DSGE models and the financial crisis: what central bank models could not see

What are DSGE models, and how do central banks use them?

DSGE stands for dynamic stochastic general equilibrium. These models apply general equilibrium theory and microeconomic principles to growth, business cycles, and the effects of policy. They were first proposed in the 1980s by Kydland and Prescott and by Long and Plosser, and their real business cycle form is the precursor of today’s New Keynesian versions. Policy makers use them to explain historical data, to analyze policy, and to forecast.

Central banks adopted them widely. The European Central Bank developed the Smets-Wouters model to analyze the euro area as a whole, the Bank of England has evaluated forecasts from an estimated DSGE model, and Rochelle Edge of the Federal Reserve Board describes DSGE models as prominent tools at many policy institutions. That reach explains why their blind spots matter for markets: many decisions rest on the same family of tools.

Which DSGE assumptions matter most in a stock market crash?

The base assumptions are compact. Households and firms are identical and infinitely lived, a representative consumer stands for the whole population, expectations are rational, competition is perfect, and information is complete, with no asymmetric information. New Keynesian versions let prices stick, since firms set them under monopolistic competition and cannot change them at no cost.

Each assumption removes a crash ingredient. A representative consumer cannot panic against another consumer, rational expectations leave no room for herd belief, and complete information leaves no hidden exposure. Extensions add frictions such as habit persistence and adjustment costs for investment and labor, and these frictions shape smooth business cycles. A sudden collapse needs different pieces: debt, opacity, and thresholds.

What did DSGE models say about the 2008 financial crisis?

According to the EQOS Asia study, the DSGE models of the IMF, the Fed, and the ECB saw no systemic risk before 2008 (EQOS Asia 2008 back-test, 2026). The study traces the cause to the model design: agents who maximize their own utility, perfect information, and a linear system in which ruptures cannot occur by construction.

Academic voices agree on the gap after the fact. Kocherlakota said DSGE models were not very useful for analyzing the 2008 financial crisis, Stiglitz called them ill-suited for predicting or responding to a financial crisis, and Solow told Congress in 2010 that popular DSGE models do not pass the smell test. The Bank of England admitted that none of the models it used and evaluated coped well.

Which data did DSGE models leave out before the 2008 financial crisis?

Public data showed what the models left out. According to the EQOS Asia study, banks ran leverage of 30:1 in 2007, so a fall of 3.3% in asset values reached the breaking point, while the real decline in 2007 and 2008 was about 20% (EQOS Asia 2008 back-test, 2026). A linear model has no breaking point to find, since a small shock always produces a small response.

The study adds the BIS figure for credit default swaps, 62,000 billion dollars, set against a world GDP of 54,000 billion, and the fact that 80% of CDO tranches were rated AAA. It estimates the real risk, off-balance-sheet exposure included, at about 5 times the declared exposure. A model that assumes perfect information has no variable for that gap, so the gap stayed outside the forecast.

How have DSGE models changed since the financial crisis?

Economists have not left the models unchanged. Kocherlakota reports a growing consensus that DSGE models need financial market frictions next to price stickiness, and Chari argues that current models frequently incorporate financial market imperfections, frictional unemployment, and sticky prices and wages. The direction is toward models in which credit matters.

Critics still see limits. Buiter argued that the models rely too much on complete markets and cannot describe highly nonlinear dynamics, which is the kind of dynamics a crash shows. The sensible reading is that DSGE models keep their value for routine policy analysis, and that a crash needs a second view built for thresholds, opacity, and debt.

What does the 2008 back-test measure that DSGE models do not?

Per the EQOS Asia study, the global financial system scored φ(S) = 0.234 on public 2006 and 2007 data, below the 0.3 critical threshold, with 6 of 9 dimensions in the critical zone (EQOS Asia 2008 back-test, 2026). Its clarity measure, X(S), came to 0.185: the data existed in SEC, BIS, and Case-Shiller sources, yet it was scattered and opaque, with more noise than signal.

The study also names a point of no return about 1.1 years after mid-2006, and an ethical safeguard of -0.187, which it reads as an absolute block. These are its own calculations on public data, offered as a second view next to DSGE output. The EQOS Asia method page explains how the scores are built, and the Λ simulator lets you score a case of your own.

How should a decision maker use DSGE output alongside other tools?

Use DSGE output for what it does well, baseline policy paths in normal times, and add checks for what it leaves out. Four questions help: Where is leverage highest? What exposure sits off the balance sheet? At which fall does a key borrower fail? Who has to sell first? Record the answers next to the model forecast each quarter. This page is general information, not investment advice.

Teams in Singapore, ASEAN, and Hong Kong can bring one decision to a 30-minute pre-diagnostic through the contact page, and the pricing page lists the diagnostic options. The session applies the scoring to your case and returns a score with your options in priority order.

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile