Machine Learning Techniques for Trading › Complete Worked Example

# Complete ML pipeline for return prediction
import warnings

import numpy as np
import pandas as pd

warnings.filterwarnings("ignore")

# Generate more realistic synthetic data
np.random.seed(42)
n_days = 1000

# Simulate price with some predictable patterns
price = [100]
for i in range(n_days - 1):
    momentum_effect = 0.001 * (
        price[-1] / np.mean(price[-min(20, len(price)) :]) - 1
    )
    mean_rev_effect = -0.0005 * (
        price[-1] / np.mean(price[-min(50, len(price)) :]) - 1
    )
    noise = 0.02 * np.random.randn()
    ret = momentum_effect + mean_rev_effect + 0.0003 + noise
    price.append(price[-1] * (1 + ret))

price = np.array(price)
volume = np.random.lognormal(15, 0.5, n_days)

# Create DataFrame and features
data = pd.DataFrame({"close": price, "volume": volume})
data["return"] = data["close"].pct_change()

# Feature engineering
data["momentum_5"] = data["close"].pct_change(5)
data["momentum_10"] = data["close"].pct_change(10)
data["momentum_20"] = data["close"].pct_change(20)
data["vol_10"] = data["return"].rolling(10).std()
data["vol_20"] = data["return"].rolling(20).std()
data["vol_ratio"] = data["vol_10"] / data["vol_20"]
data["ma_ratio"] = data["close"] / data["close"].rolling(20).mean()
data["zscore"] = (data["close"] - data["close"].rolling(20).mean()) / data[
    "close"
].rolling(20).std()
data["volume_ratio"] = data["volume"] / data["volume"].rolling(10).mean()

# Target: next-day direction
data["target"] = (data["return"].shift(-1) > 0).astype(int)
data = data.dropna()

feature_cols = [
    "momentum_5",
    "momentum_10",
    "momentum_20",
    "vol_10",
    "vol_20",
    "vol_ratio",
    "ma_ratio",
    "zscore",
    "volume_ratio",
]

from sklearn.base import clone

# Prepare data
X = data[feature_cols].values
y = data["target"].values

# Scale features
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

# Time series cross-validation with multiple models
tscv = TimeSeriesSplit(n_splits=5)

models = {
    "Logistic Regression": LogisticRegression(C=0.1, max_iter=1000),
    "Random Forest": RandomForestClassifier(
        n_estimators=100, max_depth=4, random_state=42
    ),
    "Gradient Boosting": GradientBoostingClassifier(
        n_estimators=100, learning_rate=0.1, max_depth=3, random_state=42
    ),
}

results = {name: {"accuracy": [], "auc": []} for name in models}

for train_idx, test_idx in tscv.split(X_scaled):
    X_train, X_test = X_scaled[train_idx], X_scaled[test_idx]
    y_train, y_test = y[train_idx], y[test_idx]

    for name, model in models.items():
        # Clone model for fresh training
        m = clone(model)

        m.fit(X_train, y_train)
        y_pred = m.predict(X_test)
        y_prob = m.predict_proba(X_test)[:, 1]

        results[name]["accuracy"].append(accuracy_score(y_test, y_pred))
        results[name]["auc"].append(roc_auc_score(y_test, y_prob))

# Placeholder cell to ensure proper notebook structure
pass