In the financial world, having a workforce with strong decision-making capabilities goes a very long way to ensure the success of an organization. Of course, in the last few decades the collective bar has been raised across the industry – digitization and computerization have provided firms with more advanced ways to assess risk-reward situations and derive an appropriate set of action.
But there is still room for improvement.
As technology keeps progressing and innovative methods mature, it’s vital that leading firms progress and innovate along with them so their competitive edge stays sharp. Here we briefly review why stress testing tools have become more strategically important, take stock of current stress testing and optimization methods, and posit what the future may hold for this progressive and evolving area of risk management.
Looking back to move forward
Although it was 14 years ago, the reverberation of the Global Financial Crisis is still being felt – no more so than in the field of risk management. It is hard to imagine now that the emphasis on worst-case scenario testing will ever return to the previous focus of variability analysis. The idea that market dynamics could be reduced to relatively straight-forward processes has simply been abandoned.
The changes have been felt in several ways. Traditional measures of variance and quantiles are now complemented by additional probability metrics. Governments (via the G20 in particular) now expect financial firms to include subjective quantification of any strongly adverse conditions they believe are both material and possible when analyzing systemic and ‘tail’ risks. And systemically important financial institutions (SIFIs) are liable to a laundry list of Financial Stability Board regulations regarding their management and reporting of ‘extreme’ risks.
For all three of these examples, stress testing has emerged as the tool of choice. And while some may argue that assessing worst case scenarios is more an art than a science, one look at the insight firms are now expected to produce should make it clear that very advanced financial simulations capabilities – some of which are still being developed – are essential in today’s world.
Striving for perfection
At the heart of any stress testing tool is a powerful simulation engine. Thanks to the availability of cost- effective devices and the progress of high-performance, distributed computing techniques, it is now possible to calculate projections of positions and exposures with breakneck speed and pinpoint accuracy. Defining future states that are both plausible and consistent, however, is trickier.
Luckily, future states can be made up of a combination of multiple drivers that have decades of research and engineering behind them. The following drivers have a large body of methods and models already available:
- Economic and market conditions
- Potentiality of counterparties to default or fail on their obligations
- Potentiality of conditional events to be triggered
- Appetite and behavior of economic agents
- Cost and effectiveness of recovering from an adverse event
But not all drivers are quite as well explored, and while there is still an intense academic interest in them, some domains remain imperfect such as:
- Implementing machine learning techniques to capture behavioral characteristics
- Modelling contagion and amplification-through- propagation of individual shocks
- Incorporating loopback effect in long range simulations
- Better representing and assessing cross-dependencies between individual risk elements
- Modelling assets, markets and counterparties for which knowledge is incomplete or adjustment mechanisms are ineffective or distorted
Research into these factors continues, but even if we were able to resolve all the shortcoming of existing simulation methods, realistically we would still be a long way from being able to model any potential future crisis. What has puzzled economists and finance professional alike for generations is the phenomenon by which they develop: What makes bubbles burst? Inflation accelerate? Consumption collapse? Investor sentiment shift? And when such events occur, what makes the cascade of reactions go one way rather than another?
Retrospective analysis humbles even the best experts. Stress testing therefore is not an ultimate tool to mitigate extreme situations, but rather a method to prepare oneself to confront unpredictable scenarios by building a reference map of catastrophic possibilities.
Retrace your steps - before you take them
In simple terms, a stress scenario relates a highly undesirable outcome to the set of conditions (initial values and evolution of risk ‘factors’) that have led to it. It is therefore natural to try identify the particular scenarios that may lead to a selected list of particularly unwanted outcomes. The term ‘reverse’ is then used, although the relationship between scenarios and outcomes is not really reversible (at least in the sense of a ‘well posed’ problem). As it is generally difficult to perform a reverse analysis, a wide range of methods have been considered. More and more, machine learning algorithms come to the rescue to facilitate the delivery of practical solutions.
But high dimensional problems can prove to be very tough nuts to crack. Reverse stressing on a complex portfolio with many interconnected drivers may require extremely large computing power. One common approach to mitigate this issue consist in replacing the portfolio with a simplified version that maintains its key risk characteristics. This reduction process itself may however turn to be quite laborious, but recent experiments have shown that modern technologies can overcome the challenge.
The other side of the stress testing coin: profitability optimization
So far, we have explored how risk can be quantified via stress testing. However, profitability can also be quantified in the same way. Simulating both returns and unexpected losses simultaneously gives a good indication on how much of the incremental risk undertaken is compensated by an excess return. For instance, an institution may decide to implement an aggressive credit policy to gain market share at the price of deteriorating its risk profile. Projecting the effect of such decisions is generally called strategic business planning. Extending the approach to systemic and ‘tail’ risks is at the heart of risk appetite management.
All financial institutions continuously adjust their business parameters (portfolio composition, pricing, and risk policies, etc.) in the quest to maximize risk-adjusted performance while containing the ‘extreme risks’ budget within acceptable boundaries. Essentially, this is a reverse stress testing optimization problem. Unfortunately, if one considers the entire balance sheet of an institution, the degree of complexity of the problem becomes very high, and the system used to optimize it contains too many degrees of freedom for any conventional method to be applied successfully. At least this is the case for now: this may change with the advance of quantum computing technology, but we will have to wait a few years for practical solutions to be available.
This does not mean that optimization cannot be applied to financial institution’s business problems. On the contrary, if one decomposes the balance sheet into manageable elements, it is always possible to explore well defined sets of scenarios with the aim of maximizing a composite metric that represents the organization goals. This prospect defines the next frontier for asset and liability management (ALM) systems.
More than a feature or function
In summary, stress testing and optimization are not ‘features’ that a risk system either does or does not have. They should instead be treated as multifunctional domains where continuous improvements can be made to enable institutions to better understand, prepare for, assess, and report the situations that might severely impact their business. They should also be seen as powerful mechanisms that can adjust the structure of a balance sheet and the composition of a firm’s portfolio to maximize business opportunities, while at the same time controlling exposure to systemic and ‘tail’ risks.
Current capabilities in computational finance already support these processes quite satisfactorily, and active research and experimentation keeps pushing the boundaries of what is achievable. But what we know, now and in the future, is that a robust foundation within a stress testing and optimization system is essential to ensure the best possible standards of exposure representation and simulation.
