{
  "id": "8ec9c589-c3b4-561c-b9f4-6850e255ad13",
  "slug": "eigenvalue-decomposition",
  "term": "Eigenvalue Decomposition",
  "aliases": [],
  "category": "Financial Mathematics",
  "category_slug": "financial-mathematics",
  "difficulty": "advanced",
  "definition": "Eigenvalue decomposition is a linear algebra technique that factorizes a square matrix into its eigenvectors (directions) and eigenvalues (scaling factors), representing the matrix's transformation in terms of its principal axes. In finance, it is central to Principal Component Analysis (PCA) of covariance matrices, enabling dimensionality reduction of risk factor spaces, yield curve analysis, and correlation structure decomposition.",
  "key_takeaways": [
    "For a matrix A, eigenvalue decomposition gives A = QΛQ⁻¹, where Q is the matrix of eigenvectors and Λ is a diagonal matrix of eigenvalues.",
    "Applied to a covariance matrix (symmetric, positive semi-definite), eigenvectors represent uncorrelated risk factors (principal components) and eigenvalues represent their explained variance.",
    "The first principal component of a yield curve covariance matrix typically explains 70–90% of yield variance (parallel shift); the second explains twist; the third explains curvature.",
    "Eigenvalue analysis is used to detect near-multicollinearity (near-zero eigenvalues) in factor models and to regularize covariance matrices for portfolio optimization.",
    "Positive semi-definiteness of a covariance matrix is verified by confirming all eigenvalues are non-negative; negative eigenvalues indicate an invalid (numerically corrupted) covariance matrix."
  ],
  "detailed_explanation": "Eigenvalue decomposition (EVD, also called spectral decomposition) is a fundamental operation in linear algebra with pervasive applications across quantitative finance, risk management, and portfolio optimization. For a square matrix A, the eigenvalue decomposition finds scalars λ (eigenvalues) and corresponding non-zero vectors v (eigenvectors) satisfying: Av = λv. Geometrically, eigenvectors represent the special directions that the linear transformation A scales rather than rotates, and eigenvalues are the corresponding scaling factors.\n\nFor symmetric matrices (such as covariance and correlation matrices), the eigenvalue decomposition produces an orthogonal factorization: Σ = QΛQᵀ, where the columns of Q are orthonormal eigenvectors and Λ is a diagonal matrix with eigenvalues λ₁ ≥ λ₂ ≥ ... ≥ λₙ. Since covariance matrices are positive semi-definite (all eigenvalues ≥ 0), this decomposition is always valid and interpretable.\n\nThe most prominent financial application is Principal Component Analysis (PCA) of yield curves. When EVD is applied to the covariance matrix of daily yield changes across multiple tenors (2Y, 5Y, 10Y, 30Y), the first eigenvector (corresponding to the largest eigenvalue) represents the most important source of yield variation across tenors—typically a near-uniform loading (the 'parallel shift' factor). The second eigenvector captures the next most important orthogonal variation—the yield curve 'twist' (short rates moving opposite to long rates). The third captures 'curvature' (short and long rates moving in the same direction, opposite to intermediate rates). Together, these three components typically explain 95%+ of yield curve variance.\n\nFor equity portfolio management, PCA of a large covariance matrix (e.g., 500 stocks) is used to identify the dominant systematic risk factors in a universe. The first few principal components typically correspond to broad market exposure, sector effects, and factor tilts—providing a data-driven factor decomposition that complements model-based approaches like Barra or Axioma. Hedge funds can use PCA-derived factors to construct market-neutral portfolios by ensuring near-zero loading on the first 3–5 principal components.\n\nCovariance matrix regularization using eigenvalue decomposition is essential in portfolio optimization with many assets (where the sample covariance matrix is often poorly conditioned or singular). The Ledoit-Wolf shrinkage estimator is fundamentally an approach to adjusting eigenvalues toward their grand mean, producing a better-conditioned covariance matrix. Alternatively, setting small eigenvalues (below a threshold) to a minimum value (or to zero for the corresponding principal components) truncates noise from the covariance estimate, improving out-of-sample portfolio performance relative to using the raw sample covariance matrix.",
  "example": "A fixed income hedge fund applies PCA to the daily changes in U.S. Treasury yields across 8 tenors (3M, 6M, 1Y, 2Y, 5Y, 7Y, 10Y, 30Y) using 5 years of daily data. The eigenvalue decomposition of the 8×8 yield covariance matrix produces eigenvalues: λ₁ = 45.2, λ₂ = 8.3, λ₃ = 2.1, λ₄–λ₈ < 1.0 (summing to 3.0). Explained variance: PC1 = 45.2/(45.2+8.3+2.1+3.0) = 77.5%, PC2 = 14.2%, PC3 = 3.6%. The first eigenvector has near-equal loadings across all tenors (parallel shift); the second has positive loadings at short tenors and negative at long (twist); the third has positive loadings at short and long with negative in the middle (curvature). A yield curve trader wanting to bet on a flattening (short 2Y yield rising relative to 10Y) can construct a DV01-neutral trade that has zero PC1 exposure (no directional rate risk) but positive PC2 (twist) exposure, using the PCA decomposition to cleanly separate the desired risk from unwanted directional exposure.",
  "formula": "Σ = QΛQᵀ where Σv = λv; Explained Variance of PCₖ = λₖ / Σλᵢ",
  "formula_latex": null,
  "interactive_type": null,
  "calculator_id": null,
  "related_terms": [
    "copula",
    "correlation",
    "correlation-vs-causation",
    "covariance",
    "covariance-matrix",
    "discount-rate",
    "dv01",
    "equity",
    "hedge-fund",
    "jensens-inequality",
    "ledoit-wolf-shrinkage",
    "modified-internal-rate-of-return",
    "portfolio-optimization",
    "principal-component-analysis",
    "shrinkage-estimator"
  ],
  "backlinks": [
    "continuous-compounding",
    "correlation-matrix",
    "gaussian-copula"
  ],
  "cross_references": [
    "correlation",
    "covariance",
    "covariance-matrix",
    "dv01",
    "equity",
    "hedge-fund",
    "ledoit-wolf-shrinkage",
    "portfolio-optimization",
    "principal-component-analysis",
    "shrinkage-estimator",
    "systematic-risk",
    "variance",
    "yield",
    "yield-curve"
  ],
  "tags": [
    "level:advanced",
    "cat:financial-mathematics"
  ],
  "asset_classes": [],
  "regulators": [],
  "see_also": [],
  "sources": [],
  "wordcount": 751,
  "checksum": "fc73c5fe99b3f5b1",
  "version": "2026.05.03",
  "license": "CC-BY-4.0",
  "updated_at": "2026-09-07T02:15:24+00:00",
  "_links": {
    "self": "https://hedgefund.wiki/api/v1/terms/eigenvalue-decomposition",
    "jsonld": "https://hedgefund.wiki/api/v1/terms/eigenvalue-decomposition?format=jsonld",
    "markdown": "https://hedgefund.wiki/api/v1/terms/eigenvalue-decomposition?format=md",
    "graph": "https://hedgefund.wiki/api/v1/graph/eigenvalue-decomposition",
    "category": "https://hedgefund.wiki/api/v1/categories/financial-mathematics",
    "schema": "https://hedgefund.wiki/schema/term.schema.json",
    "html": "https://hedgefund.wiki/#/terms/eigenvalue-decomposition"
  }
}