Appearance

Personalize your experience

Mode

Accent Color

Direction

Execution Modelling

Backtesting

Commission, slippage, spread and fill assumptions — the difference between a backtest that means something and one that flatters. Costs are where most strategies actually die.

Costs are not a detail

A zero-cost backtest is not an approximation, it is a different . Costs are charged per trade, so they scale with frequency while the edge does not — which means the same signal can be excellent at 50 a year and unprofitable at 600.

One 18% gross edge, four trade frequencies, 0.03% commission per side. The last row is a losing strategy.

Caution:Read the bottom row. Two thousand trips a year at 0.03% per side costs 120% of capital in commission — the is unrunnable regardless of how good the signal is. A backtest with costs at zero shows the same signal as a comfortable winner, and nothing on the page hints otherwise.

Commission

The straightforward cost, and the easy one to get : your broker publishes it. What people get wrong is applying it per side rather than per trade, and forgetting the components.

Component	Type	Description
Brokerage	% or flat	The headline charge. Often a with a per-order cap, which makes the effective rate depend on order size.
Exchange fees	%	but per-trade, so they matter at frequency.
Statutory charges	%	Transaction tax, duty, regulatory fees. Jurisdiction-specific and non-negotiable.
GST / VAT	% of brokerage	Charged on the brokerage rather than on turnover, so it scales with the brokerage component.

effective round-trip cost

1// what to actually enter as "commission" in the backtest2function roundTripCostPct(orderValue: number) {3  const brokeragePerSide = Math.min(orderValue * 0.0003, 20);  // 0.03%, capped4  const exchangePerSide  = orderValue * 0.0000325;5  const statutoryPerSide = orderValue * 0.0001;6  const gst              = brokeragePerSide * 0.18;78  const perSide = brokeragePerSide + exchangePerSide + statutoryPerSide + gst;9  return (perSide * 2) / orderValue * 100;   // both sides, as a percentage10}1112// note the cap: at large order sizes the effective rate falls,13// so a strategy's cost profile depends on its position size14roundTripCostPct(50_000);    // ≈ 0.13%15roundTripCostPct(500_000);   // ≈ 0.04%

★

Pro tip:The per-order cap matters more than it looks. A trading positions pays a much higher effective than the same strategy trading large ones, so a backtest at one position size does not tell you the cost profile at another. Re-check costs whenever you change .

Slippage

Slippage is the gap between the price the signal fired at and the price you actually got. Unlike commission it is not published — it has to be measured, and it varies by , , time of day and market condition. It is conventionally quoted in basis points, which keeps it comparable between instruments at different price levels.

A 2% stop, three outcomes. The level is an instruction; the fill is whatever the market offers.

Source	Description
Spread	A market buy pays the ask and a market sell receives the bid, so crossing the spread is a guaranteed cost on every . On liquid instruments it is a ; on illiquid it can exceed the edge.
Queue position	A at the is behind everyone already resting there. The backtest it when price reaches the level; reality may not fill it at all.
Latency	Time between the bar closing and the order arriving. Price moves in that , and it moves against you more often than not — the signal fired because price was moving.
Market impact	orders consume the book and move the price. Negligible at retail size on liquid instruments; not negligible on thin ones.
Gaps	The worst case, and the one no slippage setting captures. Price jumps past the stop and wherever the market reopens.

Caution:Slippage is asymmetric. It costs you on entries and exits, and it is worst precisely when you most need a fill — on a stop during a fast move, which is when spreads widen and liquidity withdraws. Modelling it as a symmetric average understates the tail.

Fill assumptions

The engine at the , completely, at one price. Each part of that is a simplification worth knowing.

Two intra-bar paths, one identical candle. If both a stop and a target sit inside the bar, the backtest cannot know which was reached first.

Assumption	Description
Fill at bar close	, the order is placed after the close and at the next available price. On a fast-moving bar those differ materially.
Complete fill	No . limit orders in a thin book fill partially, leaving a position size the did not for.
Single price	A order consuming several book levels at a volume-weighted average worse than the .
Stop fills at the level	The most optimistic assumption in the engine. A stop is a market order once triggered, so it at whatever is available — and a gap ignores the level entirely.

Caution:If a 's stop and target are both inside typical bar range, it is not properly testable on that . The engine resolves the ambiguity by convention rather than by evidence, and roughly half of those are recorded as the wrong outcome. Either widen the levels or test on a finer timeframe.

Setting realistic values

Instrument class	Type	Description
Index futures (liquid)	0.02–0.05% cost, 1–2 ticks slippage	NIFTY, BANKNIFTY. The friendliest case, and where tight spreads make higher frequency viable.
Large-cap equity	0.05–0.15% cost, 2–5 ticks	Cash equity attracts more statutory charges than derivatives in many jurisdictions.
Mid and small-cap equity	0.15–0.4% cost, 5–20 ticks	Wider spreads and real . Frequency strategies rarely survive here.
Options (near the money)	0.1–0.3% cost, wide spreads	Spreads dominate. Model them explicitly rather than folding them into a slippage average.
Options (far OTM)	spread frequently exceeds the edge	A spread of several percent of makes most systematic strategies uneconomic.
Crypto (major pairs)	0.05–0.2% cost, low slippage	Deep books on major pairs; funding costs apply to perpetuals and are easy to forget.

★

Pro tip:When uncertain, be pessimistic. A that survives twice your slippage estimate is one you can deploy; a strategy that needs your optimistic estimate to work will discover the real number with capital. Overestimating costs loses you a marginal strategy — underestimating them loses you money.

Reconciling live against backtest

When diverge, the cause is measurable rather than mysterious. Work through it in this order.

realised slippage from twenty against the bar closes they should have executed at. Feed the mean back into the backtest and re-run. This usually closes most of the gap.
Check the exit-reason mix. If shows more stops than the backtest did, the stop is being hit intra-bar in a way the replay could not see.
trade counts. Fewer at feed gaps or latency causing missed signals; more points at a configuration difference.
Check average position size. If sizes are smaller, orders are filling partially, and every P&L figure scales accordingly.
Whatever residue remains after those four is timing, and timing is not by adjusting cost parameters — it is fixed by trading a slower timeframe or accepting the .

ℹ

Note:None of this indicates a broken . The same code evaluates conditions in both places, so a divergence is always , costs, or timing. Knowing that narrows the investigation to four measurable things instead of leaving it open-ended.

Appearance

Execution Modelling

Costs are not a detail

Commission

Slippage

Fill assumptions

Setting realistic values

Reconciling live against backtest

See also

Related articles

Appearance

Costs are not a detail

Commission

Slippage

Fill assumptions

Setting realistic values

Reconciling live against backtest

See also

Related articles

Related to Execution Modelling