5 Key Parameters That
Make or Break EA Performance

Position sizing determines how much capital is exposed on any single trade, and it is the single largest determinant of whether a losing streak is survivable. A fixed lot size might look fine on a backtest with a stable account balance, but in live trading it fails to account for equity drawdown, meaning risk as a percentage of the account actually increases the deeper you're already in a losing streak.

Most robust EAs use a percentage-risk model (commonly 0.5%–2% of equity per trade) combined with volatility-adjusted sizing — typically an ATR multiple — so that lot size shrinks automatically in high-volatility conditions and expands modestly in calmer ones. Without this, an EA calibrated on a low-volatility backtest window can be badly oversized the moment volatility regimes shift.

The relationship between stop-loss distance and take-profit distance defines the EA's required win rate to break even, and it's one of the most commonly over-optimized parameters. A tight 1:3 risk-reward setup found by an optimizer might produce a stellar equity curve on historical data purely because it happened to fit that data's specific swing structure — not because the ratio reflects the instrument's real volatility.

A durable approach ties SL/TP distances to measurable market structure (recent swing highs/lows, ATR bands, or supply/demand zone edges) rather than fixed pip values. This keeps the ratio adaptive as volatility changes, and reduces the chance that the "optimal" ratio was simply reverse-engineered from a handful of historical outlier trades.

Backtests frequently assume perfect fills at the requested price, but live execution introduces slippage and variable spread — especially around news releases, session opens, and low-liquidity periods. An EA that scalps small targets (say, under 10 pips) can see its entire edge consumed by 1–2 pips of average slippage plus spread widening, even though the backtest showed a healthy edge.

Robust EAs include a maximum allowable spread filter (rejecting entries when spread exceeds a defined threshold) and a slippage tolerance setting on order execution, so trades are skipped rather than filled at a materially worse price. Running the backtest with realistic slippage and commission modeling — not the platform default of zero — is essential before trusting any reported performance figures.

When and how often an EA is permitted to trade has a direct effect on both edge quality and cost drag. A strategy that performs well during the London/New York overlap can perform very differently during the thin Asian session, where spreads widen and price behaves more erratically relative to the strategy's assumptions.

Session filters, day-of-week restrictions, and news-event blackout windows (avoiding entries in the minutes around high-impact releases) all reduce exposure to conditions the strategy wasn't designed for. Equally, over-restricting trade frequency in pursuit of a smoother backtest equity curve can create a false sense of consistency — it's worth testing whether performance holds up with the filters loosened, not just tightened.

How parameters were derived matters as much as the parameters themselves. Brute-force optimization across dozens of inputs on a single historical dataset will almost always find a combination that looks exceptional in-sample — but that combination is frequently fitted to noise rather than a repeatable market inefficiency, and tends to degrade sharply out-of-sample.

Walk-forward analysis (optimizing on one period, then validating on the next unseen period, rolled forward repeatedly) and out-of-sample testing on data the optimizer never touched are the standard checks against this. A useful heuristic: if performance falls off a cliff when parameters are nudged by 10–20% from their "optimal" values, the strategy is likely overfit rather than robust — a genuinely durable edge tends to perform reasonably across a broad neighborhood of parameter values, not just a single narrow peak.