» Introduction.

» What Is Economic Capital ?

• We aim to arrive at a figure which usefully represents the overall risk of a portfolio of financial exposures.
• We try to capture all the key risk effects.
  - There are various candidate measures that we can use.
  - There are different ways of computing these measures.

» Background - An Evolution of Risk Measurement.

• Exposure Analysis.
• Expected Loss.
• Economic Capital.

» Portfolio Loss Distribution.

• Move the focus :
• From Expected Loss,
• To Unexpected Loss.

» Applications of Economic Capital.

• Quantify Risk and Risk Appetite.
• Efficient holistic portfolio-level risk control.
• Effective contextual sub-portfolio analysis.
• Risk/reward analysis.
• Active portfolio management and optimisation.

» Applications, For Example:

• Risk-based pricing of loans.
  - We can price for risk in the context of a portfolio.
• Guide strategic business decisions.
  - Which types of new business will give the best incremental risk-reward performance ?
• Pro-active portfolio re-balancing.
  - We can adjust exposure profiles dynamically.

» 'Simple' or 'Complex' Risk Measures ?

• Simple measures are useful, but often do not really capture 'Risk'.
• We want to understand how risky a portfolio really is.
  - We want to understand 'unexpected loss'.
  - Take into account concentration and correlation.
  - Identify 'toxic' risk combination effects.

   •  Exposure Analysis

       →    Expected Loss

           →     Variance-Based Measures

               →      Credit Value-At-Risk

                   →     Beyond Credit VAR
          

» Exposure Analysis and Expected Loss.

 - Exposure Analysis

     Exposure is a very simplistic measure.

• This is useful to an informed analyst, but is not, in itself, a measure of 'risk'.
• This requires a basic set of information, and minimum modelling assumptions.
• We must have internally consistent data sets.

 • Expected Loss

  - Expected loss takes into account how much of each exposure might be lost.

• E-L = Exposure * PD * (1 - Recovery)
• E-L is the best estimte of the loss likely to be incurred.
• 'Expected' is in the statistical sense, not 'anticipated'.
• Is E-L a measure of risk ? - Surely 'Risk' is about uncertainty !

» Variance-Based Risk Measures.

• We can use the standard deviation of a loss distribution as an estimator of uncertainty.
• This tells us about the 'width' of the distribution.
• For asymmetric distributions we can use their 'root semi-variances'.
• This is very useful, but does not necessarily capture information about the tails of the distribution.

» Credit Value-At-Risk, and Beyond.

 • We can use 'confidence level' VAR measures

• Suitable confidence levels are risk sensitive.
• Confidence levels define a 'Loss Threshold', they do not tell us how big losses might be beyond this value, so :
  - 'Value-At-Risk' can be a bit of a misnomer !
  - Values can be unstable and hard to decompose.

Beyond Credit VAR

 - We prefer coherent statistical risk estimators :

• These tend to be more computationally stable.
• 'Sub-additivity' means that we can often do more sensible risk decomposition.
• Their properties mean that we can have more confidence in doing risk-trend analysis.

» What do Model Results Really Mean ?

• We should consider the possible impact of both data errors and modelling assumptions.
• It does not always necessarily matter that data and model assumptions are not exactly correct.
  - We compute risk indicator statistics that combine and distill the key risk factors in a sensible way.
  - We can do sensible drill-down analysis and trend analysis.
  - We can continually improve data, models and processes.

» A Risk Management Process Model.

Resulting in :

• Improved risk management.
• Better use of capital.
• Enhanced risk/reward.

» The Impact of Basel-2.

• It should not need regulators to convince banks that portfolio risk analysis is worth doing !
• Basel-2 recognises the benefits of 'advanced' models, and their use in practice.
• Regulatory requirements are driving changes in risk reporting, data management and business processes.

» Conclusions.

• Use a range of risk measures, that capture both expected and unexpected loss.
• Prefer coherent risk measures, especially for drill-down and trend analysis.
• The fact that approximations and assumptions are used does not invalidate the process.
• Use risk models proactively, and continuously review the risk management process.