A decade ago, the global financial crisis triggered a rethink of the accuracy and predictability of statistical pricing and risk-forecasting models, which had failed to pick up on the significant build up of credit risk that ultimately triggered the crisis. The Value at Risk model was one of many models that were widely criticised for failing to take into consideration the possibility of “fat tails” and Nassim Taleb’s Black Swan: the Impact of the Highly Improbable could not have been better timed, published on the eve of the market’s collapse.
Fast forward ten years and, paradoxically, the prospect of another fat tail seems increasingly probable. At a minimum, the broad expectation is that markets are due for another change of regime, though opinion is divided as to when or what that might look like.
In this context, investment risk professionals are increasingly questioning the validity of the portfolio risk models they apply, as well as looking for new methods to identify and manage tail risk.
Dr Laura Ryan, former Senior Vice President Quantitative Research at PIMCO who has recently taken a job as Head of Portfolio Construction at a soon-to-be-announced asset manager, told Fund Business that the problem with standard approaches to portfolio risk management is that they rely on empirically derived correlations that are measured over some relatively arbitrary time period. These correlations are not stable over time and can change suddenly in the event of a regime change. While the typical response is to include correlations obtained during previous market events, such as the global financial crisis of 2008 or synthetic “worst case” scenarios, these correlations may not bear any relevance to future crisis scenarios.
“These approaches may lead portfolio managers to be overly confident in their approach to managing risk since future crisis may not resemble any pre-existing data sample, while on the other hand excessively conservative “worst case” scenarios may simply be ignored if portfolio managers do not believe in their relevance. In any case traditional approaches provide no insight in to when or how markets shift regime,” Ryan said.
Back to basics
Part of the solution to be better prepared for potential crises is to make sure you have the basics covered.
“From a risk model perspective, it’s important to minimise all known sources of error. While there is no one perfect model, it’s very important that when you’re comparing your assets and your liabilities, you’re using the same model. That sounds obvious, but it’s very common for there to be a mismatch – for example when liabilities are calculated by an asset owners’ actuaries, while assets are being measured by a fund manager using a different model. It just means you end up measuring apples and oranges,” Ryan said.
In addition to minimising sources of error arising from asset and liability model mismatches, it’s important to understand the different techniques available and to compare different models and different data sources. “’You’ll never find a model that will give you a 100% hit rate. But what you can do is look at the world through a multi-model lens. You could have a vendor internally that you subscribe to, but also lean on your fund managers to cross check your internal views with their analysis,” Ryan said, adding that stress testing and scenario analyses are also critical.
“Try modelling scenarios that haven’t happened before. Tweak your covariance matrix and model a breakdown in risk factor relationships. Try to break your portfolio and your model. I am starting to see a lot of portfolio construction experts building portfolios within crisis periods and then tweaking the asset allocation around that structure rather than building portfolios using normal state or steady times,” she said.
Identifying regime change
While traditional risk models hold significant value, particularly over short time horizons, they are less equipped to pick up on regime change and this is where, some say, machine learning techniques could make a difference.
In particular, so-called ‘turning point models’ are seen as one of the more reliable applications of big data and artificial intelligence. The idea is that by entering variables that might predict a turning point or regime switch, such as for example volatility, changes in consumer confidence or bond prices, you can get a better idea of when and under what conditions, the probability of transition peaks.
“Both academics and practitioners are applying modern machine learning techniques to the problem of identifying regime shifts or economic turning points. The out of sample tests are showing improvements when compared to traditional techniques. Nowcasting (identifying what regime the market is currently in; when the regime begins to shift and most importantly if the new regime fits a previously encountered pattern) seems to be one of the most successful applications. These new tools allow portfolio managers to proactively manage risk in ways not previously possible,” Ryan said.
She said that, while these new techniques are no substitute for traditional models, it is worth paying attention. “Should you throw away all of your old models and just use machine learning to spit out “the answer”? No, but you shouldn’t ignore the researched and documented improvements that new techniques bring to understanding the problem. Every little bit helps in this low return environment,” Ryan said.
However, others have urged caution regarding the application of machine learning models to identify change. Stefano Cavaglia, Senior Consultant to Invesco and Founder of the Global Equity L/S strategy managed out of UBS O’Connor, the $40 billion hedge fund arm of UBS, told Fund Business that the risk inherent in models like these is that they produce many false negatives. “You think there is a disaster coming and then it doesn’t happen, and that is the issue with all these statistical relationships,” Cavaglia said.
Cavaglia said a particular challenge with respect to turning point models relates to the lack of major disaster data points. “If you’re running statistical analyses on 20,000 stocks and you’re getting the direction right on 55% of these, that’s a lot of data points. When you’re looking at financial crises, there simply aren’t enough of these disasters and data points to provide confidence in the reliability of the results. That’s why, as much as I like statistics, any output from these models should be taken with a significant grain of salt,” Cavaglia said.
Tail hedging in scope
While market participants may not yet be able to reliably predict the next point at which the market will change regime, what is clear is that many more fund managers are looking to prepare for the possibility.
Traditionally, investors have looked to protect against tail risk events by using a variety of strategies – diversification being one common tool, followed by risk budgeting techniques and managed volatility strategies. Alternatively, they’ve often switched to bonds in times of stress.
In the current market, investors are also increasingly looking to tail risk hedges as an alternative to buying bonds. While bonds have historically reduced long term expected returns, tail risk hedges allow investors to buy protection while maintaining higher overall equity allocations. Typically structured as long term put options, these hedges will profit in the event of a downturn and may provide investors with a source of liquidity at a time when overall liquidity in the market dries up.
These strategies do come at a cost – fixed fees charged by funds that offer this type of protection as part of equity exposure are currently in the range of 1.5 to 2 per cent – but compared to the opportunity costs from investing in bonds, it’s a discussion worth having.
“If you have a long horizon of creating wealth for your clients and there is a big fall in the market value of your assets, you are not going to be able to compound, so anything that mitigates those drawdowns is really important. A protection that has zero expected return, but that protects you on the downside is something you should be willing to pay for,” Cavaglia said.
Ryan agreed, “You have to balance it up and figure out how much upside you’re willing to forego to avoid a huge drawdown, but it’s definitely a discussion worth having,” she said.