Investing & Quant ML Glossary
Clear definitions of the investing, quantitative, and machine-learning concepts behind hedgewing.ai.
- Walk-Forward Backtesting — An out-of-sample model validation method that trains on a rolling historical window, predicts the next unseen period, then slides forward — preventing look-ahead bias.
- Look-Ahead Bias — A backtesting flaw where a model is accidentally fed information that would not have been known at the time of the prediction, inflating apparent performance.
- Overfitting — When a model learns noise specific to its training data instead of the underlying signal, producing strong in-sample results that collapse on new data.
- Ensemble Model — A model that combines predictions from several independent models, typically producing more robust forecasts than any single model alone.
- LSTM (Long Short-Term Memory) — A recurrent neural network architecture designed to learn long-range dependencies in sequential data such as price history.
- Transformer Model — A neural architecture based on self-attention that processes an entire sequence in parallel, capturing relationships between any two time steps directly.
- Sharpe Ratio — A measure of risk-adjusted return: excess return over the risk-free rate divided by the standard deviation of returns.
- Sortino Ratio — A variant of the Sharpe ratio that measures return only against downside volatility, ignoring upside swings.
- Value at Risk (VaR) — An estimate of the maximum expected loss over a given horizon at a chosen confidence level, e.g. 95% or 99%.
- Beta — A measure of how much a stock moves relative to the broad market — beta of 1 moves with the market, above 1 is more volatile, below 1 less.
- Factor Analysis — Decomposing a stock's returns into exposures to systematic factors such as market, size, value, profitability, and investment.
- Alpha — The portion of return not explained by exposure to market or other systematic factors — a measure of genuine skill or edge.
- Calibrated Confidence — A confidence score is calibrated when its stated probability matches the observed frequency — 70% confident predictions should be right about 70% of the time.
- RSI (Relative Strength Index) — A momentum oscillator from 0 to 100 that flags overbought (above 70) and oversold (below 30) conditions.
- MACD (Moving Average Convergence Divergence) — A trend-following momentum indicator built from the difference between two exponential moving averages and a signal line.
- Bollinger Bands — Volatility bands plotted a number of standard deviations above and below a moving average, widening and narrowing as volatility changes.
- Neural Network — A machine-learning model made of interconnected layers of simple computing units (neurons) that learn to map inputs to outputs by adjusting weighted connections during training.
- Deep Learning — A branch of machine learning that uses neural networks with many stacked layers to automatically learn hierarchical patterns directly from raw or lightly processed data.
- Recurrent Neural Network (RNN) — A neural network designed for sequences that processes data one step at a time while carrying forward a hidden state, letting earlier inputs influence later outputs.
- Gated Recurrent Unit (GRU) — A streamlined recurrent neural network cell that uses update and reset gates to retain long-range information, achieving much of the LSTM's power with fewer parameters.
- Temporal Convolutional Network (TCN) — A sequence model that applies stacked, dilated, causal convolutions to time-series data, capturing long histories in parallel without the step-by-step recurrence of an RNN.
- Attention Mechanism — A neural-network technique that lets a model dynamically weight which parts of its input matter most for each output, enabling it to focus on the most relevant information in a sequence.
- Feature Engineering — Feature engineering is the process of transforming raw data into informative input variables that a machine-learning model can use to make better predictions.
- Supervised Learning — Supervised learning is a machine-learning approach where a model learns from labeled examples, mapping known inputs to known outputs so it can predict outcomes for new data.
- Time-Series Forecasting — Time-series forecasting is the task of predicting future values of a sequence from its past, accounting for trends, seasonality, and the order-dependence of observations.
- Regularization — Regularization is a set of techniques that constrain a model's complexity during training to prevent it from memorizing noise and to improve how well it generalizes to new data.
- Cross-Validation — Cross-validation is a model-evaluation method that repeatedly trains on part of the data and tests on the rest to estimate how well a model will generalize to unseen data.
- Hyperparameter Tuning — Hyperparameter tuning is the process of searching for the configuration settings of a model—such as learning rate or depth—that produce the best generalization performance.
- Gradient Descent — Gradient descent is the optimization algorithm that trains most machine-learning models by repeatedly nudging parameters in the direction that most reduces prediction error.
- Stacking (Meta-Learner Ensemble) — Stacking is an ensemble method that combines several base models by training a meta-learner to weigh their predictions, typically outperforming any single model.
- Information Ratio — The information ratio measures a portfolio's excess return over a benchmark per unit of tracking error, gauging how consistently a manager beats that benchmark.
- Maximum Drawdown — Maximum drawdown is the largest peak-to-trough percentage decline in a portfolio's value before a new peak is reached, capturing the worst loss an investor would have endured.
- Calmar Ratio — The Calmar ratio divides an investment's annualized return by its maximum drawdown, measuring return earned per unit of worst-case loss.
- Risk-Adjusted Return — Risk-adjusted return measures investment performance relative to the amount of risk taken, so strategies with different volatility can be compared fairly.
- Standard Deviation — Standard deviation measures how widely a set of returns spreads around its average, serving as the most common statistical proxy for volatility and risk.
- Correlation — Correlation measures how closely two assets' returns move together, ranging from +1 (perfectly in sync) through 0 (unrelated) to -1 (perfectly opposite).
- Volatility — Volatility measures how much an asset's price fluctuates over time, usually expressed as the annualized standard deviation of returns; higher volatility means larger and less predictable price swings.
- Implied Volatility — Implied volatility is the market's forward-looking estimate of an asset's future price variability, reverse-engineered from current option prices rather than measured from past returns.
- Capital Asset Pricing Model (CAPM) — CAPM is a foundational model that estimates an asset's expected return as the risk-free rate plus its beta times the market's risk premium, linking expected reward to non-diversifiable risk.
- Efficient Market Hypothesis (EMH) — The Efficient Market Hypothesis holds that asset prices already reflect available information, making it difficult to consistently earn excess returns without taking on additional risk.
- Monte Carlo Simulation — Monte Carlo simulation estimates the range of possible outcomes by running thousands of randomized scenarios, producing a distribution of results rather than a single point forecast.
- Hierarchical Risk Parity (HRP) — Hierarchical Risk Parity is a portfolio construction method that groups assets by similarity using machine learning, then allocates risk across the resulting hierarchy without inverting an unstable covariance matrix.
- Mean Reversion — Mean reversion is the tendency of an asset's price or other financial variable to drift back toward its long-run average after deviating from it, forming the basis of contrarian trading strategies.
- Momentum (Investing) — Momentum is the empirical tendency of assets that have performed well (or poorly) over the recent past to continue performing in the same direction over the near term.
- Algorithmic Trading — Algorithmic trading is the use of computer programs to automatically generate, route, and execute orders according to predefined rules or models, with little or no manual intervention.
- Quantitative Analysis — Quantitative analysis is the use of mathematical, statistical, and computational methods to model financial markets, value securities, and make systematic, data-driven investment and risk decisions.
- Diversification — Diversification is the practice of spreading capital across assets whose returns are imperfectly correlated, reducing portfolio risk for a given level of expected return.
- Moving Average — A moving average smooths price data by continuously averaging the most recent N periods, producing a single trend line that filters out short-term noise.
- Exponential Moving Average (EMA) — An exponential moving average is a moving average that weights recent prices more heavily, so it reacts faster to new information than a simple moving average of the same length.
- Average True Range (ATR) — Average True Range measures market volatility by averaging the daily true price range over a period, capturing how much an asset typically moves regardless of direction.
- Stochastic Oscillator — The stochastic oscillator is a momentum indicator that measures where the current close sits within its recent high-low range, signaling potential overbought or oversold conditions.
- Support and Resistance — Support and resistance are price levels where buying or selling pressure has repeatedly halted a move, acting as floors and ceilings that traders watch for reversals or breakouts.
- Volume-Weighted Average Price (VWAP) — VWAP is the average price an asset traded at over a period, weighted by volume, serving as a benchmark for execution quality and intraday fair value.
- Price-to-Earnings (P/E) Ratio — The price-to-earnings ratio divides a stock's share price by its earnings per share, showing how much investors pay for each dollar of company profit.
- Earnings Per Share (EPS) — Earnings per share is a company's net profit allocated to each outstanding share of common stock, a core measure of per-share profitability.
- Market Capitalization — Market capitalization is the total market value of a company's outstanding shares, calculated by multiplying share price by the number of shares outstanding.
- Technical Analysis — Technical analysis studies historical price and volume data, using charts and indicators, to forecast future price movements based on patterns and market behavior.
- Fundamental Analysis — Fundamental analysis estimates a security's intrinsic value by examining financial statements, business quality, and economic conditions to judge whether it is mispriced.
- Sentiment Analysis (Markets) — Market sentiment analysis measures the collective mood of investors toward an asset or market, often using news, social media, and survey data to gauge bullish or bearish bias.
- Brier Score — The Brier score measures the accuracy of probabilistic forecasts as the mean squared difference between predicted probabilities and actual outcomes; lower is better, and 0 is perfect.
- Reliability Diagram (Calibration Curve) — A reliability diagram plots predicted probability against observed frequency; a well-calibrated model's curve lies on the 45-degree line.
- Regime Detection — Regime detection identifies which of several distinct market states (for example low-volatility trending versus high-volatility choppy) currently prevails, because models and strategies behave differently in each.
- Volatility Z-Score — A volatility z-score expresses current volatility as the number of standard deviations it sits above or below its own historical average, flagging unusually calm or stressed conditions.
- Fama-French Five-Factor Model — The Fama-French five-factor model explains stock returns using market, size, value, profitability, and investment factors, extending the earlier three-factor model.
- Feature Importance — Feature importance measures how much each input variable contributes to a model's predictions, helping explain which signals actually drive a forecast.
- Survivorship Bias — Survivorship bias is the distortion that arises when a backtest or study includes only securities that still exist today, excluding the delisted, bankrupt, or acquired ones that failed along the way.
- Point-in-Time Data — Point-in-time data records what was actually known on each historical date, including later-restated figures as originally reported, so that backtests do not use information from the future.
- Ensemble Disagreement — Ensemble disagreement is the spread between the predictions of the individual models in an ensemble; high disagreement is a signal of uncertainty even when the combined forecast looks decisive.
- Hit Rate (Directional Accuracy) — Hit rate, or directional accuracy, is the percentage of forecasts that correctly predicted the direction of a move; it ignores the size of moves and the confidence attached to each call.
- Expected Value — Expected value is the probability-weighted average outcome of a decision; a trade with a positive expected value is favorable over many repetitions even if any single instance loses.
- Position Sizing — Position sizing is the decision of how much capital to commit to a trade, typically governed by a fixed risk budget per position rather than by conviction alone.
- Kelly Criterion — The Kelly criterion is a formula for the bet size that maximises long-run growth of capital given the probability of winning and the payoff ratio; in practice investors use a fraction of it.
- Conditional Value at Risk (Expected Shortfall) — Conditional value at risk, or expected shortfall, is the average loss in the worst cases beyond the VaR threshold, answering 'how bad is it when it is bad' rather than 'how often is it bad'.
- Data Leakage — Data leakage occurs when information from outside the training period, or from the target itself, contaminates a model's inputs, producing results that cannot be reproduced in live use.
- Out-of-Sample Testing — Out-of-sample testing evaluates a model on data it never saw during training or tuning, which is the only way to estimate how it will perform on genuinely new data.
- Tracking Error — Tracking error is the standard deviation of the difference between a portfolio's returns and its benchmark's returns, measuring how closely the portfolio follows the benchmark.
- Downside Deviation — Downside deviation measures the volatility of returns that fall below a chosen threshold, such as zero or a minimum acceptable return, ignoring upside volatility.
- Fat Tails (Kurtosis) — Fat tails describe return distributions in which extreme moves occur far more often than a normal distribution predicts; kurtosis is the statistic that measures this.
- Cointegration — Cointegration is a statistical relationship in which two or more non-stationary price series move together over the long run so that a combination of them is stationary; it is the basis of pairs trading and distinct from correlation.
- Log Returns — Log returns are the natural logarithm of the ratio of consecutive prices; they add across time, are approximately symmetric, and are the standard return definition in quantitative modeling.
- Post-Earnings Announcement Drift — Post-earnings announcement drift (PEAD) is the documented tendency of stock prices to continue moving in the direction of an earnings surprise for weeks after the announcement.