Pipsgrowth EX09007 MeanReversion

EX09007 looks for a very specific candle shape: a body in the upper third of the bar with a lower wick at least twice as long as the body, and a lower wick longer than the upper wick. That geometric definition is what most retail traders call a Hammer (bullish reversal) or a Hanging Man (bearish reversal) — but here it is computed in the SignalPattern() function on the just-closed M5 bar (shift=1), with a precise threshold: the body must sit in the upper third of the bar's range (bodyBot > high - range/3.0) and lowerWick >= 2.0 * bodySize. That last clause is what disqualifies a lot of "looks like a hammer" candles — the lower wick has to be at least double the body, not merely noticeable.

Pattern alone is not enough. The EA also reads a 5-period SMA on the same timeframe for trend context: a Hammer must close below the SMA(5) — meaning the bar printed in an existing micro-downtrend — and a Hanging Man must close above the SMA(5). That SMA gate ensures the EA is fading short-term momentum rather than buying strength. After the trend gate comes the third filter: the Money Flow Index with period 37. A Hammer must print with MFI below 40 (sell-pressure exhaustion), and a Hanging Man must print with MFI above 60 (buy-pressure exhaustion). The two inputs InpMFIBuyMax=40.0 and InpMFISellMin=60.0 are the literal thresholds; the confidence output of SignalPattern() is computed as 50 + |offset|*1.2 capped at 100, but it is logged only and not gating — only the pattern match plus the threshold comparisons actually matter for the entry decision.

A high-confidence pattern is not by itself enough to fire. The EA runs a full 7-state regime classifier via ClassifyRegime() using ADX(14), a 100-bar ATR(14) percentile, and a Bollinger-Band width z-score (period 20, deviation 2.0). The seven states are StrongTrend, WeakTrend, Range, Breakout, Compress, Expand, and Choppy. RegimeAllowsEntry() permits entries only in three of them: Range, Compress, and WeakTrend. StrongTrend, Breakout, Expand, and Choppy are blocked. The intent is straightforward — mean reversion only works when the market is not trending hard and not in a volatility expansion. A Hammer forming inside a StrongTrend regime is a continuation signal, not a reversal, and is ignored.

Before the order is even sized, three more confirmations must pass in ComputeConfirm(). The HTF EMA filter pulls a 50-period EMA on H1 (HTF_TIMEFRAME = PERIOD_H1): a Hammer is only confirmed if the M5 close is below the H1 EMA — a pullback context — and a Hanging Man only if the close is above it. That single cross-timeframe gate is what stops the EA from fading a clean H1 trend on a 5-minute "fake" Hammer. The spread gate allows a 35-point absolute cap (MAX_SPREAD_POINTS=35) or 3× the rolling 20-bar average spread, whichever is tighter, when InpMaxSpreadOverAvg=3. The R:R gate enforces tpDist / slDist >= 1.2 (the constant MIN_RR_RATIO) — a thin trade is never taken.

Risk math is straightforward. CalcLotByRisk() reads the SL distance in price, divides effective capital by tick_value / tick_size to get the per-lot risk, and sizes the order so that 0.5% * effective_capital is at risk per trade (InpRiskPercent=0.5). The lot is then clamped to the symbol's volume step and to InpMaxLots=5.0. The capital base is the minimum of equity and InpCapitalCapAmount (when the cap is non-zero), which is the EA's way of letting you cap exposure at a fixed dollar figure on a larger account — setting the cap to 0 disables it.

Five exit paths run on every tick through ManageOpenPosition(). First, a one-shot break-even at 1.0R (BE_TRIGGER_R=1.0): once profit equals the initial risk, the stop is moved to entry + 2 points. Second, a partial close at 1.0R (PARTIAL_TP_R=1.0): 50% of the position is closed (PARTIAL_FRACTION=0.5), with the heuristic that the lot must be at least twice the symbol's volume minimum — otherwise the partial is skipped to avoid a too-small residual. Third, an ATR(14) trailing stop at 2.5×ATR (TRAIL_ATR_MULT=2.5) with a 10-point step (TRAIL_STEP_POINTS=10), forward-only, that only ratchets in the profitable direction. Fourth, an opposite-pattern exit: if a Hammer prints against an existing long, the position is closed immediately, and vice versa. Fifth, a regime-change exit: if the classifier flips into StrongTrend, Breakout, or Expand — the states that block entries — the position is force-closed, because the mean-reversion thesis is no longer valid. A 60-bar time stop (MAX_BARS_IN_TRADE=60) layers on top of that stack, which is roughly 5 hours on the M5 default.

The no-trade system around all of this is one of the more comprehensive in the corpus. The NoTradeReason() function returns a string tag for the first failure and short-circuits: kill switch on, broker not allowing trades, terminal disconnected, equity below the 50-dollar floor (InpCapitalCapFloor=50.0), outside the 07:00-20:00 server-time session (SESSION_START_HOUR=7, SESSION_END_HOUR=20), weekend (Saturday or Sunday), current spread above 35 points, a ±30-minute window around each hour boundary treated as a generic news guard (NEWS_WINDOW_MIN=30), regime blocked, 5 trades already taken today (MAX_DAILY_TRADES=5), the 5-bar cooldown after a loss (COOLDOWN_BARS_AFTER_LOSS=5), or 1 position already open (InpMaxConcurrent=1). The session window is in broker server time, not auto-detected GMT — a user on a GMT+2 broker sees the live window at 05:00-18:00 UTC. The news guard is a primitive time-based proxy, not an MQL5 economic-calendar feed.

Risk management outside the trade is enforced after entries: a daily 3% loss of effective capital (InpDailyLossLimitPct=3.0) and a weekly 6% loss (InpWeeklyLossLimitPct=6.0) trigger a force-close of any open position via the RealizedPnLToday() and RealizedPnLThisWeek() history walkers. These walk the deal history from start-of-day and start-of-Monday to now, filtering by InpMagic and symbol, summing profit + swap + commission. The OnTester custom criterion multiplies net profit by profit factor and divides by 1 + drawdown_percent, with a 30-trade minimum to prevent over-fit curves from scoring well. Trade-execution transient errors are absorbed by a 1-retry loop with a 150ms sleep on REQUOTE, PRICE_OFF, PRICE_CHANGED, CONNECTION, and TIMEOUT retcodes; close operations are wrapped in a 3-retry 200ms loop that also covers PRICE_OFF and PRICE_CHANGED, but TryModify_EX09007 only retries on REQUOTE and TIMEOUT — a small gap where a price-change reject on a stop modification will not retry.

Two defaults deserve a callout. InpDryRun=true ships enabled — every order attempt is logged with the [DRYRUN] prefix and no live order is sent. That is a deliberate safe default for forward testing but it means the EA does nothing in live until the user flips the input. InpKillSwitch=false is the other binary that needs to be off for trading; turning it on hard-blocks all new entries regardless of conditions. The portfolio-heat cap of 6% (PORTFOLIO_HEAT_CAP=0.06) is estimated from CountOurPositions() * InpRiskPercent / 100.0, a crude but effective size-based proxy.

For backtests, expect this EA to fire on Gold in the 07:00-20:00 server window most often on bars that print the geometric pattern with MFI exhaustion — typically 1 to 3 times per London-NY overlap on a 5-minute chart, more on volatile days. The 60-bar time stop is a hard ceiling: a trade that hasn't paid by 5 hours closes regardless of MFI. The two combination gates that most often kill an otherwise valid pattern are the H1 EMA pullback requirement (which filters out "buy the breakout" patterns) and the regime classifier (which filters out the strongest trend days, where the Hammers print as continuation candles). A user wanting more trades should look at lowering MIN_RR_RATIO or widening the regime allowlist in RegimeAllowsEntry() — both reduce selectivity in measurable ways. The pyramid constants PYRAMID_MAX_LEVELS=5 and PYRAMID_ATR_SPACING=1.5 are defined but unused at runtime, because InpMaxConcurrent=1 forces a single open position; they are leftover scaffolding from a strategy variant.