The resource loading... loading...

Dynamic Filter Quant Trading Strategy

Author: ChaoZhang, Date: 2023-12-25 11:10:09
Tags:

img

Strategy Overview

This strategy named Dynamic Filter Quant Trading Strategy mainly uses Range Filter indicator combined with multiple technical indicators to implement automated trend tracking trading of the cryptocurrency BTCUSDT. The strategy is suitable for high-frequency quant trading by dynamically adjusting stop loss and take profit to lock in profits and reduce drawdowns.

Strategy Logic

The core indicator of this strategy is Range Filter, which generates a median line based on the statistical price movement range. Trading signals are generated when the price breaks through this median line. In addition, the strategy also combines RSI indicator to judge overbought and oversold, moving average to determine the trend, MACD to judge momentum and other indicators for combined filtering to form more reliable trading signals.

Specifically, the median line of the Range Filter is obtained from the exponential moving average of the price movement range, and the directional judgement is based on the strength and speed of breaking through this median line. When the price breaks through the median line continuously over several candlesticks, a strong breakout signal is generated.

The RSI indicator that judges the overbought and oversold state is used to confirm the filter signal. When the moving average points up, the trend is judged to be up, and when it points down, the trend is judged to be down. The MACD indicator judges whether the market momentum is sufficient to form a trend.

By combining the judgments of these indicators, relatively reliable trend breakthrough points can be identified as opportunities to establish positions.

Advantage Analysis

The biggest advantage of this strategy is that it combines multiple indicators for decision making instead of relying on a single technical indicator, which can effectively reduce the probability of wrong trades and ensure that trading signals are more reliable. In addition, the dynamic adjustment of parameters also enables the strategy to adapt to market changes.

Another advantage is that high-frequency trading can be performed. The Range Filter indicator is very sensitive to price changes over small periods, which means that the strategy can open and close positions in a relatively short period of time, so it is very suitable for high-frequency trading and allows profits to be made in the volatile cryptocurrency market.

Risk Analysis

This strategy still has some risks. The first is the risk that technical pattern judgments fail because indicators cannot guarantee price movements 100%. When prices reverse, it may lead to stop loss.

Another major risk is that the median line of the Range Filter cannot completely filter out price fluctuations. When there is a larger price fluctuation beyond the range of the median line, the median line will fail, resulting in the risk of generating wrong signals. In this case, the parameters can be appropriately relaxed to expand the range of the median line.

Finally, high-frequency trading itself also carries some risks. When the trading frequency is too high, transaction costs will be relatively large, which may offset some profits. In this case, the trading frequency and holding time can be appropriately reduced.

Optimization

There is still room for further optimization of this strategy. For example, more indicators can be considered, such as volatility indicators to confirm trends and establish stricter filtering criteria to ensure more precise trading signals. Or study the price behavior patterns of different cryptocurrencies and stocks, and set indicator parameters that best suit them.

From the trading logic, dynamic stop loss and take profit ranges can also be set. That is to say, when the position size increases, the stop loss range can be expanded to lock in more profits. Or when the profit is relatively large, accelerate the take profit speed. This can reduce drawdowns to some extent.

Finally, the filter parameters can be optimized to find a set of parameters so that the median line range can effectively filter out fluctuations while capturing trend turning points as much as possible. This requires a lot of backtest data for iterative analysis.

Summary

This strategy successfully combines multiple indicators for judgment to form a highly reliable trading strategy suitable for high-frequency quantitative trading. With continuous optimization and improvement, it is believed that stable returns can be obtained and it is worth further development.


/*backtest
start: 2022-12-18 00:00:00
end: 2023-12-24 00:00:00
period: 1d
basePeriod: 1h
exchanges: [{"eid":"Futures_Binance","currency":"BTC_USDT"}]
*/

//@version=5
strategy(title='5cel Scalp Strategy BTCUSDT Long & Short 30 Min', shorttitle='BTCUSDT Long & Short Scalp 30m', precision=1, overlay=true)

//Swing Call - Based on RSI Overbought & Oversold
//#### Starts Here #####
ema_value = input(5)
sma_value = input(50)
ema1 = ta.ema(close, ema_value)
sma2 = ta.sma(close, sma_value)
rs = ta.rsi(close, 14)

iff_1 = high < sma2 ? color.red : color.yellow
iff_2 = low > sma2 ? color.lime : iff_1
mycolor = rs >= 85 or rs <= 15 ? color.yellow : iff_2

//For Main Strategy
bool swingCallGreen = false
bool swingCallRed = false
bool swingCallYellow = false

if rs >= 85 or rs <= 15
    //color.yellow
    swingCallGreen := false
    swingCallRed := false
    swingCallYellow := true
    swingCallYellow
else
    if low > sma2
        //color.lime
        swingCallGreen := true
        swingCallRed := false
        swingCallYellow := false
        swingCallYellow
        //color.red
    else if high < sma2
        swingCallGreen := false
        swingCallRed := true
        swingCallYellow := false
        swingCallYellow
    else
        //color.yellow
        swingCallGreen := false
        swingCallRed := false
        swingCallYellow := true
        swingCallYellow

hlong = input.int(80, title='Overbought limit of RSI', step=1)
ll = input.int(20, title='Oversold limit of RSI', step=1)

buyexit = ta.crossunder(rs, hlong)
sellexit = ta.crossover(rs, ll)

sellcall = ta.crossover(sma2, ema1) and open > close
buycall = ta.crossunder(sma2, ema1) and high > sma2
//#### Ends Here #####


//Parabolic SAR -  Trend Circles
//#### Starts Here #####
start = input.int(2, minval=0, maxval=10, title='Start - Default = 2 - Multiplied by .01')
increment = input.int(2, minval=0, maxval=10, title='Step Setting (Sensitivity) - Default = 2 - Multiplied by .01')
maximum = input.int(2, minval=1, maxval=10, title='Maximum Step (Sensitivity) - Default = 2 - Multiplied by .10')
sus = input(true, 'Show Up Trending Parabolic Sar')
sds = input(true, 'Show Down Trending Parabolic Sar')
disc = input(false, title='Start and Step settings are *.01 so 2 = .02 etc, Maximum Step is *.10 so 2 = .2')

startCalc = start * .01
incrementCalc = increment * .01
maximumCalc = maximum * .10

sarUp = ta.sar(startCalc, incrementCalc, maximumCalc)
sarDown = ta.sar(startCalc, incrementCalc, maximumCalc)

colUp = close >= sarDown ? color.lime : na
colDown = close <= sarUp ? color.red : na

parabolicSARGreen = ta.sar(startCalc, incrementCalc, maximumCalc)
parabolicSARRed = ta.sar(startCalc, incrementCalc, maximumCalc)
//#### Ends Here #####


//EMA Line
//#### Starts Here #####
ema100 = ta.ema(close, 100)
//#### Ends Here #####


// Ichimoku Cloud
//#### Starts Here #####
sCloud = input(false, 'Show Ichimoku lines')

// Colors
colorGreen = #00ff00
colorRed = #ff0000
colorTenkanViolet = #9400D3
colorKijun = #fdd8a0
colorLime = #006400
colorMaroon = #8b0000

//Periods are set to standard
tenkanPeriods = input.int(9, minval=1, title='Tenkan')
kijunPeriods = input.int(26, minval=1, title='Kijun')
chikouPeriods = input.int(52, minval=1, title='Chikou')
displacement = input.int(26, minval=1, title='Offset')

donchian(len) =>
    math.avg(ta.lowest(len), ta.highest(len))

tenkan = donchian(tenkanPeriods)
kijun = donchian(kijunPeriods)
senkouA = math.avg(tenkan, kijun)
senkouB = donchian(chikouPeriods)
displacedSenkouA = senkouA[displacement]
displacedSenkouB = senkouB[displacement]

bullishSignal = ta.crossover(tenkan, kijun)
bearishSignal = ta.crossunder(tenkan, kijun)

bullishSignalValues = bullishSignal ? tenkan : na
bearishSignalValues = bearishSignal ? tenkan : na


strongBullishSignal = bullishSignalValues > displacedSenkouA and bullishSignalValues > displacedSenkouB
neutralBullishSignal = bullishSignalValues > displacedSenkouA and bullishSignalValues < displacedSenkouB or bullishSignalValues < displacedSenkouA and bullishSignalValues > displacedSenkouB
weakBullishSignal = bullishSignalValues < displacedSenkouA and bullishSignalValues < displacedSenkouB

strongBearishSignal = bearishSignalValues < displacedSenkouA and bearishSignalValues < displacedSenkouB
neutralBearishSignal = bearishSignalValues > displacedSenkouA and bearishSignalValues < displacedSenkouB or bearishSignalValues < displacedSenkouA and bearishSignalValues > displacedSenkouB
weakBearishSignal = bearishSignalValues > displacedSenkouA and bearishSignalValues > displacedSenkouB
//#### Ends Here #####


//Higher High Lower Low Strategy
//#### Starts Here #####
lb = input.int(5, title='Left Bars', minval=1)
rb = input.int(5, title='Right Bars', minval=1)
showsupres = input.bool(true, title='Support/Resistance', inline='srcol')
supcol = input.color(color.lime, title='', inline='srcol')
rescol = input.color(color.red, title='', inline='srcol')
// srlinestyle = input.string(line.style_dotted, title='Line Style/Width', options=[line.style_solid, line.style_dashed, line.style_dotted], inline='style')
srlinewidth = input.int(3, title='', minval=1, maxval=5, inline='style')
changebarcol = input.bool(true, title='Change Bar Color', inline='bcol')
bcolup = input.color(color.blue, title='', inline='bcol')
bcoldn = input.color(color.black, title='', inline='bcol')

ph = ta.pivothigh(lb, rb)
pl = ta.pivotlow(lb, rb)

iff_3 = pl ? -1 : na  // Trend direction
hl = ph ? 1 : iff_3
iff_4 = pl ? pl : na  // similar to zigzag but may have multiple highs/lows
zz = ph ? ph : iff_4
valuewhen_1 = ta.valuewhen(hl, hl, 1)
valuewhen_2 = ta.valuewhen(zz, zz, 1)
zz := pl and hl == -1 and valuewhen_1 == -1 and pl > valuewhen_2 ? na : zz
valuewhen_3 = ta.valuewhen(hl, hl, 1)
valuewhen_4 = ta.valuewhen(zz, zz, 1)
zz := ph and hl == 1 and valuewhen_3 == 1 and ph < valuewhen_4 ? na : zz

valuewhen_5 = ta.valuewhen(hl, hl, 1)
valuewhen_6 = ta.valuewhen(zz, zz, 1)
hl := hl == -1 and valuewhen_5 == 1 and zz > valuewhen_6 ? na : hl
valuewhen_7 = ta.valuewhen(hl, hl, 1)
valuewhen_8 = ta.valuewhen(zz, zz, 1)
hl := hl == 1 and valuewhen_7 == -1 and zz < valuewhen_8 ? na : hl
zz := na(hl) ? na : zz

findprevious() =>  // finds previous three points (b, c, d, e)
    ehl = hl == 1 ? -1 : 1
    loc1 = 0.0
    loc2 = 0.0
    loc3 = 0.0
    loc4 = 0.0
    xx = 0
    for x = 1 to 1000 by 1
        if hl[x] == ehl and not na(zz[x])
            loc1 := zz[x]
            xx := x + 1
            break
    ehl := hl
    for x = xx to 1000 by 1
        if hl[x] == ehl and not na(zz[x])
            loc2 := zz[x]
            xx := x + 1
            break
    ehl := hl == 1 ? -1 : 1
    for x = xx to 1000 by 1
        if hl[x] == ehl and not na(zz[x])
            loc3 := zz[x]
            xx := x + 1
            break
    ehl := hl
    for x = xx to 1000 by 1
        if hl[x] == ehl and not na(zz[x])
            loc4 := zz[x]
            break
    [loc1, loc2, loc3, loc4]

float a = na
float b = na
float c = na
float d = na
float e = na
if not na(hl)
    [loc1, loc2, loc3, loc4] = findprevious()
    a := zz
    b := loc1
    c := loc2
    d := loc3
    e := loc4

_hh = zz and a > b and a > c and c > b and c > d
_ll = zz and a < b and a < c and c < b and c < d
_hl = zz and (a >= c and b > c and b > d and d > c and d > e or a < b and a > c and b < d)
_lh = zz and (a <= c and b < c and b < d and d < c and d < e or a > b and a < c and b > d)

plotshape(_hl, text='HL', title='Higher Low', style=shape.labelup, color=color.new(color.lime, 0), textcolor=color.new(color.black, 0), location=location.belowbar, offset=-rb)
plotshape(_hh, text='HH', title='Higher High', style=shape.labeldown, color=color.new(color.lime, 0), textcolor=color.new(color.black, 0), location=location.abovebar, offset=-rb)
plotshape(_ll, text='LL', title='Lower Low', style=shape.labelup, color=color.new(color.red, 0), textcolor=color.new(color.white, 0), location=location.belowbar, offset=-rb)
plotshape(_lh, text='LH', title='Lower High', style=shape.labeldown, color=color.new(color.red, 0), textcolor=color.new(color.white, 0), location=location.abovebar, offset=-rb)

float res = na
float sup = na
res := _lh ? zz : res[1]
sup := _hl ? zz : sup[1]

int trend = na
iff_5 = close < sup ? -1 : nz(trend[1])
trend := close > res ? 1 : iff_5

res := trend == 1 and _hh or trend == -1 and _lh ? zz : res
sup := trend == 1 and _hl or trend == -1 and _ll ? zz : sup
rechange = res != res[1]
suchange = sup != sup[1]

var line resline = na
var line supline = na
//#### Ends Here #####



//Range Filter 5Min
//#### Starts Here #####

src = input(defval=close, title='Source')
per = input.int(defval=100, minval=1, title='Sampling Period')

// Range Multiplier
mult = input.float(defval=3.0, minval=0.1, title='Range Multiplier')

// Smooth Average Range
smoothrng(x, t, m) =>
    wper = t * 2 - 1
    avrng = ta.ema(math.abs(x - x[1]), t)
    smoothrng = ta.ema(avrng, wper) * m
    smoothrng
smrng = smoothrng(src, per, mult)

// Range Filter
rngfilt(x, r) =>
    rngfilt = x
    rngfilt := x > nz(rngfilt[1]) ? x - r < nz(rngfilt[1]) ? nz(rngfilt[1]) : x - r : x + r > nz(rngfilt[1]) ? nz(rngfilt[1]) : x + r
    rngfilt
filt = rngfilt(src, smrng)

// Filter Direction
upward = 0.0
upward := filt > filt[1] ? nz(upward[1]) + 1 : filt < filt[1] ? 0 : nz(upward[1])
downward = 0.0
downward := filt < filt[1] ? nz(downward[1]) + 1 : filt > filt[1] ? 0 : nz(downward[1])

// Target Bands
hband = filt + smrng
lband = filt - smrng

// Colors
filtcolor = upward > 0 ? color.lime : downward > 0 ? color.red : color.orange
barcolor = src > filt and src > src[1] and upward > 0 ? color.lime : src > filt and src < src[1] and upward > 0 ? color.green : src < filt and src < src[1] and downward > 0 ? color.red : src < filt and src > src[1] and downward > 0 ? color.maroon : color.orange

// Break Outs
longCond = bool(na)
shortCond = bool(na)
longCond := src > filt and src > src[1] and upward > 0 or src > filt and src < src[1] and upward > 0
shortCond := src < filt and src < src[1] and downward > 0 or src < filt and src > src[1] and downward > 0

CondIni = 0
CondIni := longCond ? 1 : shortCond ? -1 : CondIni[1]
longCondition = longCond and CondIni[1] == -1
shortCondition = shortCond and CondIni[1] == 1
//#### Ends Here #####


//#### Starts Here #####
source = close
useCurrentRes = input(true, title='Use Current Chart Resolution?')
resCustom = input.timeframe(title='Use Different Timeframe? Uncheck Box Above', defval='60')
smd = input(true, title='Show MacD & Signal Line? Also Turn Off Dots Below')
sd = input(true, title='Show Dots When MacD Crosses Signal Line?')
sh = input(true, title='Show Histogram?')
macd_colorChange = input(true, title='Change MacD Line Color-Signal Line Cross?')
hist_colorChange = input(true, title='MacD Histogram 4 Colors?')

res1 = useCurrentRes ? timeframe.period : resCustom

fastLength = input.int(12, minval=1)
slowLength = input.int(26, minval=1)
signalLength = input.int(9, minval=1)

fastMA = ta.ema(source, fastLength)
slowMA = ta.ema(source, slowLength)

macd = fastMA - slowMA
signal = ta.sma(macd, signalLength)
hist = macd - signal

outMacD = request.security(syminfo.tickerid, res1, macd)
outSignal = request.security(syminfo.tickerid, res1, signal)
outHist = request.security(syminfo.tickerid, res1, hist)

histA_IsUp = outHist > outHist[1] and outHist > 0
histA_IsDown = outHist < outHist[1] and outHist > 0
histB_IsDown = outHist < outHist[1] and outHist <= 0
histB_IsUp = outHist > outHist[1] and outHist <= 0

//MacD Color Definitions
macd_IsAbove = outMacD >= outSignal
macd_IsBelow = outMacD < outSignal

plot_color = hist_colorChange ? histA_IsUp ? color.aqua : histA_IsDown ? color.blue : histB_IsDown ? color.red : histB_IsUp ? color.maroon : color.yellow : color.gray
macd_color = macd_colorChange ? macd_IsAbove ? color.lime : color.red : color.red
signal_color = macd_colorChange ? macd_IsAbove ? color.yellow : color.yellow : color.lime

circleYPosition = outSignal
//#### Ends Here #####


//////////////////
// Main Strategy
/////////////////
//#### Starts Here #####
var bottomText = 'Something is not ok'

bool rangeBuy = false
if longCondition
    rangeBuy := true
else
    rangeBuy := false

bool rangeSell = false
if shortCondition
    rangeSell := true
else
    rangeSell := false

bool ema100Bullish = false
bool ema100Bearish = false
bool ichimokuBearish = false
bool ichimokuBullish = false
string statusChance = 'Who knows what will happen'
string futureIchimokuTrend = 'Anything can happen'

if close > ema100
    ema100Bullish := true
    ema100Bearish := false
else
    ema100Bullish := false
    ema100Bearish := true

if displacedSenkouA > displacedSenkouB
    ichimokuBearish := false
    futureIchimokuTrend := 'Green - chance to go up'
    ichimokuBullish := true
else
    ichimokuBearish := true
    futureIchimokuTrend := 'Red - chance to go down'
    ichimokuBullish := false
    ichimokuBullish

if ema100Bullish and parabolicSARGreen
    if ichimokuBullish
        statusChance := '100%'
    else
        statusChance := '95%'
else
    if ema100Bullish and parabolicSARRed
        statusChance := '75%'
    else if ema100Bearish and parabolicSARGreen
        statusChance := '65%'
    else
        statusChance := '55%'

bool longTradePosition = false
bool shortTradePosition = false
string longTradeText = 'Now cannot say anything'

if (swingCallGreen or swingCallYellow) and ichimokuBullish and longCondition and ema100Bullish and parabolicSARGreen
    longTradePosition := true
    longTradeText := 'Bullish'

bottomText := longTradeText + ' Chance: ' + statusChance + '\n Future Trend: ' + futureIchimokuTrend
// Bottom Text

var tLog = table.new(position=position.bottom_right, rows=1, columns=2, bgcolor=color.blue, border_width=1)
table.cell(tLog, row=0, column=0, text=bottomText, text_color=color.white)
table.cell_set_text(tLog, row=0, column=0, text=bottomText)
//#### Ends Here #####

bool entryLongPosition = false
bool exitLongPosition = false

bool entryShortPosition = false
bool exitShortPosition = false

bool longPositionCount = false
bool shortPositionCount = false


if (strategy.position_size > 0)
    longPositionCount := true

if (strategy.position_size < 0)
    shortPositionCount := true
    
// Entry LONG
if (longCondition) and (not longPositionCount)
    entryLongPosition := true

// Exit LONG
if (shortCondition) and (longPositionCount)
    exitLongPosition := true
    
// Entry SHORT
if (shortCondition) and (not shortPositionCount)
    entryShortPosition := true

// Exit SHORT
if (longCondition) and (shortPositionCount)
    exitShortPosition := true

// LONG Entry & Exit
plotshape(entryLongPosition, style=shape.labeldown, location=location.abovebar, color=color.new(color.green, 0), size=size.tiny, title='buy label', text='5cel\nLONG Entry', textcolor=color.new(color.white, 0))
plotshape(exitLongPosition, style=shape.labelup, location=location.belowbar, color=color.new(color.blue, 0), size=size.tiny, title='sell label', text='5cel\nExit LONG', textcolor=color.new(color.white, 0))

//SHORT Entry & Exit
plotshape(entryShortPosition, style=shape.labeldown, location=location.abovebar, color=color.new(color.red, 0), size=size.tiny, title='buy label', text='5cel\nSHORT Entry', textcolor=color.new(color.white, 0))
plotshape(exitShortPosition, style=shape.labelup, location=location.belowbar, color=color.new(color.blue, 0), size=size.tiny, title='sell label', text='5cel\nExit SHORT', textcolor=color.new(color.white, 0))

//Get the Current Value
heikinashi_close = request.security(ticker.heikinashi(syminfo.tickerid), timeframe.period, close)

if entryLongPosition
    longLabel = label.new(bar_index, high, text=str.tostring(heikinashi_close, '0.00'), color=color.orange, style=label.style_label_down, yloc=yloc.abovebar)

if entryShortPosition
    shortLabel = label.new(bar_index, high, text=str.tostring(heikinashi_close, '0.00'), color=color.orange, style=label.style_label_down, yloc=yloc.abovebar)

/// SHORT Exit
strategy.close("short", when=exitShortPosition, comment="close_short_position")

/// LONG Exit
strategy.close("long", when=exitLongPosition, comment = "close_long_position")

/// LONG Enter
strategy.entry("long", strategy.long, when=entryLongPosition, comment="open_long_position")

/// SHORT Enter
strategy.entry("short", strategy.short, when = entryShortPosition, comment="open_short_position")

More