VaR

David,

During today's VaR coordination meeting we had a discussion of issues
related to mapping of the forward price curves into core locations.

Mapping is a necessity dictated by the limitations of the computer system:
we have to reduce the dimensionality of the problem to stay within the bounds
of available CPU memory. Also, in some cases the quality of price discovery
is poor
and it's difficult to model the price curves independently: we solve the
problem by mapping
them into more liquid and better behaved core locations curves.

We have agreed on the following:

1. Winston will investigate the IT side and determine to what extent we can
increase the number
of forward price curves that are simulated as basic (core) curves. He will
investigate the impact of a larger
number of the core curves on the time required to complete the VaR run.

2. The curves associated with the biggest 10-20 positions in each commodity
should be
modeled as core curves (i.e. no mapping into other locations). It makes sense
to monitor
the biggest risks separately and avoid aggregating them into less
transparent aggregates.

3. The results of an automated clustering (mapping) procedures should be
systematically
monitored by a human and corrected if they misrepresent the risks of the
trading positions.
This responsibility should be vested with one person (right now the
responsibility is
dispersed through the organization and this means in practice that nobody
is responsible). Research can allocate one person to this task;
cooperation of trading and RAC will be critical.

Vince