What are the three types of uncertainty in CME analysis, and which one is the most dangerous?

The three types of uncertainty form a hierarchy — and understanding that hierarchy is critical for CME development.

1. Model Uncertainty — The Foundational Risk:

Model uncertainty asks: 'Is the chosen model structurally and conceptually correct?' This is the most serious type because if the model is wrong, no amount of data or careful estimation can rescue your conclusions.

Example — Thornbridge Endowment, Late 1990s:

Many institutional investors used a simple model: 'The expected equity return is a constant μ, and the best estimate is the historical mean over the longest available sample.' As the tech bubble inflated, this model became self-reinforcing:

• Rising prices → higher historical mean → higher expected return → more allocation to equities → rising prices

The model was structurally flawed because it assumed stationarity (constant μ) when in reality, expected returns vary with valuations, the business cycle, and risk appetite. When the bubble burst, the model's predictions were not just slightly off — they were catastrophically wrong.

2. Parameter Uncertainty — Estimation Noise:

Even with the correct model, the parameters (coefficients, means, variances) must be estimated from finite samples. Every estimate comes with standard errors.

Example — Irongate Capital:

Irongate uses the CAPM to estimate expected returns. Their equity beta estimate is 1.15 with a standard error of 0.20. The 95% confidence interval for beta is 0.75 to 1.55. Using the point estimate of 1.15 with a 6% market risk premium gives an expected excess return of 6.9%. But the true return could range from 4.5% to 9.3% just from parameter uncertainty in beta alone.

Mitigation: Use longer data series (if stationary), apply shrinkage estimators (pull extreme estimates toward the mean), and report sensitivity ranges rather than point estimates.

3. Input Uncertainty — The Proxy Problem:

Input uncertainty arises when the model requires an unobservable variable and you must use a proxy. The classic example is the CAPM's 'market portfolio' — theory says it includes ALL investable assets worldwide (stocks, bonds, real estate, human capital, etc.), but in practice we proxy with a stock index like the MSCI World.

When it matters: If you're testing whether the CAPM is theoretically valid (as Roll's critique argues), the proxy problem is devastating — you can never truly test the model.

When it doesn't matter as much: If you're simply using the CAPM as a practical tool to generate reasonable expected return estimates, the proxy issue is less critical. A good equity benchmark may not be the theoretical market portfolio, but the resulting estimates may still be useful for allocation purposes.

The Hierarchy in Practice:

Key Exam Takeaway: Model uncertainty is the most dangerous because it can lead to fundamentally flawed conclusions. Parameter and input uncertainty degrade precision; model uncertainty destroys the entire analytical framework.

Explore model uncertainty scenarios in our CFA Level III question bank.