Sie sind auf Seite 1von 5

Attribution Analysis for Equities

Craig B. Wainscott, CFA Manager, Plan Sponsor Products Frank Russell Company

Performance attribution has a mixed reputation, but properly conducted, attribution can assure clients that apparent positive results do indeed stem from manager skill rather than luck. Attribution that consistently ties added value to manager decisions over time is an important source of management information.

Performance attribution is a methodology to quantify the success of or value added by a strategy. In that sense, it is not much different from traditional performance measurement, in which one set of returns is subtracted from another and the result is the value added. Performance attribution goes one step further, however, to identify the sources of value, and if used properly, it can thus distinguish between skill and luck. When the sources of added (or lost) value have been identified, the sponsor or client can decide where to focus attention. If things are going well in one area but not so well in another area, the sponsor can spend more time on the area where performance is below expectations. Performance attribution is defined in many ways. The aim of this presentation is to describe a powerful and broadly accepted approach to performance attribution and also to provide some warnings about its careful and proper use.

Approaches to Equity Performance Attribution

The two main approaches to performance attribution are factor models and return decomposition. The factor model approach typically takes the list of securities in an equity portfolio at a point in time and, on a security-by-security basis, tries to attribute performance to various factors (beta and size, for example); the approach is then to sum those factor effects at the portfolio level and create a report of which factors were responsible for what amounts of return. Few systems use the actual return that the portfolio achieved over time; they focus on the returns of specific securities at a particular time. Return decomposition takes the opposite ap- Dissecting the Net Management Effect proach by asking how the actual rate of return on a The ultimate aim of performance attribution is to portfolio was achieved. Starting with that actual re73

turn, this approach focuses on the sources of added value relative to a benchmark and examines the different segments that contributed to the return. It separates the return by primary sources and can be based on the most detailed, accurate data available. The return decomposition approach has a number of benefits. First, it is fairly easy to interpret because the numbers add up: Here is the index, here is the portfolio return, and the attribution effects explain the difference. It is also fairly easy to calculate; the math is no more complicated than adding, subtracting, and multiplying. It is flexible; as long as the necessary underlying rates of return and cash flows are provided, the data can be examined and used in any number of ways. Finally, the data used for return decomposition are familiar-standard rates of returns, cash flows, and weights in either sectors or asset classes. One drawback is that, although the numbers all add up nicely when the method is applied for a single period, that additive property is lost when compounding over time is undertaken. For users who are comfortable with the geometric nature of returns, this issue is not important. But for those who are new to complex performance analysis, the aesthetics of an additive solution can make the difference between enlightenment and frustration. Another issue is the "interaction effect," which is a cross-product that must be dealt with in this type of performance attribution. Volume of data can also be a challenge; each time the attribution analysis is cut finer, the amount of data required to do the analysis compounds.

describe the net management effect-the difference between the portfolio return and the benchmark. The point is not to explain the total portfolio return. If the portfolio achieves 12 percent and the benchmark 10 percent, where did the 2 percent come from? What are the activities that actually added value? What was the net management effect? The net management effect-positive or negative-for each equity portfolio can be dissected into the value added by two or three pieces, depending on whether the portfolio is purely domestic or global: (1) allocation, which for a traditional equity portfolio is the industry groups or economic sectors, (2) security selection, which is the stocks picked, and for a global portfolio, (3) currency management. Keep in mind in relation to these three pieces that to attempt performance attribution without some idea about the style or the intention of the manager being measured is dangerous and potentially useless. Attribution is a tool that helps interpret the results of decisions, but one must be certain that the results come from intentional decisions, not simply luck. To illustrate the concept of performance attribution, Table 1 contains a typical attribution report for a U.s. small-capitalization equity portfolio. This type of presentation clearly shows differences in returns and the sources of those differences. It uses portfolio and benchmark weights and returns by industry group to show allocation, selection, and total effects. The basic formulas for the allocation, security selection, and interaction effects are as follows. These formulas are for a single period only, howevertypically one month-and ignore the effects of compounding; thus, the formulas will not directly generate the results shown in Table 1. The benchmark returns in the formulas are local currency returns:

Allocation effect = (Portfolio weight - Benchmark weight) x (Benchmark segment return - Benchmark total return);

Security selection effect = (Portfolio segment return Benchmark segment return) x (Benchmark weight); Interaction effect = (Portfolio weight - Benchmark weight) x (Portfolio segment return - Benchmark segment return)

Allocation Effect
To analyze the effect of allocation, performance attribution compares the weight of the portfolio in a sector or industry with the benchmark weight (the average weight during the measurement period is used) and compares the return to that sector in the benchmark with the total return of the benchmark. For example, focus on the allocation effect for the "health care" industry group in Table 1. The weight for this group is 16.6 percent in the portfolio and only 13.5 percent in the benchmark, so the portfolio manager overweighted the health care group. The return in the benchmark for health care was 85.2 percent versus a total return for the benchmark of 46.5 percent. The portfolio manager thus overweighted an industry group that significantly outperformed the benchmark as a whole. The result is a positive allocation effect of 1.78 percent. A similar calculation can be made for each sector for the portfolio, and the results add up to the total effect of allocation on this portfolio of 5.15 percent; that is, the total value added from allocation to this portfolio's performance is just over 5 percent.

Security selection Effect


Security selection is the other effect typically examined in attribution analysis for U.s. equity port-

Table 1. Equity Performance Attribution: Allocation, Security Selection, and Interaction Effects: One Year Ending December 31
Portfolio Industry Technology Health care Consumer discretionary Consumer staples Integrated oils Other energy Materials and processing Producer durables Autos and transportation Financial services Utilities Other Total aNet management effect. Weight 16.8% 16.6 31.0 5.0 0.6 3.9 0.3 8.9 16.4
~

Benchmark Weight 16.1% 13.5 17.3 3.1 0.0 5.3 14.2 7.0 0.0 15.2 6.9 1.3 100.0% Return 52.56% 85.20 51.79 28.50 8.92 -3.83 29.84 29.42 41.06 65.27 28.59 56.08 46.46%

Attribution Effects Allocation 0.26% 1.78 0.85 -0.37 2.99 1.49 1.32 -0.79 -2.32 -0.08 5.15% Selection -1.07% 6.09 -2.63 0.70 3.61 2.66 -5.13 5.00 1.65 1.77 12.64% Interaction -0.51% 0.10 -2.33 0.24 -3.35 -1.62 4.66 -2.54 1.62 -1.03 --4.77% Total -1.32% 7.98 --4.11 0.58 3.25 2.53 0.84 1.67 0.95 0.65 13.02%a

Return 46.80% 133.73 37.37 44.67 45.15 51.31 -25.04 94.30 51.12 201.01 59.48%

100.0%

Source: Craig B. Wainscott.

74

folios. In performance attribution for security selection, allocation to the sector does not matter; what matters is how well the stocks perform in that sector compared with the benchmark. The size of the negative or positive effect, however, does depend on the sector weight. Note from Table 1 that this portfolio manager achieved a return of about 133.7 percent for the health care group, which was weighted at 16.6 percent in the portfolio. This sector's return for the benchmark was about 85.2 percent. Therefore, the portfolio performed considerably better with its health care stocks than the benchmark did. The resulting security selection effect was nearly 6.1 percent. Indeed, for the total portfolio, security selection contributed a positive 12.64 percent; that is, this manager added more value from security selection (nearly 13 percent) by picking better stocks than the benchmark within each industry group than the manager added from allocation (about 5 percent) by picking better industry groups.

implemented passively. If a manager is focusing on stock picking, the allocation effect can be considered something of a residual. Most or all of the attribution for the cross-product should be in the security selection because that is what the manager is trying to do well. If the interaction effect is not explicitly identified in a report, it is more than likely combined into the selection effect. Asking where it is included is a fair question; the answer gives some insight into the manager's focus.

Compounded Effects
Attribution effects for a single period can be compounded for multiple periods. Table 1 illustrated an annual period, and Table 2 provides an example of compounding for four years. The table includes the interaction effect as well as the allocation and selection effects, the actual annualized returns for the benchmark and the portfolio, and the net management effects. These results are straightforward and easy to interpret; they represent a tradeoff, however, between complete precision and ease of understanding. Compounding the attribution effects generates precise results but does not produce totals that are simply additions of the period-by-period effects; the figures could be left in the compound/untotaled (and more precise) form, but Frank Russell Company has found that explaining performance attribution to clients is easier if the effects add up than if they are absolutely precise. The size of the difference in the two approaches is typically small (often no more than 10 basis points), and that level of detail is less important than the conclusions clients should reach. In the Frank Russell Company approach, therefore, the effects are calculated on a monthly basis and compounded; then, the differences are rounded off and smoothed so that the effects appear to be additive again.

Interaction Effect
The interaction effect is a residual when the allocation and selection effects are calculated in their purest forms. It is not merely a leftover, however, with no information content. The interaction effect shows the impact of the allocation and selection effects working together. For example, recall from Table 1 that the health care segment had a positive allocation effect of 1.78 percent and a positive selection effect of 6.09 percent. This manager apparently had confidence in being able to pick superior stocks in this sector and backed that confidence by overweighting it. These two positive results combined to create an interaction effect of 0.1 percent. When the interaction effect is buried, where it is placed should depend on the manager's approach. If a manager is truly approaching the portfolio composition in a top-down fashion, focusing on picking industry groups or countries, the interaction effect is usually combined into the allocation effect. For this manager, picking stocks is either not a focus or is

Currency Effect
Allocation, selection, and interaction effects apply to both domestic and global portfolios. Global or inter-

Table 2. Equity Performance Attribution: Total Effects for Four Years


Annualized Returns 3 6 December Months Months Russell 2000 (modified) Allocation Selection Interaction Actual return Net management effect
Source: Craig B. Wainscott.

Calendar Years 4 Years 14.2 2.3 7.8 -3.7 20.7 6.4

1 Year 46.5 5.1 12.6 -4.8 59.5 13.0

2 Years 8.5 2.3 10.2 -7.3 13.7 5.2

3 Years 11.2 2.0 11.0 -5.4 18.8 7.6

19XX -19.6 0.6 8.1 -8.0 -18.9 0.7

19XX 16.7 1.4 12.7 -1.3 29.5 12.8

19XX 24.0 3.01 -2.5 2.1 26.7 2.6

19XX 19XX -8.4 1.9 22.0 -D.l 15.4 23.8 5.2

8.1 1.1 5.2


-D.4

14.0 5.9

6.0 1.8 4.5 0.5 12.8 6.8

14.6 3.6 8.7 0.1 27.0 12.4

75

national portfolios, however, are also subject to the attributions to skill simply because of poor benchmark definition. effects of currency moves. Because they are important measures of global or international managers' Inappropriate benchmarks are common with decisions, currency effects on the portfolio in comglobal and international equity portfolios because parison with currency effects on the benchmark managers tend to have biases for or against certain should be isolated in performance attribution analycountries. The objective of attribution analysis, howsis. ever, is to measure the impact of decisions, so the Sponsors and managers should note that much benchmark should reflect the options among which research has been and continues to be performed in the manager can choose. For example, if the portfolio this area-using single- and multiple-currency is allowed to include emerging markets, the chosen frameworks. Brian Singer, for example, has develglobal benchmark should include them. oped a framework for global attribution that separates currency effects from market and interest rate effects. 1 Conclusion

Benchmarks
Because performance attribution is based on differences between a portfolio's return and a benchmark return-and not in terms of total returns alone but on industry-by-industry and stock-by-stock baseshaving the right benchmark is critical to the analysis. An inappropriate benchmark, such as a large-cap benchmark for a small-cap portfolio or a broad market benchmark for a growth portfolio, can result in

Performance attribution is an important source of management information. If a client can define the types of decisions the portfolio managers are making and where they are trying to add value, and if the client can see that value is added as a result of those decisions, the information generated by attribution analysis can reassure the client that the good results are coming from skill rather than luck. The client will be even more reassured if the results of the performance attribution analysis show consistency over time.

1 See Mr. Singer's presentation, pp. 86-90.

76

Question and Answer Session


Craig B. Wainscott, CFA
Question: Is attribution analysis used to monitor managers more than it is used to select managers?

Wainscott: Yes, but not by a large margin. Attribution is a great tool for use in selecting a manager, and it should be used for that purpose. The reason it is not used is generally a lack of data. Manager data are usually available from the trustee or a consultant, but when numerous prospective managers are being screened, all the data on industrylevel cash flows, weights, and returns that are needed to conduct the attribution analysis often are not available. Some large consulting firms or research houses will provide that type of data, however, and many analysts are actively doing performance attribution as part of manager searches.
Question: Is attribution analysis available for market-neutral strategies?

look at risk in terms of factor exposures; large allocations to cap size, for example, or other factors that might affect the portfolio, need to be incorporated when measuring either performance attribution or a manager on an ongoing basis. With market risk, if a manager hikes up the portfolio beta, the manager may be able to construct a portfolio that is, in a sense, industry neutral. The manager or client might imagine that the manager is getting rewarded for some skill when, in fact, all the high beta means is that the manager has simply taken on more market risk.
Question: What groupings other than economic sectors are most important in attributing performance?

tual results of a strategy, which is an additional reason why knowing the underlying strategy is important. For example, the interaction effect often will become noticeable when a strategy focused heavily on an industry that had a huge cash flow and a sizable return. Such a strategy causes some of the numbers (weights and returns) to loom very large and can cause a large interaction effect. In such a case, looking closely at the individual components and becoming familiar with the math in order to understand the interaction effect is important.
Question: In calculating the value added from security selection, why not multiply the return difference by the actual weight instead of the benchmark weight, which would cause the interaction effect to disappear?

Wainscott: Market-neutral strategies introduce problems with benchmarks and data. Correct attribution is very difficult when derivatives are involved, and most market-neutral strategies are option-based strategies.
Question: In attribution analysis, does risk matter?

Wainscott: The most important two are cap size and economic sector. At Frank Russell Company, we regularly analyze attribution for those two factors, and for some managers, we also carry out performance attribution on the bases of price-to-book ratios, price-to-earnings ratios, dividend yields, and growth in earnings per share.
Question: As the number of time periods being joined together increases, does the interaction effect become a huge factor, perhaps even larger than the other two effects?

Wainscott: If you use the actual weight, you are automatically burying the interaction effect in the selection effect.
Question: Some people can trade only at high trading costs and others can trade at low cost. How do differences in trading costs show up in attribution analysis?

Wainscott: Risk matters. Risk can be considered in terms of the volatility of returns; we know we should never look at returns without looking at the risk experience through time. We also need to

Wainscott: The interaction effect can be large, but it is not so much a function of time and how much you compound as of the ac-

Wainscott: Trading costs are simply part of the return achieved by the portfolio, so they will all go into the security selection effect. Low costs will raise returns for the stocks in an industry group or portfolio.

77

Das könnte Ihnen auch gefallen