Pricing of Catastrophe Insurance Options under Immediate Loss Reestimation Author(s): Francesca Biagini, Yuliya Bregman and Thilo Meyer-Brandis Source: Journal of Applied Probability, Vol. 45, No. 3 (Sep., 2008), pp. 831-845 Published by: Applied Probability Trust Stable URL: http://www.jstor.org/stable/27595989 . Accessed: 30/04/2013 05:54 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. . Applied Probability Trust is collaborating with JSTOR to digitize, preserve and extend access to Journal of Applied Probability. http://www.jstor.org This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions J.Appl. Prob. 45, 831-845 (2008) Printed in England ? Applied ProbabilityTrust2008 INSURANCE PRICING OF CATASTROPHE OPTIONS UNDER IMMEDIATE LOSS REESTIMATION * **and FRANCESCA BIAGINI YULIYA *** BREGMAN,* THILO Universit?t Ludwig-Maximilians MEYER-BRANDIS,**** University M?nchen of Oslo Abstract the initial estimate of each loss index, where for a catastrophe specify a model loss is reestimated starting from the immediately by a positive martingale catastrophe insurance options random time of loss occurrence. We consider the pricing of catastrophe written on the loss index and obtain option pricing formulae by applying Fourier transform works for loss distributions is that our methodology techniques. An important advantage We with heavy tails, which discuss the case when and provide Keywords: formula; We also tail behavior for catastrophe modeling. factors are given by positive affine martingales of positive affine local martingales. is the appropriate the reestimation a characterization Catastrophe insurance tail; positive heavy loss index; Fourier option; affine martingale transform; option pricing 2000Mathematics Subject Classification:Primary46N30; 60G07; 91B30 1. Introduction Over thepast decades the rise in insured losses has exploded from2.5 billion dollars per year to an average value of the aggregated insurance losses of 30.4 billion dollars per year (2006 prices); see [20] formore details. Table 1 gives a summary of the 10most expensive natural catastrophes for the last 30 years. In order of the vast options. to securitize potential the catastrophe of capital markets Exchange-traded for example, reinsurance, insurance they offer risk, by insurance introducing instruments low companies several present transaction have exchange-traded costs, because tried to take advantage insurance catastrophe advantages are they with respect standardized, to and includeminimal credit risk, because the obligations are guaranteed by the exchange. See [18] and [19] for the comparison of insurance securities. In particular, catastrophe options are standardized contracts based on an index of catastrophe losses, for example, compiled by the Property Claim Service (PCS), an internationally recognized market authority on property losses from catastrophes in theUS. The first index-based catastrophe derivatives were introduced at theChicago Board of Trade in 1992, but therewas only littleactivity in themarket. A second version of the index, compiled by the PCS, was introduced in 1995. At the peak, the total capacity created by this version Received 28 February 2008; revision received 18 July 2008. * Theresienstrasse Universit?t M?nchen, Postal address: Department of Mathematics, 39, Ludwig-Maximilians D-80333 Munich, Germany. ** Email address: [email protected] *** Email address: [email protected] **** Postal address: CMA, University of Oslo, Postbox 1035, Blindem, Norway. Email address: [email protected] 831 This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions F. BIAGINI ETAL. 832 10 insured catastrophe 1: Top Table Insured losses (source: Swiss Re, sigma Nr. 2/2007). loss (xlO9 dollars) 2005 66.3 Event Year Hurricane Country US Katrina: Gulf ofMexico floods dams 1992 23.0 2001 21.4 Bahamas burst to oil rigs damage Hurricane Andrew: North Atlantic US Bahamas flooding Terrorist attack on World Trade US Center, Pentagon, and other buildings 13.7 1994 2004 Northridge Hurricane earthquake Ivan: US 13.0 2005 to oil rigs damage Hurricane Wilma: US 19.0 US Caribbean torrential rain Mexico floods 2005 10.4 Hurricane Jamaica Haiti US Gulf ofMexico Rita: floods damage to oil rigs 8.6 2004 HurricaneCharley 8.4 1991 1989 TyphoonMireille Cuba US Caribbean 7.4 Hurricane Japan US Hugo Puerto Rico of the PCS options amounted to 89 million. Trading in these options slowed down in 1999. In a separate initiative, theBermuda Commodities Exchange (BCE) was launched in 1997 to tradeproperty catastrophe-linked option contracts. The BCE suspended itsoperations in 1997. Trading in the PCS options slowed down in 1999, because of the lack ofmarket liquidity and of qualified personnel (see, e.g. [18]). However, the record losses caused by hurricanes Katrina, Rita, andWilma a in creating catalyst new derivative instruments to trade kets. InMarch 2007, theNew York Mercantile Exchange futures and options again. These new contracts were risks the catastrophe (NYMEX) designed in2005 have been in the capital mar began trading catastrophe to bring the transparency and liquidity of the capital markets to the insurance sector, providing effectiveways of protecting against property catastrophe risk and providing the investorswith the opportunity to trade a new asset class which has little or no correlation to other exchange-traded asset classes. The NYMEX catastrophe options are settled against theRe-Ex loss index, which is created from the data supplied by thePCS. Following the descriptions in [12], [17], and [18], the structureof catastrophe insurance options is described as follows. The options are written on a loss index that evolves over two periods: the loss period [0, T\] and the consecutive development period [T\,T2]. During the contract occur. However, loss period, the index measures events that may specific catastrophic at the time of are the losses estimates of the true losses, occurrence, catastrophe reported only and these estimates are consecutively reestimated until the end, T2, of the development period. This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions Pricing insurance of catastrophe 833 options The loss index thus provides, at any t e [0, T2], an estimation of the accumulation of the final time (T2) amounts of catastrophe losses thathave occurred during the loss period. Let Nt, t e denote [0, T\], of catastrophes the number to time up let C//, /= t, and 1, ..., denote Nt, < corresponding final amounts of losses at time T2 (which are unknown at time 0 Then the value Lt of the loss index can be expressed as the t < T2). NtATx Lt= E[Ui ]P t e[0,T2], Ift], (1.1) /=i where the filtration {!Ft, t e compound a Poisson sum with martingales Poisson If the number Nt [0, T2]} represents the information available. to follow is assumed of catastrophes as the structure process, of index the is thus a summands. In the literature,a fewmodels have been proposed in order tomodel the catastrophe loss index and to price catastrophe options written on this index. In [12], [13], and [14] theunder was index catastrophe lying as represented a Poisson compound with process nonnegative jumps. In thismodel, reestimation is not taken into account at all. In [2] and [11] the authors distinguished between a loss and a reestimation period andmodeled the index as an exponential over each is assumed reestimation However, process period. Levy common is not realistic because reestimation factor. This assumption at T\ by a to start exactly loss reestimation happens individually for each catastrophe and begins almost immediately after the catastrophic event. In addition, the assumption of an exponential model for accumulated losses during the loss is rather period than earlier For unrealistic. and catastrophes, are more that later catastrophes implies starts at a positive value (instead this example, that the index severe at 0). of starting In [1] a more realisticmodel for the loss indexwas proposed and analytical catastrophe option were formulae pricing but developed, was reestimation also done by a common over factor the development period only. In [16] a model including immediate reestimation was assumed, where the reestimationwas modeled through individual reestimation factorsgiven by geometric Brownian motion. no efficient However, were methods pricing obtained. In this paper we specify a realistic model for the loss index that is consistent with (1.1). a case, particular occurrence is modeled As it comprises the model a Poisson process, by assume in [16]. We that catastrophe proposed for each and consider reestimation individual catastrophe where the initial estimate of the zthcatastrophe loss is reestimated immediately by a martingale positive from starting the random time of loss occurrence. then consider We the pricing of catastrophe options written on the index. The main contribution of this paper is to Fourier employ of an transform to reduce manage of expectation random independent in order techniques to obtain option the characteristic variables. We of function mention, pricing function of the characteristic the calculation the reestimation in particular, To formulae. this end, we to the computation of the index factor, in two evaluated that our methodology also works for loss distributions with heavy tails, which is the appropriate tail behavior for catastrophe modeling. positive We then proceed affine martingales. to discuss In this the case situation we when provide the reestimation factors a characterization are given of positive by affine (local) martingales. We believe that the use of exchange-traded insurance derivatives will play a crucial role in the securitization of the increasing catastrophe risk in the future. For this purpose, one essential task is to develop quantitative tools thathelp to establish liquid trading of these instruments. We hope that this paper contributes to this aim in that it sets new insights in the pricing of catastrophe options. This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions F. BIAGINI ETAL. 834 The remaining parts of thepaper are organized as follows. In Section 2 we present themodel for the loss index. In Section 3 we consider thepricing ofEuropean options in thegeneral model by using Fourier transformtechniques, and in Section 4 we discuss the special case of positive affinemartingales as reestimation factors. We conclude with Section 5, where we explicitly compute prices for spread options, which are the typical instruments in themarket. 2. Modeling the loss index Let (Q, 7, P) be a complete probability space. We consider a financialmarket endowed with a risk-freeasset with deterministic interest rate rt, and the possibility of trading catastrophe insurance options, written on a loss index. Following the description in the introduction,we distinguish two time periods: a loss period [0, T\\\ a development T\ < [T\,T2l, period During the contract specific loss period period, can users option choose either T2 < oo. [0, T\], the catastrophic events occur. After the loss a or six month a twelve month development period [T\,T2], where the reestimates of catastrophe losses that occurred during the loss period continue to affect the index. The contract option matures at the end of the chosen development period T2. Motivated by the index structure(1.1) discussed in the introduction,we model the stochastic = process L (Lt)o<t<T2i representing the loss index as follows: Lt=J2 = 7 1 te[0,T2], YJALr where the following notation and hypotheses have been used. (HI) Ns, s e [0, T2], is a Poisson process with intensityX > 0 and jump times ry thatmodels the number of catastrophes occurring during the loss period. (H2) Yj, j = 1, 2, ..., are positive independent and identically distributed (i.i.d.) random variables with distribution functionFY that represent thefirst loss estimation at the time occurs. the y'th catastrophe (H3) Ai, s e [0, T2], and j = AJ0 (H4) AJ, Yj, j = are 1, 2,..., = 1 i.i.d. martingales positive for all /'= such that 1,2,.... 1, 2, ..., and N are independent. In the sequel we will drop the index j and simplywrite Y and A in some formulae, when only theprobability distribution of the objects matters. The martingales AJt represent the unbiased reestimation factors. Here we assume that reestimation begins immediately after the occurrence of the 7thcatastrophe with initial loss estimate Yj, individually for each catastrophe. Here we suppose thatmarket participants observe the evolution of the individual catastrophe losses. Note that observing themarket quotes of the catastrophe index L alone is in general not sufficientfor theknowledge of the single reestimation factors,A. However, it might not be This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions Pricing of catastrophe insurance to assume unrealistic 835 options that market are able participants to obtain additional about information the evolution of individual catastrophes. Therefore, we assume thatthemarket informationfiltration (!Ft)o<t<T2 is the right continuous completion of the filtration generated by the catastrophe occurrences N, factors AK initial the corresponding loss estimates Yj, and the corresponding reestimation 3. Pricing of insurance derivatives We now consider the problem of pricing a European option with payoff depending on the value of the loss Lp2 at maturity index measure In the catastrophe 72. insurance the underlying market, themarket is highly incomplete and the choice of the pricing index L is not traded. Hence, is not clear. we Here the common suppose that the risk neutral approach pricing measure is structure preserving for themodel, except for the fact that thepricingmeasure might introduce a drift into the reestimation AJ', martingales j = we Here 1,2,.... do not discuss further the choice of the pricing measure, and, without loss of generality, perform pricing with themodel specification given under P, where we substitutehypothesis (H3) with (H3;) s G A{, [0, 72], and j = I, 2, AJ0 ..., are = 1 i.i.d. positive for all = j semimartingales such that 1,2,.... Consider a European derivative written on the loss index with maturity 72 and payoff > 0 for a payoff function h :R i-> R+. Since we have assumed that the interest h(Lp2) rate r is deterministic, insurance derivative without loss in discounted we can of generality, = i.e. we can set r terms, express the price 0. Then option is given by 7Tt=E[h(LT2) I !Ft], te[0,T2]. the price process process of the of the (3.1) In the following we will calculate the expected payoff in (3.1) using Fourier transform tech niques. To this end, we impose the following conditions. (Cl) The payoff function h(-) is continuous. (C2) h(-) - ke L2(R) = {g :R -> C measurable | f^ \g(x)\2dx < oo} for some keR. Remark 3.1. In [1] we could have considered more general options thatdid not necessarily fulfill condition (C2) by considering dampened payoffs. However, the cost of this approach is that treating heavy-tailed distributions of claim sizes Y becomes more complicated. The approach in this paper allows for general claim size modeling, including distributions with heavy tails. Furthermore, as we will see in Section 5, condition (C2) is, for example, satisfied by call and put spread catastrophe insurance options, the typical options traded in themarket. Now let h(u) =? 1 2n 1?? c~mz(h(z) / J_00 - k) dz for all u e R ? k and assume that be theFourier transformof h(-) (C3) h(') e Ll(R). This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions F. BIAGINI ETAL. 836 Note that condition (C2) does not necessarily imply condition (C3). Since conditions (C2) and (C3) are in force, the following inversion formula holds (cf. [9, Section 8.2]): h(x) i: (3.2) cluxh(u)du. Remark 3.2. Note thattheequality in (3.2) is everywhere, not only almost everywhere, because of condition (Cl). If the probability distribution of Lj2 had a Lebesgue density, an almost everywhere equality in (3.2) would have been sufficientfor the following computations. How ever, since the loss index is driven by a compound Poisson process, the distribution of Lj2 has and we atoms an need everywhere to guarantee equality (3.3), below. By (3.2), and condition (C3), we obtain 7Tt= E[h(LT2) |rt] = \Ft] + k E[h(LT2)-k /oo -oo exp{iuLT2}h(u) du (3.3) Ft | !Ft]h(u)du + k, E[exp{iuLT2} (3.4) -L where in the last equation we could apply Fubini's theorem, because condition (C3) holds. in order to calculate the price process (7Tt)te[oj2] in (3.4), the essential task is to compute the conditional characteristic function of Lj2, Hence, NT] ct(u) := = | !Ft] E E[exp{iwL;r2} exp Ft l 7= 1 \iuJ^YjAJT2__Tj u e (3.5) for t e [0, T2]. We define the conditional characteristic function of the reestimationmartingale A-7 as := E[exp{uM?} 0 < t < s < T2, |FtAJ], ftAI(s, u) where !FtAJ := o (AJS,0 < s < t) is thefiltrationgenerated by the jth reestimation factor.Then our main result is as follows. Theorem 3.1. The conditional characteristic function, (3.5), of the loss index Lj2 1. fort < T\, by ct(u) = U, uY)])} 0(1 exp{-?(ri E[^oV2 Nt x sj, uyj)\Sj=Tj,yj=Yj, fi= tf-sj(r2 2. for te u e M; 1 7 [TU T2], by NT] ct(u) = ["J ftA-Sj(T2 = 7 1 - Sj, uyj)\Sj=tj,yj=Yj, u E This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions isgiven, Pricing of catastrophe insurance 837 options Here U is a uniformlydistributed random variable on [t,T\] and Y is a random variable with distributionfunction FY, independent ofU. Note that, estimates, Yj, in Theorem up the 3.1, to time times t are known of catastrophe occurrence, Xj, and the initial loss data. 3.1. Proof of Theorem 3.1 We distinguish the computations over the two periods. 1. For t g [0, T\], we obtain, by hypothesis (H4) and by the independent increments ofNt, ct(u) ? E J2 ' ' 1 exp{i?(5>42_r.+ V = l j=N, + \YJAk-r)\ Nl NTl } Ft i i E rexp ?i? Er ? explwJ2YjAJT2_Tj\ ? -l;?i -I ! L ?;?w ii } F, YjAJT2_T.\ J (3.6) :=bt{u) :=at{u) We compute separately the termsat and bt in (3.6). By hypothesis (H4) we have, forat (u), u e = E at(u) Nt exp in | Y,YJAT2-Tj Ft y=i = CXp^iuJ2yjAT2-sj] Ntt n=Nt,Sj=Tj,yj=Yj |Ft]Sj=Tjtyj=Yj Y\E[cxp{iuyjAJT2_Sj} = 7 Ft 1 Nt = \ F?JSj]Sj=Tj,yj=Yj Y\E[exp{iuyjAJT2_Sj} = 7 1 Nt Wtf-sjWi-Sj.uyj) = 7 Note that, for at(u), the 1 YjS, the tys, and Nt sj=Tj>yj=Yj are known data, because the corresponding catastrophes happened before time t. For the second term,bt (u), u G R, we obtain, again by hypothesis (H4) and the independent increments bt(u) = E = E " of the Poisson NTl ? expliu J2 process N, 1 Ft J 1 j=Nt+ \YJAT2-Tj\ exp J2 YJAT2-Tj 1 j=Nt+ \ jiM This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions F. BIAGINI ETAL. 838 - Nti txpliuj^yj^-sj = 1 .. ., Tnt. -Nt, Nt,Y\,T\, , YNt -Mt , ^r, - 7 = E -Sj,uyj) n^o4"7^ = 1 7 - A^ri Nt,Y\,x\,..., , xnT{ -n( w, J - = yj Yj S j=T; yy=yy 7Vri =E P[ + t?(T2-Tj,uYj) l (3.7) L7=/V, By Theorem 5.2.1 of [15] we obtain the following result. Lemma 3.1. Let t < T, 0 < a Poisson be Nt with process INT (rNt+l,...,tNt has the same as distribution the order times jump statistics i.i.d. uniformlydistributed on the interval [t,T]. - Nt ..., (?/(i), Zj, = = j Then, 1,2,.... for all n) where U(n)), j Uj, ? 1, ..., n, are Using Lemma 3.1, and hypothesis (H4) again, we can replace the r7s in (3.7) with the order statisticsU(j) of i.i.d. uniformly distributed random variables on the interval [t,T\] and obtain NT] bt(u) = E PI ^(T2-Uu)9uYj) U j=Nt + \ Next, we note the following simple, helpful lemma. Lemma Un and 3.2. Consider a bounded statistics the order measurable function ... E[f(U{i), Proof. We denote E[/(i/(i),..., i/(i), f(x\, ,U{n))] ..., ... U(n) ,xn) ... =E[f(U{, ?/(n))]=E J2 /(^or(l), of {1, random i.i.d. symmetric the set of all permutations by Ew ofn variables in its arguments. ,Un)l ..., n}. We have - -, Ua{n)) l{Ua{l)<-<Uain) oeHn f(Ul,...,Un) = ]T 1{Ucr(\)<-<Ucr{n) E[f(Ui,...,U?)]. By the i.i.d. assumption of the F/s and A-'s, we see that the function /Bn(s,,...,iB):=E Y[^(T2-Sj,uYj) u e 7=1 This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions U\, Then ..., Pricing insurance of catastrophe is symmetric in s\,..., 839 options It is then not difficult sn. bt(u)= e\e\ Y\^(T2-Sj,uYj) = = = ..., E[f?(s\, NTl to see, using Lemma that 3.2, -Nt,Uri),...9UfNTi-Nt) n=NTl-Nt,Sj=Uu). Sn)\n=NTl-N,,Sj=UU)] . . . , U(n))\n=NTl-Nt] E[/^(?/(i), E[f2(Ui,...,Un)\n=NTl-Nt] NT] = E *q(T2-Uj,uYj) Il \ j=Nt + NTl exp\iuJ2 YJAT2-Uj 1 +l (3.8) j=Nt where we substituted the order statisticsU(j) with the i.i.d. uniform variables Uj. Note that (3.8) coincides with the characteristic function of a compound Poisson process of the form NT] *> ? j=Nt + \ where Z-7 = this case well ? j YjAJT_u_, known. Thus, are 1,2,..., we can NTl Y AJ ^ IJ/?T2-Uj 1 j=Nt expjiw +l rewrite = = i.i.d. The (3.8) as exp{-?(Fi form - exp{-?(F! 0(1 0(1 of the characteristic - function is in E[exp{iuZ[}])} - E[i?f?(T2 U, uY)])}. This completes the proof for the case inwhich t <T\. 2. For the case t> in which Ct(u) = T\, we obtain nT{ W tf-sj (T2 - 7= 1 as for the term at in the case in which 0 < Sj, uyj)\Sj=Tjiyj = Yj , t< u G R, T\. This completes theproof of Theorem 3.1. Remark 3.3. In [17] a special case of our model was presented, where the reestimation factor A was geometric Brownian motion. In this case the conditional characteristic function of the reestimation factor can be computed by numeric integrationby i/^(s, u) ? | E[exp{iwexp{/?? ^s}} Ft] ? ? ? ? = E[exp{iwexp{#r ^t}} txp{Bs Bt ^(s 0}] ? = E[exp{iiiu;?exp{5J_? / expfiwu^e^Jexpj - ?(s - 0}}]L,=exp{B,_,/2) - (y-(l/2)(s-t))2 2(s t) dy Wt=exp{Bt-t/2} This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions F. BIAGINI ETAL. 840 For further of function the characteristic about results random lognormal we variables, refer the reader to [10]. the next In turn our we section to a class attention of reestimation where processes the conditional characteristic function is numerically tractable and in some cases analytically For affine processes. obtainable: on affine processes information further their applications and tomathematical finance, we refer the reader to [5], [6], and [7]. 4. Reestimation with positive affine processes suppose that the reestimation factors are given by positive affine processes. We constitute processes a reach class of processes to model suitable a wide Affine of phenomena. range At the same time, the advantage is that the conditional characteristic function can be obtained explicitly up to the solution of twoRiccati equations. Definition 4.1. A Markov process A = (At, Px) on [0, oo] is called an affineprocess if there exist C-valued functions (?)(t,u) and \?/(t,u),defined on R+ x R, such that = | !Ft] exp{0(r2 E[exp{iwAr2} assume We t, u) + ir(T2 - for t > 0. t, u)At] (4.1) that (Al) A is conservative, i.e. for every t > 0 and x > 0 = PX[A, <oo] l; (A2) A is stochastically continuous for every Px. 1.1 of [8], assumption (A2) is equivalent to the assumption that (p(t, u) and By Proposition \?r(t, u) are continuous in t for each u. In the framework of our model, the computation of the conditional characteristic function reduces to the computation of 0 and t/>.In some cases these are explicitly known, otherwise they can be obtained In the particular numerically. affine martingales under the pricing case when we measure, the reestimation are able to prove factors the following remain positive characterization, which provides a useful simplification of the conditional characteristic function. Theorem 4.1. Let A be an affineprocess satisfying assumptions (Al) and (A2). The affine process A is a positive local martingale ifand only ifA admits thefollowing semimartingale characteristics (B, C, Bt = ? v): I Asds, Ct Jo = a and / As ds, Jo where v(dt, dy) = Atp(dy) dt, /oo yp{?y), a > 0, and ?jl is a Levy measure on (0, oo). Proof Since At satisfies assumptions (Al) and (A2), by [6, Theorem 2.12] and [8, The orem 1.1], At is a positive affine semimartingale if and only if At admits the following characteristics (B,C,v)\ Bt = / Jo (b + ?As)ds, Ct=a Jo Asds, and v(dt, dy) = (m(dy) + Atp(dy)) This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions dt, Pricing insurance of catastrophe for every 841 options Px, where b = b+ / 7(0, oo) (1A y)m(dy), a, b > 0, ? G R, and m and /xare Levy measures on (0, oo), such that / 7(0,oo) < oo. (vA l)ra(d_y) See also [8]. By assumption (A2) and Theorem 7.16 of [3], the operator L, (f(x+ y) f(x))m(dy) Lf(x)= \uxf"(x)+ (b+ ?x)f'(x)+z f 7(0, oo) + * /" (/(*+ y)- f(x) - f(x)(\ A y))p(dy) (4.2) 7(0,oo) on C2(R+), is a version of therestrictionof theextended infinitesimalgenerator ofA toC2(R+). (An operator L with domain ?>i is said to be an extended generator forA if?>l consists of those Borel functions / forwhich thereexists a Borel function Lf such that theprocess L{ = f(At)-f(A0)- [ Lf(Xs)ds Jo is a local martingale.) Then A is a local martingale ifand only if Lf(x) Substituting f(x) = 0 forf(x) =x. = x into (4.2), we obtain Lx = b+ ?x + ym(dy)+x / ?/(0,oo) = I 7(0,oo) (y (1A y))p(dy) + b+ / ym(dy). (y- l)v>(dy))x V / 7i 7(o,oo) (?+ Hence, A is a local martingale ifand only if \?+ (y-l)ii(dy))x + b+ / V7l 7(0,oo) ym(dy) = 0 foranyxGR+. (4.3) Since b > 0 and m is a nonnegative measure, condition (4.3) means that /oo yfi(dy). (4.4) Let A be an affine process satisfying assumptions (Al) and (A2). By Theorem 4.3 of [7], the conditional characteristic function of A satisfies (4.1), where 4>(t,u) and \//(t,u)solve the equations (?)(t,u)= i F(x//(s,u))ds,(4.5) 7o dtx?/(t,u) = R(f(t, u)), V(0, u) = iu, (4.6) This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions F. BIAGINI ETAL. 842 where, for z e {C | Rez < 0}, R(z) = \az2 + ?z+ 2 f (zzy-1-z?A .7(0,00) (4.7) l))/x(dy), F(z) = bz+ [ (ezy l)m(dy),(4.8) .7(0, oo) and a, ?, b,m, and /xare parameters of infinitesimal generator (4.2) of A. martingale thenby (4.4) we can simplify (4.7) and (4.8) as follows: If A is a local + f -zy- Dp(dy), (4.9) R(z)= \az2 (tzy 2 J(0,oo) F(z)=0. (4.10) From (4.5) and (4.10), we immediately obtain, for positive affine local martingales, (j)(t, u) = 0. In order to determine i/?we have to, in general, solve (4.5) numerically. For some special cases, however, we can compute We y?ranalytically. two examples. give Example 4.1. If A has no jump part thenA is called a Feller diffusion (see, e.g. [6]). In this case thepositive affinemartingale dynamics are given by dA, = Ja~?tdWt, where Wt is a standard Brownian motion. Consequently, we have /x= 0 in (4.9), and we can rewrite (4.6) as f't = \af}. (4.11) Solving the differential equation (4.11), we obtain \J/(t,u)= 0 where C(u) the expression or \/f(t,u)=-, mgM, at/2 + C(u) can be found from the boundary condition \j/(0,u) = for we \?/, \u. Substituting C(u) into obtain iJf(t,u)==Q or \jf(t,u) =-, at/2 + \/u u e R. Note that ifwe have no jump part then A has positive probability of being absorbed at 0. However, itmay stillbe of interestto also consider the case of positive probability of absorption at 0, ifwe wish to include thepossibility of fraud or falsified reportingof claims into themodel. In this case, reestimationmight discover the fraud and the previous fake evaluation will be set toO. Example 4.2. In order to give an example of a positive affinemartingale including jumps where we can solve for j//explicitly,we specify the jump density /x(dy) in the semimartingale characteristics inTheorem 4.1 as 3 4V^ dy " y5/2 This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions of catastrophe Pricing insurance 843 options Then some calculations give R(z) in (4.7): 1 ? 3 R(z) = -az2 + ?= 2 = 4V7T {az2 + dy 1)-^ f (^ -zy\ y5/I 7(0,00) (~z)3/2 for z g {C | Rez < 0}. Consider r?(t,u) := ?i/r(t, u). By (4.6) we thenhave -v?t The solutions to (4.12) are n(t, u) =0 r?(i, u) = + \oLr)2 r?3/2. (4.12) and = -,(1 or + W(-C(w)e-r/?;))-2, (4.13) is defined to be where W(-) is the Lambert W function. (The Lambert W function, W(z), = z, z G C. See also [4] formore details on theLambert the function satisfyingW(z)QW^ = = iu function.) The boundary condition n(0, u) ?\?/(0,u) yields = + C(M) -(-1^)exp{-I = into (4.13), we obtain, for \?r(t, u) ?rj(t, u), Substituting C(u) ^(t,u)=0 + ^) or \lf(t,u)=-2(1 (l + + + + W/fl~1 w((-\ -7-) expi-- - 1+ -7 5. Pricing of call and put spreads We conclude thispaper by showing thatwe can apply thedeveloped pricingmethod to spread options, which are the typical catastrophe options traded in themarket. A call spread option with strikeprices 0 < K\ < K2 is a European derivative with thepayoff function atmaturity given by 0 h(x) x - K\ if0<x<^i, ifK\ < x < K2, K2-Ki ifx>K2. is satisfied for k := K2 ? K\. In particular, The integrabilitycondition h(-) ? k g L2(R+) h(-)-ke L,(M+). To satisfy conditions (Cl) and (C3), we continuously extend h fromR+ toR by h(x) ifx<0, :? \h(-x) { ifx>0. \h(x) Note that theprice processes of the two corresponding options with payoffs h(Lp2) and h(Lp2) remain the same, because Let h(u) Lp2 :=? > 0. / 27T 7-oo e-mz(h(z) - k) dz for all u e R This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions F. BIAGINI ETAL. 844 ? k. Then be theFourier transformof h ?w-?(?' 1 1 =-~(exp{ 1 1 (Re(cxp{iu K2}) 1 11 ? =-x(cosuK2 n uz - - K2)dx K2)dx ? + cxp{iuK2] ?iuK2} 2tt uz 71 UZ c(Ki + K2)dx J-K\ c(x =-^ rKx - *(-x ? exp{?iuK\} ? cxp{iuK\}) Rc(cxp{iuK\})) e L cosuK\) (K), and, by applying the inversion formula (3.2) to h(x) forx > 0, we find that (3.2) also holds for h, since h(x) = h(x) forx > 0. In particular, since cs n^ > Lj2 0 almost surely :/: spread, we can write \Ft] + k dw expfiwL^j/^w) E[txp{iuLr2} Ft | !Ft]h(u)du + & du -\-k ct(u)h(u) JT J-oo of the call |Ft]+k =E[h(LT2)-k = E[h(LT2)-k = E for the price (cos uK2 M2 ? cos ? K\ uK\) du + K2 (5.1) = where ct(u) is defined in (3.4). Note that the integral in (5.1) is real, since Im(cr(?u)) ? definition of ct. ln\(ct (u)) by Analogously, for the put spread catastrophe option with payoff at thematurity given by 8(x) if0 <*<#!, ifK\ < x < K2, ifx > K2, K2-K\ - x K2 0 we obtain nt PS =? 1 f00 ct(u) I ?5? IT J_00 UZ ? (cosuK\ cos uK2) du. Note that the call-put parity is satisfied: 7Tt = K2 ? K\ ? 7Tt . Acknowledgements We wish to thank theNew YorkMercantile Exchange for the informationprovided concerning catastrophe insurance options and Damir Filipovic for interesting comments This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions and remarks. Pricing of catastrophe insurance options 845 References T. (2008). Pricing of catastrophe insurance options written on [1] BiAGiNi, F., Bregman, Y. and Meyer-Brandis, a loss index. To appear in Insurance Math. Econom. C. V. (1999). A new model for pricing catastrophe insurance derivatives. CAF Working Paper [2] Christensen, Series No. 28. E., Jacod, J., Protter, P. and Sharpe, M. J. (1980). Semimartingales and Markov processes. Prob. [3] ?inlar, Theory Relat. Fields 54, 161-219. R. M. et al. (1996). On theLambert W function. Adv. Comput. Math. 5, 329-359. [4] Corless, K. (2000). Transform analysis and asset pricing for affine jump diffusions. [5] Duffie, D., Pan, J. and Singleton, Econometrica 68, 1343-1376. W. (2003). Affine processes and applications in finance. Ann. [6] Duffie, D., Filipovic, D. and Schachermayer, Appl. Prob. 13,984-1053. [7] Filipovic, D. (2001). A general characterization of one factor affine term structuremodels. Finance Stoch. 5, 389^12. S. (1971). Branching processes with immigration and related limit theorems. K. and Watanabe, [8] Kavazu, Theory Prob. Appl. 16, 36-54. K. (1993). Analysis 2. Springer, Berlin. [9] K?nigsberger, [10] Leipnik, R. B. (1991). On lognormal random variables. I. The characteristic function. J.Austral. Math. Soc. Ser. B 32, 327-347. M. (1996). Pricing PCS-options with the use of Esscher-transforms. AFIR Colloquium N?rnberg, [11] M?ller, October 1-3 1996. Available at http://www.actuaries.org/AFIR/colloquia/Nuernberg/papers.cfm. [ 12] M?RMANN, A. (2001 ). Pricing catastrophe insurance derivatives. Financial Markets Group Discussion Available at http://ssrn.com/abstract=31052. [13] M?RMANN, A. (2003). Actuarially consistent valuation of catastrophe derivatives. Working Paper 400. paper. Available http ://knowledge.wharton. upenn.edu/muermann/. [14] M?RMANN, A. (2006). Market price of insurance risk implied by catastrophe derivatives. Working Available at http://irm.wharton.upenn.edu/muermann/. at paper. J. L. (1999). Stochastic Processes for Insurance and [15] Rolski, T., Schmidli, H., Schmidt, V. and Teugles, Finance. JohnWiley, Chichester. [ 16] Schmidli, H. (2001 ).Modeling PCS options via individual indices.Working paper 187, Laboratory ofActuarial Mathematics, University of Copenhagen. new instrument for managing catastrophe insurance options?a H. R. (1996). PCS Catastrophe [17] Schradin, risk. Contribution to the 6th AFIR International Colloquium N?rnberg, October 1-3 1996. Available http://www.actuaries.org/AFIR/colloquia/Nuernberg/papers.cfm. [18] Sigma Nr. 3 (2001). Capital Market Innovation in the Insurance Industry. Swiss Re, Z?rich. [19] Sigma Nr. 7 (2006). Securitization?New Opportunities for Insurers and Investors. Swiss Re, Z?rich. in 2006. Swiss Re, Z?rich. [20] Sigma Nr. 2 (2007). Natural Catastrophes and M an-Made Disasters This content downloaded from 193.227.29.18 on Tue, 30 Apr 2013 05:54:40 AM All use subject to JSTOR Terms and Conditions at
© Copyright 2026 Paperzz