Insights

Position Sizing and Concentration Limits for Shorts

Alphanume Team · March 6, 2026

Capping single-name and sleeve risk.

The short side has structural risk asymmetries that demand stricter concentration discipline than long-side strategies. Maximum loss on a long position is the position itself; maximum loss on a short position is unbounded. Single-name short losses can substantially exceed initial position size in adverse scenarios. Position sizing rules that work on the long side are inadequate for the short side.

The asymmetry frame

For a 1% position:

Long: Maximum loss is 1% (stock goes to zero).
Short: Maximum loss is unbounded; a 5x rally produces 5% loss on a 1% position.

The asymmetry suggests two responses: tighter starting size, and active management of in-flight positions to prevent the loss from compounding.

Standard per-name limits

For a systematic event-driven short book in small caps:

Risk class	Standard size	Maximum
Low-risk (large cap, GC borrow)	1.5-2%	2.5%
Standard (small-cap, modest HTB)	1-1.5%	2%
Elevated (deep HTB, low float)	0.5-1%	1.5%
High (squeeze-prone, structural)	0.25-0.5%	1%

The classification depends on per-name borrow cost, float, squeeze risk, and overall portfolio concentration.

Sleeve-level concentration

Beyond per-name, sleeve-level limits prevent over-exposure to single event types:

Maximum per sleeve: 25-35% of total portfolio.
Maximum across two correlated sleeves (e.g., dilution + ATM): 50% of portfolio.
Maximum across all short sleeves: typically 90-100% of portfolio (allowing modest cash buffer).

Sector and vintage limits

Within and across sleeves:

Per-sector maximum: 20-25% of portfolio.
Per-vintage maximum (for time-clustered events like de-SPAC and IPO lock-ups): 30% of portfolio.
Per-investor counterparty maximum (for structured-financing shorts): 20% of portfolio.

The borrow-cost budget

Position sizing also gates on borrow cost. Per-position daily borrow budget:

Daily borrow cost = Position size × Annualized borrow rate / 360

A position whose daily borrow consumes more than the expected-alpha-per-day at the current sizing should be reduced or refused. This is the gating mechanism that prevents over-exposure to deep-HTB names — the borrow cost crowds out the expected alpha.

Adaptive sizing

Position sizes can adjust based on in-flight observations:

Increase on confirmed pre-event drift consistent with thesis.
Hold on neutral price action; let the event play out.
Reduce on adverse moves exceeding pre-defined thresholds.
Close on confirmed thesis-invalidation signals.

The adaptive layer prevents the worst-case losses by exiting positions before they reach maximum-loss scenarios.

Stop discipline

Pre-defined per-position stops:

Adverse-move stops: Close position on a defined percentage adverse move (typically 15-25% from entry).
Volatility-adjusted stops: Adverse-move thresholds calibrated to per-name volatility.
Event-specific stops: Close if the structural thesis is invalidated by post-entry events.
Squeeze-condition stops: Reduce or close on confirmed squeeze setup development.

Stops are not strategy alpha; they are insurance against the asymmetric tail. The expected cost of frequent stop-outs is the price paid for limiting catastrophic-loss scenarios.

Portfolio-level drawdown controls

Beyond per-position management:

Daily drawdown circuit-breaker: If portfolio loses more than X% in a single day, reduce overall gross exposure.
Weekly drawdown threshold: If cumulative loss over a week exceeds threshold, suspend new entries pending review.
Monthly drawdown protocol: Larger drawdowns trigger formal review of sleeve allocations and individual sleeve performance.

The over-sizing trap

The most common short-side error is over-sizing into high-conviction names with poor borrow conditions. The combination of large size + high borrow cost + squeeze risk + small float regularly produces severe drawdowns. Discipline:

Conviction does not justify oversizing.
The harder the borrow, the smaller the position.
Tail risk is the binding constraint, not expected value.