Model risk

Many of the issues outlined above primarily exist or are exaggerated in the spot
markets where many quantitative estimation tools are less effective and the resultant
model risk becomes very significant. One of the major reasons for hedging into
forward markets should be to avoid this model risk.
Model risk can come from a number of sources, but are principally driven from
forward curve models (particularly the volatility forward curve) and pricing models
(option, simulation and optimization models). It is generally recommended that
traders and risk managers have a preset (and audited) library of pricing tools to work
from, but in many cases longer-term contracts will require the correct building blocks
to correctly price a particular offering. One example of where such pricing can go
wrong is taking account of correlations between different components of a deal in
hedging a variable retail load.
It is common in both the gas and power markets to offer retail hedges where
the volume sold is linked to actual consumption of a customer group. In these
circumstances there is a positive correlation between the volume being hedged and
market clearing prices (the customer group’s consumption itself being part of the
demand driver in prices). This correlation changes at different consumption levels
following a generally exponential curve.
A simplistic approach could make a mistake on two fronts. If you ignore the
extent of the positive (non-linear) correlation between price and demand you could
significantly under-price the hedge. Above certain overall market demand levels,
prices will rise exponentially, which will occur exactly as the volume requirement of
the hedge increases. While most models will take account of the cost of increased
demand at high prices, they often do not simulate the non-linear relationship, thus
under-pricing the option.
A number of early power deals reportedly made this mistake when taking on such
positions. They hedged full requirement positions (supply all the load a group of
customers take) at expected volumes. With increasing volatility in the marketplace
and a positive correlation between prices and the customer’s consumption they then
found themselves long at lower than expected prices and short at higher than
expected prices, resulting in a significant overall loss compared to the hedged
position.
The alternative problem is to ignore the fact that the call option is not being
optimized against high prices (customers use energy when they need it, not when it
has most value in the market) and it is a positive but by no means perfect correlation.
If the customer’s ‘call option’ to increase their volume is priced, ignoring this the
option will be significantly over-priced. A diverse portfolio of different customer types
dissipates much of this call risk.
As well as the problem associated with volume uncertainty, energy contracts
abound with complexity reflecting the immaturity of the market (many older contracts
still in force reflect the realities of a regulated utility world) and physical realities of
the market price. Combinations of average price options, look-back index options,
knock-out features, swing contracts and time spread options are common. In addition,
it is not uncommon to see contracts against terms such as ‘average market
prices’ rather than a specific published index, making the translation from contract
to model more complex (as well as opening up a number of other pricing issues).
To minimize model risk, an extensive testing program needs to be put in place to
ensure this is quantified on an on-going basis. The backtesting of model estimates
should be used to estimate potential model errors and incorporated into the hedging
strategy. In particular the assumptions surrounding mean reversion in the volatility
curve (which significantly reduces the volatility over time) need to be carefully tested
against actual marks wherever possible given the lack of implied curves that can be
derived. Market illiquidity will make this process difficult, often resulting in arbitrary
allocations between liquidity and modeling risk allocations. A risk manager needs to
be very careful of such allocations to ensure consistency across different products
and markets. 546