Questions: DSGE Models: Dynamic Stochastic General Equilibrium

Question 1

A central bank wants to evaluate how GDP and inflation would respond if it adopted a more aggressive interest rate rule. A colleague suggests using a DSGE model instead of a historical VAR model. What is the main advantage of the DSGE approach for this specific question?

• A. DSGE models are more accurate because they incorporate more historical data
• B. DSGE models are built from optimization problems, so their structure doesn't change when the policy rule changes
• C. DSGE models are simpler and faster to estimate than reduced-form models
• D. DSGE models use machine learning to identify causal relationships from observed patterns

Answer: (B)

This is the Lucas critique in action. Reduced-form VAR models estimate statistical relationships from historical behavior under the *existing* policy rule. If the rule changes, agents adjust their behavior, invalidating those historical correlations. DSGE models derive behavioral equations from household and firm optimization, so agent behavior is re-derived for any policy rule — the model's structure doesn't assume the old rule will continue. This is why central banks use DSGE models specifically for policy experiments.

Question 2

After log-linearizing a DSGE model, an economist finds that the matrix A has 3 explosive eigenvalues. The model has 2 forward-looking (jump) variables. What do the Blanchard-Kahn conditions imply?

• A. The model has a unique stable solution — more explosive eigenvalues ensure faster convergence
• B. The model is indeterminate — too few explosive eigenvalues relative to jump variables
• C. The model has no stable solution — there are more explosive eigenvalues than forward-looking variables
• D. The eigenvalue count is irrelevant; only the sign of the largest eigenvalue matters

Answer: (C)

The Blanchard-Kahn conditions require the number of explosive eigenvalues to equal the number of forward-looking variables exactly. Here 3 explosive eigenvalues exceed 2 forward-looking variables — the 'too many' case, which means no stable solution exists. If there were too few explosive eigenvalues (fewer than jump variables), the solution would be indeterminate (infinitely many). This is a direct application of eigenvalue analysis: the model's solvability depends on the spectral structure of A.

Question 3

A DSGE model's behavioral equations remain valid for policy analysis even after a major policy regime change.

• T. True
• F. False

Answer: True

This is precisely the appeal of DSGE models. Because behavioral equations come from household utility maximization and firm profit maximization — not from fitting historical data under a specific regime — they reflect deep structural parameters (preferences, technology) that don't change when policy changes. A new Taylor rule changes only the policy block; the Euler equation and Phillips curve are re-derived from the same optimization, not re-estimated. This is what makes DSGE models uniquely suited for policy experiments.

Question 4

The 2008 financial crisis confirmed that standard DSGE models were well-specified for capturing systemic financial risk.

• T. True
• F. False

Answer: False

The crisis exposed critical limitations. Standard DSGE models with representative agents cannot capture heterogeneous household balance sheets or wealth inequality. Rational expectations frameworks struggle with the sudden belief shifts that characterize crises. Most importantly, financial frictions were omitted — the standard model treated the financial sector as a veil that didn't affect real outcomes. The crisis catalyzed research into models with financial frictions (Bernanke-Gertler-Gilchrist) and heterogeneous agents (HANK), precisely because the standard framework missed what mattered most.

Question 5

What is the Lucas critique, and why do DSGE models claim to be immune to it?

Model answer: The Lucas critique argues that reduced-form econometric models estimated under one policy regime fail when policy changes, because agents adjust their behavior to the new rule, invalidating historical correlations. DSGE models claim immunity because they derive behavioral equations from explicit optimization — households maximize lifetime utility, firms maximize profits — so the equations reflect structural preferences and technology, not policy-dependent patterns. When the policy rule changes in the model, only the policy block changes; household and firm equations remain valid because they come from the same optimization problem regardless of the rule.

The key distinction is between 'structural' parameters (preference curvature, production technology) that are policy-invariant, and 'reduced-form' correlations that shift with policy. DSGE models are structural in this sense. The caveat: immunity holds only if the structural assumptions are correct. Mis-specified structural forms (wrong utility function, wrong market structure) produce their own bias that survives regime changes — the model is structurally wrong rather than structurally right but policy-sensitive.