Volume 3 — Valuation and Risk Metrics © Copyright 2002, CCRO. All rights reserved. 22 Maintaining the Model The model needs to be maintained and periodically validated to ensure the quality of the results. • Back-test the model by comparing the forecast results with the historical actuals and note any significant differences. • Identify any factors that may have caused the significant differences. These factors may be present in the forecast methodology, the exposure mapping, or in the difference between assumptions used to calculate the target and the earnings metric. • Refine the model to minimize the variations. 5.0 Asset Risk Metrics 5.1 Discussion of Metrics For assets, the greatest risk-management benefit is obtained by analyzing not only the MTM and fair value, as discussed in Section 2.0, but also some metric of the dispersion around this value. For measuring risk, the approach needs to be different from that applied to proprietary trading. The risk metric offered here is a variant of the one normally used for trading and marketing activity. In addition, the risk can be measured in terms of the dispersion of the value over the life of the asset that is under consideration and does not connote the value that will be available through liquidation. The objective is to give a measure of the degree of variability over the period when market liquidity exists. 5.2 Calculation To measure the risk of the market years, the user calculates the fair value at risk, which is the VaR associated with the portfolio over the period when market liquidity exists while simulating risk factors influencing price and volume. The following are general principles useful in this calculation: • Based on the VaR methodology employed, the user calculates the VaR by commodity, location, asset, tenor, etc. The data used in the calculation can include the data from the market years for the assets, the hedges, or the entire portfolio to show where the risks are coming from by commodity, by tenor, with hedges, without hedges, etc. • The holding period is associated with the length of time estimated to elapse before hedges can be put in place and reduce to a minimum the variability in the expected revenue projections. This calculation considers only the relevant tenor. • When the fair value of the operating revenues from the assets is calculated, it generates not only an expectation but also a distribution, which can be discrete (in most cases when either a scenario-based projection or a simulation-based model is used) or continuous (when a rough tolling model is used, with many operational details presumed by abstraction).