The APT Approach: The Three Pillars of Risk Management
 |
There are many stakeholders for risk managers within investment firms to consider, including senior management of the firm, portfolio managers, shareholders of the firm and regulators. Each of these stakeholders will demand somewhat different analyses of risk, especially in 2009 after the market shocks of the last two years, and so we believe the job of the risk manager has become more complicated than it used to be. Today we believe that effective risk management products must allow three distinct capabilities for portfolio market risk reporting. |
First is the ability to generate “headline risk measures” such as tracking error, beta, volatility, correlation, 99% value-at-risk (VaR) etc. These headline measures must be available on different levels of aggregation from individual books (portfolios) all the way up to the enterprise level.
Second is the capability to create attribution analysis, that is to “drill-down” into the headline risk number to attribute that value to the other obvious characteristics of the portfolio, such as asset class allocation, country and industry allocation, individual position sizes and possibly to factor exposures where the factors may be macro factors, company fundamentals or styles. With this capability the risk manager can create a narrative about the headline risk measures and explain changes in the headline measure much more convincingly. However there is still a striking limitation of such attribution in that it reflects only the static or baseline view of the underlying markets.
The third capability, we believe, is that of scenario analysis (when scenarios are more extreme this is called stress testing). Only with scenario analysis can the risk manager start to question the assumptions under which the headline measures have been estimated, and present different numbers under “shocked” scenarios. The specification of these shocks, however, is not an easy business and requires a good deal of careful thought and economic awareness.
 |
RISK MeasurementThe first capability for a defensible headline measure requires that the model should have at least two desirable properties: responsiveness and robustness. Responsiveness is the ability of the model to change appropriately when the underlying market is changing (particularly the well-observed measures of volatility, risk aversion and liquidity). Robustness is the quality of the model to take account of the non-Gaussian nature of asset return distributions for nearly every class of underlying assets, and the non-linear payoffs associated with many kinds of derivatives. Early models which generated only “parametric” measures based on assumptions of Gaussian behaviour are clearly not to be relied upon. |
|
Risk attributionThe second capability of attribution is most valuable when it is flexible. Investment managers will not pay much attention to a risk attribution which is set out in terms inappropriate to their investment process. Hence the ability to generate several different types of attribution (position-based, macro-based or fundamental factor-based) is very valuable. One of the most fundamental attribution analyses is the breakdown of risk to systematic (shared) and specific parts, which is essential when using a model to evaluate the effectiveness of hedging. |
|
Scenario AnalysisIn the same way scenario analysis is only really useful when it is flexible. Creating many random simulations within a scenario provides only limited value if the simulations cannot be linked to economically intuitive market shocks. In particular the ability to characterise historical scenarios (shocks which have already happened) in a standard form for comparison with forward-looking scenarios is an important requirement for any scenario analysis tool. |