Given the events of the latter half of 2016, it's hard to deny how omnipresent volatility and uncertainty have become in the modern era. Last year, Boryana Racheva-Iotova, FinAnalytica co-founder, and Senior Vice President, Director of Risk Research and Cognity Operations at FactSet, discussed the rise of market turbulence as a "new normal" and the implications of that new normal on practical risk management.
Since this article's first publication, volatility continues to be a major concern for the financial industry (and an underpinning of FactSet's aim to innovate multi-asset class risk modeling).
Why Turbulence Is the New Normal
For many decades, up until the early signs of market tremor in 2007, financial markets were characterized by periods of “normal” behavior, followed by shorter periods exhibiting higher probability of extreme events, i.e., stressed periods. This is no longer the case.
“Normal” periods are gone. Stressed or “turbulent” periods are not necessarily the periods with highest volatility. They are the periods when—relative to the local volatility—returns that are both abrupt and large in magnitude can happen with high probability. This is measured by the volatility-of-volatility. As visualized below, based on the Cognity VaR backtesting module, prior to 2007 we clearly had normal periods (spread between normal and fat-tailed VaR is close to zero) and non-normal periods (widening spread), while the last eight to nine years have been characterized by this new “normal” turbulent regime. The level of turbulence is measured by the difference between fitted normal tail and a dynamic fat-tailed model tail.
The Demise of Normal Market Periods
There are both global irreversible factors at play in the end of normal periods and those that are related to the current economic conditions. Starting with the latter: the various stress-points in the global economic and financial markets landscape contribute to the turbulent behavior of the markets driven by the behavioral reactions of market participants to news and re-positioning towards anticipated risks (Brexit, China, oil-oversupply associated impacts, etc.).
There are simply too many hot spots burning on the map of the economic and financial risks for the foreseeable future. The irreversible universe of factors relates to the natural evolution of the financial system: technology development, evolution of investment strategies, and reach of available products (starting with those available to retail investors and the growing diversity of smart-beta funds, but also increasing complexity of investment strategies across the board, structured products). Those factors increase the speed of markets, crowding, herding, and short-term liquidity evaporation—all of which create turbulence and therefore a significant probability for large sudden drops (or spikes/soars) in the markets.
What Volatility Means for Practical Risk Management
Historical approaches now simply don’t work. It is now more true than ever that we cannot describe the future by what was observed in the past. To put it directly, historical VaR and related risk measures are simply dangerous for any kind of decision making. What is required are predictive models.
Predictive models can only be based on a smart blend of factor modeling and advanced probabilistic and econometric models to project the possible future behavior of the factors. Factor models are essential in our attempts to characterize the economic environment and related behavior of the financial markets.
Even the simplest equity factors, such as leverage and size, behave in a very non-normal way, as demonstrated in the charts below.
Daily Returns via Simulated Brownian Path Leverage Factor Returns
Growth Factor Returns Size Factor Returns
Without factor models, any kind of reasonable stress-scenarios construction (based on both subjective and model-based approaches) is doomed. Risk managers lacking that capability are disarmed and lack the ability to test their portfolios against upcoming (not past) events! The advanced modeling of factors should capture possible regime shifts, volatility clustering, and extreme events. Having the factors available, but not being able to model their true behavior properly is equally useless.