Long-short portfolio construction: how negative allocations change the optimisation problem

Most retail portfolios are long-only by default. The constraint is intuitive (you can only own what you have bought) but it is a meaningful restriction on the optimiser. Relaxing it—allowing negative weights as well as positive—opens the construction problem to more efficient solutions, at the cost of leverage, short-borrow expense, and additional execution complexity.

What long-short construction is

Long-short construction is the portfolio optimisation framework in which weights can be positive (long positions) or negative (short positions). The constraints typically applied are looser than in long-only construction: long weights have no fixed upper bound (subject to leverage constraints), and short weights are bounded below by short-borrow availability or capital constraints rather than at zero.

The unconstrained mean-variance solution—the textbook efficient frontier—does not impose a non-negativity constraint and routinely produces large short positions in the lowest-expected-return assets and large long positions in the highest. Most practical portfolio construction adds the non-negativity constraint not because the unconstrained solution is wrong, but because shorting introduces costs and risks that are not always tractable in a single optimisation. Long-short construction is the framework for cases where those costs are tractable.

A 130/30 structure—130% long, 30% short—is the most common form of constrained long-short. The portfolio is gross-leveraged (total exposure 160%) but net-long (exposure 100%); the short positions enable the additional long exposure without requiring more capital from the investor. Pure market-neutral structures (50% long, 50% short, gross 100%) take the framework further by eliminating directional market exposure entirely.

How it works

The mathematics extends mean-variance optimisation directly. The objective function and the covariance matrix are unchanged; only the constraints differ. Where long-only optimisation requires w_i ≥ 0 for all i, long-short optimisation allows w_i to be any real number, subject to a different set of bounds: a leverage constraint on the gross exposure (Σ |w_i| ≤ leverage cap), a net exposure constraint (Σ w_i = target net), and individual position bounds.

The relaxed optimisation typically produces a higher in-sample Sharpe ratio than the long-only equivalent, because the optimiser can bet against assets with low expected returns rather than simply excluding them. The improvement comes from two sources: the short side captures the low-return assets' negative expected return, and the freed capital from the shorts allows additional long exposure in higher-return assets. In an idealised setting with accurate inputs, the unconstrained solution is strictly more efficient than the constrained.

Out-of-sample, the picture is less favourable. Estimation error in the inputs becomes more damaging in long-short construction because it operates on both sides of the portfolio: a noisy expected-return estimate that overstates an asset's return drives the optimiser to long it; an estimate that understates another's return drives the optimiser to short it; both errors compound in the realised P&L. Robust techniques—shrinkage, position limits, factor-based portfolio construction—are even more important in long-short than in long-only.

What the evidence shows

The earliest implementations of long-short construction in equity markets were the dedicated short-bias hedge funds of the 1970s and 1980s. The strategy spread to long/short equity hedge funds in the 1990s, to 130/30 mutual funds in the 2000s, and to retail-accessible long/short ETFs from the 2010s onward. The strategy class has produced uneven performance over its modern history, with the best implementations capturing genuine factor edges and the worst exhibiting the overfitting tendencies that long-short structure exposes.

Frazzini, Israel, Moskowitz, and others have documented the empirical Sharpe ratio improvement from long-short factor portfolios relative to long-only equivalents over multi-decade samples. The improvement is typically meaningful (long-short Sharpe approximately 0.6–0.9 vs long-only equivalents of 0.4–0.6 for the same factor) but smaller than the in-sample efficiency gain would suggest, because of the implementation costs and the noise multiplier.

The 2018 quant shock and the more recent value-momentum drawdowns have highlighted the specific vulnerabilities of long-short factor construction. When factor returns reverse, the long-short structure amplifies the loss because both legs lose simultaneously: the long leg falls and the short leg rises. Simple long-only factor portfolios suffer the long-leg loss only, which is a meaningfully shallower drawdown.

Limitations and trade-offs

The most binding practical limitation is execution. Short positions require borrowing the underlying asset, paying a borrow fee that varies by stock and by market conditions, and posting margin against the position. Hard-to-borrow stocks command very high borrow fees that can erase the strategy's expected return; easily-borrowed names compete with all other short interest for the available supply, which compresses returns when many participants are positioned similarly.

Leverage introduces its own vulnerabilities. A 130/30 structure has 160% gross exposure on 100% of capital; the additional volatility relative to a long-only equivalent is meaningful, and the strategy can be forced to delever in drawdowns just as more direct leveraged structures can. Risk management within long-short construction must explicitly account for leverage and gross-exposure dynamics in addition to standard portfolio risk metrics.

Long-short structures are also tax-inefficient in many jurisdictions. Short positions can produce ordinary income rather than long-term capital gains, and the resulting tax bill can erode the strategy's after-tax return materially. The framework is most efficient in tax-deferred or tax-exempt accounts; for taxable retail portfolios, the after-tax case for long-short is weaker than the pre-tax case.

Long-short construction in pfolio

pfolio's Portfolio Builder allows negative allocations to assets, and the Asset Builder supports synthetic short positions via -1 leverage. Combined, these features let investors construct long-short portfolios across the multi-asset universe and analyse them within the same risk and return framework as long-only portfolios. Risk metrics are visible in pfolio Insights.

Related articles

• Market neutral strategies: generating returns independent of broad market direction
• Mean-variance optimisation: the algorithm behind optimal portfolio construction
• Leverage in investing: how borrowed capital amplifies returns and losses
• Factor investing explained: momentum, value, quality, and low volatility
• 130/30 strategies: extending the long-only framework with a controlled short overlay
• Long-short equity strategies: how taking both sides of the market changes the return profile