Skill Times the Square Root of Breadth: Grinold's Fundamental Law

Two portfolio managers walk into a room. The first is a brilliant stock picker who makes ten high-conviction bets a year and gets them right 60 percent of the time. The second runs a quantitative strategy across a thousand stocks, and its predictions are barely better than a coin flip, right maybe 51 percent of the time.

Which one is the better manager?

Most people's instinct says the first. Grinold's Fundamental Law says: almost certainly the second, and it tells you exactly why.

The problem: how do you compare skill across completely different strategies?

Active managers get judged on the information ratio: how much return they generate above their benchmark, per unit of the risk they take relative to that benchmark. It is the Sharpe ratio of the value you add. It is the right yardstick.

But the information ratio is an outcome. It tells you what happened. It does not tell you where it came from, and it certainly does not tell you how to build a strategy that will have a good one. Grinold wanted to decompose it into its underlying drivers, so that a manager could reason about what they are actually selling.

The key idea via analogy: the casino, not the poker player

Think about how a casino makes money on roulette. Its edge on any single spin is tiny, roughly 5 percent. If the casino ran one spin per year for a million dollars, it would be an extremely risky business, and in many years it would lose money.

But the casino does not do that. It runs the same tiny edge millions of times, in small amounts, over and over. The edge is unchanged. What changes is that the noise averages out across an enormous number of independent bets, and what remains is the edge, arriving with beautiful reliability.

The casino is not a great gambler. It is a mediocre gambler operating at massive breadth. That is the entire insight.

Grinold formalized it. His law says:

Information ratio is approximately the information coefficient multiplied by the square root of breadth.

Two ingredients:

Information coefficient (IC): your skill. How well your forecasts correlate with what actually happens. An IC of zero means your forecasts are worthless. An IC of 1 means you are omniscient. Real, good quantitative signals live in the range of about 0.02 to 0.10. Yes, really. A correlation of 0.05 between your forecast and the outcome is a valuable signal in this business. That is how weak real predictive power is.

Breadth (BR): how many independent bets you make per year. The word independent is doing enormous work here, and we will come back to it.

The relationship contains two lessons that pull in opposite directions on your intuition.

Lesson one: skill enters linearly. Double your skill and you double your information ratio. Skill is precious and directly rewarded.

Lesson two: breadth enters as a square root. Making four times as many bets does not make you four times better. It makes you twice as good. Breadth has diminishing returns, and you must add a lot of it to move the needle.

Put them together and you get the strategic trade-off that the whole active management industry lives inside. To match a manager with double your skill, you need four times their breadth. That is a hard climb, but it is a climb that is available to you. Skill is extremely hard to improve. Breadth, for a systematic manager, is often a question of engineering: cover more stocks, more markets, more asset classes, rebalance more often.

This is precisely why quantitative investing works. A quant strategy has a pitiful IC that would embarrass a discretionary stock picker, and it makes up for it by having breadth in the thousands. The concentrated stock picker has genuinely higher skill and almost no breadth, so their results are dominated by luck, and you cannot tell whether they are good for a very long time.

Why it mattered

• It gave quantitative investing its intellectual justification. Before Grinold, defending a strategy with a 51 percent hit rate was awkward. After Grinold, you could show precisely why a weak signal applied broadly beats a strong signal applied narrowly, and why the resulting business is more reliable.
• It turned "diversification" into an alpha statement, not just a risk statement. Everyone knew diversification reduces risk. Grinold showed that diversifying your bets, not just your holdings, directly manufactures information ratio.
• It gave managers a design language. If you want a better information ratio, you have exactly two levers: improve the signal, or make more independent applications of it. That framing is now how systematic strategies are architected, and it is the basis of Grinold and Kahn's book "Active Portfolio Management," which became the standard text for the industry.
• It explains why huge funds struggle. As you grow, capacity constraints force you into fewer, larger positions in the most liquid names. Your breadth falls. Your information ratio falls with the square root of it. Size is the enemy of breadth, and therefore of skill-adjusted performance.

The honest limitations

The limitations are so serious that Grinold and Kahn themselves, and later Clarke, de Silva and Thorley, spent substantial effort patching them.

• "Independent" is the word that ruins everything. The formula counts independent bets. If you hold 500 stocks but your signal is really one big bet on the value factor, then your true breadth is closer to 1 than to 500. Correlated bets do not average out. Nearly every naive application of the law overstates breadth wildly, which is why the formula so often predicts information ratios that nobody achieves.
• It assumes you can act on your forecasts freely. In reality, you face constraints: no shorting, position limits, sector neutrality, turnover budgets, transaction costs. Every constraint stops you from expressing your view fully, and the achieved information ratio falls below the theoretical one. The "transfer coefficient," added in later work, measures exactly how much of your signal survives the constraints, and it is often shockingly low.
• It assumes your IC is constant and known. It is neither. IC varies over time, decays as a signal gets crowded, and is itself estimated with error from a limited history.
• It invites gaming. Because breadth enters positively, a manager can inflate their apparent breadth by trading more frequently or splitting one idea into many correlated pieces. That produces more trades, not more information, and mostly produces more transaction costs.
• The derivation rests on simplifying assumptions. It is an approximation derived under a particular setup, not an identity. Treat it as a way of thinking, not as a calculator.

The one-line takeaway

Grinold showed that your performance as an active manager is roughly your skill multiplied by the square root of the number of independent bets you make, which is why a mediocre signal applied a thousand times beats a brilliant signal applied ten times, and why the word "independent" is where almost everyone goes wrong.