The pair sleeve: dollar-neutral, factor-neutral, and what's left over

The pair sleeve is the strangest object in the catalog. Net exposure pinned at zero. Factor exposure pinned at zero. No index hugs, no benchmark, no tracking error to budget. What's left, after dollar- and factor-neutralisation, is the cross-sectional residual — and a small return stream uncorrelated with the long-only book sitting next to it. The question this post answers: is that residual worth its operational weight, and how does it behave through regimes.

Tax alpha lives in the long book — in the dispersion of single names against the index, harvested over time. The pair sleeve adds a second, smaller, mostly-orthogonal return stream: the cross-sectional residual after factor exposures are neutralised. Together — a tax-aware DI core plus a small pair sleeve — the household carries two independent return streams. Sharpe-on-Sharpe, the combination dominates either piece.

max   αᵀ w   −   λ · wᵀ Σ w   −   τ(w)   −   φᵀ |w|
 w

s.t.  Σ w = 0                          (dollar-neutral)
      Bᵀ w = 0                          (factor-neutral, hard equality)
      Σ |w| ≤ 2 · NAV                   (gross-exposure ceiling)
      offsetting-position wash-sale on long/short pairs of same name

The two budget identities define the construction:

• Σ w = 0 — every long dollar has a short dollar facing it. Net exposure to the market is zero in expectation. Realised net beta drifts off zero because B is estimated, but it stays below 0.05 in backtest.
• Bᵀ w = 0 — the active weights have zero loading on each named factor (size, value, momentum, low-vol, profitability, market). What the optimizer can express is purely the cross-sectional residual.
• α as input. The signal α can be anything — analyst forecasts, an internal quant model, a third-party score. The strategy module is signal-agnostic; it just runs the construction.

The pair sleeve uses the same risk model as the long-only DI strategies — Σ from a 504-day return panel under Ledoit–Wolf shrinkage, factor loadings B from a 6-factor decomposition. The only new input is the per-name alpha signal. The backtests below substitute a synthesised, slowly-decaying score that's been calibrated to roughly match the autocorrelation pattern of 12-month price momentum without using any momentum loading directly (so the factor neutrality bites).

Because there's no benchmark, the comparison is the sleeve's cumulative NAV against cash plus borrow accrual — which is what the holder gives up by funding the sleeve from the long-only book's idle margin. The sleeve runs at $200k notional (so $100k long, $100k short) and 0.05% target single-name weight.

At first glance the sleeve loses to cash. That's the right read: in a regime where the holder's risk-free alternative pays ~5%/yr (the cash + borrow line in the chart includes the ~150 bp/yr borrow accrual saved by not running the short leg), the residual on this signal — slow-decay, unweighted — only delivers ~3%/yr. The sleeve isn't useful as a standalone return stream. It's useful as a diversifier of the long-only book's beta-loaded return stream.

Run the same construction with a higher-quality α (information coefficient ~0.06, vs ~0.025 for the slow-decay signal above) and the sleeve's annualised return roughly doubles. Run it with no signal at all — purely tax-loss-driven cross-sectional rotation — and it goes negative net of borrow. Signal quality is the load-bearing assumption; the construction is signal-agnostic but the economics aren't.

• The sleeve is not a hedge. The factor-neutral constraint removes average beta exposure, not realised beta in a drawdown. In sharp dispersion events, realised net beta can spike to 0.15+ for short windows.
• Tax efficiency is mediocre. The sleeve trades both legs aggressively against the signal; harvest opportunities exist but are smaller in magnitude than in the long-only DI book. Don't run the sleeve for the tax alpha.
• The construction uses an equality constraint Bᵀw = 0, which can be tight to satisfy in small universes. We relax to ‖Bᵀw‖∞ ≤ ε with ε = 1e-3 in production.

For the strategy this sleeve usually pairs with, see Tax-aware direct indexing in full. For an alternative way to add long-side dispersion exposure without giving up the index, see the long/short deep-dive.

Educational illustration · numbers illustrative.