1. Some calculations with exponential martingales Wojciech Szatzschneider School of Actuarial Sciences Universidad Anáhuac México Norte Mexico

2. Introduction Girsanov theorem in Practice. Under P : physical world if

3. And Risk Neutral world Q exists. Under Q law Always a martingale

4. Problem When is true martingale and not only local martingale?

5. Novikov: Local martingale is true martingale if and only if is of class (DL), if for every a>0, the family of random variables X T, ( T all stopping times < a ), is uniformly integrable. martingale Set

6. Kazamaki submartingale martingale

7. Novikov for small intervals: for some then martingale

8. … Simply note choose and use “Tower property”

9. Basic fact If under Q, is true martingale > 0 and if under P, > 0 a. e. then Q ~ P is a martingale. If possible, we will take advantage of

10. Change of measure Let Y t be a process such that: and r t =  Y t. Let X t be a BESQ  process And consider the continuous exponential local martingale

11. Where M(s)=X s -  s. It results that The local martingale is bounded (X s ≥ 0 and  < 0). Therefore, Z t is a martingale, and the change of drift via the Girsanov’s theorem is justified.

12. At time zero (to simplify the notation), we have

13. We look for F u (s) = F(s) such that F 2 (s)+F’(s)=  2 +2  for s  [0, u), F(u)= . In this part, the most suitable approach to the Riccati equation, that defines F, is via the corresponding Sturm-Liouville equation. Writing F(s)=  ’(s)/  (s), s  [0, u),  (0) =1, we get where  ¯ is the left-hand side derivative.

14. The exponential martingale that corresponds to F will be called Z, (Z corresponds to F in the same way as to  ).

15. The solution of the equation  ’’(s)/  ’(s)=c 2 in [0,u) is clearly Ae cs +Be -cs with conditions A+B=1 (because of  (0)=1) and We obtain after elementary computations, here X(0)=1, u=1. and reproduce easily Cox, Ingersoll & Ross formula.

16. The same method can be applied in the case of  2 -2  >0 with c 2 =  2 -2 . Case 2 In this case we have Lemma 1

17. Lemma 2 If then, as before we have Define Now

18. Proofs i.Here and as before Therefore, and

19. ii. Now and the bounded solution exists in [0,1] (after elementary calculations) for and using (1) we obtain the result.

20. Linear Risk Premia We will clarify what can be done and what can not in one dimensional financial market driven by Brownian motion, and asset prices that in the RW (under the law P) follow a Geometric Brownian Motion:

21. Set (discounted prices) where and is the spot IR in the RW.

22. Now, The RNW is defined as the probability law Q (Q ~ P), t < T that under Q being W* another Brownian Motion.

23. It can be shown that if r(t) is CIR (in real world and driven by the same BM), then such Q does not exist. An easy argument is based on explosion until T = 1 of the process defined by:

24. But what we really want is CIR in the RNW. We prove the following: Theorem, If under P Then for any T>0, there exists Q ~ P, for the process considered until time T such that under Q the interest rates follow:

25. Longstaff Model In 1989, Longstaff proposed the so-called double square root model defined in Risk Neutral World by: where

26. In 1992 Beaglehole & Tenney showed that Longstaff’s wrong formula for Bond Prices in his model gives the correct bond prices in the case of:

27. Longstaff uses Feynman-Kac approach and obtains the formula for bond prices of the form:

28. This calculations repeat all over the world in several textbooks

29. For some functions m,n,p and x=r t. However to apply Feynman-Kac representation, P(t,T) should be of class C 2 with respect to x. Some relaxation of this assumption is imaginable, but there is no possibility to make adjustments that could work for P(t,T). The problem is of course at zero.

30. We will show how to calculate:

31. and iff in (0,t)

32. This matching procedure does not work in the original Longstaff model, it means for calculations of: An application of Girsanov theorem leads to

33. CIR and intensity based approach Our goal is to calculate for r(s) and (s) dependent CIR processes.

34. Assume that one can observe correlations between r(s) (default free rate), and (s) (the intensity of default). We will use very special dependence structure between r(s) and (s) that can approximately generate the correlation one, and this structure will produce explicit formulas.

35. Our modelling is as follows: Set: with independent r 1 and r 2, 1 and 2.

36. Also r 1 is independent of 1. Now set 2 (t) =  r 2 (t) for some  consequently

37. With the use of Pythagoras theorem define: with the restriction that:

38. Using a well known formula for the variance of the CIR we can write: This method can be extended easily to multifactor CIR

39. Model with saturation We want to prove that is true martingale and not only local martingale.

40. satisfies The law of this process we will call P. The best way to prove that H(t) is true martingale is through some equivalences of laws of processes on,.

41. Note that V (t) is Pearl-Verhulst model. Let under the law Q, dV(t)=V(t)dt+V(t)dW(t). The law Q is equivalent to the law Q 1. Under Q 1, V(t)=e W(t)-½t. Now, under Q 1

42. Z T is clearly true martingale and by Girsanov theorem changes the law Q 1 into such that under, Moreover, under measure. Therefore on, and we have proved that Z T is P martingale. The last equivalence is obvious.