241
whose makeup is constantly changing. However, this does not apply to high fre-
quency trading or systems with high turnover by design.
● Specified in advance. The benchmark should be constructed prior to the start of
evaluation.
● Published risk characteristics. The benchmark provider should regularly publish
detailed risk metrics of the benchmark so that the investment manager can compare
the actively managed portfolio risks to the passive benchmark risks.^1Choosing the right benchmark for a portfolio is important because the benchmark estab-
lishes the risk and return parameters for managing the portfolio. There are many bench-
marks to choose from in the market, and making a choice depends on many factors that
are individual to each money manager. Choosing the right benchmark depends on iden-
tifying those that meet certain minimum requirements and then selecting the index that
best matches the investor ’ s goals for the portfolio and the level of risk the investor is will-
ing to assume in order to meet those goals.
Benchmarks should be appropriate, investable, accessible, independent, and unambigu-
ous. We recommend attribution be performed under the Brinson–Fachler method. Returns
are attributed to asset allocation and stock selection effects and presented according to the
industry sector, region credit rating of the company, security type, model rank, etc.26.2.2. Alternative Benchmark Construction
For many strategies, an appropriate index benchmark may not exist. Nevertheless, a good,
clean investable universe will. (The investable universe will have been defined in Stage 2.)
The team can create a randomly selected portfolio from the investable universe for use as
a benchmark that conforms to agreed upon portfolio constraints. This is called the Monte
Carlo portfolio.
A Monte Carlo portfolio simulates multiple portfolios from the investable universe.
These randomly generated portfolios produce a distribution of performances—returns,
standard deviations, drawdowns, etc.—using simple portfolio construction constraints.
The median, mode, upper/lower quartile, upper/lower decile, etc. portfolios can be
used as the benchmarks for risk/reward comparisons, attribution analysis, and SPC dif-
ference monitoring. An alternative way to construct a benchmark is to simply build a
random basket of securities and consider this to be the benchmark. The issue with this
method is that you could have picked the black swan winner and thrown a very good
strategy. (According to Nassim Teleb, a black swan is an event rare beyond the ability
of normal expectations to predict.) The reverse of this is you could pick the black swan
loser and keep a poor strategy. Therefore, we recommend many simulations and using the
median simulated benchmark.
This technique could also be used in qualitative funds to create a benchmark, even if
the manager objects. We recommend all trading/investment system developers and man-
agers embrace the Monte Carlo benchmark as a starting point for kaizen, the continuous
process of measuring performance and implementing solutions that boost performance of
trading/investment strategies.
Money managers must stay focused on beating the competition. By constantly meas-
uring performance versus a Monte Carlo benchmark and quickly adjusting for root causes
of special variation, trading/investment systems can consistently beat competitors that do
not benchmark performance or engage in continuous improvements to strategies.26.2. LOOP 2: DEFINE BENCHMARKS AND ATTRIBUTION