Correlations

Few, if any, energy companies trade single products at the same delivery point. The
cross-correlation between energy products this becomes crucial in estimating the
overall value and risk profile for a particular position.
As stated above, oil correlations are actively traded as crack spreads, locational
spreads and time spreads. Similarly, ‘spark spreads’ between gas and power are
commonly traded and these create a synthetic gas-fired power station. The higher
the correlation between the input gas costs and output electricity prices, the easier
it is to hedge power prices with gas prices and thus reduce the overall risk. In other
words a high correlation will reduce the risk associated with a spark spread option,
as well as reducing the potential upside.
US power markets create particularly complex correlations between power delivered
at different points. There are between twelve and fifteen actively traded hubs in the
USA that show a wide range of correlations, depending on the ability to transfer
power between the regions and similarity of weather conditions. For instance, there
is little or no correlation between COB (California / Oregon border) and Cinergy (near
Cincinnati), while a relatively close correlation is seen between COB and Palo Verde
(in southern California). A matrix of correlations for power was shown in Figure 18.4.
The inherent problem of relying on correlations to hedge risk is that they are prone
to break down particularly during significant market events. For instance, when prices
rose in Cinergy in June 1998 many participants wanted to transport power from the
PJM region to cover their positions. This proved impossible given PJM’s high demand
conditions and the correlations between PJM and Cinergy broke down, leaving some
players with significant losses despite their ‘hedged’ position. Even under ‘normal’
market conditions correlations can vary significantly over time and risk managers
need to be very careful of traders leveraging what seem to be stable correlations.
As stated above, it is also common within the power sector to treat different forward
months as individual products. If this approach is taken it is necessary to link the
contracts by their estimated correlations. While this solves one problem, it creates
another computational one given the resultant matrix and computation necessary to
keep the correlations up to date.