How the Skill Score works

1. The record is measured week by week

For a fund, the record is what you would have earned by copying its filings: its 50 largest reported US stock positions, at their reported weights, bought at the first weekly close after each filing became public and held until the next one. Funds file up to 45 days after a quarter ends, and that delay is part of the result.

Each week’s return is compared with index funds of the fund’s own style. Three things describe a style: how much market it holds, whether it leans toward small or large companies, and whether it leans toward cheap (value) or expensive (growth) ones. We find the mix of four index ETFs (SPY, a small-company fund, a value fund and a growth fund) that moves most like the fund, and that mix is its benchmark.

Why not just SPY? Because a style can beat SPY for years without anyone picking a good stock. When the largest companies led the market, any fund holding them looked skilled against SPY. Against a benchmark with the same lean, it does not. What is left after the benchmark is the active return. Its average, divided by how much it swings, is the measured skill, stated per year. (This is an information ratio.) Recent years count more: a week from three years ago carries half the weight of this week.

2. The record is adjusted for luck

Among thousands of managers, some will look brilliant by chance alone. So a measured skill is not taken at face value. We first learn how skill is actually spread across every manager on file, including those that stopped filing. The answer is lopsided: most have none, and a few have a lot.

Each record is then weighed against that spread. A short or erratic record is pulled most of the way back to the typical manager. A long, steady one mostly keeps what it measured. The result is the luck-adjusted skill, with a range around it.

3. The score is what we are 80% sure of

The score does not come from the best estimate of skill. It comes from the level we are 80% sure the true skill exceeds. That is why a long steady record outranks a short spectacular one, and why a manager needs two years on record before being ranked. The scale is fixed, not graded on a curve: if every manager improved, every score would rise.

A fund is labelled “Proven edge” only when the chance that it has no skill is under 10%. So far none is: across more than 6,000 managers’ filings since 2013, the best records are about as good as luck alone would produce, and funds that ranked well in the first half of that period did no better than average in the second. Beside the score, each fund page shows the edge in percent a year, the record as it was measured before the adjustment, and the plain return against SPY, which is not adjusted for anything.

What the score cannot tell you

• It rates a replica of reported holdings, not the fund’s own return. A 13F leaves out short positions, most options, bonds and non-US stocks.
• Stocks that have since been delisted are missing from older filings. A filing is skipped when less than 70% of its dollars are in stocks we can price.
• The style benchmark covers market, size and value. Other leans, such as holding recent winners or one industry, still count toward the score.
• A past edge is evidence, not a promise. Nothing here is investment advice.