Practice17 May 2026 · 6 min read

How Transaction Costs Silently Destroy Strategy Edge

A precise look at how spreads, impact, and turnover compound against backtested alpha and why most cost models understate the damage.

A strategy with a 12% annualized return before costs and a 9% return after costs has not lost 3%. It has lost a quarter of its edge, and with it, most of its statistical significance. Transaction costs do not subtract linearly from performance — they compound against turnover, scale with position size in nonlinear ways, and erode the very tail of the return distribution that justifies running the strategy in the first place. Most backtests fail not because the alpha is fake, but because the costs are.

The Components Nobody Models Fully

Retail backtests typically subtract a flat commission and call it done. Real execution cost is the sum of at least five distinct frictions: explicit commissions, exchange and regulatory fees, the bid-ask spread, market impact, and opportunity cost from delayed or partial fills. Each scales differently with order size, holding period, and venue.

Spread cost alone destroys more strategies than commissions ever will. On liquid US equities you pay roughly half the quoted spread on entry and half on exit — for a stock trading 100.00/100.02 that is 2 basis points round-trip, before anything else. On small-cap or international names the spread can exceed 50 bps round-trip, which means a strategy must generate more than 0.5% per round-trip just to break even on friction.

The Turnover Multiplier

The destructive power of costs is governed by turnover, not by trade count in isolation. A strategy that rebalances 100% of its book daily incurs roughly 252 round-trips per year on every dollar of capital. At 5 bps round-trip cost — an aggressive assumption for retail execution — that is 1260 bps, or 12.6% of annual drag. Most "edges" do not survive this arithmetic.

Annual cost drag = Turnover × Cost per round-trip

Net Sharpe ≈ (μ_gross − c·T) / σ

Where μ_gross is gross annualized return, c is per-round-trip cost in decimal, T is annual turnover, and σ is annualized volatility. Notice that cost reduces the numerator but leaves σ untouched — meaning a high-turnover strategy with thin edge sees its Sharpe collapse faster than its returns suggest.

A strategy with a gross Sharpe of 2.0 and 500% annual turnover, paying 10 bps round-trip, loses 50% annualized drag before considering volatility scaling. If gross return is 30% at 15% volatility, net return is −20%. The gross Sharpe was real; the strategy is still uneconomic.

Market Impact Is Not a Constant

Commission and spread are roughly linear in size, at least within the regime where you do not move the book. Market impact is not. The standard square-root model — impact proportional to √(order size / average daily volume) — captures the empirical reality that doubling your size does not double your slippage, it multiplies it by roughly 1.41.

Impact (bps) ≈ k · σ_daily · √(Q / ADV)

Where Q is order quantity, ADV is average daily volume, σ_daily is daily volatility, and k is a venue-specific constant typically between 0.1 and 1.0. The consequence: a strategy that backtests well at 100k capital may be unviable at 5M, not because the alpha decays but because impact grows as the square root of size while edge stays flat.

This is why capacity is a property of the strategy, not the trader. Two strategies with identical Sharpe ratios can have radically different capacity curves. The one trading liquid futures may scale to billions; the one trading microcap mean-reversion may saturate at six figures.

The Asymmetric Damage to Tails

Transaction costs do not just shift the mean of the return distribution — they truncate the right tail disproportionately. The biggest winning trades in most strategies are the ones the model wants to hold, scale into, or chase. These are exactly the trades where urgency increases impact, where the spread widens because volatility is elevated, and where partial fills are most likely. Costs eat winners faster than they eat losers.

The practical effect: a strategy's gross-to-net translation is not a uniform shrinkage. The Sortino ratio degrades faster than the Sharpe. The maximum drawdown gets longer because recovery trades, often taken during stressed liquidity, pay above-average costs. The skew of the net return series is more negative than the gross series — sometimes substantially so.

Building Honest Cost Models

A defensible backtest models costs as a function of three inputs at minimum: the instrument's average spread during the trade window, the order size relative to ADV, and the realized volatility of the period. Flat-bps assumptions are acceptable only as a sanity floor, not as a final answer. For strategies trading across asset classes or liquidity regimes, a single cost number is a fiction.

In Kestrel Signal, cost models attach to the execution layer rather than to the signal — meaning the same signal can be evaluated against optimistic, realistic, and adversarial cost regimes without recomputation of the underlying alpha. The question to ask of any backtest is not "what is the net Sharpe" but "at what cost assumption does the net Sharpe fall below 1.0." The distance between those two numbers is your margin of safety, and it is almost always smaller than you think.