Gross Versus Net-of-Cost Performance

Gross performance measures what a strategy's signal would have earned trading at the mid-price with zero friction — it isolates whether the underlying idea has predictive power at all. Net performance subtracts everything it actually costs to act on that signal: commissions, the bid-ask spread paid on every entry and exit, market impact from moving the price, and any financing or borrow fees. The gap between the two isn't a rounding adjustment — for a high-turnover strategy it can be the entire difference between a strategy worth trading and one that loses money on every single trade despite having a genuinely predictive signal.

Why the gap can be large

Cost scales with how often a strategy trades, not with how good its signal is. A signal that correctly predicts a 0.05% expected move but requires trading in and out of a position to capture it will lose money if round-trip costs (spread plus commission plus impact) exceed 0.05% — regardless of how statistically significant the underlying prediction is. This is why gross performance alone tells you almost nothing about whether a strategy is investable: a signal can be genuinely real, well-validated, and statistically robust, and still be economically useless once realistic costs are applied, simply because it trades too often relative to the size of the edge it's capturing.

What a proper net calculation subtracts

A defensible net-of-cost number accounts for: the bid-ask spread crossed on every trade (roughly half-spread per side for a market order), commissions per share or per trade, an estimated market-impact cost that scales with order size relative to available liquidity (not a flat number — larger trades cost proportionally more), and any holding costs like margin interest or borrow fees for short positions. Skipping any one of these — most commonly market impact, since it's the hardest to estimate — produces a net number that's still optimistic, just less so than gross.

Worked example: a signal that dies at realistic costs

A mean-reversion signal trades 200 times a year, capturing a gross average of 8 basis points per round trip, for a gross annual return before compounding of roughly $200 \times 0.0008 = 16\%$. Estimated costs per round trip: half-spread 3bp each way (6bp round trip) plus commission 1bp plus market impact 4bp at the strategy's typical size — 11bp total round-trip cost. Net return per round trip: $8 - 11 = -3$bp. Net annual return: $200 \times (-0.0003) = -6\%$. A strategy showing +16% gross is actually losing 6% a year once realistic costs are applied — not because the signal is fake, but because its 8bp edge doesn't clear an 11bp cost hurdle.

What this means in practice

Never evaluate a strategy — or compare two strategies — on gross returns alone. The right question is always net-of-realistic-cost performance, and for high-turnover strategies especially, cost estimation deserves as much rigor as the signal research itself, since it's just as capable of determining whether the strategy is viable. A strategy report that shows only gross numbers, or that uses a flat, low, unjustified cost assumption, should be treated with the same skepticism as one with no out-of-sample testing at all.