Can you walk through how flawed models contributed to the tech bubble and the 2007-2009 financial crisis?

Both crises are textbook cases of model uncertainty — not parameter errors or bad inputs, but fundamentally wrong conceptual frameworks that gave investors false confidence.

Crisis 1 — The Tech Bubble (Late 1990s):

The specific model failure: The implicit model assumed expected returns are stationary — a fixed constant that can be estimated by averaging historical returns. But expected returns are NOT constant. They vary with:

• Valuations (higher P/E → lower expected returns)
• Risk premiums (which compress in euphoria and expand in panic)
• Monetary conditions and the business cycle

As prices rose, the historical average rose with them, making the model self-reinforcing. Ironically, this coincided with the opposite belief — that the 'new economy' had invalidated historical relationships. Investors simultaneously believed that history was the best guide to returns AND that history was irrelevant. The contradiction went unnoticed because both views supported the same conclusion: buy more equities.

Crisis 2 — The Global Financial Crisis (2007-2009):

Component 1 — Geographic diversification of housing risk:

Risk models assumed housing price declines would be geographically isolated — a bust in one city wouldn't spread nationwide. Historical data supported this for localized events. But the model failed to account for a systemic cause (nationwide loose lending standards + nationwide rate resets) that would produce correlated declines everywhere simultaneously.

Component 2 — Originate-to-sell incentive structure:

Lenders who planned to immediately sell loans had minimal incentive to assess borrower quality. The model assumed loan quality would remain reasonable because lenders had reputational capital at stake. In reality, the volume incentives overwhelmed quality controls.

Component 3 — Securitization as risk diversification:

Computer simulations showed that pooling thousands of mortgages diversified individual default risk effectively. This was true for MICRO risk (idiosyncratic borrower defaults). But it completely missed MACRO risk — when the entire housing market declines, every mortgage in every pool is affected. Diversification across pools provides zero protection against systematic risk.

The Common Thread:

Both crises demonstrate that model uncertainty is catastrophic precisely because it's invisible from inside the model. The tech bubble model produced internally consistent results — it just happened to be conceptually wrong. The GFC risk models passed every backtest — they just happened to miss the scenario that actually occurred.

Lessons for CME:

1. Always ask: 'What if this model is fundamentally wrong?' — not just 'What if the parameters are off?'
2. Use multiple models with different structural assumptions
3. Stress-test against scenarios that the model says are impossible
4. Be especially wary when a model becomes self-reinforcing (positive feedback loops)

Test your understanding of model breakdowns in our CFA Level III question bank.