Português PT
Ver no GitHub

Gselector Pattern Probability Strategy

Overview

The Gselector Pattern Probability strategy is a StockSharp port of the MetaTrader 4 "Gselector" expert. It studies direction changes of synthetic price series built from multiple step sizes, keeps probability statistics for every observed pattern, and trades when the probability of a continuation move is high enough. Stop-loss and take-profit distances are simulated in software to mirror the original expert behaviour.

Learning Process

  1. Synthetic ladders – For each configured delta multiple the strategy constructs a step-based series by recording the last closing price every time the market moves by the required distance.
  2. Pattern encoding – A bit mask is created by comparing every pair of neighbouring values inside the ladder. Rising steps get bit 0, falling steps get bit 1, which reproduces the Ncomb encoding from the MQL implementation.
  3. Event tracking – When a new pattern appears, the strategy starts watchers for every configured stop level. A watcher stores the origin price and waits until price moves up or down by the threshold.
  4. Probability update – Once a watcher completes, upward moves increase the "growth" statistic, downward moves increase the "decline" statistic. A forgetting factor emulates the decay logic (forg) of the original expert.
  5. Persistence in memory – All statistics are kept in memory and reset on strategy start, matching the behaviour of the MQL version when ReadHistory is disabled.

Trading Logic

  1. Continuation probabilities are calculated for the current pattern on every delta ladder.
  2. A buy signal requires:
    • Probability ≥ ProbabilityThreshold.
    • Observations ≥ MinSamples.
    • Cooldown elapsed since the previous buy.
    • If a short position exists, the new probability must exceed the stored sell probability plus the ProbabilityBuffer.
  3. A sell signal mirrors the buy rules with the growth/decline roles swapped.
  4. Entries use BuyMarket / SellMarket to emulate OrderSend. When the opposite position is open the strategy closes it first, reproducing the reversal behaviour of the expert adviser.
  5. Protective exits are handled internally: stops and takes are expressed in price units derived from the point value and stop level.

Parameters

Name Description Default
CandleType Candle data type used for the backtest/live session. 1-minute time frame
ProbabilityThreshold Minimum continuation probability required to open a trade. 0.8
BaseDeltaPoints Base point distance that defines the first synthetic ladder. 1
DeltaSteps Number of delta ladders to evaluate. 20
PatternLength Number of elements in the ladder history. 10
StopLevels Count of stop/take levels. 1
StopDistancePoints Base stop/take distance in points. 25
ForgetFactor Decay applied to growth/decline counters after each observation. 1.05
MinSamples Minimum number of completed observations. 10
ProbabilityBuffer Extra probability required to close the opposite position. 0.05
FixedVolume Base trade volume. 1 lot
UseReinvest Enables balance-proportional volume adjustment. true
VolumeMode 0 – fixed, 1 – percent per 10k, 2 – ladder, 3 – linear. 1
PercentPer10k Percentage per 10 000 units in mode 1. 3
BaseDeposit Base deposit for modes 2 and 3. 500
DepositStep Deposit increment for modes 2 and 3. 500
MaxVolume Maximum volume cap. 10000
CooldownFactor Number of candle intervals used as the reactivation cooldown. 2

Differences from the MQL Expert

  • File-based persistence was removed; statistics are rebuilt from scratch whenever the strategy starts.
  • Orders are simulated through BuyMarket/SellMarket and software stop management instead of MT4 pending orders.
  • The position sizing helpers were adapted to StockSharp portfolio data. If equity values are not available the strategy falls back to the fixed volume.
  • Trailing stop inputs from the original code are ignored because the MT4 version never applied them.

Usage Notes

  • Attach the strategy to a security with a valid PriceStep. If the step is unknown the strategy falls back to 0.0001.
  • The learning process needs a minimum number of ladder activations; expect a warm-up phase before trades start.
  • Increasing DeltaSteps or PatternLength raises memory usage exponentially because the pattern dictionary grows quickly.
  • The default probability threshold (0.8) is very strict. Lower the value for more frequent trades.
using System;
using System.Collections.Generic;

using StockSharp.Algo.Indicators;
using StockSharp.Algo.Strategies;
using StockSharp.BusinessEntities;
using StockSharp.Messages;

namespace StockSharp.Samples.Strategies;

/// <summary>
/// Gselector Pattern Probability strategy - SMA crossover with RSI filter.
/// Buys when fast SMA crosses above slow SMA and RSI is below overbought.
/// Sells when fast SMA crosses below slow SMA and RSI is above oversold.
/// </summary>
public class GselectorPatternProbabilityStrategy : Strategy
{
	private readonly StrategyParam<int> _fastPeriod;
	private readonly StrategyParam<int> _slowPeriod;
	private readonly StrategyParam<int> _rsiPeriod;
	private readonly StrategyParam<decimal> _overbought;
	private readonly StrategyParam<decimal> _oversold;
	private readonly StrategyParam<DataType> _candleType;

	private decimal _prevFast;
	private decimal _prevSlow;
	private bool _hasPrev;

	public int FastPeriod { get => _fastPeriod.Value; set => _fastPeriod.Value = value; }
	public int SlowPeriod { get => _slowPeriod.Value; set => _slowPeriod.Value = value; }
	public int RsiPeriod { get => _rsiPeriod.Value; set => _rsiPeriod.Value = value; }
	public decimal Overbought { get => _overbought.Value; set => _overbought.Value = value; }
	public decimal Oversold { get => _oversold.Value; set => _oversold.Value = value; }
	public DataType CandleType { get => _candleType.Value; set => _candleType.Value = value; }

	public GselectorPatternProbabilityStrategy()
	{
		_fastPeriod = Param(nameof(FastPeriod), 10)
			.SetDisplay("Fast SMA", "Fast SMA period", "Indicators");

		_slowPeriod = Param(nameof(SlowPeriod), 30)
			.SetDisplay("Slow SMA", "Slow SMA period", "Indicators");

		_rsiPeriod = Param(nameof(RsiPeriod), 14)
			.SetDisplay("RSI Period", "RSI lookback", "Indicators");

		_overbought = Param(nameof(Overbought), 75m)
			.SetDisplay("Overbought", "RSI overbought level", "Levels");

		_oversold = Param(nameof(Oversold), 25m)
			.SetDisplay("Oversold", "RSI oversold level", "Levels");

		_candleType = Param(nameof(CandleType), TimeSpan.FromHours(4).TimeFrame())
			.SetDisplay("Candle Type", "Candle timeframe", "General");
	}

	public override IEnumerable<(Security sec, DataType dt)> GetWorkingSecurities() => [(Security, CandleType)];
	protected override void OnReseted() { base.OnReseted(); _prevFast = 0m; _prevSlow = 0m; _hasPrev = false; }

	protected override void OnStarted2(DateTime time)
	{
		base.OnStarted2(time);

		_hasPrev = false;

		var fast = new SimpleMovingAverage { Length = FastPeriod };
		var slow = new SimpleMovingAverage { Length = SlowPeriod };

		var subscription = SubscribeCandles(CandleType);
		subscription
			.Bind(fast, slow, ProcessCandle)
			.Start();
	}

	private void ProcessCandle(ICandleMessage candle, decimal fast, decimal slow)
	{
		if (candle.State != CandleStates.Finished)
			return;

		if (!_hasPrev)
		{
			_prevFast = fast;
			_prevSlow = slow;
			_hasPrev = true;
			return;
		}

		// Fast crosses above slow = buy
		if (_prevFast <= _prevSlow && fast > slow && Position <= 0)
		{
			if (Position < 0)
				BuyMarket();
			BuyMarket();
		}
		// Fast crosses below slow = sell
		else if (_prevFast >= _prevSlow && fast < slow && Position >= 0)
		{
			if (Position > 0)
				SellMarket();
			SellMarket();
		}

		_prevFast = fast;
		_prevSlow = slow;
	}
}