Skill × Breadth

Start hereThe one equation that decides everything

Richard Grinold's Fundamental Law of Active Management says the risk-adjusted performance you can expect from an active strategy — its information ratio — is roughly your skill on each bet (the information coefficient, the correlation between your forecast and what happens) multiplied by the square root of the number of independent bets you make. That is the whole game, and it is unforgiving in a useful way.

Look at what it implies for us. Our macro direction signal is weak — we measured a liquidity-to-forward-return correlation of about +0.35, and that is a generous read of our skill. With low skill, the equation leaves exactly one lever: breadth. To reach a respectable information ratio on low per-bet skill, you need many roughly-independent bets. One levered bet on the direction of the whole market is the opposite of breadth — it is a single, low-skill wager, amplified. That is precisely the structure that blew up "conviction sizing alone" at −30% in our long-only work; leverage would only deepen the hole.

This is why the fork we discussed isn't really a matter of taste. Cross-sectional — long the regime's winners, short its losers, dozens of positions rebalanced over time — manufactures breadth. It takes our modest skill and multiplies it by a large √BR. Directional leveraged timing starves the equation of the only input we can supply. The Law doesn't tell us the cross-sectional book will work; it tells us it is the only structure where our kind of edge could.

Where the bets come fromThe premia that are long/short by nature

Breadth is worthless if the individual bets have no edge. Fortunately, the most robust return sources in all of finance are relative — they are long/short by construction, which is exactly what we need. Two papers are the anchors.

Asness, Moskowitz & Pedersen (2013) — Value and Momentum Everywhere — shows that buying cheap / recent-winner assets and shorting expensive / recent-loser assets earns a premium across every market they test: stocks, countries, bonds, commodities, currencies. Crucially, value and momentum are negatively correlated, so combining them is far smoother than either alone — more independent bets, again. Moskowitz, Ooi & Pedersen (2012) — Time Series Momentum — shows the trend-following effect that our own fast-cut relies on is a pervasive, sign-based premium in its own right.

The honest caveat lives here too. Harvey, Liu & Zhu (2016) — …and the Cross-Section of Expected Returns — catalogs a "factor zoo" of hundreds of published signals, most of which are statistical mirages that fail once you correct for how many things were tested. The lesson is the one our own peer review just taught us: a handful of economically-motivated, decorrelated bets beats a pile of data-mined ones.

The connective tissueTurning a macro regime into a ranking

Our six-cycle gauge already does the hard part — it reads the growth-and-inflation regime. The bridge to a long/short book is to convert that regime into a cross-sectional expected-return ranking: which sleeves, sectors, or factors are favored now, and which are penalized.

The canonical map is Merrill Lynch's Investment Clock (Greetham & Hartnett) — it rotates asset and sector leadership around the four growth-inflation quadrants, the same 2×2 our gauge produces. For the deeper "why" behind every premium and whether it can be timed at all, Ilmanen (2011) — Expected Returns — is the single best synthesis ever written on the subject; it is the book to keep on the desk.

Concretely, our existing RISK_SPLIT table already ranks sleeves by regime. The long/short version is almost a re-reading of it: go long the top of the ranking, short the bottom, size by the spread. We are not starting from a blank page — we are extending a ranking we already trust.

From forecast to bookWhy shorting is what makes the Law pay

Having ranks is not having a portfolio. Grinold & Kahn, Active Portfolio Management is the standard text for the translation from forecasts to position sizes, risk budgeting, and the optimizer that turns "alphas" into weights. But the most practical insight for us is a refinement.

Clarke, de Silva & Thorley (2002, 2004) add the transfer coefficient — the correlation between the bets you wanted to make and the ones your constraints let you make. Their finding is direct: a long-only constraint quietly throws away much of the Fundamental Law's promise, because you can express a bullish view (buy) but only weakly express a bearish one (you can at most not-own something). Removing the short constraint — even partly, as in a 130/30 book — sharply raises the transfer coefficient and lets more of your breadth reach the portfolio. In their framing, the ability to short is not a bell or whistle; it is how a modest edge is allowed to compound.

Leverage & survivalThe engine we must build first

Leverage changes the failure mode. A long-only book falls; a levered book can be liquidated at the bottom, turning a drawdown into a permanent loss. So the risk engine is not a finishing touch — it is the foundation, and it has a literature of its own.

• Volatility targeting. Moreira & Muir (2017) — Volatility-Managed Portfolios — show that scaling exposure down when realized volatility rises improves risk-adjusted returns, because high-vol regimes don't pay you enough for their risk. This is the vol engine our long-only model conspicuously lacked, and with leverage it becomes mandatory.
• Position sizing. The Kelly criterion (and Thorp's fractional Kelly in practice) sets the growth-optimal bet size and, more importantly, shows why over-betting an edge is ruinous even when the edge is real — the single most common way leveraged strategies die.
• Drawdown as a hard constraint. The CPPI and drawdown-control literature (Grossman–Zhou and successors) formalizes the "cut on the break" discipline our fast-cut already implements — now as a governor on gross exposure, not just net.

The honest priorsWhat usually goes wrong

Keep these pinned to the wall, because they are the base rates. Welch & Goyal (2008) — A Comprehensive Look at the Empirical Performance of Equity Premium Prediction — is the sobering classic: predictors that dazzle in-sample overwhelmingly fail out-of-sample. Most long/short funds do not beat the index net of fees, borrow costs, and financing. Shorting adds asymmetric risk (unbounded loss, squeezes, recall). And every leverage point multiplies both the edge and the mistakes.

None of this is a reason not to build it. It is the reason to build it in the order above, to insist on genuine out-of-sample discipline and point-in-time data, and to judge success against an honest bar — a real improvement in risk-adjusted return, net of everything, that survives outside the window we fit it in.

If you read only one thingThe shelf, in order

If you truly want one: read the Fundamental Law pair (#1). It's the shortest, it's the most decision-relevant, and everything else on this page is a consequence of it.

The bridgeHow this maps to what we've already built

The encouraging part: almost every piece of the long/short design already exists in our long-only engine, waiting to be re-pointed.

So when you're ready, "where do we begin" has a clean answer that falls straight out of this reading: define the mandate, build the vol-and-drawdown risk engine, then prove the cross-sectional edge unlevered before adding a single turn of leverage. The Fundamental Law told us the shape; the risk literature tells us the order.