This action might not be possible to undo. Are you sure you want to continue?

# THE FLORIDA STATE UNIVERSITY COLLEGE OF ARTS AND SCIENCES

VALUATION OF MORTALITY DERIVATIVES

By SHELDON ROBINSON

A Thesis submitted to the Department of Mathematics in partial fulﬁllment of the requirements for the degree of Master of Science.

Degree Awarded: Summer Semester, 2010

. .1 Lee-Carter Model . . . . . . . . . . . . . 3 4 Bibliography . . . Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Conclusion . . . . . . . . . . . . . . i . . . . . . . . . . . . . . . . . . . . . .2. 1 Introduction 1. . Background 2. . . . . . . . . . . . . . . . . .2 Stochastic Mortality Models . .1 Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Blake and Dowd Model Framework Results and Conclusion 4. . . . . . . . . . . .1 Stochacstics Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii iii 1 1 2 2 4 4 4 5 6 6 8 9 2 . . . . . . . . . . . . . . . . . . 2. . . . . . . . . . . . . .CONTENTS List of Figures .1 Viatical and Longevity Risk Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . .2 Cairns. . 2.

. . . Lee-Carter forecast for Survival rate and Mortality rates at (65) and (85) . . 6 7 7 8 ii . . . . .4 Parameter estimates for the Lee-Carter model . . . . . . . . . . .2 4. . . . .3 4. . . . . . . . . . . . . . . . . . . . . . . . C-B-D forecast for Survival rate and Mortality rates at (65) and (85) . . . . . . .LIST OF FIGURES 4. Parameter estimates for the C-B-D model . . . . . .1 4.

iii . Longevity risk is seen as one of the greatest challenge facing pension funds. liable for pensions of fund members and nations. Mortality derivatives provide mechanisms for transfer of longevity risk amongst market participants. liable for old-age beneﬁt schemes.ABSTRACT We look at a framework for pricing mortality products using stochastic mortality models analogous to that for credit products found in.

1. pioneered by JPMorgan is Lifemetrics [5]. The reasons for emergence of such a market. access to data sources monitoring state of insureds and transparency in pricing methodogies 2.CHAPTER 1 INTRODUCTION The discussion of current practice in the viatical market is taken from [3] with particular interest in the sections Life Settlements Securitization by Emmanuel Modu. A life settlement is an insurance policy sold by the owner for an amount greater than the surrender value but less than the face value of the policy. along with a description of a product likely to be traded. hence. The strongest reason for life insurance policies being sold is a need for cash by the insured. The factors that will affect its growth and adoption are 1. A framework under which mortality derivatives are currently priced is the Lifemetrics Toolkit discussed in [5].A Primer by Jonathan Sadowsky and Matthew Browndorf. The purchaser of the life settlement become the new owner and beneﬁciary of the life insurance policy and is repsonsible for making future premium payments and collecting the death beneﬁt of the insured. A further development in the viatical market is life settlement securitization where a life settlement portfolio is aggregated and securitized in a manner similiar to collaterized debt obligations. Commoditization This involves standardization of the mortality models for predicting life expectancies. allows portfolio diversiﬁcation.1 Viatical and Longevity Risk Markets A viatical market is a secondary market for life insurance policies. is found in [6]. Life Settlement Pricing by James Cavoli and Synthetic Longevity Swap . A framework for creation of a longevity risk market. 1 . An example of a mortality derivatives that may be price using the Lifemetrics Toolkit is a q-Forward discussed in [6]. The reason for life settlements being purchased is that it is uncorrelated with other ﬁnancial assets. Regulation involving safeguards of identities and ﬁnancial status of insureds.

ω) ∈ B.d ).3 (Adler [1]).d ≥ 1. t ∈ I} differ by a subset of N Deﬁnition 2. if • P[B0 ∈ Γ] = µ(Γ). Let B = {Bt . t ∈ RN . the tj are points of RN . ω) ∈ B.1.d the σ-ﬁeld containing all sets of the form {g ∈ GN. G N. A random ﬁeld is a collection of random variable {Xt .1.1 Stochacstics Processes Deﬁnition 2. d)-dimensional random ﬁeld or stochastic process to be a measurable mapping X from (Ω. m} where m is an arbitrary integer. where Id is the (d × d) identity matrix. the increment Bt − Bs is independent of Fs and is normally distributed with mean zero and covariance matrix equal to (t − s)Id . j = 1. • for 0 ≤ s < t. t ∈ I ∩ D} (2. t ∈ A}. P). An alternate deﬁnition is Deﬁnition 2. . This process is called a d-dimensional Brownian motion with initial distribution µ. . .d : g(tj ) ∈ Bj . A random ﬁeld is X(t). and the Bj are half-open intervals of Rd . . Then we deﬁne an (N. We further assume that lives are uncorrelated. Let d be a positive integer and µ a probability measure in (Rd . B(Rd )). 2. The aim is to model the underlying mortality process for the insureds as d-dimensional Brownian motion.d denote the set of all Rd -valued functions on RN . adapted process with values in Rd . deﬁned on some probability space (Ω.1.N .2 (Karatzas & Shreve [10]). t ≥ 0} be a continuous. F) into (GN. is said to be separable if there exists a countable set D ⊂ RN and a ﬁxed event N for which P (N) = 0.4 (Karatzas & Shreve [10]). Let GN. F. where A is partially ordered set Deﬁnition 2.1.1) 2 . for all Γ ∈ B(Rd )).CHAPTER 2 BACKGROUND We present a brief overview of Stochastic Random Fields and Levy Processes following the development found in [8] and [1]. such that for any closed interval B ⊂ RN and open interval I ⊂ RN the two sets {ω : X(t. Ft . and {ω : X(t.d .1 (Adler [1]). and G N.

F).k = 0. nd = 0. Kd is a complete separable Hilbert space. . 1 2 d Hence. . d2 ∈ N. i = 1. H) becomes a Banach space. Y >Kd = (2. . T ]. Y >Gd by d T 2 αk (s)βk (s)Bs dPds k=1 0 Ω < X. and. So. Kd becomes a pre-Hilbert Space. Φ(·)k is measurable as a mapping from (Ω. . 3 . Bs independent of Bs . . .1 (Liu [12]).. . >Kd .Y = (β (s)B 1 . . Deﬁnition 2. . let Gd be the collection of d-dimensional Brownian Motion deﬁned by 1 2 d Xs = α1 (s)Bs + α2 (s)Bs + · · · + αd (s)Bs (2. Now. ±2. . in particular. In particular. Let Kd be the collection d-dimensional Brownian Motion deﬁned by 1 2 d Xs = (α1 (s)Bs . β (s)B 2 . L(Kd . T ].< ·. A mapping Φ(·) from Ω into L(K. into (H.5 (Liu [12]). H) is said to be strongly measurable if for arbitrary k ∈ K. Gd . by construction. . Similarly. . n2 . we have the Banach space L(Kd1 . respectively. . . . we deﬁne the inner product < X. n→∞ Thus. . . we can obtain a complete separable Hilbert space. 0 ≤ 1 . . forms a basis for Kd .2) j i i where αi (s) ∈ L2 [0. · >H . for i = j. Y >Gd = (2. ±2.3) With the inner product < . Gd2 ). Kd is dense in Kd [13]. ek > ek || → 0 as n → ∞.5) Following the same arguments used for Kd . . equipped with the usual operator norm · . ek > ek . . d ∈ N. · >K . d. 1 We note that the functions ek (s) = √T eiks .We proceed by constructing a separable random ﬁeld. . k∈Z It follows that x = k∈Z < lim xn . a measurable space. Thus. B(H)).1. . We denote by L(K. ±1. n1 . For Xs = (α1 (s)Bs 2 1 2 d d s s s s s Kd . β (s)B d ) ∈ s ≤ T . form a basis for L2 [0.Y = β1 (s)Bs +β2 (s)Bs +· · ·+βd (s)Bs ∈ Gd . there exist a sequence xn ∈ Kd such that ||x − xn ||Kd → 0 as n → ∞. we have K1 = G1 Applying the following theorem from [12] Theorem 2. . ±1. en1 (s)Bs ⊗ en2 (s)Bs ⊗ · · · ⊗ end (s)Bs . α (s)B d ). Bs are standard Brownian Motion. d1 . We deﬁne the Hilbert space Kd as the completion of Kd . Let K and H be two separable Hilbert spaces with · K .4) 1 2 d d 2 1 and for Xs = α1 (s)Bs +α2 (s)Bs +· · ·+αd (s)Bs ). H) the set of all linear bounded operators from K to H. αd (s)Bs ) (2. . for any x ∈ Kd . K1 ). α (s)B 2 .1. Then set L(K. Y >Kd by d T 2 αk (s)βk (s)Bs dPds k=1 0 Ω < X. Hence. . deﬁne the inner product < X. α2 (s)Bs . . . ||x − < xn . · H and inner products < ·.

k ∈ K. Blake and Dowd Model Here. H) such that Φ(ω)kP (dω) = Ψk. κt are stochastic processes deﬁned by (i) x. H)) be the smallest σ-ﬁeld of subsets of L(K.6) then Φ : Ω → L(K. η 2 ). F ) into the space (L(K.9) where t mx is the mortality rate of (x) at age x + t. dκt = µ(i) + σ (i) dWt .t (1) (2) x.2 2. let F(L(K. t γt = 0. H))). 2. A ∈ B(H) (2.2. i = 1. H) containing all sets of the form {Φ ∈ L(K. This is a two-factor model involving a 2-dimensional stochastic process deﬁned by logit(t qx ) = κt + κt (x − x) + where x is the mean age over the age range used. 2 (2.βx is inﬂuence of cohorts effect on (x).7) Ω . ηt ). (1) (2) κt .Continuing as described in [12]. αx is the average mortality of (x). x βx = 1 (2. H) : Φk ∈ A}. [2] and [9]. H).t (2.t . k ∈ K (2.8) is called the strong Bochner integral of Ψ.10) is a stochastic error assumed to be N (0. Mapping Φ is said to be Bochner integrable. with respect to the measure P if for arbitrary k.t is a stochastic error assumed to be 2 N (0. The operator Ψ denoted by Ψ= Ω Φ(ω)P (dω) (2. 2.2. H) is a strongly measurable mapping from (Ω. a more rigourous discussion is found in [7]. the mapping Φ(·)k is Bochner integrable and there exists a bounded linear operator Ψ ∈ L(K.2 Cairns. The Lee-Carter Model is a single-factor model deﬁned by log(t mx ) = αx + βx γt + x. F(L(K.1 Lee-Carter Model Stochastic Mortality Models This is a brief overview of of the Lee-Carter Model [4]. we provide a brief overview of CBD Model.γt represents a cohort effects and x. see [14].11) 4 .

T ] and Bt is standard Brownian motion.6 to L1 and L2 . a complete separable Hilbert space.4) with k dκi. 2 (3. d (3.1 + κi. . respectively. we can ﬁnd Bt given by ˆ i αx + βx γt + B i Bt = t ηi such that ˆi log t mi = η i Bt x Hence.CHAPTER 3 FRAMEWORK Given a portfolio of d lives. the of bounded linear operator equipped the operator norm || · ||. we have F1 and F2 as σ-ﬁelds of strongly measurable mapping from (Ω. ηt ∈ L2 [0.2) (3. F2 ). ˆi Using Girsanov’s Theorem. hence logit(t qx ) ∈ G2 ⊗d k = 1. K1 ) and L2 = L(G2 .k + σ i. we can deﬁne Banach spaces L1 = L(Kd .5) . d (3. K1 ). 5 . Similarly. x i = 1.1. γt . . . βx are integrable with respect to the measure P .2 (xi − x) + η i Bt . t i i We note that logit(t qx ) ∈ G2 .1. log t mx ∈ Kd . . . Consider the Lee-Carter Model given by i log t mi = αx + βx γt + η i Bt . . ⊗d Using 2.3) i = 1. t t (3.k = µi. . Applying 2. . F1 ) and (L2 .1) i where αx . consider the C-B-D Model given by i logit(t qx ) = κi.k dBt . F) into (L1 .

The results are found in the following pages.1: Parameter estimates for the Lee-Carter model 6 .1 Result We applied the methodolgy outllined in [11] to calibrate the Lee-Carter and Cairns-Blake-Dowd Model to a deaths and population exposures for the United States 1968-2003.CHAPTER 4 RESULTS AND CONCLUSION 4. Figure 4. ages 60-89 (60-84 before 1980).

2: Parameter estimates for the C-B-D model Figure 4.Figure 4.3: Lee-Carter forecast for Survival rate and Mortality rates at (65) and (85) 7 .

2 Conclusion We have constructed spaces of linear bounded linear operators acting on portfolio of d lives with Lee-Carter and Cairns-Blake-Dowd Mortality Model.4: C-B-D forecast for Survival rate and Mortality rates at (65) and (85) 4.Figure 4. 8 .

Sinha J. LifeMetrics: Software user guide. [9] Alen Ong Amit Sinha Javier Hevia-Portocarrero Emily Gingrich Marwa Khalaf-Allah Praveen Joseph Guy Coughlan. Coughlan. JPMorgan. Joseph. 9 . D. Carter and Ronald D. G. 2007. D. 2007. Sinha G. http://gking. Modeling and forecasting us sex differentials in mortality. Population Forecasting. [2] David Blake Andrew J. [5] A. [13] Willard Miller. Lee. Springer. September 2002.harvard. Springer. David Epstein. 8(3):393 – 411. Chapman and Hall. [11] LifeMetrics. A two-factor model for stochastic mortality with parameter uncertainty: Theory and calibration. [4] Lawrence R. 1991. including an application to fractal image compression. Hevia-Portocarrero E. Honig. Epstein and P. [10] Ioannis Karatzas and Steven Shreve. [3] Vishaal Bhuyan. Lecture notes and background materials on lebesgue theory from a hilbert and banach space perspective. Epstein and P. Gingrich M. Journal of Risk and Insurance. 2009. JPMorgan. Brownian Motion and Stochastic Calculus. JPMorgan. 2007. Stability of Inﬁnite Dimensional Stochastic Differential Equations. 73(4):687–718. International Journal of Forecasting. 2007. Coughlan. [12] Kai Liu. Random Fields and Geometry. [6] A. LifeMetrics: A toolkit for measuring and managing mortality and longevity risks Technical Document.edu/ﬁles/abs/lcabs. 1992. 2007. Life Markets: trading mortality and longevity with life settlements and linked securities. 2006. LifeMetrics: A toolkit for measuring and managing mortality and longevity risks. Understanding the lee-carter mortality forecasting method. 2009.BIBLIOGRAPHY [1] Robert Alder and Jonathan Taylor.shtml. SpringerVerlag. JPMorgan. q-Forwards: Derivatives for transferring longevity and mortality risk. Ong A. Malliavin Calculus for Lévy Processes with Applications to Finance. John Wiley and Sons. [7] Federico Girosi and Gary King. Khalaf-Allah G. 2007. [8] Bernt Øksendal Giulia Nunno and Frank Proske. 2007. Cairns and Kevin Dowd.

Topics in Banach Space Integration. World Scientiﬁc. 2005. 10 .[14] Stefan Schwabik and Ye Guoju.