being here in Connecticut, a farmer!”
“Thank heaven I am!” and there followed one of those foolish but
apparently enjoyable scenes which no dignified historian is expected
to describe. Stepping away from the rocking-chair Molly turned with
a frown upon its remaining occupant as she pressed an escaping
lock into position. Through the open window the setting sun sent a
bar of light across the attic that illumined her hair with a golden
touch.
“We must find that book,” she exclaimed, with an impatient gesture.
“It will tell us the very things we wish to know. Come, get up, and
hunt!”
Slowly rocking, with his head resting against the chair, he regarded
her with admiring eyes, but showed no signs of haste. “There is but
one book I care to study, and that is a poem in pink, about five feet
six in length, with gilt edges at the top.”
She smiled sadly. “No, not a poem, but very ordinary prose, and you
will get precious little wisdom from studying it.”
“On the contrary, every page is a revelation. Why, the binding alone
is a poem! Merely to hold it in one’s lap and look at the cover is a
gentle intoxication.”
Wavering between a smile and a frown, she answered:
“I wonder if all rajahs are such transparent flatterers. But come! Find
the book! It must be downstairs in the library.”
“No, it is not down there. I know every book among them.”
“Where can it be, then? tucked away in some trunk or drawer?”
“Probably.”
“Could it be in that?” and she pointed to an old cherry-wood desk just
behind him. He turned and regarded it.
“As likely there as anywhere. It is the desk he used until he died.”
Molly opened the slanting top and found an array of pigeonholes
filled with old papers. There were some very small drawers, all of
which she opened, but they contained no book, so she closed the
top and opened the long upper drawer. It was almost empty, the only
contents being a few envelopes of seeds, some tools, scattered
cards, and a couple of marbles that ran about as the drawer was
opened.
“I rather think you know this place,” and she lifted up a bladeless
jackknife. “Only a boy could treat a knife in such a way.”
“Yes, I remember all those things. That wooden pistol has killed lots
of Indians.”
The second drawer held among other things a camel’s-hair shawl, a
bed-cover, a pair of woman’s slippers, a huge shell-comb elaborately
carved, some black mits, and a package of letters; almost everything
except a book. The third drawer and the fourth were equally
disappointing. The lowest drawer was deeper and heavier, and it
stuck. Amos sprang to help her, and together they pulled it open,
then sat down upon the floor in front of it. The character of its
contents was much like the others, but Molly delved thoroughly
among its treasures and she received her reward. As her hand was
exploring a farther corner she looked up into his face with a look of
excitement.
“Here is a book! It must be the one!” and a little volume was drawn
forth.
“‘The Heroes of India!’ aren’t we in luck!”
It was a handsome little book, with a blue morocco cover and gilt
edges, published in Calcutta. Turning over the leaves with eager
fingers she came to a bookmark opposite a portrait, a steel
engraving, showing the head and shoulders of a bejewelled prince.
“Why, it might be you! It is exactly like you! Look!” and she held it
before him.
“So it is, but perhaps they all are. Let’s hear about him if you are
sure he is our man.”
“Oh, I am sure of it! He is the image of you and the others are not;”
and she began to read.
 “He is the image of you”
“Of all the royal families in India, none claim an existence
more remote than that of the Maharaja Sirdar Oumra Sing.
According to accepted history and tradition, this princely
 house not only dates back to the earliest centuries of
Eastern history, but owes its origin to the immortal Vishn’u
himself. It is a romantic story, in fact the survival of an
ancient fable, poetic and supernatural, but, curiously
enough, seems to be substantiated by the extraordinary
attributes of a recent ruler. The Rajah Sirdar Sing, whose
portrait heads this article, was perhaps the most popular
hero of Northern India, and unless we eject the evidence
of all his contemporaries, was possessed of powers that
brought him the most startling victories both in peace and
war, and over adversaries that were considered invincible.
His kingdom, during his reign of thirty years, was nearly
doubled in territory and enormously increased in wealth. In
his own country to-day there are none who question his
prophetic powers: men of science and of letters,
historians, high priests, lawyers, soldiers, all firmly believe
in his immortal gifts. To us Europeans, however, these
tales are more difficult of acceptance.
“In the very centre of Sirdar Sing’s forehead the reader
may have observed a faint spot scarcely half an inch in
diameter, and this appeared, we are told, like a scar or a
burn, of a lighter color than the skin and, except under
certain conditions, was barely noticeable. But the tradition
runs that when exercising his prophetic faculty this little
spot increased in brilliancy and almost glowed, as if of
flame.”
“And so does yours!” and she regarded him with a look of awe.
“Go ahead,” he said, looking down at the book. “Let us hear the
rest.”
“The legend is this:
“When Vishn’u in his Kr’ishn’a-Avatâra, or eighth
incarnation, was hard-pressed in his war against the
Kurus, he received great assistance from Arjuna, a
Pân’d’u prince who, after a four days’ battle, and at great
risk to himself, delivered to his immortal ally the sacred
 city of Dwârakâ. For this service and in token of his
undying gratitude, Vishn’u laid his finger upon the
forehead of Arjuna and endowed him with a knowledge of
future events, also promising that once in a hundred years
a descendant should possess this priceless gift. Although
we may not accept this romantic tale, there is no doubt
whatever that Sirdar Sing, the original of our portrait, was
guided by a knowledge of the future, either earthly or
divine, which neither scientists nor historians have yet
explained. The next in order to inherit this extraordinary
faculty, if there is truth in the legend, will be the son of the
present rajah, whose nuptials have just been celebrated
with such lavish and magnificent festivities.”
She paused for a moment, then with trembling fingers turned back to
the title-page. The book was printed twenty-eight years ago, the year
before Amos was born.
For a long time they sat on the floor talking; she asking many
questions and he answering, until the listening objects in the attic
began to lose their outline and become a part of the gloom. The
sunlight along the rafters dwindled to a narrow strip, then
disappeared; and the voices of the haymakers were long since gone
when Amos and Molly finally climbed to their feet and descended the
stairs.
 IX
SEPTEMBER brought other guests, and with their arrival Amos Judd
and Molly Cabot found the easy, irresponsible routine of their happy
summer again disturbed. To his own fierce regret Amos could invent
no decent pretext for escaping a visit he had promised early in the
summer, and a more unwilling victim never resigned himself to a
week of pleasure. To the girl he was to leave behind him, he
bewailed the unreasonable cruelty of his friends. “This leaving you,
Soul of my Soul, is worse than death. I shall not eat while I am gone,
and nights I shall sit up and curse.”
But at the end of a week he returned, promptly on the minute. His
moments of depression, however, seemed rather to increase than
diminish, and, although carefully repressed, were visible to a pair of
watchful eyes. Upon his face when in repose there had always been
a melancholy look, which now seemed deepening as from an inward
sorrow, too strong to conquer. This was betrayed occasionally by a
careless speech, but to her questioning he always returned a
cheerful answer. In spite of these heroic efforts to maintain a joyful
front, Molly was not deceived, and it was evident, even to Mr. Cabot,
that the young man was either ill in body or the victim of a mental
disturbance that might be disastrous in its results. Of this he was
destined to have a closer knowledge than his daughter. It came
about one Sunday morning, when the two men had climbed a
neighboring hill for a view which Mr. Cabot had postponed from week
to week since early June. This was his last Sunday in Daleford and
his final opportunity.
The view was well worth the climb. The day itself, such a day as
comes oftenest in September, when the clear air is tempered to the
exact degree for human comfort by the rays of a summer sun, was
one in which the most indifferent view could shine without an effort.
Below them, at the foot of the hill, lay the village of Daleford with its
single street. Except the white spires of the churches, little of it could
be seen, however, beneath the four rows of overhanging elms. Off to
their left, a mile or two away, the broad Connecticut, through its
valley of elms, flowed serenely to the sea; and beyond, the changing
hills took on every color from the deepest purple to a golden yellow.
A green valley on their right wandered off among the woods and
hills, and in it the stately avenue of maples they both knew so well. A
silence so absolute and so far-reaching rested upon the scene that,
after a word or two of praise, the two men, from a common impulse,
remained without speaking. As thus they sat under the gentle
influence of a spell which neither cared to break, the notes of an
organ came floating upward from the trees below them, and mingled
with the voices of a choir. Mr. Cabot’s thoughts turned at once to the
friend at his side, whom he felt must experience a yet deeper
impression from these familiar scenes of his childhood. Turning to
express this thought, he was so struck by the look upon Amos’s
face, an expression of such despairing melancholy, that he stopped
in the middle of his sentence. While well aware that these tragic eyes
were always most pathetic objects in repose, he had never seen
upon a human face a clearer token of a hopeless grief.
“What is it, my boy?” he asked, laying a hand upon the knee beside
him. “Tell me. I may be able to help you.”
There was a slight hesitation and a long breath before the answer
came. “I am ashamed to tell you, Mr. Cabot. I value your good
opinion so very much that it comes hard to let you know what a weak
and cowardly thing I have been, and am.”
“Cowardly—that I do not believe. You may be weak; all of us are
that; in fact, it seems to be the distinguishing attribute of the human
family. But out with it, whatever it is. You can trust me.”
“Oh, I know that, sir! If you were only less of a man and more like
myself, it would be easier to do it. But I will tell you the whole story.
By the fourth of November I shall not be alive, and I have known it
for a year.”
Mr. Cabot turned in surprise. “Why do you think that?”
But Amos went on without heeding the question.
“I knew it when I asked Molly to be my wife; and all the time that she
has gone on loving me more and more, I have known it, and done all
I could to make things worse. And now, as the time approaches and I
realize that in a few weeks she will be a broken-hearted woman—for
I have learned what her affection is and how much I am to her—now
I begin to see what I have done. God knows it is hard enough to die
and leave her, but to die only to have played a practical joke on the
girl for whom I would joyfully give a thousand lives if I had them, is
too much.”
He arose, and standing before her father, made a slight gesture as of
surrender and resignation. The older man looked away toward the
distant river, but said nothing.
“Listen, sir, and try to believe me.” Mr. Cabot raised his glance to the
dark face and saw truth and an open heart in the eyes fixed solemnly
upon his own; and he recognized a being transformed by a passion
immeasurably stronger than himself.
“When I found she loved me I could think of nothing else. Why
should I not be happy for the short time I had to live? Her love was
more to me than any earthly thing, than any possible hereafter.
Better one summer with her than to live forever and not have known
her. Oh! I thought of her side of it, often and often; many a night I
have done nothing else, but I could no more give her up than I could
lift this hill.” He paused, drew a long breath, as if at the hopelessness
of words to convey his meaning, then added, very calmly:
“Now I am soberer, as the end approaches, and I love her more than
ever: but I will do whatever you say; anything that will make her
happier. No sacrifice can be too great, and I promise you I will make
it. I have often wished the bull had killed me that day, then I should
have her love and respect forever; and yours too, perhaps.”
“You have both now, Amos. But tell me why you think you are to die
by November fourth?”
Amos resumed his seat upon the rock and answered: “Because I
have seen myself lying dead on that day.”
“I have sometimes wondered,” said Mr. Cabot, “if that temptation
would not prove too strong for you.”
“No, sir, it was not too strong for me under ordinary circumstances,
but it happened when I was not myself, when I came out of that fever
last October, and as I lay in bed, weak and half-conscious, I felt sure
my day had come. I thought the doctor was not telling me the truth,
so, by looking ahead for myself, I learned more than I cared to know,
and saw myself lying on a sofa in a strange room, a place I had
never been into; a public building, I should think.”
“But why do you think it is to be the fourth of November, and this
year?”
“Because I looked about and saw near a window a little day
calendar, and that was the date it bore. Then on a table lay a daily
paper of the day before, and two magazines of the same month, all
of this year.”
“But is it not possible the room is unoccupied and that these things
have been lying there indefinitely?”
Amos shook his head. “No, sir, it is a room that is lived in. There are
other papers lying about: books, and a letter on the desk waiting to
be mailed. And in the fireplace the embers are still glowing.”
Mr. Cabot looked with the profoundest sympathy toward his friend,
who was scaling bits of moss from the rock beside him; then he
turned again to the view and its tranquil beauty seemed a mockery.
In the village below them he could see the congregation pouring out
from a little white church like ants from a loaf of sugar. Mr. Cabot was
not a religious man, and at present there was nothing in his heart
that could be mistaken for resignation. His spirit was in revolt, his
pugnacity aroused, and with this quality he was freely endowed.
Rising to his feet he stood for a moment in silence, with folded arms,
frowning upon the distant hills.
“Amos,” he said, finally, “in spite of bygone defeats I am inclined to
resist this prophecy of yours. You were not absolutely master of your
own mind at the time, and under such conditions nothing would be
easier than to confuse your own imagining with a vision of another
character. At least it is not impossible, and if by good luck you did
happen to confound one with the other we are having our panic for
nothing. Moreover, even if this vision is correct, it need not
necessarily signify an undeviating fulfilment in every detail. It may
indicate the result to be expected in the natural order of events; that
is, if things are allowed to take their course without obstruction or
intervening influences. But it is difficult for me to believe this faculty
is to continue infallible through all your mental and physical
developments and fluctuations of faith, and never, under any
possible conditions, vary a hair’s-breadth from the truth. It is a law of
nature that a disused faculty shall weaken and lose its power, and for
years you have done your best to repress and forget it.”
“Yes, sir, but whenever employed it has been correct.”
“That may be, and its day of failure still remain a probability. In this
present case the prophecy, aside from its uncertain origin, is one
whose fulfilment is more easy to avert than some of the others. You
say the room in which you saw yourself is one you are unfamiliar
with, and consequently is not in Daleford.”
“Oh, no! There is nothing like it in this vicinity.”
“Well, suppose you were to remain in Daleford during the critical
period with two men, nominally visitors at your house, to watch you
day and night and see that you do not escape? Or, better still, let me
send you to an institution in which I am a director, where you will be
confined as a dangerous patient, and where escape, even if you
attempted it, would be as hopeless as from a prison.”
Amos doubted the success of any attempt at foiling fate, or, in other
words, giving the lie to a revelation once received, but he was willing
to do whatever his friend desired. As they walked home they
discussed the plan in detail and decided to act upon it; also to take
every precaution that Molly should be kept in ignorance.
The first week in October the house at the north end of the avenue
was empty and the Cabots were in New York. As the end of the
month approached a little tale was invented to explain the cessation
for a time of Amos’s visits, and early one afternoon the two men got
into a cab and were driven to the outskirts of the city. They entered
the grounds of a well-known institution, were received by the
superintendent and one or two other officials, then, at the request of
the elder visitor, were shown over the entire building and into every
room of any size or importance. When this inspection was over Mr.
Cabot took his companion aside and asked if he had seen the room
they sought. Amos shook his head and replied that no such room
could be within the grounds. A few minutes later the young man was
shown to a chamber where his trunk had preceded him. The two
friends were alone for a moment, and as they separated Amos gave
the hand in his own a final pressure, saying: “Don’t think I am
weakening, Mr. Cabot, but I cannot help feeling that I have seen
Molly for the last time. And if you and I never meet again, you may
be sure my last thoughts were with you both.”
In a cheerful tone the lawyer answered: “I shall listen to no such
sentiments. If your prophecy is correct you are to be lying in a room
outside these grounds on November fourth. No such prophecy can
be carried out. And if the prophecy is incorrect we shall meet for
several years yet. So good-by, my boy. I shall be here the third.”
During ten days Amos was to remain under the strictest watch, to be
guarded by two men at night and by two others in the day-time, and
to be permitted under no conditions to leave that wing of the building.
By the subordinate in charge and by the four guardians he was
believed to be the victim of a suicidal mania. As the fourth of
November approached Mr. Cabot’s thoughts were less upon his
business than with his imprisoned friend. He remembered with what
inexorable force he himself had been held to the fulfilment of a
prediction. He had felt the hand of an unswerving fate; and he had
not forgotten.
But the fourth of November came and went with no serious results,
and when the five succeeding days had safely passed he
experienced a relief which he was very careful to conceal. With
friendly hypocrisy he assumed a perfect confidence in the result of
their course, and he was glad to see that Amos himself began to
realize that anything like a literal fulfilment of his vision was now
improbable.
One week later, the last day of durance, the prisoner and Mr. Cabot
had an interview with Dr. Chapin in the latter’s private office. Dr.
Chapin, the physician in charge, an expert of distinction in mental
disorders, was a man about sixty years of age, short, slight, and
pale, with small eyes, a very large nose, and a narrow, clean-shaven
face. His physical peculiarities were emphasized by a complete
indifference as to the shape or quality of his raiment; his coat was a
consummate misfit, and his trousers were baggy at the knees. Even
the spectacles, which also fitted badly, were never parallel with his
eyes and constantly required an upward shove along his nose. But a
professional intercourse with this gentleman led to a conviction that
his mental outfit bore no relation to his apparel. Mr. Cabot had known
him for years, and Amos felt at once that he was in the presence of a
man of unusual insight. Dr. Chapin spoke calmly and without
pretension, but as one careful of his speech and who knew his facts.
“That you should have made that visit against your will,” he said to
Mr. Cabot in answering a question, “is not difficult to explain as Mr.
Judd unconsciously brought to bear upon your movements a force to
which he himself has repeatedly yielded. If he happens to remember,
I think he will find that his thoughts were with you at that time,” and
he smiled pleasantly on Amos.
“Yes, sir, but only as a matter of interest in the novel experience I
knew Mr. Cabot was going through.”
“Certainly, but if you had forgotten the visit and if you believed at that
moment that he was to go in another direction, Mr. Cabot would have
followed the other thought with equal obedience. This unconscious
control of one intelligence over another is well established and within
certain limits can be explained, but in these affairs science is
compelled to accept a barrier beyond which we can only speculate.
In this case the unusual and the most interesting feature is the
unvarying accuracy of your visions. You have inherited something
from your Eastern ancestors to which a hypothesis can be adjusted,
but which is in fact beyond a scientific explanation. I should not be at
all surprised to find somewhere in the city the room in which you saw
yourself lying; and it is more than probable that, if unrestrained, you
would have discovered it and fulfilled your prophecy, unconsciously
obedient to that irresistible force. A blow, a fall, a stroke of apoplexy
or heart disease; the sudden yielding of your weakest part under a
nervous pressure, could easily bring about the completion of your
picture. Some of the authenticated reports of corresponding cases
are almost incredible. But before you are forty, Mr. Judd, you will find
in these visions a gradual diminution of accuracy and also, as in this
case, that their fulfilment is by no means imperative.”
For Amos there was immense relief in hearing this, especially from
such a source, and he left the building with a lighter heart than he
had known for months. Now that the danger was over, he wished the
wedding to take place at once, but Molly would consent to no
undignified haste. He found, however, an unexpected and influential
ally in her grandmother Jouvenal, just arrived from her home in
Maryland for a month’s visit, and who insisted upon the wedding
taking place while she was with them. Mrs. Jouvenal was a slender
person of sprightly manners, whose long life had been sweetly
tempered by an exaggerated estimate of the importance of her own
family; but in other matters she was reasonable and clear-headed,
endowed with quick perceptions, a ready wit, and one of those
youthful spirits that never grow old. She was interested in all that
went on about her, was never bored and never dull. It was of course
a little disappointing that a girl with such an ancestry as Molly’s, on
her mother’s side, should give herself to an unknown Judd from an
obscure New England village; but her fondness for Amos soon
consoled her for the mésalliance. Molly had a strong desire to
acquaint her grandmother with the ancestral facts of the case, but
Amos refused to give his consent. Those discoveries in the attic he
insisted they must keep to themselves, at least while he was alive.
“When I am transplanted I shall be beyond the reach of terrestrial
snobs, and you can do as you please.”
The first week in December Mrs. Jouvenal was to visit her son in
Boston. “And really, my child,” she said to Molly, “it is the last
wedding in the family I shall be alive to see, and with such an exotic
specimen as you have selected, I shall not be sure of a Christian
ceremony unless I see it myself.”
As her father remained neutral Molly finally yielded, and there was a
wedding the first Wednesday in December.
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookfinal.com