[Paper Review] Rational Decisions, Random Matrices and Spin Glasses

Stefano Galluccio, Jean‐Philippe Bouchaud+1 more|RePEc: Research Papers in Economics|Jan 21, 1998

Computability, Logic, AI Algorithms4 citations

TL;DR

This paper applies random matrix theory and spin glass models to analyze rational decision-making in portfolio optimization under nonlinear constraints. It demonstrates that the number of optimal solutions is generically exponential, undermining the practical utility of rational optimization, while showing that long-ranged (Lévy-like) spin glass couplings do not exhibit exponential ground state degeneracy.

ABSTRACT

We consider the problem of rational decision making in the presence of nonlinear constraints. By using tools borrowed from spin glass and random matrix theory, we focus on the portfolio optimisation problem. We show that the number of ``optimal'' solutions is generically exponentially large: rationality is thus de facto of limited use. In addition, this problem is related to spin glasses with Lévy-like (long-ranged) couplings, for which we show that the ground state is not exponentially degenerate.

Motivation & Objective

• To investigate the structural properties of optimal solutions in rational decision-making under nonlinear constraints.
• To examine the implications of randomness and disorder in portfolio optimization using tools from statistical physics.
• To determine whether rational optimization remains practically useful when multiple optimal solutions exist.
• To analyze the ground state degeneracy in spin glasses with Lévy-like (long-ranged) couplings.
• To connect the portfolio optimization problem to disordered systems and random matrix theory.

Proposed method

• The authors model portfolio optimization as a spin glass system with quenched disorder.
• They apply random matrix theory to analyze the eigenvalue spectrum of the covariance matrix in the optimization problem.
• The number of optimal solutions is assessed via the complexity of the energy landscape, derived from statistical mechanics techniques.
• The system is mapped to a spin glass with Lévy-distributed couplings to study long-range interactions.
• The ground state degeneracy is evaluated using replica and cavity methods, adapted to heavy-tailed interaction distributions.
• Analytical and numerical techniques are used to compare the behavior of short- and long-ranged interaction models.

Experimental results

Research questions

• RQ1 How many optimal solutions exist in a rational portfolio optimization problem with nonlinear constraints?
• RQ2 To what extent does the exponential degeneracy of optimal solutions limit the practical value of rational decision-making?
• RQ3 How does the presence of long-ranged (Lévy-like) couplings affect the ground state degeneracy in spin glass models?
• RQ4 What is the role of random matrix theory in characterizing the solution space of portfolio optimization?
• RQ5 Does the solution space of rational decision-making remain tractable when the number of optimal solutions is exponentially large?

Key findings

• The number of optimal solutions in the portfolio optimization problem is generically exponentially large, implying that rationality alone cannot select a unique solution.
• The exponential degeneracy of optimal solutions renders rational decision-making practically ineffective in high-dimensional settings.
• In contrast to short-ranged spin glasses, spin glasses with Lévy-like (long-ranged) couplings do not exhibit exponential ground state degeneracy.
• The ground state of the long-ranged spin glass model is non-degenerate, suggesting a more stable and unique solution structure.
• Random matrix theory successfully characterizes the solution space, revealing a bulk of eigenvalues consistent with random fluctuations and a few large eigenvalues indicating dominant risk factors.
• The analysis shows that the complexity of the solution space is fundamentally different in systems with heavy-tailed interactions compared to those with finite-variance couplings.