Beruflich Dokumente
Kultur Dokumente
Outline
Objective Objecti e Collateralized Debt Obligation Basics CDO on the Cell/B.E. A preliminary result Conclusion
Objective j
Objective Demonstrate the competitive edge of the Cell/B.E. on CDO pricing using Monte Carlo simulation with Gaussian Copula No intention to develop new models for CDO pricing No Why CDO? The fastest growing sector of the asset-backed securities market. cco d g SIFMA, g oba C O issuance increased to $488.6 , global CDO ssua ce c eased $ 88 6 According to S billion in 2006, nearly twice the $249.3 billion issued in 2005. CDO is challenging to price. Monte Carlo simulation has been the most popular method for CDO valuation. Monte Carlo simulation can be very resource intensive for large CDOs CDOs. Seems to be the good fit for the Cell/B.E.
CDO Basics
A Collateralized Debt Obligation (CDO) is an asset-backed security backed by a diversified pool of defaultable instruments like loans, junk bonds, mortgages, etc. If the portfolio contains only credit default swaps (CDS), it is called a synthetic CDO CDO. It is structured as multiple tranches and sold to investors. Each tranche has different priority to claim on the principal. Separate out the risks by prioritize the receipt of principal among the investors.
Assets sold to the SPV Principal & interest
Cash
Cash
Funding
Loss
Originating Bank
SPV
Senior 30-70%
Detachment point - d
Attachment point - a
Distribution of Losses
Loss given default amount of the ith reference obligation:
Li = (1 Ri ) N i
where Ni is the notional amount and Ri is the recover rate. The accumulated portfolio loss is
n
L(t ) = Li 1{ i t }
i= i =1
CDO Pricing g
Losses due to defaults (the issuer fails to satisfy the terms of the obligation) are the main source of risk as payoffs. Estimate the present value of tranche losses due to defaults default leg (floating leg) T r (u ) du
DL = E e 0
t 0
dLa ,d (t )
Calculate the present value of the premium payments weighted by the outstanding capital premium leg (fixed leg)
i r ( u ) du w 0 PL = sa ,d E i e min{max[d L(ti ) 0] d a} ), ], i =1 T
The fair price of the CDO tranche is defined to be spread such that the expected value of both legs is equal.
* sa , d =
T r ( u ) du E e 0 dLa ,d (t ) 0
t
Default intensity or hazard rate of a given firm determines its default time. time
First Generation Cell/B.E. 90 nm 241M transistors 235mm2 9 cores, 10 th d threads >200 GFlops (SP) >20 GFlops (DP) Up to 25 GB/s memory B/W Up to 75 GB/s I/O B/W >300 GB/s EIB Top frequency >4GHz
(observed in lab)
Cell/B.E. Features
Heterogeneous multi-core system architecture SPE
SPU SPU SPU SPU SPU SPU SPU SPU
Power Processor Element for control tasks Synergistic Processor Elements for dataintensive processing
Synergistic Processor Element (SPE) consists of
SXU LS MFC
16B/cycle
SXU LS MFC
SXU LS MFC
SXU LS MFC C
SXU LS MFC C
SXU LS MFC C
SXU LS MFC C
SXU LS MFC C
Synergistic Processor Unit U it (SPU) Synergistic Memory Flow Control (MFC) Data movement and synchronization Interface to highperformance Element Interconnect Bus
11
PPE
PPU L2 L1
MIC
BIC
PXU
Dual XDRTM FlexIOTM
32B/cycle 16B/cycle
Generate Normals
Random Numbers: O(N2p) ( p) Generate Default Times: O(Np) Sort: O(pN logN) Calculate Payments: O(Np)
13
N = 624, M=397 Vector t ti f V t starting from location l ti (i+1) or (i+M) may not be quadword aligned. Computation of latter part of array requires updated data from the first M entries Data dependence
14
Optimization on SPE
U t Use two random number vectors a & b d b t Redo if condition fails for any pair of random numbers
Overheard due to skipping of perfectly normal random numbers
15 CDO Pricing on Cell/Lurng-Kuo Liu/Virat Agarwal 2007 IBM Corporation
Time (in seconds) to generate 100 million random numbers in sequential and block pattern on various architectures.
* Source: http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/SFMT/speed.html
16 CDO Pricing on Cell/Lurng-Kuo Liu/Virat Agarwal 2007 IBM Corporation
2 2 .1 2 0 .0 0
Time (seconds)
1 .0 0 .8
1 2 .4 1 0 .6 6 9 .9 8 .3 6 .6 6 .3
0 .6 0 .4 0 .2 0 .0 In te l_ 1 .4 In te l_ 3 .0
A M D _ 2 .4
P P C _ 1 .3 3
C e ll
17
Performance comparison of our optimized RNG (Mersene Twister) as compared with other Cell/B.E. implementations
7.7
Running Time (seconds)
3
2.7
1
Performance comparison of our optimized RNG (with Normalization) as compared with other Cell/B.E. implementatoins
20
Time (in seconds) to generate 100 million normalized random numbers on a single SPE.
18 16 14 12 10 8 6 4 2 0
32-bit 64-bit
2.3
2.2
* Vectorized Random Number generation ecto ed a do u be ge e at o available with Cell SDK 2.1
18 CDO Pricing on Cell/Lurng-Kuo Liu/Virat Agarwal
Cholesky decomposition on correlation matrix C -> LLT , where L is a NxN lower triangular matrix
M difi d version of the Gauss Algorithm Modified i f th G Al ith
19
Also working on utilizing the lower triangular property of the matrix L, to achieve , better performance.
21
Conclusions
CDO pricing is computationally intensive instead of communications intensive. We use Monte-Carlo simulation
Highly scalable among various SPEs g y g
Initial Performance results Show substantial speedup for Mersenne Twister and Normalization
as compared to other architectures
Cell is a good fit for financial workloads. Double precision is essential for FSS workloads
22 CDO Pricing on Cell/Lurng-Kuo Liu/Virat Agarwal 2007 IBM Corporation
23