Michael Olson and Russell Walker look at what plan sponsors need in terms of portfolio risk analysis

THE concept of risk management enjoys central importance in modern portfolio management. Since Markowitz first codified the principle of portfolio diversification in 1959, the analysis of the risk/reward tradeoff in managed asset pools has allowed the financial consulting industry to flourish worldwide. However, the need for a unified framework of risk management that explains performance for multi-asset-class portfolios has only recently been tackled in the financial community.

This type of all-encompassing model, with explanatory power across asset classes, will be particularly useful for pension fund management, where plan sponsors and fund managers have increasingly sought analytical tools to decompose performance across many pools of actively- and passively-managed assets. Certainly, many measures of risk exist, and each has its particular application to fund management. An understanding of the concepts, their strengths and weaknesses, and their appropriate use in risk analysis is key. The goal of the portfolio or fund manager should be the successful integration of those appropriate risk measures into a model to analyse total fund performance.

Value at Risk is a well-known risk measure utilised widely in the banking industry, where exposure to financial risk must be managed over very short time horizons. The concept is immensely attractive: one piece of data estimating the maximum dollar loss on a pool of assets over a given time horizon with a 95% or higher confidence interval. However, successful application of VaR analysis requires sophisticated calculation techniques and a thorough understanding of the concepts behind the number. To calculate VaR correctly for multi-asset portfolios, variance/covariance analysis of large amounts of returns data is necessary.

VaR computation becomes more complex as exposure to assets with non-normal return distributions increases; Monte Carlo performance simulation techniques are generally necessary to generate possible asset valuations under multiple scenarios. A 95% confidence interval VaR will still yield one period in 20 on average where the loss will exceed the VaR. For investors with longer time horizons, reliance on VaR analysis may lure them into lower-risk investments than necessary; VaR is a “lower-tail” measure, and upside potential is basically ignored in the focus on short-term downside volatility. From the standpoint of determining the source of performance volatility, VaR has no real role. However, it remains a valuable tool in short time horizon risk management, particularly in the banking industry where asset protection is a vital concern.

Stock fund management was revolutionised by the application of the concept of beta in the 1960s. Regression analysis allowed analysts to quantify historical return volatility for individual stocks and for portfolios, as well as introducing the concepts of systematic versus non-systematic risk and diversification to reduce portfolio volatility. However, the limitations of beta as an analytical tool include its reliance on exclusively historical data, the dependence of its results on the analyst’s definition of “the market”, and beta’s inability to explain the management process or the source of the risk it reports.

Similarly, the application of duration to fixed income portfolio management enabled fund managers to quantify bond market risks and to construct portfolios with acceptable return/risk profiles. One important distinction between duration and beta is that the former is a fundamental characteristic of fixed income instrument – a point-in-time risk factor, and is not described in terms of historical performance. The increasing complexity of fixed income markets has led to the need for more sophisticated tools to describe bond performance. Duration’s role in decomposing bond market risk is still large, but meaningful decomposition of fixed income risk requires more sensitive, multi-factor analysis. The use of duration was an important first step.

In fact, multi-factor analysis of stocks and bonds has created a wide range of highly sophisticated models of equity and fixed income pricing, and has evolved from its roots in the measurement of “size and maturity” factors found in early equity factor models. Such models begin by defining quantifiable, meaningful risk factors, determine periodic returns that can be attributed to these factors for a given market (as often as daily), then calculate an asset or pool of assets’ exposures to these factors.

The next logical step in quantifying total portfolio risk is to unify the disparate models for stocks and bonds into one that meaningfully incorporates the determinants of return inherent in each asset class while also isolating asset allocation risk factors. Indeed, this is key, as asset allocation is by far the most powerful determinant of portfolio return.

Our factor risk models allow fund managers to utilise variance/covariance matrix analysis to isolate point-in-time risk factor exposures. Wilshire’s global risk model decomposes into five major factors:

q Currency risk

q Equity factor risk

q Fixed income factor risk

q Specific (non-diversifiable) risk

q Covariance (factor interaction)

The first three risk factors can be further decomposed into market and asset class-specific risk factors; for equity, these would include market capitalisation and valuation ratios, and for fixed income, duration and other term structure factors. This allows fund managers to effectively pinpoint the sources of their portfolio risk, in terms of the actual portfolio construction process. Managers can view their risk exposures and isolate where they have taken active bets in relation to their benchmark.

A related consideration when approaching risk analysis deals with the performance attribution question, specifically the factors used to decompose return.

Many existing performance attribution models focus on sector/ industry or country/currency analysis, where return is decomposed into weighting and issue selection factors. The fundamental weakness of such models is the lack of insight into the actual management process; simply stated, these models cannot shed light upon the factors the manager uses to build and maintain the portfolios. For performance attribution to have true validity in portfolio analysis, the risk factors (ie, selection factors used by the manager) must be the same as the return factors used to explain the resultant performance. Put another way, for plan sponsors to truly understand their plans’ risk exposures, they must utilise the same factors the portfolio managers use.

In our unified framework for performance attribution, fund managers decompose portfolio performance attribution over time periods as small as one day into the following factors:

q Allocation among the asset classes

q Currency exposures

q Risk factors within each asset class

q Selection

Each risk factor in the model will yield a contribution to the total return that is the product of the asset’s exposure to that factor and the factor return for the time period.

As institutional money management has become truly global in scope, and as the information revolution moves markets to increased efficiency, fund managers must find tools that afford comprehensive risk management and return attribution on a global basis. They should be able to decompose risk and performance into common terms that illuminate the portfolio construction process; the relationship between bets taken and value added can be directly made only when the risk factors and return attribution factors are identical.

Such a global risk management tool, combined with other risk measurement and management tools, will serve fund managers well as they face increasingly complex capital markets into the next century.

Michael D Olson is vice president and principal of Wilshire Associates in London and Russell J Walker is an associate at Wilshire Associates in Santa Monica

## No comments yet