quantra-go-algo

# Obtain the prediction feature with the above function
data['y'] = triple_barrier_method(data, holding_period=10, upper_lower_multipliers=[2, 1])

quantra-go-algo

# Import the Apple dataframe
data = yf.download('AAPL', start='1990-01-01', end='2024-04-04', auto_adjust=True)

# Compute the daily volatility
data['vol'] = get_Daily_Volatility(data)

# Drop the rows that have NaN values
data.dropna(inplace=True)

quantra-go-algo

def triple_barrier_method(df, holding_period=10, upper_lower_multipliers=[2, 1]):
    # Set the close price as a cupy array
    close = cp.array(df['Close'].values, dtype=cp.float64)
    # Set the high price as a cupy array
    high = cp.array(df['High'].values, dtype=cp.float64)
    # Set the low price as a cupy array
    low = cp.array(df['Low'].values, dtype=cp.float64)
    # Set the daily volatility as a cupy array
    daily_volatility = cp.array(df['vol'].values, dtype=cp.float64)
    # Set the barriers empty array as a cupy array

quantra-go-algo

@cuda.jit
def triple_barrier_method_cuda(close, high, low, daily_volatility, upper_lower_multipliers, holding_period, out):

    # Set the number days passed to zero
    days_passed = 0
    # Set the vertical barrier initial value to NaN
    vert_barrier = math.nan
    # Set the top barrier initial value to NaN
    top_barrier = math.nan
    # Set the bottom barrier initial value to NaN

quantra-go-algo

def get_Daily_Volatility(df,span0=20):
    # simple percentage returns
    df0=df['Close'].pct_change()
    # 20 days, a month EWM's std as boundary
    df0=df0.ewm(span = span0, adjust = False).std()
    # Round the column values to six decimals
    df0 = df0.round(6)
    return df0

quantra-go-algo

def dropLabels(events,minPct=.05):
    # apply weights, drop labels with insufficient examples
    while True:
        # Count the number of observations the prediction feature has for each label
        df0=events['y'].value_counts(normalize=True)
        # If the label with minimum number of observations is lower than the minPct threshold 
        # or the number of labels in the prediction features is 2, then break the while loop
        if (df0.min()>minPct) or (df0.shape[0]<3):break
        # Drop the label with minimum number of observations
        events = events[events['y']!=df0.index[df0.argmin()]]

quantra-go-algo

import numpy as np
from numba import cuda
import math
import cupy as cp
import cudf
import yfinance as yf

quantra-go-algo

# Set the figure size
plt.figure(figsize=(15,7))

# Plot both the benchmark and strategy returns
plt.plot(df_results.index, df_results['Benchmark'], label = "Benchmark Cumulative Returns")
plt.plot(df_results.index, df_results['Stra_cum_returns'], label = "Strategy Returns", color='g')

# Set the title of the graph
plt.title('Benchmark and Strategy Cumulative Returns using a 1-year span', fontsize=16)

quantra-go-algo

# Create the signal
df_results['signal'] = np.where(df_results['forecast']>=0.0,1,-1)  

# Create the Buy and Hold returns
df_results['returns'] = np.log(df_results['Close']/df_results['Close'].shift(1))

# Create the Buy and Hold cumulative returns
df_results['Benchmark'] = df_results['returns'].cumsum()

# Create the strategy returns

quantra-go-algo

# Subset the data
plot_data = data.iloc[initial_iloc:]

# Buy and hold strategy cumulative returns
plot_data['buy_and_hold_cum_returns'] = (1+plot_data['returns']).cumprod()

# Strategy returns
plot_data['strategy_returns'] = plot_data['returns'] * plot_data['signal']

# Strategy cumulative returns