Fills, stops & modeled P&L
How simulated fills and stop exits are modeled, and how the model is validated against real executions rather than quoted data alone.
A backtest is only as honest as its fill assumptions. This page states ours.
How simulated fills work
- Entries and exits are priced from the quoted market at that second, with slippage and commissions accounted for — never at an idealized mid.
- The treatment is asymmetric: exits — especially stop exits, where you're crossing the spread under pressure — cost more than entries. That asymmetry is what real execution data shows, and pretending otherwise flatters a backtest badly.
- Stop exits are recorded at the stop level as the modeling convention: when the cost-to-close crosses the configured stop, the loss is booked there. Live stops can fill worse — treat a backtested stop loss as the intended loss, not a guaranteed fill. (When the market gaps clean through a stop, the loss isn't the stop at all — that's the Black Swan scenario.)
How we keep the model honest
Most backtests are built on exchange quote data alone — and quote data can only ever tell you what was theoretically available, not what you would actually have gotten. That gap is where most backtests quietly become fiction.
We close it differently: CashFlow Engine trades these strategy families on its own automated accounts, and we use those real executions — our own fills, in the real market, at real times — to validate and calibrate the model behind the catalog. The methodology itself is ours and stays in-house, but the principle is simple and worth stating plainly: the numbers are checked against reality, not just against a quote database.
That's also the honest reason to expect some gap between simulated and live results. We work to keep it small and measured rather than unknown.
What this means for your expectations
- Simulated results sit somewhere between "theoretical ceiling" and "live reality" — closer to reality than naive mid-fills, but never a promise.
- The gap is strategy-dependent: configurations that stop out often carry more execution drag than configurations that mostly settle.
- If an edge in the catalog is so thin that a few dollars of per-contract slippage erases it, it is not an edge — filter accordingly.
The full fill-by-fill comparison of live versus modeled execution ships as Reality Check.