Fundamental Law of Active Management: Skill and Breadth

Two research decks land in the same week. The first belongs to a manager who called four big macro turns over three years and got every one of them right. The second is a quant screen running a thin, steady edge across roughly two hundred names a month. Both decks lean on the same word: skill. The question that actually decides which edge survives at size isn’t who looks more impressive in the meeting. It’s how the fundamental law of active management scores each one. That law gives you a way to weigh raw forecasting skill against the number of separate chances a strategy gets to use it.

What the fundamental law of active management actually says

The law is a compact relationship. The information ratio you can expect from a strategy is approximately the information coefficient times the square root of breadth. The information ratio measures active return per unit of active risk, so it’s the number a serious allocator cares about. The information coefficient stands in for skill. Breadth stands in for how many independent chances you get to apply that skill.

Put numbers on it and the shape gets obvious. A signal with an information coefficient of 0.05 applied across a breadth of 100 gives 0.05 times the square root of 100, which is 0.05 times 10, or 0.50. Halve the breadth to 25 and the same skill produces 0.05 times 5, or 0.25. Skill didn’t change. The number of independent bets did, and the result moved with the square root of that count.

That square root matters more than it first looks. Doubling breadth doesn’t double the information ratio. It multiplies it by about 1.41. To double the ratio from breadth alone you’d need four times as many genuinely independent decisions. Traders who chase scale by piling on names run into that ceiling fast.

The information coefficient: skill you can measure

The information coefficient is a historical measure of association between what a forecast said and what the market later did. In most setups it’s the correlation between the forecast, or the rank a screen assigned, and the return that followed over the horizon you care about. When your top-ranked names tend to outperform your bottom-ranked names, the coefficient is positive. When the ordering is closer to noise, it sits near zero. The guide to how the information coefficient scores a forecast works through the calculation step by step.

The values are humbling. A sustained coefficient of 0.05 is already a real edge. I treat anything above 0.10 as a claim to verify before I trust it, because most durable signals I’ve measured settle between 0.02 and 0.06, and the ones that flashed 0.15 in a backtest usually drifted back toward 0.03 once I tested them on data they’d never seen. A high in-sample coefficient proves little on its own. It has to survive out of sample first, on data the signal never touched during fitting, before it counts as skill.

Breadth counts independent bets

Breadth is the piece almost everyone overstates. It’s the number of genuinely independent decisions a strategy makes, not the count of securities, observations, or trades. Two hundred positions look like two hundred bets. If all two hundred rise and fall together with one factor, the strategy is really making something much closer to a single bet, dressed up two hundred times.

When I audit a screen’s breadth, I count how many of the top twenty ranks are separate, independent bets and how many are the same trade wearing different tickers. On one momentum screen I went through, fourteen of the twenty names were the same high-beta software cohort moving as a block, so the honest breadth was nearer five or six than twenty. The effective number of holdings puts a figure on exactly that gap. Define your investable universe too loosely and you’ll inflate the count without adding a single independent decision.

Why modest skill can beat brilliant-looking skill

Go back to the two decks. The macro manager who nailed four calls has, in the language of the law, very high apparent skill and almost no breadth. Four independent decisions over three years is a breadth of four. Even a spectacular information coefficient of 0.15 on those calls produces 0.15 times the square root of four, which is 0.15 times 2, or 0.30.

The quant screen with the thin edge carries a coefficient of maybe 0.05, but it acts on a hundred independent decisions a month. That’s 0.05 times the square root of 100, or 0.50. The unassuming screen outscores the star manager, and the reason is structural. Small skill compounds when you get many independent chances to express it. Large skill measured a handful of times can be luck, and it never gets enough repetitions to prove otherwise.

This is the same arithmetic that sits under systematic trading. Ralph Vince’s work on edge and exposure makes the case from the money-management side: a small, real edge repeated across many independent bets is what turns a hunch into a track record, while a large edge you rarely get to use stays a story you can’t check.

The transfer coefficient: from forecast to position

The clean version of the law assumes every forecast becomes a position at full size. Real portfolios never manage that. The transfer coefficient captures how much of the forecast actually survives the trip into implemented positions, and it enters the law as another multiplier: the information ratio is approximately the transfer coefficient times the information coefficient times the square root of breadth.

A transfer coefficient of 1.0 would mean flawless implementation. Real books run well below that, and a set of ordinary constraints is what drags it down:

• Concentration limits that cap how large any single position can get.
• Turnover ceilings and portfolio turnover costs that keep you from acting on every fresh signal.
• Liquidity, which stops a forecast on a thin name from ever reaching full size.
• Benchmark rules that force weights you wouldn’t otherwise hold.
• Transaction costs and the slippage between decision price and fill, the gap the literature calls implementation shortfall.
• Missing data, which quietly drops names from the forecast set.
• Correlated signals that pull weight toward the same few exposures.

Take the screen scoring 0.50 before costs. Apply a transfer coefficient of 0.6, common once real limits bite, and the achievable information ratio falls to 0.6 times 0.50, or 0.30. That one multiplier just erased the whole advantage the screen held over the macro manager. Every item on that list is friction you have to price in before you trust a headline number, and none of it prescribes how to build a book.

How apparent breadth gets overstated

Because breadth enters under a square root and drives so much of the score, overstating it is the most flattering error a research pitch can make. It tends to happen in one of a few ways.

Securities move together. In a risk-off week, correlations across a book converge toward one, and the fifty independent bets you thought you held behave like five. That’s the moment correlations break down and hidden concentration surfaces. Factors overlap in the same way. A value screen and a quality screen often pick overlapping names, so stacking them adds far less breadth than the raw signal count implies, a trap that multi-factor composite screening runs into constantly. Ranks can be stable without being independent. If this month’s top decile is mostly last month’s top decile, you’re re-betting one decision and calling it two. And a screen can keep selecting the same economic exposure. A momentum screen in 2020 that filled up with stay-at-home software names held one macro bet under the hood, whatever the ticker count claimed.

Auditing a scalable-edge claim

The best use of this framework is defensive. When someone claims a research process will scale, the law tells you what to ask. Where does the skill come from, and has the information coefficient held up out of sample? How many of these bets are truly independent, or is the breadth padded with names that move as one? What’s the transfer coefficient after real limits, costs, and liquidity take their cut? A pitch that survives those three questions has earned some attention. One that dodges them is selling breadth it doesn’t have.

The limits of the law deserve equal weight. Every input is an estimate. The information coefficient is measured on the past and can shift when the regime changes, so a value that held through a trending market can decay in a choppy one. Independence is genuinely hard to establish in market data, where almost everything correlates with almost everything else under stress. Strategy capacity sets a ceiling that no amount of paper breadth can lift. Treat the law as a lens for interrogating a claim and it earns its keep. Treat it as a formula that guarantees a number and it’ll mislead you, because none of its inputs stay still enough to promise anything.

Where each edge is fragile

Come back to those two decks one last time. The law locates where each claim is fragile instead of declaring a winner. The manager’s edge rests on too few independent bets to prove itself, and the screen’s edge lives or dies on whether its breadth is real and how much of it survives implementation. That’s the value of holding skill, breadth, and transfer in a single frame. You stop being impressed by a high hit rate on a handful of calls, and you start counting independent bets and pricing the friction between a forecast and a filled order. Learn the pattern. Ride the trend. Keep the gains.

Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.