A Unified Framework for Equilibrium Selection in DSGE Models

This paper characterizes DSGE models as fixed-point selection devices for self-referential economic specifications. We formalize this structure as $(S, T, Π)$: specification, self-referential operator, and equilibrium selector. The framework applies to any DSGE model through compositional pipelines where specifications are transformed, fixed points computed, and equilibria selected. We provide formal results and computational implementation for linear rational-expectations systems, reinterpreting Blanchard-Kahn conditions as a specific selection operator and verifying that standard solution methods (such as QZ decomposition and OccBin) realize this operation. We show that alternative selectors (minimal-variance, fiscal anchoring) become available under indeterminacy, revealing selection as a policy choice rather than a mathematical necessity. Our framework reveals the formal structure underlying DSGE solution methods, enabling programmatic verification and systematic comparison of selection rules.

💡 Research Summary

The paper proposes a novel, unified formalism for dynamic stochastic general‑equilibrium (DSGE) models by treating them as fixed‑point selection devices. The authors introduce a three‑component triad (S, T, Π): a specification S that encodes admissible economic relationships (resource constraints, optimality conditions, market‑clearing equations), a self‑referential operator T that embeds rational expectations and generates the fixed‑point condition x = T(x), and a selection operator Π that chooses a single equilibrium from the (generally non‑singleton) set of fixed points Fix(T).

The central insight is that the classic Blanchard‑Kahn (BK) conditions, traditionally presented as existence‑and‑uniqueness theorems for linear rational‑expectations (LRE) systems, are in fact a concrete implementation of a particular selection rule Π_BK. Π_BK selects the unique saddle‑path equilibrium that satisfies stability (transversality) and boundedness criteria. The authors demonstrate that standard solution algorithms—QZ‑decomposition based gensys (Sims 2002), Klein’s method, time‑iteration, and the OccBin algorithm for occasionally binding constraints—all compute the same Π_BK, merely differing in computational steps. This unifies disparate solution techniques under a single semantic operation.

When the BK conditions fail (indeterminacy), the fixed‑point set contains multiple admissible paths: explosive trajectories, sun‑spot equilibria, and possibly several stable solutions. In such cases Π_BK is either inapplicable or yields a set rather than a single point. The paper argues that this is not a model flaw but a manifestation of an under‑specified selection problem. It then formalizes alternative selection operators that can be employed by policymakers:

• Minimum‑variance selector (Π_MV) – chooses the equilibrium that minimizes long‑run variance of key variables, aligning with a policy objective of volatility reduction.
• Fiscal‑anchoring selector (Π_FA) – uses the government budget constraint to pin down the price level (following Leeper 1991, Woodford 1994), thereby selecting an equilibrium when monetary policy alone does not determine it.

Both selectors are defined mathematically as mappings from Fix(T) to a singleton (or empty set if no equilibrium satisfies the rule). The authors provide linear‑algebraic characterizations for Π_MV and Π_FA in the LRE context, showing how they can be implemented by augmenting the standard state‑space representation with additional linear constraints or objective functions.

A further contribution is a programmatic pipeline that makes the (S, T, Π) decomposition explicit. Starting from a model’s textual specification, the pipeline parses equations, linearizes around the steady state, constructs the expectation operator T, computes Fix(T) (via eigenvalue decomposition), and then applies any user‑specified Π. This architecture can be embedded into existing tools such as Dynare as a plug‑in, enabling automatic diagnostics about which selection rule is being applied and how alternative rules would alter the solution.

The paper’s four main claims are:

1. Axiomatic foundation – The triad (S, T, Π) captures the minimal semantic structure required for a model to be classified as DSGE, independent of any particular economic interpretation.
2. Reinterpretation of BK – BK conditions are a specific Π_BK; solver diagnostics (“determinacy”, “indeterminacy”, “no stable solution”) are statements about the applicability of Π_BK, not about the underlying specification S.
3. Unification of solution methods – QZ, time‑iteration, and OccBin are shown to be algorithmic realizations of the same semantic operation Π_BK, explaining why they produce identical equilibria when the same selection rule is used.
4. Policy‑driven alternative selectors – By formalizing Π_MV, Π_FA, and potentially others (e.g., robust‑control worst‑case selectors), the framework turns equilibrium selection into a transparent policy choice rather than an implicit mathematical necessity.

The authors illustrate the framework with three case studies: (i) a standard New‑Keynesian model where Π_BK succeeds; (ii) a model with parameter values that generate indeterminacy, where Π_MV yields a low‑variance equilibrium; and (iii) a piecewise‑linear model with occasionally binding constraints solved by OccBin, showing how regime‑specific Π_BK can be iterated across regimes.

In conclusion, the paper makes the hidden selection mechanism of DSGE models explicit, provides a rigorous mathematical language for discussing alternative selection criteria, and offers a practical computational architecture for systematic comparison. This advances both the theoretical understanding of equilibrium multiplicity and the applied practice of macro‑policy modeling, allowing researchers and policymakers to deliberately choose equilibrium selection rules aligned with their objectives rather than relying on opaque defaults embedded in legacy solvers.