Differences

This shows you the differences between two versions of the page.

--- backtest_w_python [2025/03/23 07:31] – bruno
+++ backtest_w_python [2025/03/27 17:02] (current) – [Sinon] bruno
@@ Line 1: / Line 1: @@
-=== Deathcross ===
+==== Deathcross ====
 <code python>
@@ Line 107: / Line 107: @@
 </code>
-=== Trend following RSI ===
+==== Trend following RSI ====
 <code python>
@@ Line 183: / Line 183: @@
 </code>
-[[https://support.kraken.com/hc/en-us/articles/360047124832]]
-=== Sinon ===
-Un vieux code de 2022 (si pas plus). Chrypowatch n'existe plus, c'est dire :-)
+==== Advanced trend following RSI ====
 <code python>
 # coding: utf-8
-import requests
 import numpy as np
 import pandas as pd
+import talib as ta
-import talib
-url = 'https://api.cryptowat.ch/markets/kraken/btceur/ohlc'
-ohlc = requests.get(url).json()['result'][str(12*60*60)]
-columns = ['time','open','high','low','close','volume','count']
-df = pd.DataFrame(ohlc, columns=columns).astype(float)
-df = df.iloc[-1000:]
-df['RSI'] = talib.RSI(df['close'], timeperiod=14)
+df = pd.read_csv('Téléchargements/btceur-2h.csv')
-#df['RSI'] = df.RSI.fillna(value=df.RSI.loc[14])
+#df = pd.read_csv('https://raw.githubusercontent.com/brulint/backtesting/main/btceur-2h.csv')
-df['long'] = talib.SMA(df.close, timeperiod=200)
-#df['long'] = df.long.fillna(value=df.long.loc[200])
-df['short'] = talib.SMA(df.close,timeperiod=14)
-#df['short'] = df.short.fillna(value=df.short.loc[14])
-df['trend'] = df.long < df.short
-# signal
-df['sig_in'] = (df.RSI > 60) & df.trend
-df['sig_out'] = (df.RSI < 40)# | 1-df.trend
-#df['sig_in'] = (df.RSI.shift() < 70) & (df.RSI > 70)
-#df['sig_out'] = (df.RSI.shift() > 30) & (df.RSI < 30)
-#df['sig_in'] = (df.RSI.shift() < 25) & (df.RSI > 25)
-#df['stoploss'] = df.low.rolling(5).min().where(df.sig_in==1).ffill()
-#df['sig_out'] = ((df.RSI.shift() > 75) & (df.RSI < 75)) | ((df.RSI.shift()>25) & (df.RSI<25)) | (df.close < df.stoploss)
-#df['signal'] = df.sig_in.where(df.sig_in).fillna(1-df.sig_out.where(df.sig_out)).ffill()
+# strategy
-#df['sig_out'].loc[0] = True
+RSI = ta.RSI(df.close, timeperiod = 14)
-#df['signal'] = (1-df.sig_out.where(df.sig_out)).fillna(df.sig_in.where(df.sig_in)).ffill()# * df.trend
+signal_long_buy = RSI > 80
+signal_long_sell = RSI < 50
+signal_short_sell = RSI < 20
+signal_short_buy = RSI > 50
-df['sig_0'] = df.sig_in.astype(int) - df.sig_out.astype(int)
+# position
-df['sig_1'] = df.sig_0.where(df.sig_0!=0).ffill()
+signal_long = signal_long_buy.astype(int) - signal_long_sell.astype(int)
-df['signal'] = df.sig_1 > 0
+signal_short = signal_short_sell.astype(int) - signal_short_buy.astype(int)
-# Rendements
+position_long = signal_long.where(signal_long != 0).ffill() > 0
-df['close'] = df.close.replace(to_replace=0, method='ffill')
+position_short = signal_short.where(signal_short != 0).ffill() > 0
-df['r_0'] = df.close / df.close.shift()
+position = position_long.astype(int) - position_short.astype(int)
-df['r_strat'] = np.where(df.signal.shift(), df.r_0, 1)
-df['r_fee'] = np.where(df.signal.shift() + df.signal == 1, 1-0.0025, 1)
+# returns
-# tronquage datafame
+returns_hodl = np.log(df.close / df.close.shift())
-#df = df.iloc[-700:]
+returns_strat = returns_hodl * position.shift()
-# Rendement cumulé
+returns_netto = returns_strat - 0.0025 * (position != position.shift())
-df['R_net'] = (df.r_strat * df.r_fee).cumprod()
-# Graphiques
+# graphic
 from bokeh.plotting import figure,show
 from bokeh.layouts import column,row
-p1 = figure(height=300,width=800)
+from bokeh.models import DatetimeTickFormatter
-p1.line(df.time,df.close)
-#p1.line(df.time,df.long,color='green')
+df['date'] = pd.to_datetime(df.time, unit='s')
-#p1.line(df.time,df.short,color='red')
+xformatter = DatetimeTickFormatter(hours="%H:%M", days="%d/%m", months="%m/%Y", years="%Y")
-p2 =  figure(height=100,width=800,x_range=p1.x_range)
-p2.line(df.time,df.RSI)
+p1 = figure(height=325, width=800)
-#p3_0 = figure(height=100,width=800,x_range=p1.x_range)
+p1.xaxis[0].formatter = xformatter
-#p3_0.line(df.time,df.trend)
+p1.line(df.date, df.close)
-p3_1 = figure(height=100,width=800,x_range=p1.x_range)
-p3_1.line(df.time,df.sig_in,color='green')
+p11 = figure(height=125, width=800, x_range=p1.x_range)
-p3_2 = figure(height=100,width=800,x_range=p1.x_range)
+p11.xaxis[0].formatter = xformatter
-p3_2.line(df.time,df.sig_out,color='red')
+p11.line(df.date, RSI)
-p3_3 = figure(height=100,width=800,x_range=p1.x_range)
+p11.line(df.date, 70, color='green')
-p3_3.line(df.time,df.sig_0)
+p11.line(df.date, 50, color='red')
-p3_3_2 = figure(height=100,width=800,x_range=p1.x_range)
+p11.line(df.date, 30, color='green')
-p3_3_2.line(df.time,df.sig_1)
-p3_4 = figure(height=100,width=800,x_range=p1.x_range)
+p2 = figure(height=125, width=800, x_range=p1.x_range)
-p3_4.line(df.time,df.signal)
+p2.xaxis[0].formatter = xformatter
-p4 = figure(height=150,width=800,x_range=p1.x_range)
+p2.line(df.date, position)
-p4.line(df.time,df.r_0,color='lightgray')
-p4.line(df.time,df.r_strat)
+p3 = figure(height=325, width=800, x_range=p1.x_range)
-p4.line(df.time,df.r_fee,color='red')
+p3.xaxis[0].formatter = xformatter
-p5 = figure(height=300,width=800,x_range=p1.x_range)
+p3.line(df.date, df.close, color='gray')
-p5.line(df.time,df.r_0.cumprod(),color='lightgray')
+position_long_in = df.close.where((position == 1) & (position.shift() != 1))
-p5.line(df.time,df.r_strat.cumprod())
+position_long_out = df.close.where((position != 1) & (position.shift() == 1))
-p5.line(df.time,df.R_net,color='red')
+position_short_in = df.close.where((position == -1) & (position.shift() != -1))
-layout = column(p1,p2,p3_1,p3_2,p3_3,p3_3_2,p3_4,p4,p5)
+position_short_out = df.close.where((position != -1) & (position.shift() == -1))
+p3.triangle(df.date, position_long_in, color='cyan', size=7)
+p3.inverted_triangle(df.date, position_long_out, color='blue', size=7)
+p3.inverted_triangle(df.date, position_short_in, color='orange', size=7)
+p3.triangle(df.date, position_short_out, color='red', size=7)
+p3.line(df.date, np.where(position == 1, df.close, np.nan), color='green')
+p3.line(df.date, np.where(position == -1, df.close, np.nan), color='red')
+p4 = figure(height=150, width=800, x_range=p1.x_range)
+p4.xaxis[0].formatter = xformatter
+p4.line(df['date'], returns_hodl, color='lightgray')
+p4.line(df['date'], returns_strat)
+p5 = figure(height=325, width=800, x_range=p1.x_range)
+p5.xaxis[0].formatter = xformatter
+p5.line(df['date'], np.exp(returns_hodl.cumsum()), color='lightgray')
+p5.line(df['date'], np.exp(returns_strat.cumsum()))
+p5.line(df['date'], np.exp(returns_netto.cumsum()), color='red')
+layout = column(p1, p11, p2, p3, p4, p5)
 show(layout)
 </code>