Financial Econometrics and Empirical Market Microstructure

Stress Grades Amplify Market Based Risk Signals

As volatility based metrics, StressGrades will not directly uncover hidden structural risk or predict Black Swans. StressGrades merely amplify existing market based signals of risk, as a seismograph amplifies geological tremors.[23] And as with earthquakes, the absence of tremors does not imply the absence of risk. Or, to use a medical analogy, a stethoscope allows doctors to listen to what’s inside. An experienced doctor can use it to detect imbalances without being fooled into believing it provides a full picture of health. To be effective, StressGrades should be used within the Adaptive Stress Testing framework: (a) prioritize immediately escalating stress themes, and (b) probe for hidden risks in themes with abnormally low volatility.

StressGrades require three major steps:

1. Design Stress Indices

As discussed earlier, Adaptive Stress Testing calls for the construction of market based Stress Indices that are correlated with key scenarios. For example, we might consider an oil shock scenario due to conflict with Iran, which we could model at different levels of detail (e. g., oil price, FX prices, country and industry equity sectors, and even down to the specific company level). Note that Stress Indices might be constructed using options theory to model non-linearities (e. g., an oil call with a 130 strike + equity put struck at 90). Once we’ve modeled our Stress Indices, we can start to monitor the tremors for each fault line that indicate escalating risk.

2. Determine a Stress Point

We then determine a critical point which represents a stressed condition. In our examples, we used maximum historical drawdowns over a 10 year period. After calculating the volatility of our ETF time series, we then determine how many standard deviations it would take to achieve such a daily loss to calculate DStress. We implemented the RiskMetrics methodology to estimate daily volatility (i. e.,

exponential weighting with 0.94 decay). Note, however, that in some cases a maximum historical drawdown could be too severe a Stress Point to consider. In practice a 99 % confidence Expected Shortfall over a decade (or a full market cycle) could be reasonable to consider as a Stress Point. Subjective assessments and analysis of similar asset classes should be applied for assets with limited loss history (e. g., subprime bonds in 2006).

3. Calculate market implied Probability of Stress (PStress)

After making a distributional assumption, we can back out the implied Probability of Stress (PStress). For example, assuming Normality, a DStress of —2.33 would imply a PStress of 1 %, while a DStress of 1.65 would imply a PStress of 5 %. In the examples below (Fig. 13), we will use a Normal distribution assumption for simplicity. Clearly, accuracy of PStress could be improved by using a fat tailed distribution such a Student (Zumbach 2007). However, given that our initial objective is to flag outlier changes in market implied stress probability, the use of a Normal distribution is appropriate (future versions will consider Student t and other fat tailed distributions).

To summarize, StressGrades are volatility based metrics which can be used to monitor market implied risk sentiment. To be useful, we need to start with a macro perspective to understand key fault lines, and apply StressGrades both to monitor emerging visible risk as well as identify artificially low levels of visible risk.