- ML3T - Machine Learning for Trading

1 Machine Learning

Overfitting

When a model learns noise in the training data rather than the underlying pattern, resulting in poor generalization to new data.

Cross-Validation

A technique for assessing model performance by partitioning data into training and validation sets multiple times.

Walk-Forward Validation

Time-series cross-validation that respects temporal order, training on past data and validating on future data.

Feature Importance

A measure of how much each input variable contributes to model predictions.

Regularization

Techniques that constrain model complexity to prevent overfitting (e.g., L1, L2, dropout).

Alpha

Excess return above a benchmark, attributed to the skill of the strategy or manager.

Sharpe Ratio

Risk-adjusted return metric: (Return - Risk-free rate) / Standard deviation.

Drawdown

Peak-to-trough decline in portfolio value, measuring downside risk.

Slippage

The difference between expected and actual execution price due to market impact and timing.

Point-in-Time

Data aligned to when it was actually available, preventing lookahead bias.

p-value

Probability of observing results as extreme as the sample, assuming the null hypothesis is true.

Multiple Testing

The problem of inflated false positive rates when testing many hypotheses simultaneously.

Deflated Sharpe Ratio

Sharpe ratio adjusted for the number of strategies tested, accounting for selection bias.

Information Coefficient (IC)

Correlation between predicted and actual returns, measuring signal quality.

Lookahead Bias

Using future information in model training or backtesting that wouldn't be available in real trading.

Survivorship Bias

Bias from excluding failed or delisted securities from historical analysis.

Feature Engineering

Creating predictive inputs from raw data, critical for ML model performance.

Data Leakage

When information from outside the training set improperly influences model building.