Vysakh Nachiketus

Melita Jaric

College of Business Administration and
School of Computing and Information Sciences
Florida International University, Miami, FL, USA


Zhang Zhenhua
Yang Le

Chinese Academy of Sciences, Beijing , China
Motivation
Why financial derivatives
Why the pricing of financial derivatives is complex
Why distributed environment
Why Monte Carlo / Binomial Method / Finite Difference Method
Proposed Work
For a given criteria continuously update a diversified portfolio of European, American,
Asian and Bermuda Options
Given current price, estimate the future stock option value by implementing
Monte Carlo or Binomial Method in grid computing environment
Provide a framework for correlating the processing speed with the portfolio performance

Conclusion

2009 Financial Derivatives Proposal
Why Financial Derivatives?

Building block of a portfolio
Current Importance/Relevance
Complexity of algorithms
Spreading the market risk and control

Why is pricing of financial derivatives complex?

Uncertainty implies need for modeling with Stochastic Processes
High volume, speed and throughput of data
Data integrity cannot be guaranteed
Complexity in optimizing several correlated parameter

2009 Financial Derivatives Proposal
Why distributed environment?

Time is money
Grid computing is more economical than supercomputing
Exploit data parallelism within a portfolio
Exploit time and data precision parallelism for a given algorithm

Why Monte Carlo or Binomial Method?

Ability to model Stochastic Process
Ubiquitous in financial engineering and quantum finance
They have obvious parallelism build into them, since they use two dimensional grid
(time, RV) for estimation
For higher dimensions Monte Carlo Method converges to the solution more quickly
than numerical integration methods
Binomial Method is more suitable for American Options

2009 Financial Derivatives Proposal
Standard options
Call, put
European, American

Exotic options (non standard)
More complex payoff (ex: Asian)
Exercise opportunities (ex: Bermudian)
2009 Financial Derivatives Proposal
Black Scholes Equation &
Stochastic Processes
Integration of statistical and mathematical models

For example in the standard Black-Scholes model, the stock price evolves as

dS = (t)Sdt + (t)SdWt.
where is the drift parameter and is the implied volatility

To sample a path following this distribution from time 0 to T, we divide the time interval
into M units of length t, and approximate the Brownian motion over the interval dt
by a single normal variable of mean 0 and variance t.

The price f of any derivative (or option) of the stock S is a solution of the
following partial-differential equation:

2009 Financial Derivatives Proposal
2009 Financial Derivatives Proposal
2009 Financial Derivatives Proposal
The value of the replicating portfolio at time T, with stock price S
T
, is
S
T
+ e
rT
B where is the no of share and B is cash invested
At the prices S
T
= S
o
u and S
T
= S
o
d, a replicating portfolio will
satisfy
( * S
o
u * e
rT
) + (B * e
rT
) = Cu
( * S
o
d * e
rT
) + (B * e
rT
) = Cd
where S
o
u is the stock price when price goes up
Solving for and B and substitute in
C = S + B to get the option price





In the field of mathematical finance, many problems, for instance the problem
of finding the arbitrage-free value of a particular derivative, boil down to the
computation of a particular integral.

When the number of dimensions (or degrees of freedom) in the problem is large,
PDEs and numerical integrals become intractable, and in these cases Monte
Carlo methods often give better results.

Monte Carlo methods converge to the solution more quickly than numerical
integration methods, require less memory , have less data dependencies and
are easier to program.

The idea is to use the result of Central Limit Theorem to allow us to generate a
random set of samples as a valid representation of the previous value of the stock.

The sum of large number of independent and identically distributed random
variables will be approximately normal.


2009 Financial Derivatives Proposal



2009 Financial Derivatives Proposal
St = S
0
e
(a 0.5

2)t
+ tZ
V S
n
e V S T
rT
T
i
i
n
( , ) ( , )
0
1
0
1

If V(S
t
,t) is the option payoff at time t, then the time-0
Monte Carlo price V(S
0
,0) is
where ST
1
, , ST
n
are n randomly drawn time-T
stock prices
Option Pricing Sensitivities
The Greeks
2009 Financial Derivatives Proposal
2009 Financial Derivatives Proposal
Define Stock Input as a 7-tuple
( Ticker, Price, Low, High, Close, Change, Volume)
Implement FAST decompression to get the actual data
Select the stocks that satisfy specified criteria
Use hashing to assign each stock to a particular processor
Create a dynamic storage management database
Collect and correlate data
Update portfolio


2009 Financial Derivatives Proposal
2009 Financial Derivatives Proposal
http://www.gemstone.com/pdf/GIFS_Reference_Architecture_Grid_Data_Management.pdf
drift = mu*delt;
sigma_sqrt_delt = sigma*sqrt(delt);
S_old = zeros(N_sim,1);
S_new = zeros(N_sim,1);
S_old(1:N_sim,1) = S_init;
for i=1:N % timestep loop
% now, for each timestep, generate info for
% all simulations
S_new(:,1) = S_old(:,1) +...
S_old(:,1).*( drift + sigma_sqrt_delt*randn(N_sim,1) );
S_new(:,1) = max(0.0, S_new(:,1) );
% check to make sure that S_new cannot be < 0
S_old(:,1) = S_new(:,1);
%
% end of generation of all data for all simulations
% for this timestep
end % timestep loop
MATLAB program for Monte Carlo
2009 Financial Derivatives Proposal
Peter Forsyth 2008
function [Pmean, width] = Asian(S, K, r, q, v, T, nn, nSimulations, CallPut)
dt = T/nn;
Drift = (r - q - v ^ 2 / 2) * dt;
vSqrdt = v * sqrt(dt);
pathSt = zeros(nSimulations,nn);
Epsilon = randn(nSimulations,nn);
St = S*ones(nSimulations,1);
% for each time step
for j = 1:nn;
St = St .* exp(Drift + vSqrdt * Epsilon(:,j));
pathSt(:,j)=St;
end
SS = cumsum(pathSt,2);
Pvals = exp(-r*T) * max(CallPut * (SS(:,nn)/nn - K), 0); % Pvals dimension: nSimulations x 1
Pmean = mean(Pvals);
width = 1.96*std(Pvals)/sqrt(nSimulations);

Elapsed time is 115.923847 seconds.
price = 6.1268
MATLAB program for Asian Options
2009 Financial Derivatives Proposal
www.fbe.hku.hk/doc/courses/tpg/mfin/2007-
2008/mfin7017/Chapter_2.ppt
2009 Financial Derivatives Proposal
function MC (S, r, T, sigma, nSamples, nP)
dataRand[nSamples] = rand();
start = idP*p/nSamples; end = (idP+1)*p/nSamples;
for each idP
mean_value[idP] = findMC(S,E,r,T,sigma,dataRand[start,end));
end
reduce(mean_value[idP]);
endfunction
Iterations are independent - Embarrassingly parallel
Parallel Random Number Generation: Master Processor
SPRNG (Scalable Pseudorandom Number Generation) library
2009 Financial Derivatives Proposal
function Bin (S, E, r, T, sigma, nTime )

for step = nTime:-1:1
for i=1:nStep+1
value(i) = (pUp*value(i+1)+pDown*value(i))*exp(-r*dt);
price(i) = price(i+1)*exp(-sigma*sqrt(dt));
value(i) = max(value(i), price(i)-E)
endfunction
2009 Financial Derivatives Proposal
2009 Financial Derivatives Proposal



Explicit Finite Differences
Careful: dt and dS are not independent
Use Ghost points to minimize the communication
Communication comes from one side only

Number of processors vs. time for each method
Computation and Communication Overlap
Communication Overhead
Design and implement an algorithm for optimizing performance for any
combination of:
number of stocks
number of processors
latency


2009 Financial Derivatives Proposal
Provide this system to individual investors through cloud computing.
Implement advanced financial hedging techniques for Fixed Income,
Future Exchanges
Address system Reliability by checking for failure, introducing
redundancy and error recovery
Address system Security by implementing encryption algorithms
Introduce different sources of information (news, internet) and trigger
warning alerts to support automated trading.

2009 Financial Derivatives Proposal
We propose to develop a software system for scientific
applications in finance with following characteristics:

Runs in distributed environment
Efficiently processes and distributes data in real time
Efficiently implements current financial algorithms
Modular and scales well as the number of variables increases
Processes multivariable algorithms better than a sequential time system
Expends logically for more complex systems
Scales well for cloud computing so that even a small investor can afford to use it
Provides an efficient and easy to use infrastructure for evaluation of current research

2009 Financial Derivatives Proposal
1. Peter Forsyth, An Introduction to Computational Finance Without Agonizing Pain
2. Guangwu Liu , L. Jeff Hong, "Pathwise Estimation of The Greeks of Financial Options
3. John Hull, Options, Futures and Other Derivatives
4. Kun-Lung Wu and Philip S. Yu, Efficient Query Monitoring Using Adaptive Multiple
Key Hashing
5. Denis Belomestny, Christian Bender, John Schoenmakers, True upper bounds for
Bermudan products via non-nested Monte Carlo
6. Desmond J. Higham, An Introduction to Financial Option Valuation
7. Alexandros V. Gerbessiotis, Architecture independent parallel binomial tree option price
valuations
8. Bernt Arne Odegaard, Financial Numerical Recipes in C++
9.Paul Wilmott, Introduces Quantitative Finance

2009 Financial Derivatives Proposal