Value-at-Risk for energy

As in many markets, VaR has been accepted as the most common way of communicating
risk parameters within the energy sector. A detailed description of VaR itself
is addressed elsewhere in this book.
VaR is driven by four principal factors: volatility, correlations, holding period and
confidence interval. Given the first two parameters VaR models are going to be
susceptible to many of the problems outlined above. However, the holding period
and confidence interval assumptions have mixed effects. A holding period of ten days
or less is common in the actively traded energy markets. This allows for a relative
lack of liquidity compared to financial markets and the ‘communication delay’ – it
may take 24 hours to identify a problem, another 24 to agree on the way forward.
A five-day holding period will tend to avoid much of the mean reversion and
seasonality issues, provided the volatilities have been estimated correctly. You can
thus give a reasonable five-day estimate of VaR despite many of the issues outlined
above. The concern is that it gives a very incomplete picture of your risk profile. In
particular, the chairman and CEO is often more concerned about potential major
losses. In other words they will be more worried about the 5% rather than the 95%.
While a VaR can give some clues to the maximum loss, with significant spikes the
losses may be many multiples of the VaR estimation.
Also, many of the traded products cannot be hedged within 10 days (many cannot
be hedged at all). In this case the holding periods need to be re-assessed in light of
the particular product. The overall VaR associated will then be affected by the time
to maturity, ability to find partial hedges and mean reversion of volatility.
Also, as we discussed above, optionality is pervasive in energy and the more
traditional VaR needs to be expanded and complemented to take account of the full
range of option risks. It other words, delta, gamma, vega and theta risks need to be
quantified and added to the VaR calculation. Where significant optionality exists a
simulation approach is likely to be most effective, although this may be at the cost
of timely information.