1	Bounding Value at Risk for Energy Companies with an Interval-Based Algorithm Gerald B. Sheblé and Daniel Berleant Dept. of Electrical and Computer Engineering Iowa State University Ames, Iowa 50011 USA
2	The Changing Energy Service Company Environment Energy service company background ESCO = Energy Service COmpany ESCOs are electric power retailers Traditionally, rates are set by a regulator Increasingly, deregulation is occurring Market forces are now driving competition A new risk is the risk of going out of business
3	Value at Risk – the Concept Value at Risk = VaR VaR consists of an upper limit on acceptable financial loss p(loss will be within the acceptable limit) The limit might be how much the company can lose, and still stay in business
4	Value at Risk – the Concept
5	VaR and the ESCO Deregulated energy service companies… are operating in a “typical” competitive mode and must understand and control their VaR A model of VaR for ESCOs is needed To reliably analyze it, we propose to: avoid assuming components are independent use an interval-based analysis algorithm, DEnv
6	The Covariance Method Covariance method is relatively fast, easy It requires a correlation matrix (for the uncertain factors) It assumes the uncertain factors are Gaussian This assumption is not always valid Invalid assumptions lead to incorrect conclusions An augmented or different method is needed
7	The Historical Method Relies on historical data as problem input But historical data may be insufficient, and The current environment may have changed The historical approach is not always best
8	The Monte Carlo method MC involves randomly generating events Events are derived from distribution functions Distributions are often assumed normal (But they don’t have to be) Distributions are often assumed independent (But they don’t have to be) MC is tricky when dependencies among distributions cannot be assigned
9	A Model for ESCO VaR We combine three contributors to risk: 1) Market Price Risk Price fluctuation can cause losses in any business 2) Supplier Contract Failure The ESCO’s energy suppliers may not deliver 3) ESCO Contract Default The ESCO itself may be unable to deliver
10	A Model for ESCO VaR
11	The Model – Market Price Risk Uses a covariance matrix of fluctuations This may be derived from historical data VaR of market price fluctuation is: P is proportion of overall value of... each portfolio component is the volatility (determines confidence) C is the covariance matrix
12	The Model – Supplier Contract Failure Failures may be the fault of GENCOs (generating companies) TRANSCOs (transmission companies) DISTCOs (distribution companies) ICA (Independent Contract Authority) This 2^nd VaR component contributes to losses due to ENS (energy not served) model the 3^rd VaR component also contributes to loss due to ENS...
13	The Model – ESCO Contract Default ESCOs (Energy Service Companies) may fail to deliver due to supplier failures fail to deliver due to operational failures Major operational failure sources are human error computer error strategic error Operational risk combines those sources
14	Energy Not Served (ENS)
15	Energy Not Served (ENS) Power undelivered to customer due to Operational failure of the ESCO Failure of suppliers to deliver to the ESCO Supply chain includes GENCOs TRANSCOs DISTCOs Coordinators ESCO
16	Cost to ESCO of Energy not Served (Recall ESCO is the energy retailer) Most obviously, there is loss of income Failure to deliver may incur a penalty Customer should be compensated (This may be via rebate or direct payment) Compensation should account for: damage to the customer, and need by ESCO to retain the customer
17	Payments Due to ENS Any link in the supply chain to the ESCO ...may cause a failure of supply ...and result in compensation TO the ESCO The ESCO may suffer operational failure ...resulting in payments to the customers Note no standards for this exist yet ESCOs should have a capital fund for this
18	Market Price Risk Energy sales depend on supply contracts Contracts may be spot, forward, or futures Contracts are made by a double auction system Astute bidding strategies are a necessity
19	Combining the Factors Step 1: get a PDF for each factor Step 2: obtain the PDF for the sum of the factors That is, given random variables, what is the distribution of their sum? Complication: The factors are not independent Worse, we don’t know their dependency
20
21
22
23
24	What Next? Strengthen effects of constraints due to partial information about dependency Address calculation of individual components Continue to develop applications of an interval-based algorithm, DEnv Questions?