Benchmarking and Measuring Performance

Scope. MarketClue selects no benchmark on a reader's behalf, publishes no performance rating, and does not describe any result as good or bad. This article explains what a benchmark has to be to mean anything, and why two correct measures of the same portfolio over the same period can differ by more than twenty percentage points a year. All figures are worked illustrations and are not forecasts.

What a benchmark has to be

Five properties, and a benchmark failing any of them will produce a verdict rather than a measurement.

Specified in advance. A benchmark chosen after the period is a benchmark chosen because of the answer it gives.

Investable. If the comparison could not actually have been held — because it excludes costs, or contains things that cannot be bought — then beating it establishes nothing.

Measurable and unambiguous. Its constituents and weights must be knowable, so that the comparison can be reconstructed by someone else.

Appropriate. It has to reflect the same kind of exposure. Comparing a portfolio to a benchmark carrying different risk measures the difference in risk and calls it skill.

Agreed in advance by whoever is being measured. Otherwise the choice becomes an argument conducted after the fact.

Two correct returns for the same portfolio

This is the finding most worth taking from the article, because both numbers are right and they answer different questions.

Time-weighted return measures what the investments did, removing the effect of money going in and out. Money-weighted return — the internal rate of return — measures what happened to the money, including the timing of contributions.

Consider a portfolio over two years: $10,000 invested, up 50% in year one to $15,000; then $85,000 added, taking it to $100,000; then down 20% in year two, ending at $80,000.

Worked example — the same two years, twenty-four percentage points apart. Time-weighted, the portfolio returned +9.54% a year. Money-weighted, it returned −14.49% a year. The gap is 24.0 percentage points annually, and neither figure is wrong. The investments genuinely did well: up 50% then down 20% compounds to +20% over two years. The money genuinely did badly: most of it arrived immediately before the decline, so the large balance experienced the loss and the small balance experienced the gain. Which measure is appropriate depends entirely on who controlled the timing. A manager who did not choose when contributions arrived should be measured time-weighted; the investor whose money it is experienced the money-weighted result and has $80,000. The number quoted in any performance discussion should always be identified as one or the other, and a figure presented without that label is not interpretable.

Tracking error, and what it does and does not say

Tracking error is the variability of the difference between a portfolio and its benchmark. It measures how closely one follows the other, and it is symmetric — it treats outperformance and underperformance identically.

A low tracking error says a portfolio behaves like its benchmark. It says nothing about whether either did well, and a portfolio can track a falling benchmark impeccably.

Three ways performance comparisons mislead

Survivorship. Comparisons against a peer group compare against the funds that still exist. Those that closed are not in the average, and they did not close because they were doing well. This is the same problem Pillar 22 covers, arriving here through the denominator.

Period selection. A start date is a choice, and moving it by a year can reverse a conclusion. A comparison whose start date was chosen after the data existed is not evidence.

Risk mismatch. A portfolio taking more risk than its benchmark should outperform on average, and does so without any skill being involved. Comparing returns without comparing variability is comparing two different things and reporting the difference as performance.