## Are you sure?

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

Alexander Novikov, University of Technology, Sydney Nino Kordzakhia, Macquarie University, Sydney

Abstract. We provide an analytic expression for the variance of ratio of integral functionals which arises as an asymptotic variance of Pitman estimators for a location parameter of independent identical distributed observations. The expression is obtained in terms of derivatives of a logarithmic moment of the integral functional of Limit Likelihood Ratio Process (LLRP). In the particular case, when the LLRP is a geometric Brownian motion we show that the established expression leads to the representation of the asymptotic variance of Pitman estimator in terms of the Riemann zeta-function.

1. Introduction. The Pitman estimator (Pitman (1939) [?]) for a location parameter θ of independent random variables (r.v.) ξi , i = 1, ..., n with a density fθ (x) = f (x − θ) has the form ∞ (n) n uLu du −∞ n) ˆ L( := f (ξi − u). (1) θn = ∞ (n) , u Lu du i=1

−∞

Existence and admissibility of Pitman estimators have been discussed in [?]. The integrals in (??) exist for suﬃcently large n under (for example) the condition E |ξi |p < ∞ for some p > 0. An exposition of some important propertes of Pitman estimators is presented in the texbook by Borovkov [?], see also [?]. The form of a limit distribution for the Pitman estimator as n → ∞ has been derived by Ibragimov & Has’minskii (1970) [?], (1981) [?], who have shown that under some regularity assumptions ∞ uLu du γ d ˆn − θ) → as n → ∞, n (θ ζ = −∞ ∞ Lu du

−∞

where Lu is a limit likelihood ratio process (LLRP) and the parameter γ depends on a form of discontinuities of derivatives of the underlying density fθ (x). The structure of the LLRP Lu , or equivalently, the structure of the process Yu := log(Lu ) has been described by Ibragimov & Has’minskii [?], [?] in several important cases. In particular, it has been shown in [?] that under some regularity conditions 1 H Yu = W u − |u|2H , 2 1 (2)

5 . It was ﬁrstly Ibragimov The special case H = 1 2 & Has’minskii (1975) [?] who obtained through simulation that V ar(ζ ) = 19. 1] in (??) the In this paper we show (see Theorem 2 in Section 2) that for the case H ∈ [ 2 ∞ value of V ar(ζ ) can be found in terms of derivatives of the expectation of log( 0 (e−m1 u Lu + e−m2 u L−u )du+δ )) with respect to parameters m1 and m2 . Set qt = ∞ −∞ Lt Lu du . 1]. see (??) below). Fujii (2010) [?]). Brownian motion i. 1].5 ± 0.e. (3) where Zeta[k ] is the Riemann-zeta function.276 ± 0. when H = 1 2 In this paper we discuss the problem of calculation of the variance V ar(ζ ) only for the 1 case (??) with H ∈ [ 2 . √ Theorem 1. when fθ (x) is a density of Weibull-like distributions. 2 . However..06 . The r. 1] appears. ζ with Yu speciﬁed in (??) appears also as a limit in a change 1 point problem for a Brownian motion (the case H = 2 . ∞)..H where Wu is a fractional Brownian motion (fBm) with the Hurst parameter H ∈ (0. H H E (Wu ) = 0. |ζ |2H is exponentially bounded i.v. [?]. The general case with H > 1 2 0 involve Levy processes are still under progress (see Remarks below). This result and some elements from its proof will be used below in the proof of Theorem 2. we expect that our method (see details in Section 2 and Remarks below) may be used to ﬁnd analytical or numerical expressions for V ar(ζ ) also in other limit schemes (for example. Then we illustrate the application 1 (see Section 3) where the distribution of functionals of this result in the case H = 2 ∞ −mu and other similar cases which e Lu du is well known. Kutoyants (2004) [?]. Golubev (1979) [?] using the Feyman-Kac formulae found an analytical expression for V ar(ζ ) in terms of integrals of products of Bessel functions and then using numerical integration obtained V ar(ζ ) = 19.e.v. there exists constant ε > 0 such that E exp{ε|ζ |2H } < ∞.v. 1 The form (??) for Yu with H ∈ ( 2 . see monographs by Ibragimov & Has’minskii (1981) [?]. s ∈ R. Proof. Rubin & Song (1995) [?] using some of results by Golubev (1979) [?] and Ibragimov & Has’minskii (1975) [?] found that V ar(ζ ) = 16Zeta[3] = 19.2329. 1 . [?]) and in a cusp estimation 1 problem in ergodic diﬀusion processes (H ∈ (0. 2 ). E |Wu − WsH |2 = |u − s|2H . u ∈ R = (−∞. |ζ |2H . for example. in (??) was studied by several authors. in the case when Yu is a Levy process. Before proving the main result in Theorem 2 we present the following theorem about existence of exponential moments of r. Note the distribution of the random variable ζ is unknown in an explicit form even for the case of a standard . If H > 54−1 then r.

2 eε|t| qt dt. by convexity arguments (Jensen’s inequality) we have for any c ≥ 0 ∞ exp{ε|ζ | 2H } ≤ cε + −∞ 1 ). 2 u2H )du}. ∞ 2H e This implies εt2H E (qt )dt ≤ e E 0 ε qt dt = eε . ∞ E exp(ε|ζ |2H ) ≤ cε + eε + 2 1 eεt E (qt )dt. u ∈ [0. 2H To estimate E (qt ) note ∞ t Lu du ≥ 0 0 H exp(Wu u2H 1 − }du ≥ t exp{ 2 t t t H (Wu − 0 u2H )du}. t H The random process 0 Ws ds is a zero-mean Gaussian process with the variance t t V ar( 0 WsH ds) = E ( 0 WsH ds)2 = 1 2 t 0 0 t [s2H + u2H − |u − s|2H ]dsdu = t2H +2 . 2H + 1 2 2H + 2 which is ﬁnite for some positive ε when ε< H 1 1 4H 2 + 2H − 1 − = 2H + 1 2 2H + 2 2(2H + 2)(2H + 1) 3 .Since qt is a (posteriori) density function. 2H + 2 Therefore ∞ 1 ∞ eεt E (qt )dt ≤ 1 2H ∞ exp{εt2H − exp{εt2H − t2H ( H 1 t2H + 2 V ar( 2H + 1 2t t WsH ds)}dt 0 = 1 H 1 1 − )}dt. t]. 2 (4) and hence qt ≤ Lt t 0 Lu du ≤ 1 t2H 1 exp{− − t 2 t H − WtH − (Wu 0 Thus we obtain for t ≥ 1 ∞ 1 ∞ e εt2H ∞ E (qt )dt ≤ 1 e εt2H t2H 1 E exp{− − 2 t t H (Wu − WtH − 0 t u2H )du}dt = 2 exp{εt 1 2H H 1 − t2H }E exp{− 2H + 1 t H Wu du}dt. 0 d H − WtH = WtH where the last equation is due to the property Wu −u . t > 0. 2H (the constant cε = 0 for the case H ≥ E exp(ε|ζ | Note 0 1 2H Due to symmetry of WsH we have also ∞ ) ≤ cε + 2 0 eεt E (qt )dt.

−∞ −∞ uLu du = α1 (0) − α2 (0).309. To formulate the main result we use the following function ∞ d d (7) g (m1 .m2 →0 [ ∂ 2 g (m1 . Recall that we discuss here the case that the process Yu = log(Lu ) 1 has the representation (??)). First note that the function g (m1 . H ∈ [ 2 . Theorem 2. This implies that ζ has a symmetric distribution and E (ζ ) = 0. With use of these notations we obtain ∞ ∞ Lu du = β1 (0) + β2 (0). 2 It will be convenient to use the following parametrised processes: ∞ ∞ β1 (m) = 0 ∞ e−mu Lu du. 2 ∂m (6) α1 (0) − α2 (0) . The main result. α2 (m) = 0 ue−mu L−u du.It remains to note that the 4H 2 + 2H − 1 is positive for H > 54−1 = 0. m2 ) := E (log(β1 (m1 ) + β2 (m2 ))) = E (log( 0 (e−m1 u Lu + e−m2 u L−u )du)) where m1 > 0 and m2 > 0 are parameters. m2 ) − ]. For the proof of Theorem 2 we use the following representation: 1 V ar(ζ ) = E ( 2 ∞ u2 Lu du −∞ ). where m ≥ 0 is an auxiliary parameter.. . ∂m2 ∂m1 ∂m2 1 Proof. β1 (0) = β2 (0). β1 (0) + β2 (0) H H = W− Note that since Wu u (in distribution) we have α1 (0) = α2 (0). i = 1. m2 ) ≤ E 0 (e−m1 u Lu + e−m2 u L−u )du) = 4 1 1 + <∞ m1 m2 .. ∞ L du −∞ u √ (5) This was proved by Golubev [?] for the case H ∈ [ 1 . m2 ) is ﬁnite because due the inequality log(x) ≤ x ∞ g (m1 . ∞ α1 (m) = 0 ue−mu Lu du. m2 ) is twice continuously diﬀerentiable and V ar(ζ ) = 2 lim m1 →0. 2. β2 (m) = 0 e−mu L−u du. αi (m) = − and ζ= d ∂ βi (m). The function g (m1 . 1]. m2 ) ∂ 2 g (m1 . 1] in (??).

g (m1 . obviously. As seen from the proof β2 (m) of Lemma 1. ∞ H 1 1 E (Z ) = t exp{−t2H ( − )}dt < ∞ 2H + 1 2 2H + 2 0 5 . Theorem 1) and in view of (??) we have V ar(ζ ) = E ( = −2E ( α1 (0) − α2 (0) 2 ) = β1 (0) + β2 (0) (8) α1 (0)α2 (0) α1 (0)2 α2 (0)2 ) + E ( ) + E ( )]. To ﬁnd the ﬁrst expectation in the RHS of (??). m2 > 0).g. m2 ) > −∞ due to the estimate (??). (β1 (0) + β2 (0))2 (β1 (0) + β2 (0))2 (β1 (0) + β2 (0))2 Note that the last two terms are equal due the property of symmetry (??). m2 ) := ∂ 2 (log(β1 (m1 ) + β2 (m2 )) α1 (m1 )α2 (m2 ) . By direct calculations we obtain Q(m1 . We shall use some of calculations presented above in the prove of Theorem 1. 0 2 (m) 2 The similar estimate is valid for the second term ( α ) in (??).and. Applying the Holder inequality we have Q(m1 . Set e−mt Lt pt (m) = ∞ e−mu Lu du 0 Since pt (m) is a density function. by Cauchy–Bunyakovsky–Schwarz inequality we have α1 (m) 2 ) ≤ ( β1 (m) ∞ t2 pt (m)dt 0 To estimate pt (m) we apply the following inequalities pt (m) ≤ and thus obtain ∞ ∞ e−mt Lt t −mu e Lu du 0 ≤ t exp{ 1 t t Lt t H (Wu 0 − u2H )du} 2 t2 pt (m)dt ≤ 0 ∞ 0 t exp{− t2H 1 + 2 t t H + (WtH − Wu 0 u2H )du}dt = 2 ≤ 0 t exp{− H 1 t2H + 2H + 1 t H (WtH − Wu )du}dt =: Z . = − (β1 (m1 ) + β2 (m2 ))2 ∂m1 ∂m2 (9) (recall we assume m1 > 0. β1 (m1 ) β2 (m2 ) (10) Now we show that the right-hand side of this inequality is uniformly bounded by an integrable random variable. The variance V ar(ζ ) = E (ζ 2 ) < ∞ (see e. m2 ) ≤ ( α1 (m1 ) 2 α2 (m2 ) 2 ) +( ). ﬁrstly note that the random functions βi (m) are continuously diﬀerentiable.

By Lebesgue’s theorem on majorised convergence and the theorem on monotone convergence.m2 →0 E( α1 (m1 ) − α2 (m2 ) 2 ) = β1 (m1 ) + β2 (m2 ) ∞ −2 lim m1 →0. which does not dependent of parameters m1 and m2 . m2 ) is unformly bounded by an integrable r. ) = − (β1 (m1 ) + β2 (m2 ))2 ∂m1 ∂m2 (11) Finding the second term in (??) is similar but a little bit more tricky. which does not dependent of parameters m1 and m2 .m2 →0 ( u2 Lu du ∂ 2 g (m1 . m2 )) = E ( α1 (m1 )α2 (m2 ) ∂ 2 g (m1 . β1 (m1 ) + β2 (m2 ) β1 (m1 ) β2 (m2 ) ∞ Note ( where as shown above the right-hand side is unformly bounded by an integrable r. β1 (0) + β2 (0) 6 ∞ 0 .v. m2 ) has a continuous mixed derivative and E (Q(m1 .1 at least for the case H ≥ 2 . This implies that the random function Q(m1 . m2 ) ) = −2 + 2 β1 (m1 ) + β2 (m2 ) + δ ∂m2 ∂m1 ∂m2 1 ∞ 2 −m1 u ue Lu du +2E ( 0 ). ∂m2 ∂m1 ∂m2 β1 (0) + β2 (0) 1 (12) On the other side. β1 (m1 ) + β2 (m2 ) + δ α1 (m1 ) − α2 (m2 ) 2 α1 (m1 ) 2 α2 (m2 ) 2 ) ≤ 2( ) + 2( ). m2 ) . Applying the expectation to the both parts of (??) and well-known theorems about differentiability of expectations over parameters we can conclude that the function g (m1 . the formula in (??) can be rewritten as follows u2 Lu du V ar(ζ ) = E ( ). due to the result of Golubev (??) and the symmetry of fBm. m2 ) ∂ 2 g (m1 . (β1 (m1 ) + β2 (m2 ))2 β1 (m1 ) + β2 (m2 ) Now combining all above equations we obtain E( α1 (m1 ) − α2 (m2 ) 2 ∂ 2 g (m1 .v. m2 ) 0 − ) + 2 E ( ). m2 ) ∂ 2 log(β1 (m1 ) + β2 (m2 )) = E ( )= ∂m2 ∂m2 1 1 ∞ 2 −m1 u ue Lu du α1 (m1 )2 0 = −E ( ) + E ( ). we obtain V ar(ζ ) = lim m1 →0. + ∂m2 (β1 (m1 ) + β2 (m2 ))2 β1 (m1 ) + β2 (m2 ) 1 This implies ∂ 2 g (m1 . One can check by direct calculations that u2 e−m1 u Lu du ∂ 2 log(β1 (m1 ) + β2 (m2 )) α1 (m1 )2 0 =− . m2 ) ∂ 2 g (m1 .

1. Calculation of V ar(ζ ) for the case of H = 2 . (13) − + are independent Levy processes with some speciﬁc Levy measures for and Ju where Ju jumps. Gushchin. We present here an elementary derivation of the explicit formula (??) based on the result of Theorem 2 and the well-known result on the distribution of exponential functional of a standard Bm. In particular.g. Dachian and Negri (2011) [?]) and autoregressive processes (Chan & Kutoyants (2010) [?]). Other representations for V ar(ζ ) can be given under some additional assumptions. ξk ∼ N (− . 3. Remarks. [?]. Contrary. 1) (the private communication of A. see e. In those settings the authors obtained the following representation + − Yu = Ju I {u ≥ 0} + Ju I {u < 0}.g. For the case under consideration ∞ β1 (m) = 0 eWu 7 1/2 −(1/2+m)u du . 2. 2 Thus V ar(ζ ) can be found when the expectation E (log( −∞ e−mu Lu du)) is known in an explicit form as a function of m. Another form of the limit process Lu or. [?]. 1 3. One can see that the RHS of (??) can be written in the following form ∂E (log( 1 V ar(ζ ) = − lim 2 m→0 2 ∞ e−mu Lu du)) −∞ − + ∂m ∞ . ∂m2 (14) then V ar(ζ ) = lim m→0 (15) The proof of this formula can be obtained similarly to the proof of Theorem 2. there are many results on distributions of integral exponentials ∞ like 0 e−mu+Ju du where Ju is a Levy process. To our knowledge there are not results of this type in the current literature.To complete the proof we need only to substitute the last expression in (??) and thus we have proved Theorem 1. as shown in [?]. if for some ε > 0 ∞ E log( 0 eεu Lu du) < ∞ ∂ 2 E (log( ∞ emu Lu du)) −∞ . For example. 2 k=1 Nu The representation (??) holds also for the case (??) under the condition E (ezJu ) = E (e(1−z)Ju ) for all z ∈ (0. the process Yu appears in the change-point problems for Poisson processes (see e. 1). [?]. Example. − d Ju = + Ju = 1 ξk . 2011). equivalently.

. ∂m1 ∂m2 hence ∂ 2 g (m1 . 2) = −8Ψ(2. m2 ).g. 2m + 1) − log(2) Γ(2m + 1) for any m ≥ 0 (this can be checked with Mathematica@ ) where the function Ψ(n. note ∞ E (log(1/β1 (m))) = 0 log(x) 2(2x)2m e−2x dx = Ψ(0. 2m1 + 1) − Ψ(0. 2] β1 (m) (16) (see e. z ) = P olyGamma[n. 1)−8Ψ(2. We have ∂2 E (log(β1 (m1 ) + β2 (m2 ))) = 4Ψ(2. 2] and hence E (log(1/β1 (m1 ) + 1/β2 (m2 ))) = Ψ(0. the random variables β1 (m1 ) and β2 (m2 ) are independent. 2)−8Ψ(2. m2 ) ∂ 2 g (m1 .v. 1). To ﬁnd the function g (m1 . z ) is the (n + 1)th derivatives of logarithm of the Gamma function Γ(z ): Ψ(n. Note that due to the properties of a standard Bm. Now we are ready to calculate V ar(ζ ) using Theorem 1 with δ = 0.v. we have 1/β1 (m1 ) + 1/β2 (m2 ) = Gamma[2m1 + 2m2 + 2.and 1 P( < x) = β1 (m) x 0 2(2y )2m −2y e dy Γ(2m + 1) i. 1) = (−1)k+1 k !Zeta[k + 1] 8 d . 2m1 + 2m2 + 2) − log(2). 2m1 + 2m2 + 2) − 4Ψ(2. 2m1 + 2m2 + 2).’s is a Gamma r. 2m1 + 1). β1 (m) has the reciprocal Gamma distribution: 1 ∼ Gamma[2m + 1. ∂m2 ∂m1 ∂m2 m1 =0. 2m2 + 1) + log(2). ∂ 2 m1 ∂2 E (log(β1 (m1 ) + β2 (m2 ))) = 4Ψ(2. Hence E (log(β1 (m1 ) + β2 (m2 ))) = E (log(1/β1 (m1 ) + 1/β2 (m2 ))) − E (log(1/β1 (m1 )) − E log(1/1/β2 (m2 ))) = Ψ(0.m2 =0 1 It remains to note that Ψ(k. z ] = dn+1 log(Γ(z )).e. 2m1 + 2m2 + 2) − Ψ(0. m2 ) V ar(ζ ) = 2[ − ]| = 8Ψ(2. Dufresne (1990) [?] ). dz n+1 Due to (??) and the fact that sums of independent Gamma r.

no. 9-10.: Mathematical statistics. K. A. J. Remark. 80 (2010). 3. The authors are indebted to Alexander Gushchin for discussions and suggestions which allowed to improve the presentation of the paper. 9 . [8] Fujii. One can check that condition (??) does hold and so it is also possible to obtain the last result using formula (??). F. Actuar. 71 (1997). : On Threshold Estimation for Some TAR Models. no. [6] Dachian.4. 61–69. no. Ilia.. [3] Borovkov. ISBN: 90-5699-018-7 [4] Carmona. J. no. Probab. 1990. 1998. A. Inference Stoch. and Stegun.212. Appl. Petit. [5] Chan. Acknowledgment. and Yor.: Present value distributions with applications to ruin theory and stochastic equations. I. The authors are thankful to Yuri Kutoyants for the suggestion to study the problem of calculations of asymptotic variances of Pitman estimators. J. and Yor. Transmission 15 (1979). 191. M. References [1] Abramowitz. 123–144. 14 (2011). 2005. : Computation of eﬃciency of maximum-likelihood estimate when observing a discontinuous signal in white noise. Lett. 1) = 16Zeta[3].: On compound Poisson processes arising in changepoint type statistical models as limiting likelihood ratios. and Kutoyants. Gordon and Breach Science Publishers.2. with applications to risk theory and pension funding. Probab. 3. no. G. Statist. Tenth Printing.: Handbook of Mathematical Functions.: Exponential functionals of Levy processes. (formula 6.. (Russian) Problems Inform. Survey. 39–79. 2001. H. D. 2.for any k=1. 255–271. Scand. [7] Dufresne. H. Stat. Process. Birkhauser Boston. M. T. : The distribution of a perpetuity.2 in Abramowitz & Stegan (1972) [?]) and so V ar(ζ ) = −8Ψ(2. preprint.: An extension of cusp estimation problem in ergodic diﬀusion processes. [9] Gjessing. 2010. : Exponential functionals of Levy processes. and Paulsen. 41–55. 1-2. [2] Bertoin. 779 783. Y. [10] Golubev. Stochastic Process. 1. M. P. Levy processes. MA. 1972. Sergue¨ ı and Negri. Amsterdam. Boston.

[11] Ibragimov. (1939) 391-421 [16] Port. C. Dokl.: The estimation of the location and scale parameters of a continuous population of any given form. and Stone. R.G. 31–43. H.J. R. 2004. I. 23 (1995). 3. 732–739. no. Math. Y. SpringerVerlag. K. [13] Ibragimov. Ann. [15] Pitman. 2 (1974). : The asymptotic behavior of generalized Bayesian estimates. v. 225–247. C.: Fisher information and the Pitman estimator of a location parameter. (Russian) Problemy Peredaci Informacii 11 (1975). 10 . Akad.: The admissibility of Pitman’s estimator for a single location parameter. [18] Stein. 518–522.: Exact computation of the asymptotic eﬃciency of maximum likelihood estimators of a discontinuous signal in a Gaussian white noise. Ltd. [14] Kutoyants. I. 194 1970. : Statistical estimation. and Has’minskii. Asymptotic theory. Ann. 1981. 257–260. [12] Ibragimov. 30 (1959). J. Biometrika 30. and Has’minskii. London. Springer Series in Statistics. and Song. S. and Has’minskii..Ann. Statist. no. Statist. [17] Rubin. no. Nauk SSSR. R. E. : Statistical inference for ergodic diﬀusion processes. C. I. : Estimation of the parameter of a discontinuous signal in Gaussian white noise. New York-Berlin. 2. 3. Springer-Verlag London. Statist.

- Chapter 6 Practice Test
- Chapter 6
- Algebra II 2nd Semester Study Guide
- Fractura de Mineral Simulacion
- Exponentials & Logarithms
- Mark Scheme June 2007 6666 Core Mathematics C4
- SQL
- 01 TRB 2009 Fatigue Paper
- Graphs
- Design of an aqueduct
- Chapter 1- Graph Analysis Edit 20130711
- Gas
- Add Math Project 2013
- Jak
- 20e-moreTaylor
- di1
- MA1301 Help Sheet
- riemann1859P
- L18.cumulants
- section 5 3 notes
- John Fox An R
- KINETICS.docx
- Mean Square Error Method 123
- Lamp Iran
- 8.7 Modeling With Exp & Power Functions
- 5th IRTbin2013_MAIN-Reviewed.doc
- Annals Rysunki
- Boyle Broadie Glasserman Mc Overview Jedc
- Math Stats Text
- Al II H Final Review PDF

- Full metal alchemist episode 2
- f&sBGK
- 我是日語語法書
- 50 Challenging Problems in Probability With Solutions
- UMC 0.000% 18-May-2020
- Tchaikovsky-Op20pf1895.pdf
- Phil Hunt , Joanne Kennedy , Antoon Pelsser - Markov-functional interest rate models.pdf
- Full metal alchemist episode 4
- Ravel_-_La_Valse__Piano_.pdf
- T.E. Duncan - Prediction for Some Processes Related to a Fractional Brownian Motion
- Rosalyn Tureck - An Introduction to the Performance of Bach
- Full metal alchemist episode 3
- Tu Reck Tribute Program
- Factory Method .NET Design Pattern in C# and VB - Dofactory
- Nsp Premia Manual
- Robert Elliott - Pricing Volatility Swap and Variance Swap under Heston Model with Regime Switching.pdf
- Duffie - AFFINE PROCESSES AND APPLICATIONS IN FINANCE.pdf
- DuffiePanSingleton jumps Econometrica 00.pdf
- Comment on ‘Pricing double barrier options using Comment on 'Laplace transforms’ by Antoon Pelsser.pdf
- Dilip Madan - On Pricing Contingent Capital Notes.pdf
- Simulation of fBm thesis nl.pdf
- Print - Lesson 6_ Multivariate Conditional Distribution and Partial Correlation.pdf
- Antoon Pelsser - Pricing double barrier options using Laplace transforms.pdf
- Dilip Madan - Pricing S&P 500 Options using Hilbert Space Basis.pdf
- C.H. Hui - Using first-passage-time density to assess realignment risk of a target zone.pdf
- Simona Svoboda - Libor Market Model with Stochastic Volatility.pdf
- Patrick Cheridito - Pricing and Hedging CoCos.pdf
- Robert Jarrow - smiles_JF_07.pdf
- David Lando - Cox Process and Risky Securities.pdf

Sign up to vote on this title

UsefulNot usefulRead Free for 30 Days

Cancel anytime.

Close Dialog## Are you sure?

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

Loading