
이 전략은 가격 분산 신호에 기반한 트렌드 거래 전략이다. RSI, MACD, Stochastics 등과 같은 여러 지표들을 사용하여 가격 분산 신호를 검출하고, 머레이 매트 진동기를 통해 확인한다. 가격 분산 신호가 발생하면, 진동기가 현재 트렌드 방향이라고 확인하면, 입장이 이루어진다.
이 전략의 핵심은 가격 분산 이론이다. 가격 혁신이 높지만 지표가 혁신이 높지 않은 경우를 곰 시장 가격 분산이라고 하며; 가격 혁신이 낮지만 지표가 혁신이 낮지 않은 경우를 황소 시장 가격 분산이라고 한다. 이것은 트렌드가 역전될 수 있음을 나타낸다. 전략은 상위 하위 유형을 사용하여 오징어와 결합하여 거래 신호를 확인한다.
이 전략의 입시 조건은 다음과 같습니다.
출전 조건은 진동기가 중선을 통과할 때 평행한다.
이 전략은 가격 분산 이론과 트렌드 확인을 결합하여 다음과 같은 장점을 가지고 있다:
주요 위험은 다음과 같습니다.
스톱로스를 설정하고, 포지션을 조정하고, 위험을 줄이기 위해 변수 조합을 최적화하는 것이 좋습니다.
이 전략에는 더 많은 최적화 가능성이 있습니다:
이 전략은 가격 분산 이론과 트렌드 분석 지표를 통합하여 잠재적인 트렌드 전환점을 효과적으로 발견할 수 있습니다. 최적화된 위험 관리 조치와 결합하여 더 나은 전략 수익률을 얻을 수 있습니다.
/*backtest
start: 2024-01-02 00:00:00
end: 2024-02-01 00:00:00
period: 2h
basePeriod: 15m
exchanges: [{"eid":"Futures_Binance","currency":"BTC_USDT"}]
*/
//@version=2
//
// Title: [STRATEGY][UL]Price Divergence Strategy V1
// Author: JustUncleL
// Date: 23-Oct-2016
// Version: v1.0
//
// Description:
// A trend trading strategy the uses Price Divergence detection signals, that
// are confirmed by the "Murrey's Math Oscillator" (Donchanin Channel based).
//
// *** USE AT YOUR OWN RISK ***
//
// Mofidifications:
// 1.0 - original
//
// References:
// Strategy Based on:
// - [RS]Price Divergence Detector V2 by RicardoSantos
// - UCS_Murrey's Math Oscillator by Ucsgears
// Some Code borrowed from:
// - "Strategy Code Example by JayRogers"
// Information on Divergence Trading:
// - http://www.babypips.com/school/high-school/trading-divergences
//
strategy(title='[STRATEGY][UL]Price Divergence Strategy v1.0', pyramiding=0, overlay=true, initial_capital=10000, calc_on_every_tick=false,
currency=currency.USD,default_qty_type=strategy.percent_of_equity,default_qty_value=10)
// || General Input:
method = input(title='Method (0=rsi, 1=macd, 2=stoch, 3=volume, 4=acc/dist, 5=fisher, 6=cci):', defval=1, minval=0, maxval=6)
SHOW_LABEL = input(title='Show Labels', type=bool, defval=true)
SHOW_CHANNEL = input(title='Show Channel', type=bool, defval=false)
uHid = input(true,title="Use Hidden Divergence in Strategy")
uReg = input(true,title="Use Regular Divergence in Strategy")
// || RSI / STOCH / VOLUME / ACC/DIST Input:
rsi_smooth = input(title='RSI/STOCH/Volume/ACC-DIST/Fisher/cci Smooth:', defval=5)
// || MACD Input:
macd_src = input(title='MACD Source:', defval=close)
macd_fast = input(title='MACD Fast:', defval=12)
macd_slow = input(title='MACD Slow:', defval=26)
macd_smooth = input(title='MACD Smooth Signal:', defval=9)
// || Functions:
f_top_fractal(_src)=>_src[4] < _src[2] and _src[3] < _src[2] and _src[2] > _src[1] and _src[2] > _src[0]
f_bot_fractal(_src)=>_src[4] > _src[2] and _src[3] > _src[2] and _src[2] < _src[1] and _src[2] < _src[0]
f_fractalize(_src)=>f_top_fractal(_src) ? 1 : f_bot_fractal(_src) ? -1 : 0
// ||••> START MACD FUNCTION
f_macd(_src, _fast, _slow, _smooth)=>
_fast_ma = sma(_src, _fast)
_slow_ma = sma(_src, _slow)
_macd = _fast_ma-_slow_ma
_signal = ema(_macd, _smooth)
_hist = _macd - _signal
// ||<•• END MACD FUNCTION
// ||••> START ACC/DIST FUNCTION
f_accdist(_smooth)=>_return=sma(cum(close==high and close==low or high==low ? 0 : ((2*close-low-high)/(high-low))*volume), _smooth)
// ||<•• END ACC/DIST FUNCTION
// ||••> START FISHER FUNCTION
f_fisher(_src, _window)=>
_h = highest(_src, _window)
_l = lowest(_src, _window)
_value0 = .66 * ((_src - _l) / max(_h - _l, .001) - .5) + .67 * nz(_value0[1])
_value1 = _value0 > .99 ? .999 : _value0 < -.99 ? -.999 : _value0
_fisher = .5 * log((1 + _value1) / max(1 - _value1, .001)) + .5 * nz(_fisher[1])
// ||<•• END FISHER FUNCTION
method_high = method == 0 ? rsi(high, rsi_smooth) :
method == 1 ? f_macd(macd_src, macd_fast, macd_slow, macd_smooth) :
method == 2 ? stoch(close, high, low, rsi_smooth) :
method == 3 ? sma(volume, rsi_smooth) :
method == 4 ? f_accdist(rsi_smooth) :
method == 5 ? f_fisher(high, rsi_smooth) :
method == 6 ? cci(high, rsi_smooth) :
na
method_low = method == 0 ? rsi(low, rsi_smooth) :
method == 1 ? f_macd(macd_src, macd_fast, macd_slow, macd_smooth) :
method == 2 ? stoch(close, high, low, rsi_smooth) :
method == 3 ? sma(volume, rsi_smooth) :
method == 4 ? f_accdist(rsi_smooth) :
method == 5 ? f_fisher(low, rsi_smooth) :
method == 6 ? cci(low, rsi_smooth) :
na
fractal_top = f_fractalize(method_high) > 0 ? method_high[2] : na
fractal_bot = f_fractalize(method_low) < 0 ? method_low[2] : na
high_prev = valuewhen(fractal_top, method_high[2], 1)
high_price = valuewhen(fractal_top, high[2], 1)
low_prev = valuewhen(fractal_bot, method_low[2], 1)
low_price = valuewhen(fractal_bot, low[2], 1)
regular_bearish_div = fractal_top and high[2] > high_price and method_high[2] < high_prev
hidden_bearish_div = fractal_top and high[2] < high_price and method_high[2] > high_prev
regular_bullish_div = fractal_bot and low[2] < low_price and method_low[2] > low_prev
hidden_bullish_div = fractal_bot and low[2] > low_price and method_low[2] < low_prev
plot(title='H F', series=fractal_top ? high[2] : na, color=regular_bearish_div or hidden_bearish_div ? maroon : not SHOW_CHANNEL ? na : silver, offset=-2)
plot(title='L F', series=fractal_bot ? low[2] : na, color=regular_bullish_div or hidden_bullish_div ? green : not SHOW_CHANNEL ? na : silver, offset=-2)
plot(title='H D', series=fractal_top ? high[2] : na, style=circles, color=regular_bearish_div or hidden_bearish_div ? maroon : not SHOW_CHANNEL ? na : silver, linewidth=3, offset=-2)
plot(title='L D', series=fractal_bot ? low[2] : na, style=circles, color=regular_bullish_div or hidden_bullish_div ? green : not SHOW_CHANNEL ? na : silver, linewidth=3, offset=-2)
plotshape(title='+RBD', series=not SHOW_LABEL ? na : regular_bearish_div ? high[2] : na, text='R', style=shape.labeldown, location=location.absolute, color=maroon, textcolor=white, offset=-2)
plotshape(title='+HBD', series=not SHOW_LABEL ? na : hidden_bearish_div ? high[2] : na, text='H', style=shape.labeldown, location=location.absolute, color=maroon, textcolor=white, offset=-2)
plotshape(title='-RBD', series=not SHOW_LABEL ? na : regular_bullish_div ? low[2] : na, text='R', style=shape.labelup, location=location.absolute, color=green, textcolor=white, offset=-2)
plotshape(title='-HBD', series=not SHOW_LABEL ? na : hidden_bullish_div ? low[2] : na, text='H', style=shape.labelup, location=location.absolute, color=green, textcolor=white, offset=-2)
// Code borrowed from UCS_Murrey's Math Oscillator by Ucsgears
// - UCS_MMLO
// Inputs
length = input(100, minval = 10, title = "MMLO Look back Length")
quad = input(2, minval = 1, maxval = 4, step = 1, title = "Mininum Quadrant for MMLO Support")
mult = 0.125
// Donchanin Channel
hi = highest(high, length)
lo = lowest(low, length)
range = hi - lo
multiplier = (range) * mult
midline = lo + multiplier * 4
oscillator = (close - midline)/(range/2)
a = oscillator > 0
b = oscillator > 0 and oscillator > mult*2
c = oscillator > 0 and oscillator > mult*4
d = oscillator > 0 and oscillator > mult*6
z = oscillator < 0
y = oscillator < 0 and oscillator < -mult*2
x = oscillator < 0 and oscillator < -mult*4
w = oscillator < 0 and oscillator < -mult*6
// Strategy: (Thanks to JayRogers)
// === STRATEGY RELATED INPUTS ===
//tradeInvert = input(defval = false, title = "Invert Trade Direction?")
// the risk management inputs
inpTakeProfit = input(defval = 0, title = "Take Profit Points", minval = 0)
inpStopLoss = input(defval = 0, title = "Stop Loss Points", minval = 0)
inpTrailStop = input(defval = 100, title = "Trailing Stop Loss Points", minval = 0)
inpTrailOffset = input(defval = 0, title = "Trailing Stop Loss Offset Points", minval = 0)
// === RISK MANAGEMENT VALUE PREP ===
// if an input is less than 1, assuming not wanted so we assign 'na' value to disable it.
useTakeProfit = inpTakeProfit >= 1 ? inpTakeProfit : na
useStopLoss = inpStopLoss >= 1 ? inpStopLoss : na
useTrailStop = inpTrailStop >= 1 ? inpTrailStop : na
useTrailOffset = inpTrailOffset >= 1 ? inpTrailOffset : na
// === STRATEGY - LONG POSITION EXECUTION ===
enterLong() => ((uReg and regular_bullish_div) or (uHid and hidden_bullish_div)) and (quad==1? a[1]: quad==2?b[1]: quad==3?c[1]: quad==4?d[1]: false)// functions can be used to wrap up and work out complex conditions
exitLong() => oscillator <= 0
strategy.entry(id = "Buy", long = true, when = enterLong() )// use function or simple condition to decide when to get in
strategy.close(id = "Buy", when = exitLong() )// ...and when to get out
// === STRATEGY - SHORT POSITION EXECUTION ===
enterShort() => ((uReg and regular_bearish_div) or (uHid and hidden_bearish_div)) and (quad==1? z[1]: quad==2?y[1]: quad==3?x[1]: quad==4?w[1]: false)
exitShort() => oscillator >= 0
strategy.entry(id = "Sell", long = false, when = enterShort())
strategy.close(id = "Sell", when = exitShort() )
// === STRATEGY RISK MANAGEMENT EXECUTION ===
// finally, make use of all the earlier values we got prepped
strategy.exit("Exit Buy", from_entry = "Buy", profit = useTakeProfit, loss = useStopLoss, trail_points = useTrailStop, trail_offset = useTrailOffset)
strategy.exit("Exit Sell", from_entry = "Sell", profit = useTakeProfit, loss = useStopLoss, trail_points = useTrailStop, trail_offset = useTrailOffset)
//EOF