この戦略は,ブル・ベア・パワー (BBP) インジケーターと,ボリューム・パーセンチルに基づいた多レベルダイナミック・テイク・プロフィート・システムを組み合わせます.価格,ボリューム,インパクトデータの多次元分析を通じて,適応性のあるリスク制御型取引システムを作成します.コア論理には,Zスコア標準化されたBBP値をトレード・シグナルトリガーとして使用し,ダイナミック・テイク・プロフィート調整のためのボリューム・パーセンチル分析を組み込むことが含まれます.
基本計算にはいくつかの重要な要素が含まれます.
この戦略は,伝統的なBBP指標と近代的な定量分析方法を組み合わせ,堅実な理論的基礎と強力な実用性を持つ取引システムを構築する.多レベルの利益と動的な調整メカニズムを通じて,利益とリスクの良いバランスを達成する.パラメータ最適化はいくつかの課題を提示しているが,戦略フレームワークの拡張性は将来の改善のための広い余地を提供します.実用的な応用では,トレーダーは市場特性と個々のリスク好みに基づいて特定の調整を行うべきです.
/*backtest start: 2019-12-23 08:00:00 end: 2025-01-04 08:00:00 period: 1d basePeriod: 1d exchanges: [{"eid":"Futures_Binance","currency":"BTC_USDT"}] */ // This Pine Script™ code is subject to the terms of the Mozilla Public License 2.0 at https://mozilla.org/MPL/2.0/ // © PresentTrading // The BBP Strategy with Volume-Percentile TP by PresentTrading emerges as a sophisticated approach that integrates multiple analytical layers to enhance trading precision and profitability. // Unlike traditional strategies that rely solely on price movements or volume indicators, this strategy synergizes Bollinger Bands Power (BBP) with volume percentile analysis to determine optimal entry and exit points. Additionally, it employs a dynamic take-profit mechanism based on ATR (Average True Range) multipliers adjusted by volume and percentile factors, ensuring adaptability to varying market conditions. // This multi-faceted approach not only enhances signal accuracy but also optimizes risk management, setting it apart from conventional trading methodologies. //@version=5 strategy("BBP Strategy with Volume-Percentile TP - Strategy [presentTrading] ", overlay=false, precision=3, commission_value= 0.1, commission_type=strategy.commission.percent, slippage= 1, currency=currency.USD, default_qty_type = strategy.percent_of_equity, default_qty_value = 10, initial_capital=10000) // ———————— // Bull Bear Power Strategy Settings // ———————— lengthInput = input.int(21, "EMA Length") zLength = input.int(252, "Z-Score Length") zThreshold = input.float(1.618, "Z-Score Threshold") // ———————— // Take Profit Settings // ———————— tp_group = "Take Profit Settings" // Enable/disable take profit function useTP = input.bool(true, "Use Take Profit", group=tp_group) // === ATR Base Settings === // ATR calculation period for determining base price movement range baseAtrLength = input.int(20, "ATR Period", minval=1, group=tp_group, tooltip="ATR period for calculating base price movement range. Shorter periods are more sensitive to recent volatility") // === Take Profit Multiplier Settings === // First take profit ATR multiplier, usually the most conservative target atrMult1 = input.float(1.618, "TP1 ATR Multiplier", minval=0.1, step=0.1, group=tp_group, tooltip="First take profit level ATR multiplier, recommended 1.5-2.0") // Second take profit ATR multiplier, medium profit target atrMult2 = input.float(2.382, "TP2 ATR Multiplier", minval=0.1, step=0.1, group=tp_group, tooltip="Second take profit level ATR multiplier, recommended 2.5-3.0") // Third take profit ATR multiplier, most aggressive target atrMult3 = input.float(3.618, "TP3 ATR Multiplier", minval=0.1, step=0.1, group=tp_group, tooltip="Third take profit level ATR multiplier, recommended 4.0-5.0") // === Position Size Allocation === // First take profit position size, usually larger for securing basic profits tp1_size = input.float(13, "TP1 Position %", minval=1, maxval=100, group=tp_group, tooltip="Position size percentage for first take profit, recommended 30-40%") // Second take profit position size, medium allocation tp2_size = input.float(13, "TP2 Position %", minval=1, maxval=100, group=tp_group, tooltip="Position size percentage for second take profit, recommended 30-40%") // Third take profit position size, usually smaller for catching larger moves tp3_size = input.float(13, "TP3 Position %", minval=1, maxval=100, group=tp_group, tooltip="Position size percentage for third take profit, recommended 20-30%") // ———————— // Volume Analysis Settings // ———————— vol_group = "Volume Analysis Settings" // Volume MA period for determining relative volume levels vol_period = input.int(100, "Volume MA Period", minval=1, group=vol_group, tooltip="Period for calculating volume moving average, recommended 20-30") // === Volume Level Thresholds === // High volume threshold relative to MA vol_high = input.float(2.0, "High Volume Multiplier", minval=1.0, step=0.1, group=vol_group, tooltip="High volume threshold multiplier, typically 2x MA or above") // Medium volume threshold vol_med = input.float(1.5, "Medium Volume Multiplier", minval=1.0, step=0.1, group=vol_group, tooltip="Medium volume threshold multiplier, typically around 1.5x MA") // Low volume threshold vol_low = input.float(1.0, "Low Volume Multiplier", minval=0.5, step=0.1, group=vol_group, tooltip="Low volume threshold multiplier, typically around 1x MA") // === Volume Adjustment Factors === // High volume adjustment factor, usually extends take profit targets vol_high_mult = input.float(1.5, "High Volume Factor", minval=0.1, step=0.1, group=vol_group, tooltip="Take profit adjustment factor for high volume") // Medium volume adjustment factor vol_med_mult = input.float(1.3, "Medium Volume Factor", minval=0.1, step=0.1, group=vol_group, tooltip="Take profit adjustment factor for medium volume") // Low volume adjustment factor vol_low_mult = input.float(1.0, "Low Volume Factor", minval=0.1, step=0.1, group=vol_group, tooltip="Take profit adjustment factor for low volume") // ———————— // Percentile Analysis Settings // ———————— perc_group = "Percentile Analysis Settings" // Percentile calculation period for evaluating price position perc_period = input.int(100, "Percentile Period", minval=20, group=perc_group, tooltip="Historical period for percentile calculations, recommended 100-200") // === Percentile Thresholds === // High percentile threshold, typically indicates relative high levels perc_high = input.float(90, "High Percentile", minval=50, maxval=100, group=perc_group, tooltip="High level percentile threshold, typically above 90") // Medium percentile threshold perc_med = input.float(80, "Medium Percentile", minval=50, maxval=100, group=perc_group, tooltip="Medium level percentile threshold, typically around 80") // Low percentile threshold perc_low = input.float(70, "Low Percentile", minval=0, maxval=100, group=perc_group, tooltip="Low level percentile threshold, typically around 70") // === Percentile Adjustment Factors === // High percentile adjustment factor perc_high_mult = input.float(1.5, "High Percentile Factor", minval=0.1, step=0.1, group=perc_group, tooltip="Take profit adjustment factor for high percentile levels") // Medium percentile adjustment factor perc_med_mult = input.float(1.3, "Medium Percentile Factor", minval=0.1, step=0.1, group=perc_group, tooltip="Take profit adjustment factor for medium percentile levels") // Low percentile adjustment factor perc_low_mult = input.float(1.0, "Low Percentile Factor", minval=0.1, step=0.1, group=perc_group, tooltip="Take profit adjustment factor for low percentile levels") // ———————— // Core Bull Bear Power Calculations // ———————— emaClose = ta.ema(close, lengthInput) bullPower = high - emaClose bearPower = low - emaClose bbp = bullPower + bearPower bbp_mean = ta.sma(bbp, zLength) bbp_std = ta.stdev(bbp, zLength) zscore = (bbp - bbp_mean) / bbp_std // ———————— // Volume & Percentile Analysis // ———————— // 成交量分析 vol_sma = ta.sma(volume, vol_period) vol_mult = volume / vol_sma // 百分位數計算 calcPercentile(src) => var values = array.new_float(0) array.unshift(values, src) if array.size(values) > perc_period array.pop(values) array.size(values) > 0 ? array.percentrank(values, array.size(values)-1) * 100 : 50 price_perc = calcPercentile(close) vol_perc = calcPercentile(volume) // 止盈動態調整系數計算 getTpFactor() => vol_score = vol_mult > vol_high ? vol_high_mult : vol_mult > vol_med ? vol_med_mult : vol_mult > vol_low ? vol_low_mult : 0.8 price_score = price_perc > perc_high ? perc_high_mult :price_perc > perc_med ? perc_med_mult :price_perc > perc_low ? perc_low_mult : 0.8 math.avg(vol_score, price_score) // ———————— // Entry/Exit Logic // ———————— longCondition = ta.crossover(zscore, zThreshold) shortCondition = ta.crossunder(zscore, -zThreshold) exitLongCondition = ta.crossunder(zscore, 0) exitShortCondition = ta.crossover(zscore, 0) if (barstate.isconfirmed) if longCondition strategy.entry("Long", strategy.long) if shortCondition strategy.entry("Short", strategy.short) if exitLongCondition strategy.close("Long") if exitShortCondition strategy.close("Short") // ———————— // Take Profit Execution // ———————— if useTP and strategy.position_size != 0 base_move = ta.atr(baseAtrLength) tp_factor = getTpFactor() is_long = strategy.position_size > 0 entry_price = strategy.position_avg_price if is_long tp1_price = entry_price + (base_move * atrMult1 * tp_factor) tp2_price = entry_price + (base_move * atrMult2 * tp_factor) tp3_price = entry_price + (base_move * atrMult3 * tp_factor) strategy.exit("TP1", "Long", qty_percent=tp1_size, limit=tp1_price) strategy.exit("TP2", "Long", qty_percent=tp2_size, limit=tp2_price) strategy.exit("TP3", "Long", qty_percent=tp3_size, limit=tp3_price) else tp1_price = entry_price - (base_move * atrMult1 * tp_factor) tp2_price = entry_price - (base_move * atrMult2 * tp_factor) tp3_price = entry_price - (base_move * atrMult3 * tp_factor) strategy.exit("TP1", "Short", qty_percent=tp1_size, limit=tp1_price) strategy.exit("TP2", "Short", qty_percent=tp2_size, limit=tp2_price) strategy.exit("TP3", "Short", qty_percent=tp3_size, limit=tp3_price) // ———————— // Plotting // ———————— plot(bbp, color=bbp >= 0 ? color.new(color.green, 0) : color.new(color.red, 0), title="BBPower", style=plot.style_columns) hline(0, "Zero Line", color=color.gray, linestyle=hline.style_dotted) plot(zscore, title="Z-Score", color=color.blue, linewidth=2) hline(zThreshold, "Upper Threshold", color=color.orange, linestyle=hline.style_dashed) hline(-zThreshold, "Lower Threshold", color=color.orange, linestyle=hline.style_dashed)