O recurso está a ser carregado... Carregamento...

Estratégia de negociação ADX bidirecional

Autora:ChaoZhang, Data: 2024-01-30 17:00:44
Tags:

img

Resumo

A estratégia de negociação ADX bidirecional é uma estratégia quantitativa que implementa negociação bidirecional usando o indicador de índice direcional médio (ADX).

Estratégia lógica

  1. Calcule o intervalo verdadeiro
  2. Calcular o movimento direcional mais e movimento direcional menos
  3. Calcule a faixa real suavizada
  4. Calcular o movimento direcional suavizado mais e o movimento direcional suavizado menos
  5. Calcular os indicadores DIPlus, DIMinus e ADX
  6. Calcular a diferença entre DIPlus & ADX e DIMinus & ADX
  7. Estabelecer limiares para as diferenças comerciais longas e curtas
  8. Gerar sinais de negociação quando a diferença exceder limiares
  9. Criar ordens de compra e venda

O núcleo desta estratégia é o uso do ADX e indicadores de movimento direcional para determinar a direção e a força da tendência, combinados com regras de limiar de diferença para filtrar sinais e automatizar a negociação.

Análise das vantagens

  1. O ADX capta com precisão a tendência do mercado
  2. As regras de limiar de diferença filtram eficazmente os falsos sinais
  3. O comércio bidirecional aproveita plenamente as oportunidades de curto e longo prazos
  4. Negociação totalmente automatizada sem intervenção manual
  5. Lógica estratégica clara, fácil de compreender e modificar

Análise de riscos

  1. ADX tem atraso, pode perder pontos de virada da tendência
  2. Aumento do risco decorrente da negociação bidirecional, perdas maiores
  3. A definição incorreta dos parâmetros pode provocar uma troca excessiva
  4. Os dados dos backtests não podem representar o mercado real, existe um risco comercial real

Soluções:

  1. Confirmar sinais com outros indicadores
  2. Otimizar parâmetros, controlar a frequência do comércio
  3. Dimensão estrita da posição para gerir o tamanho da posição

Orientações de otimização

  1. Otimizar parâmetros ADX para melhorar a sensibilidade
  2. Adicionar outros indicadores aos sinais filtrados
  3. Aplicar aprendizado de máquina para otimizar parâmetros
  4. Usar estratégias avançadas de stop loss para controlar perdas
  5. Combinar com as previsões do modelo para sinais mais precisos

Conclusão

A estratégia de negociação ADX de duas direções é uma estratégia quantitativa muito prática. Identifica tendências usando o indicador ADX e captura oportunidades de negociação em ambas as direções. Enquanto isso, usa limiares de diferença para validar a eficácia do sinal. A estratégia tem uma lógica clara e simples que é fácil de modificar e otimizar.


/*backtest
start: 2023-12-01 00:00:00
end: 2023-12-31 23:59:59
period: 1h
basePeriod: 15m
exchanges: [{"eid":"Futures_Binance","currency":"BTC_USDT"}]
*/

// This source code is subject to the terms of the Mozilla Public License 2.0 at https://mozilla.org/MPL/2.0/
// © MAURYA_ALGO_TRADER

//@version=5
strategy("Monthly Performance", overlay=true)


len = input(14)
th = input(20)

TrueRange = math.max(math.max(high - low, math.abs(high - nz(close[1]))), math.abs(low - nz(close[1])))
DirectionalMovementPlus = high - nz(high[1]) > nz(low[1]) - low ? math.max(high - nz(high[1]), 0) : 0
DirectionalMovementMinus = nz(low[1]) - low > high - nz(high[1]) ? math.max(nz(low[1]) - low, 0) : 0

SmoothedTrueRange = 0.0
SmoothedTrueRange := nz(SmoothedTrueRange[1]) - nz(SmoothedTrueRange[1]) / len + TrueRange

SmoothedDirectionalMovementPlus = 0.0
SmoothedDirectionalMovementPlus := nz(SmoothedDirectionalMovementPlus[1]) - nz(SmoothedDirectionalMovementPlus[1]) / len + DirectionalMovementPlus

SmoothedDirectionalMovementMinus = 0.0
SmoothedDirectionalMovementMinus := nz(SmoothedDirectionalMovementMinus[1]) - nz(SmoothedDirectionalMovementMinus[1]) / len + DirectionalMovementMinus

DIPlus = SmoothedDirectionalMovementPlus / SmoothedTrueRange * 100
DIMinus = SmoothedDirectionalMovementMinus / SmoothedTrueRange * 100
DX = math.abs(DIPlus - DIMinus) / (DIPlus + DIMinus) * 100
ADX = ta.sma(DX, len)

// plot(DIPlus, color=color.new(color.green, 0), title='DI+')
// plot(DIMinus, color=color.new(color.red, 0), title='DI-')
// plot(ADX, color=color.new(color.white, 0), title='ADX')
// hline(th, color=color.black)


//diff_1 = math.abs(DIPlus - DIMinus)
diff_2 = math.abs(DIPlus-ADX)
diff_3 = math.abs(DIMinus - ADX)

long_diff = input(10, "Long Difference")
short_diff = input(10, "Short Difference")

buy_condition = diff_2 >=long_diff and diff_3 >=long_diff and (ADX < DIPlus and ADX > DIMinus)
sell_condition = diff_2 >=short_diff and diff_3 >=short_diff and (ADX > DIPlus and ADX < DIMinus)


if buy_condition
    strategy.entry("Long Entry", strategy.long, comment = "Long")
if sell_condition
    strategy.entry("Short Entry", strategy.short, comment = "Short")



// Copy below code to end of the desired strategy script
///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
//                                 monthly pnl performance  by Dr. Maurya @MAURYA_ALGO_TRADER                        //
///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
show_performance = input.bool(true, 'Show Monthly Monthly Performance ?', group='Monthly Performance')

dash_loc_mp = input("Bottom Right","Location"  ,options=["Top Right","Bottom Right","Top Left","Bottom Left", "Middle Right","Bottom Center"]  ,group='Monthly Performance', inline = "performance")

text_size_mp = input('Small',"Size"  ,options=["Tiny","Small","Normal","Large"]  ,group='Monthly Performance', inline = "performance")

bg_c = input.color( color.rgb(7, 226, 242, 38), "Background Color", group='Monthly Performance')

text_head_color = input.color( color.rgb(0,0,0), "Month/Year Heading Color", group='Monthly Performance')

tab_month_c = input.color( color.white, "Month PnL Data Color", group='Monthly Performance')

tab_year_c = input.color( color.rgb(0,0,0), "Year PnL Data Color", group='Monthly Performance')

border_c = input.color( color.white, "Table Border Color", group='Monthly Performance')



var table_position_mp = dash_loc_mp == 'Top Left' ? position.top_left :
  dash_loc_mp == 'Bottom Left' ? position.bottom_left :
  dash_loc_mp == 'Middle Right' ? position.middle_right :
  dash_loc_mp == 'Bottom Center' ? position.bottom_center :
  dash_loc_mp == 'Top Right' ? position.top_right : position.bottom_right
  
var table_text_size_mp = text_size_mp == 'Tiny' ? size.tiny :
  text_size_mp == 'Small' ? size.small :
  text_size_mp == 'Normal' ? size.normal : size.large

/////////////////

strategy.initial_capital = 50000

/////////////////////////////////////////////

// var bool new_month = na
new_month = ta.change(month) //> 0 ? true : false
newest_month = new_month and strategy.closedtrades >= 1

// profit
only_profit = strategy.netprofit
initial_balance = strategy.initial_capital

// month number
var int month_number = na
month_number := (ta.valuewhen(newest_month, month(time), 0)) //and month(time) > 1 ? (ta.valuewhen(newest_month, month(time), 0) - 1) :  12 //1 to 12

//month_year
var int month_time = na
month_time := ta.valuewhen(newest_month, time, 0) - 2419200000 


var int m_counter = 0
if newest_month
    m_counter += 1



// current month values
var bool new_year = na
new_year := ta.change(year)
curr_m_pnl = only_profit - nz(ta.valuewhen(newest_month, only_profit, 0), 0)
curr_m_number = newest_month ? ta.valuewhen(newest_month, month(time), 0) : month(time)
curr_y_pnl = (only_profit - nz(ta.valuewhen(new_year, only_profit, 0),0)) 



var float [] net_profit_array = array.new_float()
var int [] month_array = array.new_int()
var int [] month_time_array = array.new_int()


if newest_month
    array.push(net_profit_array, only_profit)
    array.push(month_array, month_number)
    array.push(month_time_array, month_time)



var float [] y_pnl_array = array.new_float()
var int [] y_number_array = array.new_int()
var int [] y_time_array = array.new_int()

newest_year = ta.change(year) and strategy.closedtrades >= 1
get_yearly_pnl = nz(ta.valuewhen(newest_year, strategy.netprofit, 0) - nz(ta.valuewhen(newest_year, strategy.netprofit, 1), 0), 0)
get_m_year = ta.valuewhen(newest_year, year(time), 1)
get_y_time = ta.valuewhen(newest_year, time, 0)

if newest_year
    array.push(y_pnl_array, get_yearly_pnl)
    array.push(y_number_array, get_m_year)
    array.push(y_time_array, get_y_time)
var float monthly_profit = na
var int column_month_number = na
var int row_month_time = na

 


var testTable = table.new(position = table_position_mp, columns = 14, rows = 40, bgcolor = bg_c, border_color = border_c, border_width = 1)
if barstate.islastconfirmedhistory and show_performance
    table.cell(table_id = testTable, column = 0, row = 0, text = "YEAR", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 1, row = 0, text = "JAN", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 2, row = 0, text = "FEB", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 3, row = 0, text = "MAR", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 4, row = 0, text = "APR", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 5, row = 0, text = "MAY", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 6, row = 0, text = "JUN", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 7, row = 0, text = "JUL", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 8, row = 0, text = "AUG", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 9, row = 0, text = "SEP", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 10, row = 0, text = "OCT", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 11, row = 0, text = "NOV", text_color = text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 12, row = 0, text = "DEC", text_color =text_head_color, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 13, row = 0, text = "YEAR P/L", text_color = text_head_color, text_size=table_text_size_mp)

    for i = 0 to (array.size(y_number_array) == 0 ? na : array.size(y_number_array) - 1)
        row_y = year(array.get(y_time_array, i)) - year(array.get(y_time_array, 0)) + 1
        table.cell(table_id = testTable, column = 13, row = row_y, text = str.tostring(array.get(y_pnl_array , i), "##.##") + '\n' + '(' + str.tostring(array.get(y_pnl_array , i)*100/initial_balance, "##.##") + ' %)', bgcolor = array.get(y_pnl_array , i) > 0 ? color.green : array.get(y_pnl_array , i) < 0 ? color.red : color.gray, text_color = tab_year_c, text_size=table_text_size_mp)
    curr_row_y = array.size(month_time_array) == 0 ? 1 : (year(array.get(month_time_array, array.size(month_time_array) - 1))) - (year(array.get(month_time_array, 0))) + 1
    table.cell(table_id = testTable, column = 13, row = curr_row_y, text = str.tostring(curr_y_pnl, "##.##") + '\n' + '(' + str.tostring(curr_y_pnl*100/initial_balance, "##.##") + ' %)', bgcolor = curr_y_pnl > 0 ? color.green : curr_y_pnl < 0 ? color.red : color.gray, text_color = tab_year_c, text_size=table_text_size_mp)
    

    for i = 0 to (array.size(net_profit_array) == 0 ? na : array.size(net_profit_array) - 1)
        monthly_profit := i > 0 ? ( array.get(net_profit_array, i) - array.get(net_profit_array, i - 1) ) : array.get(net_profit_array, i) 
        column_month_number := month(array.get(month_time_array, i)) 
        row_month_time :=((year(array.get(month_time_array, i))) - year(array.get(month_time_array, 0)) ) + 1 
        table.cell(table_id = testTable, column = column_month_number, row = row_month_time, text = str.tostring(monthly_profit, "##.##") + '\n' + '(' + str.tostring(monthly_profit*100/initial_balance, "##.##") + ' %)', bgcolor = monthly_profit > 0 ? color.green : monthly_profit < 0 ? color.red : color.gray, text_color = tab_month_c, text_size=table_text_size_mp)
        table.cell(table_id = testTable, column = 0, row =row_month_time, text = str.tostring(year(array.get(month_time_array, i)), "##.##"), text_color = text_head_color, text_size=table_text_size_mp)
       
    curr_row_m = array.size(month_time_array) == 0 ? 1 : (year(array.get(month_time_array, array.size(month_time_array) - 1))) - (year(array.get(month_time_array, 0))) + 1
    table.cell(table_id = testTable, column = curr_m_number, row = curr_row_m, text = str.tostring(curr_m_pnl, "##.##") + '\n' + '(' + str.tostring(curr_m_pnl*100/initial_balance, "##.##") + ' %)', bgcolor = curr_m_pnl > 0 ? color.green : curr_m_pnl < 0 ? color.red : color.gray, text_color = tab_month_c, text_size=table_text_size_mp)
    table.cell(table_id = testTable, column = 0, row =curr_row_m, text = str.tostring(year(time), "##.##"), text_color = text_head_color, text_size=table_text_size_mp)

//============================================================================================================================================================================

Mais.