Transaction costs reduce backtest returns by a quantity that scales with turnover — spread paid on every trade, market impact growing with order size relative to volume, and timing slippage between signal and fill — and the documented consequence is strategy re-ranking: gross-return leaderboards routinely reshuffle once realistic costs apply, with the highest-turnover strategies falling hardest, sometimes from profit to loss. UZU NEWS publishes information, not investment advice, and prices the friction before believing the promise.
Every backtest is a claim about net returns — what an investor would have kept. Gross simulations ignore the three frictions that stand between the two, and the frictions are not small or uniform: they scale with exactly the quantities — turnover, size, urgency — that many attractive backtests maximize. This explainer covers the cost model components and the discipline of reporting net.
What are the components?
Four, each measurable. Commissions and fees: the explicit, scheduled part — small in modern equity trading, larger in some derivatives and markets with transaction taxes. Bid-ask spread: the immediate round-trip cost, paid at the width discussed in this site's spread explainer — a one-cent spread on a $50 stock is 2 basis points each way, trivial per trade and material at high turnover. Market impact: the price movement caused by your own order, growing with order size relative to typical volume — academic and practitioner models estimate impact as a concave function of participation rate, so doubling size does not double impact but does increase it. Timing slippage: the drift between decision price and execution price across the delay — the honest lag whose absence from a backtest is itself a lookahead problem, covered separately on this site.
How much do costs change results?
Enough to reorder strategies. A worked illustration using standard magnitudes: a strategy turning over its portfolio monthly at 100 percent annualized pays roughly the spread plus impact on a sum equal to assets each year — at 5 basis points per side in liquid large caps, about 10 basis points annually, absorbable; the same turnover in small caps at 40 basis points per side pays 80 basis points annually, material; a daily-turnover strategy multiplies the arithmetic by an order of magnitude and the impact by more, which is why the empirical literature finds that gross Sharpe advantages at high frequency routinely vanish or invert once per-trade costs of even a few basis points are applied. The pattern generalizes: cost-adjusted and gross rankings correlate imperfectly, and capacity — the size at which impact erodes the edge — is part of every honest strategy description.
How should a backtest model them?
Procedurally, in layers, each stated with its assumptions.
- Commissions at the broker's actual schedule, including exchange and regulatory fees.
- Half-spread paid per side, using historical spread data for the traded universe — spreads vary by name and regime, and stress-period spreads are the ones that matter.
- Impact via a stated functional form — linear or square-root participation models are standard — with parameters from published estimates or the trader's own fills.
- Execution lag: signals act next bar or later; fills at realistic prices, never the signal bar's close.
- Sensitivity: re-run at zero, half and double the assumed costs; the strategy's dependence on the assumption is then visible rather than hidden.
The zero-cost run is not a baseline — it is a diagnostic; strategies profitable only at zero cost are spread-payers, not strategies.
What do execution studies document about real costs?
Institutional execution research — the transaction-cost analyses published by brokers and consultants on live order flow — measures realized costs including impact: single-digit basis points for liquid large-cap orders worked patiently, multiples of that for small caps, stressed conditions or urgent orders. The academic record agrees in direction: the profitability of classic anomalies attenuates with trading frictions, and several well-known patterns are stronger among less liquid names where costs are also higher — a correlation that keeps the net-return question empirical, name by name. The Investment Technology literature formalized these measurements decades ago; the modern institutional standard reports implementation shortfall — the difference between the decision-price portfolio value and the realized executed value — which is exactly the quantity a net backtest should be estimating in advance.
| Cost layer | Scales with | Typical liquid-equity magnitude | Modeling choice |
|---|---|---|---|
| Commissions | Trade count | Fractions of a basis point | Actual schedule |
| Half-spread | Turnover | 1-5 bp per side | Historical spreads |
| Impact | Size vs volume | Concave in participation | Stated functional form |
| Timing slippage | Delay, volatility | Regime-dependent | Next-bar or later fills |
Where do cost assumptions mislead?
Three slips. Single-point cost assumptions: costs are state-dependent — spreads widen exactly when strategies trade most, in stress — so a fixed 5-basis-point assumption understates the bad states that dominate net outcomes. Ignoring capacity: a backtest at portfolio size X does not survive at 10X if impact is real; honest reporting includes the size at which the edge halves. And survivorship of cost regimes: strategies validated in the low-cost modern era inherit no evidence about higher-cost regimes. Net reporting — gross, costs, net, sensitivity — is the minimum disclosure for any performance claim, the same discipline this site demands of evaluation windows and configuration counts. A promised return is a gross number wearing a net number's clothes until the friction arithmetic is printed next to it.
For more context, read How do you detect lookahead bias in someone else's backtest?.
For more context, read What does out-of-sample mean in a forecasting paper?.
For more context, read What regime-detection models can and cannot do.
