Page 5 of 5

General Remarks and so on

Posted: February 8th, 2012, 3:47 pm
by Cuchulainn
QuoteFor the quant finance and the algos, I just need to ask one question:Do you use the PDE framework or MC in your current scheme ?Both.If you present me with bespoke algos I will try to integrate into my frameworks.

General Remarks and so on

Posted: February 8th, 2012, 3:51 pm
by Cuchulainn
QuoteOriginally posted by: outruncan't you stuff a tenor in a uBlas matrix for now? That would have the benefit for having just dependencies on boost, Altas backend etc.Maybe we can quickly write a run-time dimensional version of this old snippet I used to use myself..It would be great if uBlas would get extended with tensors some day.What you did is similar to what Boost multiarray MA does.1. MA allows slices and ranges which is nice.2. MA is STL-compatible? If not then home-grown STL algos will be needed... What we need is MA -> uBLAS projections. Bingo.

General Remarks and so on

Posted: February 8th, 2012, 3:57 pm
by frenchX
The easiest implementation schemes would be :-Recursive least square Monte Carlo for the Brigo approach. converting that into a PDE wouldn't be easy though. So I suggest for the PDE the approach of Piterbarg and his colleagues at Barclays. There is also a more realistic paper based on multiple curves but I fear it would be a pain too Interest Rates After The Credit Crunch: Multiple-Curve Vanilla Derivatives and SABR I like this one PDE Representations of Options with Bilateral Counterparty Risk and Funding Costs

General Remarks and so on

Posted: February 8th, 2012, 4:04 pm
by Cuchulainn
QuoteOriginally posted by: outrunQuoteOriginally posted by: CuchulainnQuoteOriginally posted by: outruncan't you stuff a tenor in a uBlas matrix for now? That would have the benefit for having just dependencies on boost, Altas backend etc.Maybe we can quickly write a run-time dimensional version of this old snippet I used to use myself..It would be great if uBlas would get extended with tensors some day.What you did is similar to what Boost multiarray MA does.1. MA allows slices and ranges which is nice.2. MA is STL-compatible? If not then home-grown STL algos will be needed... What we need is MA -> uBLAS projections. Bingo.I though MA uses nested containers, inefficient, in directions. UBlas uses linear mem and structures layedout in it.Yes, MA is a wee bit slow. But for batch post-processing OK. In general, FAST = O(1/FLEXIBLE) .. basic law of CS.