- In a financial crisis, a central question is how quickly to react to the change.
- For external risk-policy, it has been common to employ a conservative combination of risk models at different levels of responsiveness. Internally, there may be more discretion to select less reactive models for greater flexibility.
- Risk backtesting that is both real-time and based on a long history puts these trade-offs in historical context and demonstrates the similarities with previous crises.
While every financial crisis is unique, the impact on risk policy has been strikingly similar. Risk limits are suddenly breached across many funds. Many value-at-risk (VaR) violations occur within a short timeframe, drawing extra scrutiny to the risk models that define those violations. And investors’ governance bodies often challenge the quality of the investors’ risk models.
During the crisis sparked by the COVID-19 pandemic, discussions with MSCI clients have centered around whether the models are reacting appropriately to the dramatic changes in the market, and how to most effectively use the models in the new regime. Here we describe some answers to these questions: adaptive modeling approaches for internal and external risk-policy, and up-to-date, long-view model validation (backtesting) to support modeling decisions.
Adaptive modeling approaches
When the market is not functioning normally, risk modeling must adapt to the new regime. A central question is how quickly models (and investment decisions) should react to the change. All backward-looking risk models will, by construction, react too slowly to sudden spikes in volatility, which means that risk models may under-forecast risk at these times. However, a more reactive model (based on recent data) should produce risk estimates more in line with rapidly changing market conditions.
Different approaches may be required to meet internal and external risk-management requirements. Externally, risk models must be defended using objective metrics for regulators, investment boards, external investors, etc. So, for external risk-policy, risk models have historically been assessed using validation metrics that disproportionately penalize models that are too low over those that are too high.1 During a crisis, a decision may be made to move to a more reactive model, as this may be more conservative from a risk policy perspective. Another approach is to use a systematic combination of reactive and stable models, where the maximum risk of two models is used.
For internal risk-management, risk models have historically been assessed using a more balanced view that more equally penalizes over- and under-forecasting risk. Additionally, modeling choices have been informed by qualitative, as well as quantitative, considerations, so there is more discretion on the part of the risk manager to adapt to a crisis. For that reason, a move to a less reactive (i.e., less conservative) model may be justified to temporarily give additional flexibility to portfolio managers.
Up-to-date, long-view model validation
Risk backtesting provides the empirical grounding for changes to both internal and external risk-policy. Most valuable have been backtests that include both the current crisis and those in the past. Such a long view puts the trade-off between different levels of responsiveness into historical context (see exhibit below). This type of systematic risk backtesting has been a valuable tool during times of crisis to facilitate discussions with external stakeholders and inform internal risk-policy decisions. This historical perspective has, in turn, allowed for more flexibility in modeling decisions that can be critical during times of crisis. Finally, the long-view shows that, while every crisis is unique, each from a broad risk perspective has revealed striking similarities.
Risk models compared to US market returns over weekly and monthly horizons
+/- 2x model volatility against weekly and monthly returns (based on the MSCI USA Index) for the MSCI Multi-asset Class Factor Model, Short and Long. When the model is in line with realized returns about 95% of the returns should fall within this “risk envelope” during any given period.
Comparing risk models to market returns over a long history can highlight similarities between current and past crises, along with common trade-offs made between responsive and stable risk models. Such historical perspectives have proven valuable for setting risk policy during previous crises.
1For example, VaR-based UCITS regulation only penalizes too many VaR violations, not too few.
MSCI Multi-Asset Class (MAC) Factor Model Validation (client-access only)
MSCI Risk Backtesting Monitor (client-access only)