Pipsgrowth EX09021 MeanReversion

EX09021 is a Bollinger-Bands-and-Williams-Percent-Range mean-reversion EA that requires both indicators to agree before it fires. The WPR(32) is unusually long for a momentum oscillator (most MR EAs use 14) and the BB(58) is similarly stretched well past the default 20 — that combination is deliberate. The longer periods filter out single-bar noise and only flag exhaustion events that have actually built up over hundreds of bars, which is why the EA was tuned to M5 and up rather than M1. ATR(64) feeds the dynamic stop and target distances and tracks recent volatility on the same slow timeframe, so SL/TP scale with whatever regime the symbol is actually in.

The signal pipeline runs in two stages. SignalWPR() looks at the WPR on bar-1 versus bar-2 — a long fires when the WPR on bar-1 is above its value on bar-2 and the bar-2 value was at or below -80 (the oversold line). A short is the mirror: bar-1 below bar-2 with bar-2 at or above -20 (the overbought line). So the trigger is WPR exiting an extreme zone, not WPR being in one. SignalBB() then checks the bar-0 open against the midpoints of the outer half of the BB envelope — (lower+middle)/2 for the long side, (upper+middle)/2 for the short side. A long fires when the open is below the lower-midpoint, meaning price is in the outer half of the band, not just touching the lower rail. The two signals are AND-ed in OnTick(): only when both agree does TradeProcess() see a long or short. If WPR says oversold-exit but price is still near the middle, or vice versa, the EA does nothing.

Position management is where EX09021 differs from a vanilla band-touch bot. The TradeProcess() function checks IsPresentPosOnCurrentBar() first — only one new entry per direction per bar, which prevents over-trading when an oscillating market flips WPR on the same bar twice. More importantly, the pyramiding rule is better-price-only: for a long, the EA compares the open price of the most recent buy position against the current ask and only adds when the new ask is lower (price_last > SymbolInfoDouble(SYMBOL_ASK)). For shorts, it adds only when the new bid is higher than the previous short. That makes the scaling a quality-averager, not a martingale — it tightens the average entry on a continued move, not the other way around. There is no fixed cap on pyramid count; the gating condition is the price-improvement check, and the per-bar dedup.

Stops and targets default to dynamic ATR and BB widths. The InpStopLoss input accepts three states: 0 = no SL, positive integer = fixed points, -1 = derived from HalfSizeBB(0) * InpSLMltp (default 2.6). The InpTakeProfit input does the same: 0 = none, positive = fixed points, -1 = ATR(0) * InpTPMltp (default 1.3, in points). So at defaults, the SL is 2.6 times the half-width of the BB envelope and the TP is 1.3 times the 64-period ATR — typically a 1.5:1 to 2.5:1 reward-to-risk on M5 metals depending on the volatility regime. On netting accounts, OpenPosition() zeroes out both SL and TP (line 958) because the broker will only hold one position per symbol and a static SL/TP would block averaging — the EA explicitly warns at init that it expects a hedging account.

The safety stack is the standard 9-gate family pattern, applied via IsSafeToTrade_EX09021(). The gates, in order, are: cap-amount reached (InpCapAmount > 0 and equity >= cap), cap-floor breach (equity < InpCapFloor, default 50), 95% of initial-balance floor, 3% daily realized-loss cap, 3-consecutive-losses cooldown (default 30 minutes), 50 trades per day, market-open check (no trading Friday after 21:00 GMT, all Saturday, Sunday before 22:00 GMT), active-session check (London 07:00–16:00 GMT and New York 12:00–21:00 GMT union, with the Asia window 00:00–07:00 GMT excluded by default), and an optional ±15-minute news filter around the London and NY session open hours. The OnTradeTransaction() handler updates g_realizedToday, g_tradesToday, and g_consecLosses from the actual deal stream, not the EA's own bookkeeping — so the daily counters survive restarts within the same session and match the broker's history. The DetectGMTOffset_EX09021() function walks back through H1 bars looking for a 3+ hour weekend gap, then subtracts 22 to infer the offset, falling back to InpServerGMTOffset if you set it manually.

The entry/exit path also has a race-condition guard. IsUncertainStateEnv() walks the order pool looking for an order with the right magic and symbol but no position ID — that's the gap between a fill being accepted and the position being registered. If the EA finds that state, it sleeps 1000 ms and retries up to 3 times (ENV_ATTEMPTS=3, ENV_WAIT_ATTEMPT=1000) before refusing to act. The same pattern guards OpenPosition() directly. TryClose_EX09021() retries a position close up to 3 times at 200 ms on REQUOTE, TIMEOUT, PRICE_OFF, and PRICE_CHANGED — anything else (PRICE_INVALID, REJECT, etc.) is treated as a hard fail. Filling mode is auto-detected from SYMBOL_FILLING_MODE and falls back to RETURN → IOC → FOK depending on what the broker advertises, which keeps the EA compatible with ECN, exchange, and instant-execution setups.

The OnTester() formula is the family-standard n * p / (1 + d) — net profit times profit factor divided by relative equity drawdown plus one — with a 30-trade minimum and a guard that returns 0 if profit factor is non-positive or drawdown is zero. That's the fitness function the MT5 optimizer maximizes, so the backtest results in the Strategy Tester are ranking candidates on risk-adjusted return, not raw PnL.

For practical use, EX09021 is a M5-and-up mean-reversion system tuned for the long-period family (WPR32, BB58, ATR64). It expects a hedging account and works on both metals (XAUUSD is the default symbol in the deployable version) and FX majors. The pyramid-on-better-prices rule means it does not grid in a martingale sense — the EA only adds when the market has continued in the expected direction. That makes drawdowns materially different from a classic grid bot: instead of stacking into a losing position, the EA stacks into a winning one and tightens the average entry.