Why DSGE models saw no risk before the 2008 crash, and what EQOS measures instead

Why did DSGE models see no risk before the 2008 crash?

According to the study, DSGE models are the main tool of the IMF, the Fed and the ECB, and they returned a verdict of no systemic risk before 2008. The reason lies in their design. The models assume conditions in which a financial crisis cannot appear, so a quiet output follows from the inputs (EQOS Asia 2008 back-test, 2026).

That finding concerns structure and leaves individual economists out of the story. Skilled people operated the models carefully. The blind spot sat in the assumptions, which describe a calm economy in which agents behave predictably and information flows freely. The 2006 and 2007 system did not match that description. Models inherit the world view of their assumptions, which is why a review of assumptions matters as much as a review of data.

What is a DSGE model, and who uses it?

DSGE stands for dynamic stochastic general equilibrium. According to Wikipedia as of October 2026, it is a macroeconomic method used for policy analysis, explaining historical data and forecasting, and it combines general equilibrium theory with microeconomic foundations to study growth, business cycles and policy effects.

Central banks adopted the method widely. The European Central Bank developed the Smets and Wouters model in 2002 to analyse the euro area, and a Federal Reserve Board economist described DSGE models in 2010 as quite prominent tools at many policy institutions. Their reach explains why a blind spot matters. The method gained influence because it links policy choices to a consistent story about households and firms.

Which 5 assumptions made DSGE models blind to the 2008 crisis?

The study lists 5. Agents are rational, so every actor maximises utility, although NINJA loans, bubbles and panic showed otherwise. The models have no human dimension, so trust, care, culture and fear go unmeasured. They assume perfect information, so off-balance-sheet risk does not exist in them.

The models are linear, so tipping thresholds and ruptures cannot occur by construction. Finally, they carry no ethical safeguard, so harm to borrowers never enters the output. Each assumption removes one feature of 2006 and 2007, and together they produce the quiet reading that the study reports. Reading the list as a checklist helps you test any model you rely on.

What did economists say about DSGE models after the 2008 financial crisis?

Kocherlakota wrote in 2010 that the models were not very useful for analysing the crisis and needed financial market frictions. Stiglitz said in 2018 that they are ill-suited for predicting or responding to a financial crisis. The Bank of England acknowledged in 2015 that none of the models it evaluated coped well during the crisis. These 3 comments span 2010 to 2018, so the discussion lasted well beyond the crisis.

These comments come from within the profession and match the study’s point. Each one names a missing piece, such as the financial sector, credit or asset prices. The EQOS Asia back-test adds a measurement approach for those missing pieces, built on public data and open to recomputation. A missing financial sector matters a great deal when the event under study starts in banks.

What do the 9 EQOS dimensions measure that central bank models omit?

The study measures 9 dimensions of the system. Coherence, the gap between discourse and reality, reads 0.12. Opacity enters through X(S) = 0.185, which means noise exceeded signal by about 5 times. A rupture threshold of 3.3% is computed from leverage, so the breaking point is derived and not guessed. Each reading comes from public data described in the study.

The ethical safeguard reads −0.187, which triggers an absolute block. The result is φ(S) = 0.234, below the critical threshold of 0.3, and a stop signal from mid-2006. The study stresses that the weighting inside the engine is proprietary, while the inputs and outputs are open to inspection.

How do the DSGE and EQOS readings of the 2008 financial crisis differ?

The DSGE reading says no systemic risk, because its assumptions exclude the mechanisms that produced the crash. The EQOS reading says critical zone, with 6 of 9 dimensions below 0.3. Both start from the same period, and the contrast comes from what each approach chooses to measure. A reader can list the chosen measures for any model and see what stays out of view.

The comparison does not claim that EQOS predicts dates. The study states that the point of no return fell about 1.1 year after mid-2006, a projection from a declining score, and Lehman Brothers filed with $619 billion of debt on 15 September 2008. Readers should treat the projection as a modelled result.

Which habits should decision makers adopt when relying on models before a crash?

Use a model for what its assumptions allow, and add a second lens for what they exclude. Ask which dimensions the model leaves out, and measure those separately: leverage, hidden risk, incentives and trust. This page is general information, not investment advice, and no model gives a date. Keep a record of which lens said what, so a later review can learn from both.

The EQOS Asia method page explains the 9 dimensions, and the Λ simulator estimates a score for your own case. The pillar guide on the stock market crash shows how the 2008 lessons connect with earlier episodes, such as 1929 and 1987. A short call with the team applies the scoring to your exposure.

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