How to Model Transaction Costs in a Backtest

Every trade you take has a price beyond the quoted price. Commissions, exchange fees, the bid-ask spread, and slippage all eat into returns. A backtest that skips them overstates your edge, and the damage compounds with every trade. High-frequency strategies can look brilliant and still lose money after costs.

The three common cost models

There is no single correct way to model costs. Each approach trades simplicity for realism.

• Flat per-trade: a fixed dollar amount per fill. Simple, and reasonable for fixed-commission brokers. It ignores trade size, so it can understate costs on large orders.
• Percentage of notional: a basis-point charge on the value traded. This scales with position size and fits most crypto and FX venues. It still assumes you always trade at the quoted price.
• Liquidity-based: costs that grow as your order consumes the order book. This is the most realistic for thin markets, large size, or fast intraday systems, and the hardest to estimate well.

Slippage is the part people forget

Spread and commission are knowable. Slippage, the gap between your expected fill and your actual fill, is not. It worsens in volatile or illiquid conditions, which is exactly when many strategies trade most. A defensible backtest applies a conservative slippage assumption rather than pretending fills happen at the mid-price.

Test your assumptions, do not trust them

Costs are an estimate, not a fact. Run your backtest across a range of cost assumptions and watch how the result holds up. An edge that survives only at zero cost is not an edge. An edge that survives realistic and even pessimistic costs is worth a closer, out-of-sample look.

If your strategy only works when trading is free, it does not work.

TRION was built around an honest validation sequence rather than a promise. It is a paper-only research and validation workstation: you describe a strategy idea in plain English, read the compiled logic line by line, and backtest it against real stored market data. When a metric cannot be computed honestly, TRION shows "N/A" instead of inventing a number.

TRION does not place real orders, does not connect to a broker, and does not promise profit. The current beta is simulation-only and paper-only. AI assists with drafting and explanation; it does not approve, activate, or execute anything. Humans make every decision.