Transaction Cost Modeling in Strategy Evaluation

Many published strategy backtests assume zero transaction costs, or apply a fixed per-trade fee that does not reflect actual market conditions. This creates a systematic bias toward strategies that trade frequently and capture small returns per trade.

Our backtesting infrastructure requires a cost model for every run. The engine will not execute a backtest with a null cost configuration. This is enforced at the software level, not by policy.

Cost models include bid-ask spread estimates calibrated to historical data, market impact estimates scaled to position size, and commission structures matching the target execution venue. For strategies trading less liquid instruments, we apply wider spread assumptions and test sensitivity to cost parameter changes.

Strategies that are profitable at zero cost but unprofitable at realistic cost are recorded in the trial registry as failures, not excluded from the research record. A strategy that only works when you ignore friction is not a strategy.

Cost Model Structure

Our transaction cost model decomposes execution cost into three components: spread, market impact, and commissions. Spread is calibrated from historical tick data. Market impact scales non-linearly with trade size relative to daily volume — a trade representing 1% of average daily volume incurs measurably more slippage than one at 0.1%.

For strategies with high turnover, the difference between naive and realistic cost modeling often determines whether the strategy is viable. This is why our engine will not execute a backtest without a cost configuration.

The descriptions above are for educational and informational purposes only and do not constitute investment advice. Past performance is not indicative of future results. See full disclosures.

References

• Kissell, R. (2013). “The Science of Algorithmic Trading and Portfolio Management.” Academic Press.
• Almgren, R. and Chriss, N. (2001). “Optimal Execution of Portfolio Transactions.” Journal of Risk.