Borrow Cost and the Short Rebate
Even when a short is available, it isn't free. The rebate you earn on the cash collateral rarely covers the fee a lender charges for a hard-to-borrow name, and a backtest that ignores the fee can turn a real profit into a real loss.
Prerequisites: Short Borrow Availability
Shorting a stock means borrowing shares, selling them, and posting the sale proceeds as cash collateral with the lender. The lender pays interest on that collateral — the rebate — but also charges a borrow fee, and for anything other than a large, liquid, easy-to-borrow name, the fee is the number that matters. A backtest that models shorting as "sell the stock, earn the rebate rate, done" is quietly assuming every short is general collateral. For the small-cap, high-short-interest names many short strategies specifically target, the borrow fee alone can exceed the entire expected alpha.
Worked example: the short that's right about direction and still loses money
A signal shorts a small-cap name expecting an 8% decline over the holding period — and it delivers, the stock is down 8% when the position closes. If the backtest only tracks the fee-free case, that 8% flows straight to the bottom line.
Now add the real borrow cost. This name is a well-known crowded short, general-collateral rate is near zero, and the actual stock-loan fee quoted by the desk is 12% annualized, charged for the roughly six months the position is held:
| Component | Value |
|---|---|
| Price move (correct direction) | +8% (gain to the short) |
| Borrow fee, annualized | 12% |
| Holding period | 6 months |
| Borrow cost incurred | 6% (half of 12%) |
| Net P&L | +8% − 6% = +2% |
The trade was directionally right and still returned a quarter of what the naive backtest reported. Push the fee to 20% — not unusual for a genuinely scarce borrow — and the same correct call turns into a net loss of 2%, even though the stock did exactly what the model predicted.
Model borrow cost as an ongoing carry expense proportional to the fee rate and the holding period, applied to every short position, not just an occasional line item — for hard-to-borrow names it is frequently the single largest cost in the trade.
The names a short-selection signal ranks most attractive are disproportionately the ones with the highest borrow fees, since scarcity itself is often correlated with the same crowding the signal is picking up on. A flat, low fee assumption calibrated on easy-to-borrow names will understate cost most exactly where the signal is most active.
Where historical fee data isn't available, use short interest as a percentage of float as a rough proxy — fees rise sharply, often nonlinearly, once that ratio moves into double digits — and apply the carry cost daily rather than only at position close.
Related concepts
Practice in interviews
Further reading
- D'Avolio, The Market for Borrowing Stock