Parameter Risk

Parameter risk is the danger that the input parameters estimated for a financial model — volatility, correlation, interest rates, default probabilities — are wrong, incorrect, or unrepresentative, leading to misdecision, mispricing, or losses. It is a subset of model-risk focused on the inputs rather than the model structure itself.

How parameters are estimated (and where errors creep in)

A value-at-risk model needs three key parameters: the historical return of the portfolio, its volatility (standard deviation), and the correlation between portfolio components. These are all estimated from historical data, and each estimation introduces risk.

Example of volatility estimation error:

• You estimate a portfolio’s volatility using the past 252 trading days (one year).
• If that year was unusually calm, volatility is low, and your value-at-risk estimate is low.
• But if you had used five years of data (which includes calmer and more volatile periods), your volatility estimate would be higher.
• Which is right? Neither — both are estimates, both can be wrong.

When markets calm down after a volatile period, estimates tend to be too low. When volatility spikes, estimates lag behind. This is why value-at-risk models notoriously underestimate risk just before a crash — the parameters were estimated during a calm period.

Where parameter risk is highest

Volatility: Estimated from past price moves, volatility parameters are notoriously unstable. In 2019, US stock volatility was 12%; in March 2020 it spiked to 82%. Models that assumed 12% volatility were catastrophically wrong.

Correlation: Two assets are assumed to move together 40% of the time. But during crises, correlations break down. Stocks and bonds are assumed to be uncorrelated; they both crashed in March 2020. A diversified portfolio that should have been stable swung wildly.

Default probability: Estimated from credit rating data, default probabilities are biased toward calm periods. In recessions, defaults spike above estimates.

Interest rates: Models assume interest rates follow a particular process. But interest rate paths are hard to forecast. Models that worked when the Federal Reserve was signalling zero rates failed when it suddenly raised rates in 2022.

Consequences of parameter error

Small parameter errors can lead to large output errors if the model is sensitive:

• A value-at-risk model that is sensitive to volatility: if true volatility is 20% but estimated at 16%, the model underestimates risk by about 25%. A $1M portfolio could lose more than expected.

• An options pricing model sensitive to volatility: if you estimate volatility at 15% but it is actually 25%, you underprice options by a large margin. If you are the seller, you lose money on every option sold.

• A credit model sensitive to default correlation: if defaults are more correlated in downturns than your model assumes, credit losses in a recession are worse than expected.

Managing parameter risk

Sophisticated approaches include:

• Multiple estimation methods. Do not rely on a single historical window or method. Estimate volatility using multiple periods and methods (realized, implied, GARCH). Use the range to bound risk.

• Sensitivity analysis. Vary parameters up and down by 10-20% and see how model outputs change. If outputs are highly sensitive, parameter risk is material.

• Out-of-sample testing. Estimate parameters on one period of data, and test the model on a different period. If the model fails out-of-sample, parameter risk is high.

• Implied parameters. For some parameters, you can extract market expectations. Implied volatility from option prices, term structure from bond yields. Use these to cross-check historical estimates.

• Regime-switching models. Acknowledge that different market regimes have different parameters. Use regime-switching models that adapt parameters based on current conditions.

• Conservatism. Assume parameters are worse than estimated. If estimated volatility is 15%, use 18% for risk management. The extra buffer cushions parameter error.

For individual investors, parameter risk is mostly implicit. When you use a value-at-risk calculator or a robo-adviser, parameters are already built in. The practical defense is to:

• Understand what parameters the model uses.
• Check whether they seem reasonable for current market conditions.
• Size positions conservatively; do not trust a model completely.
• Monitor actual outcomes versus model predictions; if they diverge, reduce reliance on the model.

See also