just make the dam pdf available to those who want to read it. I really don't understand why you want to squeeze a buck out of a student report, duh...anyway...

alright unkpath, i think we all got your point.i think it's nothing to be angry about that i decided to publish that text in book format ( duh! ) - if you think that it's not worth reading, so long - i do think it is.On the other hand, if you want to start an "open-source" finance-book publishing, than tell me where to do so

QuoteOriginally posted by: Apprenticeone should agree indeed that the price is fair if the text delivers quality information on the topic. We are far from the € 50+ price range of Wiley and co.Jesper himself said it took him a few years to have the Cheyette model work properlyhttp://www.math.ku.dk/~rolf/nordea_road_show_talk.pdf (p16)From the pdf: "It took about 3 years to get the implementation working."Nice.

One feature of this PDE is that it is convection-diffusion on x and convection in y (no diffusion term). In the PDE we only need to give one boundary condition in y (for example y = 0) but not at the other end. However, in FDM you must give numerical boundary conditions and this is non-trivial to do (the problem was solved by H.O. Kreiss in the 1970's in the CFD literature).Using well-known centred difference schemes with ADI or splitting introduce reflection problems at the numerical boundary as this is the same problem as with Asian option PDE as we experienced some time ago. Very messy. The reflected waves go back into the medium. And knowing where to truncate is vital.Landgraf is aware of these problems.

so has anyone read the book and can they give us some commentary?

QuoteOriginally posted by: mjso has anyone read the book and can they give us some commentary?The book is not shipping yet (December 25 from amazon.de) but I have been privy to the thesis quite some time ago ... and have examined it in some detail. It is very specialised for the reason that it has an embedded 1st order hyperbolic PDE. If you don't know the essential problems you will waste a lot of time trying to solve it with centred-difference schemes. That's the crux in a nutshell. It's different from the somewhat easier diffusion problems.My posts have been based on the bespoke thesis. So, I am assuming that the book is the same as the thesis.This is a very good thesis, as I already said. Especially in the numerics it is comprehensive. Finally, it is also a work that handles all necessary topics. It could be a template for future theses, purely for its structure alone. And he does it all in C++ and UML.Enough.//////just to update againQuoteLet me tell you why I think this...There have been many threads and questions on this model (do a Google) and I can tell for a fact that there were a number of problems to resolved, for example the original ADI scheme that needed to be adapted. For the record, I have seen this work long before the book came out. He addressed some open problems that were not resolved till that time.The author discusses the problem from A-Z, including IR models, PDE and numerical solutions right up to UML and C++. So fair play Mr. Landgraf for this initiative. So, where this model is published (I am willing to bet you won't), and we are back at my original claim. Have a look at this thread that proves the dearth of feedback on the modelhere

Slightly off-topic but related; here is a thesis (scroll for the pdf) for the Heston and SABR models where author applies less commonly used schemes in this domain that solve a number of open problems.Sheppard

The presentation slides for Peter's thesis are athttp://www.math.uni-bayreuth.de/~lgruene/diplo ... ien.pdfAny idea what software was used to draw the figures and plot the graphs? They are pretty good.Reddy

Hi,the pictures where drawn by using METAPOST and including those in a regular Latex code...METAPOST is kind of a graphical programming language with a very straightforward and simple syntax and closely related to postscript.here you can find some simple (code) examples where i have started from:MetaPost Code Examplesthe only thing you'll need to install is the metapost compiler, which is open-source software btw: because of the easy METAPOST-syntax, you can easily write C++-Programs, which generate Metapost-code..(this is how i have "drawn" the multi-color grid picture in the presentation)

The following remarks are a little off topic as they refer to much of the literature on a particular class of FDM schemes for finance.1. With FDM we must prove their stability. In many cases the accepted practice is to use von Neumann's discrete Fourier analysis and calculate the so-called symbol of the scheme. This approach works only for initial value problems with constant coefficients. It is a leap of faith to conclude how it works for initial boundary value problems with non-constant coefficients. FD schemes for such problems may or may not work in practice, you need stronger techniques such as the maximum principle and monotone schemes. But the mathemtics does not extrapolate...2. In numerical analysis, the most robust schemes are those that strive to be monotone, using M matrices. If a scheme is not monotone, you can see negative option values when you approximate the mixed derivatives, for example.3. Von Neumann is a useful (necessary?) check for stability, but it is not sufficient. A 'workaround' is extensive numerical experimentation.

QuoteOriginally posted by: TraderJoeQuoteOriginally posted by: Apprenticeone should agree indeed that the price is fair if the text delivers quality information on the topic. We are far from the € 50+ price range of Wiley and co.From the pdf: "It took about 3 years to get the implementation working."Nice.A number of schemes have been advocated for this PDE in x and y(which has the same form as Asian PDE). It is difficult and easy at the same time, depending on how you tackle it:difficult: no diffusion in y, so 3-point centred differences give problems and are awkwardeasy: the PDE in y is first-order hyperbolic (use Wendroff/Box method), use schemes with 2-pointsThe box method leads to a monotone scheme (M matrix). The problem looks tractable and it would seem that splitting/ADI (especially with centred schemes) is not necessary.Have you looked at such schemes?

QuoteOriginally posted by: pekolaHi,the pictures where drawn by using METAPOST and including those in a regular Latex code...METAPOST is kind of a graphical programming language with a very straightforward and simple syntax and closely related to postscript.here you can find some simple (code) examples where i have started from:MetaPost Code Examplesthe only thing you'll need to install is the metapost compiler, which is open-source software btw: because of the easy METAPOST-syntax, you can easily write C++-Programs, which generate Metapost-code..(this is how i have "drawn" the multi-color grid picture in the presentation)I got the book, must say it is an comprehensive treatment of the subject. Back to the Metapost subject. How can I generate Metapost code in C++/C#? I googled but couldn't find anythig. Please any help is welcome!

QuoteOriginally posted by: CuchulainnA number of schemes have been advocated for this PDE in x and y(which has the same form as Asian PDE). It is difficult and easy at the same time, depending on how you tackle it:difficult: no diffusion in y, so 3-point centred differences give problems and are awkwardeasy: the PDE in y is first-order hyperbolic (use Wendroff/Box method), use schemes with 2-pointsThe box method leads to a monotone scheme (M matrix). The problem looks tractable and it would seem that splitting/ADI (especially with centred schemes) is not necessary.Have you looked at such schemes?This is what i wanted to try next: I wanted to use IMEX (unfortunately didn't have the time to try it so far..), which would yield an explicit treatment of the hyperbolic PDE part in the second space direction..for discretisation i would use upwind (2point) - there are also upwind schemes which have 2nd/3rd order such as 5-point-upwinding...but for those oscillatory behavior may come back. Lax-Wendroff would be another alternative for upwinding (and is of second order)

QuoteOriginally posted by: arfI got the book, must say it is an comprehensive treatment of the subject. ...thanks...QuoteBack to the Metapost subject. How can I generate Metapost code in C++/C#? I googled but couldn't find anythig. Please any help is welcome!I had a look at my code (which i wrote without looking on the web), and it would be difficult to convert this code into a more generic setup..Nevertheless, simply to the following:The MetaPost-Syntax is kind of pretty simple...now make use of the "ofstream"/"sprintf" in C++/C and write all the metapost syntax which you have to use as strings into a file.The dynamical elements (which come directly out of you code, such as calculated grid points (e.g. for dense transformation), etc), have to be included at the right places..I give you an example code fragment - hope this helps you to get the general idea of how to do it:void CreateMetaPostFile(char* fname){ // generate a new file and open it... //... double xscale = 4; // dynamical variable double yscale = xscale / (states[1][gridDim[1]]-states[1][0]); ofstream<<"beginfig(1)"<<endl; // Here we start the metapost-figure ofstream<<"u:="<<xscale<<"cm;"<<endl; // Now write to file the scales of your figure which you have assigned above ofstream<<"v:="<<yscale<<"cm;"<<endl; // Furthermore you may think of calculating some colors which depend on the meshsize( which you calcuate from the grid class) }Hope this helps a bit...

QuoteOriginally posted by: pekolaQuoteOriginally posted by: CuchulainnA number of schemes have been advocated for this PDE in x and y(which has the same form as Asian PDE). It is difficult and easy at the same time, depending on how you tackle it:difficult: no diffusion in y, so 3-point centred differences give problems and are awkwardeasy: the PDE in y is first-order hyperbolic (use Wendroff/Box method), use schemes with 2-pointsThe box method leads to a monotone scheme (M matrix). The problem looks tractable and it would seem that splitting/ADI (especially with centred schemes) is not necessary.Have you looked at such schemes?This is what i wanted to try next: I wanted to use IMEX (unfortunately didn't have the time to try it so far..), which would yield an explicit treatment of the hyperbolic PDE part in the second space direction..for discretisation i would use upwind (2point) - there are also upwind schemes which have 2nd/3rd order such as 5-point-upwinding...but for those oscillatory behavior may come back. Lax-Wendroff would be another alternative for upwinding (and is of second order)I think that these schemes are useful but they do have 'side-effects' that can be messy, 5-point upwinding...Ideally, we want unconditional stable and second order. I have worked the maths for the usual FDM schemes in S and Keller box in A (2nd order, no OSC). Another issue is the imposition of BCs and we can seek an exact solution of a 1st order PDE on the boundary instead of having to use some FD scheme.In another thread, there was a discussion of PDE where NO BC is allowed/needed on a boundary (using a transformation).http://www.wilmott.com/messageview.cfm? ... DBTABLE=So, there is a whole bunch of solutions to this PDE. The best one will be the most efficient and robust.