This is a summary of the specification of work discussed at the meeting at
Askari on 12 May 97. Present were Peter Davies, Maria Berenguer and Zoran 
Hruskar of Askari, Paul Coddington and Chao-Wei Ou of NPAC. This work would
be done by NPAC for the initial phase of the project, funded by a $25K 
Infomall membership fee from Askari. Work would be completed by July 1, 
possibly running into early July if there are unforseen problems.

The following requirements, in order of priority, are to be met:

1) The three basic programs - the calibration program to fit historical data
(Laplace), the standard Monte Carlo simulator (Markov), and the path integral
Monte Carlo simulator (PIMCS) - will be ported to Windows NT using the Visual
C++ 5.0 compiler and programming environment. 
Currently some of the existing codes compile and run under this environment 
and some do not.

2) Benchmarking of the current programs will be done using the J.P. Morgan 
reference data set of approximately 500 rates. This data set is available
from the J.P. Morgan web site, Askari will provide other necessary initial
data such as a correlation matrix. NPAC will write routines for parsing this
data. Initial benchmark will be a 5 year simulation using monthly intervals.
This initial benchmark will provide a baseline for future efforts to improve 
and optimize the program. 
A priority of the benchmarking effort will be to identify what simulation
times and parameters (number of paths etc) are required to produce a given
level of statistical error. 

3) Specification of an interface to a library of high-level Monte Carlo 
routines, so that these routines can be called directly from Askari's program. 
In particular, this should include a routine for returning the status of the 
simulation (i.e number of paths computed, and an error estimate).

4) Implement multi-threading to run the simulation in parallel on a 
multi-processor NT machine.

5) Allow variable timesteps, which will be specified in the input data (e.g.
daily or weekly timesteps for the first year, monthly timesteps for the rest
of the simulation).

6) Add log-normal and binomial distributions.


The final two points are optional extras. In this phase, NPAC will determine
the amount of effort required to add these capabilities to the current code. 
If in either case this turns out to be a simple exercise, and if time allows, 
the routine will be added to the existing program. However it is likely that
these will not be implemented until the next phase of the project.
 
7) Compute interest rates using the Hull and White model.

8) Add checkpointing.


A weekly status report will be sent to Askari by email.

Milestones:

May 31
- routine for returning program status 
- identify (and hopefully fix by now) problems with porting code to NT 
- initial runs using J.P. Morgan data set 

June 15
- working NT code (possibly with multi-threading by now)
- specification of library interface 
- add variable timestep
- determine effort required for checkpointing
- preliminary benchmarking results

July 1

- multi-threading
- add lognormal and binomial distributions
- determine effort required for Hull and White model
- final benchmarking results