Trumped Up Model Risk—a Black Swan or Red Swan Event?

With nearly every poll getting the election wrong, did we just witness a black swan event in the US Presidential election? It certainly was a shock that democrats did not see coming and the last minute FBI intervention added to the uncertainty.

However, there were some polls that saw this coming and that suggests model risk was critical error in forecasting the election. The data was there, but the wrong assumptions about voter turn-out were used. That would suggest this is more of a red swan event—a modeling error that causes a large distortion in the forecast—than a black swan event.

If model risk can cost you your job, in this case, the highest office, perhaps it deserves closer attention. Similarly in portfolio risk management, bad data going in affects the reporting at the other end.

Do you have some fundamental problem with your model risk that could cost your job or perhaps even threaten your firms survival? Do your models contain assumptions or errors that will only surface at the most critical moment?

Before you dismiss this too quickly, think back to Long Term Capital and the one underlying flaw in their risk models that sunk their firm (hint, they made assumptions about volatility that turned out to be wrong).

Here are a few examples of model risk that can throw off your forecast dramatically:

1. Using a look-back period dominated by low correlations for a predictive stress scenario. This will underestimate losses significantly if we experience a highly correlated market under stressed conditions.
2. Over-use of low volatility periods when estimating VaR. Think about mortgages entering into 2006. Are we experiencing that again? How are you handling it?
3. Failure to account for negative interest rates. Risk models have only recently been introduced to account for negative rates. Have to properly implemented them?
4. Using cash proxies for securitized products due to the complexity of properly modeling them.
   This kind of thinking is borderline negligence today given what we learned in 2007.
5. Ignoring tail events not captured by standard VaR measures.
   Stress testing needs to address scenarios beyond those observed within “normal distributions.” These can only be exposed if you are shocking the portfolio well beyond 3 sigma events.

Model risk errors are certainly not unique to risk managers, but they do bring great consequences when they occur. Fortunately, Red Swan Risk has perfected the tools, consultancy, and best practices that make our clients the gold standard of Model Risk Management.

Contact Red Swan Risk today for a demonstration of our efficient solutions for these challenges and more.