create_strategy_with_rapids.py

import cudf
import cuml
import pyfolio as pf
import numpy as np
import pandas as pd
import yfinance as yf
import matplotlib.pyplot as plt
from hmmlearn import hmm
from sklearn.utils import check_random_state

import warnings
warnings.filterwarnings("ignore")

# Obtaining the EURUSD data
data = yf.download('EURUSD=X', start='2000-01-01', end='2024-01-13', auto_adjust=True)

# Initial index number to start forecasting and trading
initial_iloc = data.index.get_loc(data.loc['2019-01'].index[0])

# Drop the volume column
data.drop('Volume', axis=1, inplace=True)

# Set the columns names
columns_names = data.columns.tolist()

# Create the prediction feature
data['y'] = np.where(data['Close'].pct_change().shift(-1) > 0,1, np.where(data['Close'].pct_change().shift(-1) < 0, -1, np.nan))

# Create the input features
for j in range(1,5):
    data[[f'{column}_{j}' for column in columns_names]] = data[columns_names].pct_change().shift(j)

# Compute the Buy-and-Hold returns
data['returns'] = data['Close'].pct_change()

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

# Function to create signal based on next-day return
def create_signal(data, window_size=initial_iloc):
    # Copy the dataframe
    data = data.copy()

    # Create the signal column
    data['signal'] = 0.0
    # Create the train signal column
    data['train_signal'] = 0.0
    # Create the leverage column
    data['leverage'] = 0.0
    # Loop to create the signal for each day
    for i in range(len(data) - window_size):
        # Get the input features
        data_sample = data.iloc[i:(i+window_size)]
        # Get the input features to train the model
        X_train = data_sample[[f'{column}_{j}' for column in columns_names for j in range(1,5)]].iloc[:-2]
        # Get the prediction feature
        y = data_sample['y'].iloc[:-2]
        # Create the random forest classifier object
        rf_classifier = cuml.ensemble.RandomForestClassifier(n_estimators=10, n_streams=1, random_state=100)
        # Fit the model
        rf_classifier.fit(X_train, y)
        # Get the input features to get the signal prediction
        X_test = data_sample[[f'{column}_{j}' for column in columns_names for j in range(1,5)]].iloc[-1]
        # Get the signal prediction
        data['signal'].iloc[(i+window_size)] = rf_classifier.predict(X_test).values[0]

    return data

# Creating the signal using the historical stock data
data = create_signal(data)

print(data['signal'].value_counts())