Pipsgrowth EX09009 MeanReversion

Pipsgrowth EX09009 MeanReversion hunts for a specific geometric candle — the Hammer (bullish reversal) or the Hanging Man (bearish reversal) — and then asks the Stochastic oscillator to confirm the move before any order leaves the broker. Everything else in the EA exists to filter out trades that don't match the original idea: a 7-state market regime classifier, an H1 higher-timeframe trend filter, a 5-bar cooldown after a loss, daily and weekly drawdown caps, a 60-bar time stop, a server-time session gate, and a hard floor on the absolute account equity. The strategy is the same mean-reversion thesis used across the rest of the EX09 family — fade exhaustion moves at the end of a swing — but the confirmation oscillator here is Stochastic instead of the CCI used in EX09006, which makes the EA more sensitive to momentum exhaustion than to overbought/oversold reading of an unbounded oscillator.

How the candle is identified

SignalPattern() reads four price points from the just-closed M5 bar: iOpen, iClose, iHigh, iLow. It computes the total range, the body size (|open-close|), the upper wick (high minus body top), and the lower wick (body bottom minus low). The bar qualifies as a Hammer or Hanging Man only when two geometric conditions are both true: the body must sit in the upper third of the range (bodyBot > high - range/3.0), and the lower wick must be at least twice the body size AND longer than the upper wick. That combination is the textbook rejection-candle signature — small real body near the high of the bar, with a long lower tail showing that sellers pushed the price down hard but the buyers recovered before the close.

How the trend context is established

A rejection candle by itself is not enough. The bar must also appear on the correct side of a 5-period simple moving average to be a valid signal. If the closed bar's close is below the SMA(5), the EA treats the surrounding context as a short-term downtrend and looks for a Hammer (long signal). If the close is above the SMA(5), the context is an uptrend and the EA looks for a Hanging Man (short signal). The 5-period SMA is held as the indicator handle hMA = iMA(_Symbol, _Period, 5, 0, MODE_SMA, PRICE_CLOSE). This 5-bar window is short enough to react to a recent change of character but long enough to filter the kind of one-bar noise that would otherwise print endless reversals in a flat market.

How the Stochastic confirmation is applied

Once a geometric candle is found on the right side of the SMA(5), the EA checks the Stochastic signal line value at the same closed bar. The default thresholds in the source are InpStochBuyMax = 30 and InpStochSellMin = 70 — these are the input fields, even though the source-file brief mistakenly quotes 20/80. A Hammer is confirmed for a long entry only if the Stochastic signal is below the buy threshold; a Hanging Man is confirmed for a short entry only if the Stochastic signal is above the sell threshold. The Stochastic itself is built with non-standard periods: K=47, D=9, slowing=13 (defined as STOCH_K_DEFAULT, STOCH_D_DEFAULT, STOCH_SLOWING_DEFAULT at the top of the file), which makes the oscillator very slow and very smooth compared to the usual 14/3/3 setting. The slow build means the signal line only prints values outside the 30/70 zone when the market has been trending in one direction long enough to actually exhaust.

A confidence score is logged for every confirmed signal — 50 + (InpStochBuyMax - stoch) * 1.2 for buys, mirrored for sells, capped at 100 — but the score is informational only and does not gate the entry.

How the 7-state regime classifier decides whether the market is tradeable

ClassifyRegime() reads ADX(14), a Bollinger Band width z-score (20-period bands, 2.0 deviations), and an ATR(14) percentile over the last 100 bars. The seven states are STRONG_TREND, WEAK_TREND, RANGE, BREAKOUT, COMPRESS, EXPAND, and CHOPPY. The thresholds are explicit: ADX >= 28 with positive BB-width z-score is BREAKOUT; ADX >= 28 alone is STRONG_TREND; ADX in the 20-28 band is WEAK_TREND; BB-width z-score <= 0.8 is COMPRESS; >= 1.5 is EXPAND; ATR percentile <= 20 is RANGE; >= 80 is BREAKOUT; anything in between is CHOPPY. The EA is a mean-reversion system, so RegimeAllowsEntry() returns true only for RANGE, COMPRESS, and WEAK_TREND. STRONG_TREND, BREAKOUT, EXPAND, and CHOPPY block new entries entirely. A live position that crosses into a blocked regime is force-closed by the REGIME_CHANGE exit reason inside ManageOpenPosition.

How the higher-timeframe trend filter works

ComputeConfirm() calls HTFEMAVal(1), which reads an H1 EMA(50) (defined as hHTFEMA = iMA(_Symbol, PERIOD_H1, 50, 0, MODE_EMA, PRICE_CLOSE)). For a Hammer long signal, the EA wants the M5 close to be below the H1 EMA(50) — the pullback interpretation. For a Hanging Man short signal, the EA wants the M5 close to be above the H1 EMA(50). When InpHTFBarAgree = 1 (the default), this side check must pass or the entry is refused. Setting it to 0 disables the filter.

Sizing, stops, and targets

The stop loss is ATR(14) * 1.8 from the entry, and the take profit is ATR(14) * 3.0. That gives a base reward-to-risk ratio of 1.67:1, which the MIN_RR_RATIO = 1.2 guard in ComputeConfirm enforces as a minimum. Lot size is computed by CalcLotByRisk() from the SL distance, the SYMBOL_TRADE_TICK_VALUE and SYMBOL_TRADE_TICK_SIZE of the broker, and the configured risk percentage — InpRiskPercent = 0.5 of effective capital by default. Effective capital is MathMin(InpCapitalCapAmount, equity) when a cap is configured, or full equity otherwise. The result is then clamped to the broker's volume step, the broker's min/max, and InpMaxLots = 5.0. A portfolio-heat cap of 6% (estimated as CountOurPositions() * riskPct) blocks the new entry if adding the position would push total heat above the cap.

In-trade management

ManageOpenPosition() runs on every tick and applies three independent exit overlays, plus a regime flip. When the position is in profit by 1R (BE_TRIGGER_R = 1.0), the stop is moved to entry plus 2 points and the move is one-shot — it only happens once. When the position is in profit by 1R and the remaining volume is at least 2x the broker minimum, half the position is closed (the 50% partial). An ATR-based trailing stop then tracks at ATR(14) * 2.5 behind the current bid (or in front, for shorts), ratcheting forward in 10-point steps and never loosening. If a valid opposite candle+stochastic pattern forms, the position is closed with reason OPP_SIGNAL. If the bar count since entry exceeds MAX_BARS_IN_TRADE = 60 (five hours on M5), the position is closed with reason TIME_EXIT. If the regime flips into STRONG_TREND, BREAKOUT, or EXPAND, the position is force-closed with reason REGIME_CHANGE.

The 12 layers in this build

The header describes the EA as a 12-layer system: REGIME, SIGNAL, ENTRY, CONFIRM, NO-TRADE, CAPITAL CAP, RISK, SIZING, MANAGE, EXIT, SCALING, and OnTester. The SCALING layer is mostly aspirational in this version — InpMaxConcurrent = 1 and there is no CheckPyramidSafety function — so the EA runs one position per symbol at a time and does not pyramid into winners. The trade execution path (SendOrder) retries once on REQUOTE, PRICE_OFF, PRICE_CHANGED, CONNECTION, and TIMEOUT with a 150ms sleep and a RefreshRates() between attempts. The close and modify paths (TryClose_EX09009, TryClosePartial_EX09009, TryModify_EX09009) each retry three times on the same transient codes, with 200ms sleeps for close and 100ms for modify.

No-trade gate

Before any new order is sent, NoTradeReason() walks a list of guards and returns the first one that fails: kill switch, broker trade disabled, terminal not connected, equity below the InpCapitalCapFloor = 50 floor (or below 50 if no cap), outside the 7:00-20:00 server-time window, weekend, spread above 35 points, inside the ±30-minute news window around each hour boundary, regime disallows entries, daily trade count at 5, still inside the 5-bar post-loss cooldown, or concurrent positions at the limit. The news filter is a time-based proxy that blocks entries near every full hour on the broker clock; it is not connected to a real news feed.

Custom optimization criterion

OnTester() returns (net * profit_factor) / (1.0 + equity_dd_percent) provided the strategy executed at least 30 trades during the test. The formula rewards high net profit and high profit factor while penalizing any non-zero drawdown — it is a quality-weighted ranking that prefers stable returns over lucky blowups.

Dry-run and kill switch

InpDryRun defaults to true, which means the EA logs every intended order to the journal with [DRYRUN] prefix instead of actually sending it. To run live, the user must set this to false. InpKillSwitch defaults to false and acts as a hard stop on all new entries when flipped to true.

What to expect in the backtest

Hammer and Hanging Man candles form a few times per trading day on M5 XAUUSD during the London and New York sessions when swing points are tested. Stochastic with the 47/9/13 build sits in the 30/70 range most of the time and only breaks out during extended directional moves, so signal frequency will be low — usually 1-3 trades per day on XAUUSD M5. Most of those trades should be filtered out by the regime classifier when the market is trending, and the loss-limit and cooldown logic should cap any losing streak. As with all reversal systems, the worst case is a fast market where the rejection candle turns into a continuation bar — the ATR-based stop is wide enough to absorb a normal wick but cannot protect against a gap or a flash event.