Causal Network Representations in Factor Investing – Key Vocabulary

Flashcards summarizing essential vocabulary and concepts from the lecture on applying causal networks to factor investing.

Factor Investing

A systematic portfolio approach that targets specific characteristics (“factors”) shown to explain differences in asset returns.

Causal Discovery Algorithm

A data-driven method that seeks to uncover cause-and-effect relationships rather than mere correlations within observational data.

DYNOTEARS

A score-based causal discovery algorithm that extends NOTEARS to dynamic Bayesian networks, enabling identification of instantaneous and lagged causal links.

Directed Acyclic Graph (DAG)

A graph with directed edges and no cycles, commonly used to represent causal structures.

Dynamic Bayesian Network (DBN)

A Bayesian network that incorporates time-lagged variables, allowing modeling of sequential data.

Structural Vector Autoregressive (SVAR) Model

A multivariate time-series model capturing contemporaneous and lagged relationships among variables, often used to represent DBNs.

Score-Based Method

Causal structure learning approach that searches for the highest-scoring DAG according to a chosen criterion (e.g., BIC), avoiding strong faithfulness assumptions.

Constraint-Based Method

Causal learning technique that relies on conditional independence tests; consistency often requires the strong faithfulness assumption.

Strong Faithfulness Assumption

The requirement that all d-connected variables show non-zero associations above a threshold; often violated in financial data.

Peer Group Neutralization

Long–short portfolio construction that removes sector or peer-group exposures to isolate desired factor bets.

Global Industry Classification Standard (GICS)

A discretionary sector/industry taxonomy widely used by practitioners for grouping stocks.

Systematic Classification Scheme

A quantitative, data-driven method (e.g., clustering, network analysis) for forming stock peer groups.

Discretionary Classification Scheme

Expert-defined grouping of firms based on qualitative criteria, such as SIC, NAICS, or GICS codes.

Statistical Clustering (SC)

Grouping assets via distance measures (often return correlations) and clustering algorithms like hierarchical clustering.

Node2Vec

A graph-representation algorithm that learns vector embeddings of nodes by simulating biased random walks.

Eigenvector Centrality

A measure of a node’s importance that weights connections to highly connected neighbors more heavily.

Low Centrality Factor

A long–short strategy that buys peripheral (low-centrality) stocks and shorts highly central stocks.

Network Density

The overall connectedness of a graph, here proxied by average eigenvector centrality across stocks and used as a market-timing signal.

Market Timing Indicator

A variable intended to forecast future market returns, enabling allocation shifts between risky assets and cash.

Out-of-Sample R-Squared (R²_OS)

A measure comparing forecast accuracy of a predictive model to the historical average in unseen data.

Certainty Equivalent Return (CER) Gain

Annual fee an investor would pay for a forecast over the historical mean, reflecting economic value added.

Minimum Spanning Tree (MST)

A filtered correlation network connecting all nodes with the smallest possible total edge weight and no cycles.

Planar Maximally Filtered Graph (PMFG)

A correlation network retaining more information than MST while remaining planar to control complexity.

Minimum-Variance Portfolio

Portfolio that minimizes return variance for a given set of assets, often affected by asset centrality.

Sharpe Ratio

Risk-adjusted return metric calculated as mean excess return divided by standard deviation of returns.

Value-Weighted Portfolio

Portfolio whose asset weights are proportional to market capitalization or another value metric.

Quintile Long–Short Strategy

Portfolio that goes long the top 20 % of ranked assets and short the bottom 20 %, rebalanced periodically.

Book-to-Price Ratio

Accounting value divided by market price; core variable in the value factor.

Momentum (12–1M)

Past 12-month return excluding the most recent month; basis for the momentum factor.

Short-Term Reversal (1M Reversal)

Tendency for last month’s losers to outperform and winners to underperform next month.

Beta (60M)

Sensitivity of a stock’s returns to market returns, estimated over a 60-month window.

High-Minus-Low (HML)

Fama-French value factor: long high book-to-price stocks, short low book-to-price stocks.

Small-Minus-Big (SMB)

Fama-French size factor: long small-cap stocks, short large-cap stocks.

Up-Minus-Down (UMD)

Fama-French momentum factor capturing past-winner outperformance.

Robust-Minus-Weak (RMW)

Profitability factor: long high-profitability firms, short low-profitability firms.

Conservative-Minus-Aggressive (CMA)

Investment factor: long low-investment firms, short high-investment firms.

Betting-Against-Beta (BAB)

Factor that is long low-beta stocks and short high-beta stocks, aiming to exploit the low-beta anomaly.

Principal Component Analysis (PCA)

Dimensionality-reduction technique that transforms correlated variables into orthogonal components.

Herfindahl–Hirschman Index (HHI)

Concentration measure; in networks, used to quantify distribution of centrality or connections.

Node Centrality Measure

Any quantitative metric (degree, closeness, eigenvector, etc.) indicating a node’s prominence within a network.