Slippage and Spread — The Hidden Costs Every Backtest Underestimates

Slippage and spread are the two execution costs that quietly erode the edge of any trading strategy, and they are the most consistently underestimated numbers in quantitative trading. A backtest can show a beautifully profitable model, and the live version can lose money — not because the signals were wrong, but because the price the strategy actually filled at was not the price the backtest assumed.

Understanding these costs is not optional. For low-frequency, large-stop trading models the impact is small. For higher-frequency or thin-market models the impact can swallow the entire edge.

What spread really costs

Spread is the gap between the bid and the ask. Every market trade crosses it — you sell at the bid and buy at the ask — and the cost is paid the moment the position is opened. The wider the spread, the larger the immediate drawdown before the trade has even moved.

This matters in proportion to the size of the average winner. A spread of one pip on EURUSD is barely noticeable when the average winner is 30 pips. The same spread on a strategy whose average winner is 5 pips eats 20% of every gross profit. The breakeven win rate moves correspondingly, and a model that looked viable in a zero-cost backtest becomes a slow bleed.

Spreads are also not constant. They widen at session changes, during news events, and in low-liquidity windows like late Asian session for European pairs. A backtest that uses a single fixed spread for all trades is implicitly assuming away one of the genuine sources of strategy failure: the moments when the cost of doing business spikes are often the moments a model is most active.

What slippage adds on top

Slippage is the difference between the price a strategy intended to fill at and the price it actually got. It comes from latency, from the market moving between signal and execution, and from order size relative to available liquidity at the chosen level. A market order in a fast-moving market does not get the price on screen — it gets the next available price after the order arrives.

For liquid majors traded in modest size, slippage on entries is small. The serious cost is usually on stop losses, especially during the kind of fast moves where many models are exiting at once. A strategy that backtests with a 10-pip stop can routinely realise a 13-pip loss when the market gaps through the level. Across a few hundred trades that adds up to a different equity curve entirely.

The interaction between slippage and volatility clustering is where the real damage happens. The moments when slippage is worst are precisely the moments when many stops fire at once — and a high-volatility regime tends to produce both more stop fills and worse fills on each one.

How execution costs interact with model design

The practical takeaway is that execution costs are not a flat tax. They penalise certain model types far more than others.

High-frequency entries with tight targets are hit hardest. The cost is paid on every trade and the gross profit per trade is small, so the cost-to-edge ratio is poor. This is one of the structural reasons darwintIQ favours position managers like ATR, SMA Trail and SupRes — they typically produce wider stops and longer-held trades, which dilutes execution costs across more pips of edge.

Strategies that trade illiquid hours suffer disproportionately. A model that fires during the late New York to early Tokyo window crosses wider spreads on average and faces thinner books. The same logic that produces clean signals in the London open can produce decaying signals in the same market eight hours later.

Any strategy whose backtest looks great but lacks a meaningful margin between gross and net is structurally fragile. A small change in average spread, or a single year of slightly worse fills, can flip the system from profitable to unprofitable.

Final thoughts

Slippage and spread are not glamorous topics, and that is exactly why they get ignored. Every quant has run a backtest that looked excellent in theory and crumbled the moment execution costs were added — usually because no one budgeted for the real distribution of fills, only the average.

The useful discipline is to assume execution is worse than your assumptions and to test what happens to the equity curve under stress. Models that still produce a real edge after a generous cost assumption are the ones worth running live. Models that need optimistic execution to show profit are models that will not survive contact with a real broker.