Managing Volume Risk in a Retail Energy Business.

1
Managing Volume Risk in a Retail Energy Business.

• Jon Stamp
• Head of Portfolio Management, npower Commercial
• 29th January 2008

2
Introduction

• Aims of this presentation
• To focus on Gas Swing as an example of a factor
  that can create substantial volume risk for a
  retail energy business.
• Outline what gas swing is and illustrate it
  graphically.
• Highlight the risks that arise as a result of gas
  swing and that influence the process of modelling
  it.
• Discuss some ways to model gas swing risk so
  that it can be forecast, priced and hedged.
• Suggest possible ways to mitigate the volume risk
  related to gas swing.
• Put forward some areas that provide further
  mathematical challenges.

3
Gas Swing - What is it?

• More problematic on residential side of business
  where most customers are NDM (Non Daily Metered)
  and particularly weather sensitive.
• Seasonally normal demand (SND) is forecast by
  summing customer meters with End User Category
  (EUC), a profile of customer type created by
  National Grid.
• Long term position is established by hedging to
  SND. SND changes monthly as customer numbers
  fluctuate.
• Nearer real time, weather (the main driver of gas
  consumption) forecasts become more reliable and
  have considerable influence on demand levels.
• Short term trading (STT) takes place within the
  final 10 days to balance this.
• Swing is the difference in volume between the
  long term position and STT position and results
  in a change to revenue and a change to cost.
• The mathematical challenge is to model, estimate
  and mitigate the costs of swing.

4
Gas Swing change in volume
5
Gas Swing - The risks

• If demand gt hedged position, need to buy extra
  gas, greater demand pushes prices up, may have to
  buy at a price greater than tariff.
• If demand lt hedged position, need to sell gas,
  lower demand pushes prices down, may have to sell
  at a price lower than tariff.
• Swing cost DfSND ( TF - DA )
• Where TF tariff price, DfSND deviation from
  seasonal normal demand (which is assumed to be
  the hedged position), DA day ahead price

6
Gas Swing - The risks

• DfSND and DA must be simulated to calculate the
  gas swing cost. However they can not be simulated
  separately as they are correlated with each
  other.
• Temperature and demand are approximately -80
  correlated. Implying temperature drives demand,
  this must be modelled with caution because it is
  not always true. For example the increased use
  of air conditioning in the summer can increase
  demand when temperatures are high.
• Demand and price are approximately 48
  correlated. This relationship is most strongly
  recognisable 12 days from delivery when accurate
  weather forecasts become available.
• Modelling these sometimes unpredictable
  relationships can be a challenge.

7
Gas Swing - How can it be modelled?

• Gas Swing models aim to simulate forecasted spot
  price paths. They can then be used to give an
  associated deviation from SND based on the
  relationship between demand and price.
• A pricing equation can be developed using
  Geometric Brownian motion, with Poisson jump
  processes used to simulate price spikes.
• Temperature, and in turn demand, has a tendency
  to experience persistence in the data i.e. any
  days temperature has a correlation to the
  previous days temperature.
• Hence, the demand process is modelled using an
  ARMA (Auto-Regressive Moving Average) to
  establish the specification and consistency of
  the deviation trends.
• Monte Carlo modelling can be used to simulate
  thousands of price and volume deviations.
• The simulations can be used to construct a graph
  showing the distribution of the gas swing costs.
  Probability can be added to this.

8
Gas Swing - The risks (illustrative)

• Hedge to minimise, the risk of incurring and size
  of, gas swing costs. The hedging costs may be
  greater overall but the probability of being
  exposed to large swing costs will have been
  reduced.
• Difficult to estimate the exact effect of the
  hedge. It will also alter as the composition of
  the portfolio changes. Modelling these changes
  can be troublesome.

9
Gas Swing - Mitigating the risk.

• Gas Storage Contracts Physical storage
  facilities, can inject during low price periods
  and withdraw in high price periods. Protects
  against short term price spikes but not against a
  collapse in prices. Need to value the purchasing
  of storage and create a model that forecasts
  optimal timings for injection and withdrawal.
• LNG Gas Storage - Liquefied Natural Gas that is
  available for delivery at very short notice and
  in greater volumes than gas storage. Capacity is
  only available to purchase in an annual auction
  and only provides protection against rising
  prices. Need to create a model for bidding in
  the auction, and for injection and withdrawal as
  above.

10
Gas Swing - Mitigating the risks

• Weather Swaps - Financial instruments that pay
  out when weather is unseasonably high or cold.
  Can protect against selling back gas into a low
  price market or buying in high price market.
  Relies on the correlation between weather, demand
  and price remaining as expected. A model is
  required to value product.
• Swing Contracts - Financial options that payout
  when portfolio of customers demand deviates from
  seasonal normal. Focuses on demand rather than
  weather thus provides some protection when
  weather and demand correlation moves in the
  opposite way to expected. Need to create a model
  to value the product.
• Demand Side Management - Interact with customers
  to incentivise selling back of gas on high price
  days. Need to model the portfolio effect of
  customers altering their consumption behaviour.
  Only applicable to larger customers who are
  usually daily metered.

11
Further mathematical challenges

• Modelling the portfolio effect and impact of
  layering on the hedge, from using a variety of
  products to mitigate gas swing risk.
• Price is often assumed to be normally
  distributed, when it is actually fat tailed
  (Leptokinic). Using Poisson analysis helps to
  capture this, but it is just an overlay to the
  model and it would be more accurate to
  incorporate it.
• Price Volatility is not constant as would be
  expected. One method which could be explored to
  overcome this would be using ARCH (Autoregressive
  Conditional Heteroscedastic) modelling.
• Consumer behaviour is ever-changing, hence models
  need to be developed to factor in these changes
  e.g. increases in price sensitivity and demand
  destruction.
• Historic data reflects the market conditions at
  that time, e.g. doubts over the security of
  supply pushing up prices. These may no longer be
  present e.g. completion of the interconnector
  improving the supply network. To reflect this,
  models need to be created that recognise historic
  regime switches.

12
Backup slide
13
Appendices