Risk (Modelling in Global Aggregate Portfolios) in a Time of COVID

John Beck, Director of Fixed Income contemplates how risk is measured in our portfolios.

John W. Beck

John W. BeckSenior Vice President,Director of Fixed Income

Paraphrasing the title of the book written by the Nobel Laureate Gabriel Garcia Marquez in 1985, Love in the Time of Cholera, we aim to think not about the risks in the world before COVID-19, or indeed the risks in a world after our eventual release from weeks of lockdown, but rather to contemplate how risk is measured in our portfolios. In some cases during recent markets this has proved to be wildly off the mark, and in other cases, remarkably accurate.

We consider where models have been helpful and, with some transparency, acknowledge where they have failed. But a guiding principle should be to help us think about how our portfolios might be positioned as we (eventually) emerge from the current global shutdown. The non-mathematical reader will breathe a sigh of relief that the principle of understanding does not require certainty about mathematical models; in fact, it simply requires the willingness to accept that in any model there is also an element of doubt. Or as Mervyn King, the Governor of the Bank of England during 2008’s global financial crisis (GFC), refers to it, there is always an element of “Radical Uncertainty” in any model.

NO RISK MODEL FULLY CAPTURES “RADICAL UNCERTAINTY.”

During the GFC, the chief financial officer of a large Wall Street Firm said financial markets were experiencing 25th standard-deviation events days in a row. Although sounding impressive enough to convey the volatility experienced in markets, statistically the statement was inaccurate, since, to be plausible, a 25th standard deviation event taking place consecutively for two days—let alone three or even four days running—would require more observations than the number of days since the dawn of civilisation. If, however, the message was to convey that market events were being observed which no reasonable risk model or manager could predict, then it was accurate. And so, it has proved to be in the current crisis with current models.

Portfolio managers work with risk. Index managers work to minimize risk to be closely aligned to their benchmark target; active managers work to optimize risk positions according to their forecast outcomes. Both managers use optimizers, and both will report their expected variability against their chosen benchmark—under normal market conditions. But events in 2008, let alone in 2020, have been so far from normal as to render risk models rather inferior comfort blankets in both cases.

During the Spanish flu pandemic of 1918, the global population was one quarter of today’s, and with far lower global interconnectivity in trade and economies. But even making an assumption, as economists do, of “Ceteris Paribus”—“all things being equal”—the ability of any risk model, even with 100 years of data, to correctly capture market events as we have recently witnessed, is forlorn.

The risk model used in global fixed income portfolios measures risk/volatility based on historic daily observations. For one measure of risk, all days are deemed of equal importance; thus, the dark days of 1987, the Russian Crisis of 1998, 9/11 in 2001 and the GFC in 2008 are deemed as important as any mundane day when markets are quiet. This is referred to as “unweighted” risk. For other lenses of risk, we might say that the events of 1987, of 2001 or even 2008 were so long ago that what is more relevant is more recent risk.

So, while we won’t ignore those factors entirely, we will argue that more recent conditions are more relevant. The mathematicians call this method GARCH (generalised autoregressive conditional heteroskedasticity), and the statisticians and risk modelers call it “weighted” risk. And the brief bit of scientific mixing together is when we predict what will happen in our portfolios with 95% probability based around observations of the past. We can even predict with 99% certainty what would happen in the case of a “maximum drawdown.” Ah! But what about the 5%? Or what happens in 1% of cases? asks the cynic. In the words of another Nobel Laureate, “the answer my friends is blowin’ in the wind.” No risk model will fully capture what we have seen which is “Radical Uncertainty.”

SO, AT THE PORTFOLIO LEVEL, WHAT DOES THIS MEAN?

How much risk did portfolios carry? How did the market outcomes compare to predicted risk? A useful comparison is actually with 2008, when our global aggregate portfolio entered the final quarter with a “tracking error” of 1%, meaning that in normal conditions the return would be +/–1% against the benchmark return. The outcome in 2008? One percent of measured risk translated to an outcome 4% worse than the benchmark. In 2020, the portfolio entered the crisis with a little more risk—2% in this case, but eminently manageable. The outcome, at least in the last month, has been in similar proportion but slightly better than the outcome in 2008. Does this show that risk models don’t work? Or does it demonstrate “Radical Uncertainty?”

If risk models have not provided certainty of outcome, helpfully the breakdown of risk has. Risk in the portfolio is broken down into three broad categories: currency risk, country/duration risk, and spread risk, covering the main characteristics of our portfolio. Going into the current crisis, the approximate breakdown of risk was about a quarter coming from currency positioning, about a quarter coming from duration/yield curve positions, and the remaining half coming from sector (governments or corporates), credit positions (mortgages, corporate bonds, emerging markets or high yield), and bond issue selection (selection of corporate bonds).

In a global pandemic there are no winners. When the impact of the COVID-19 shock is largely equivalent across geographies, relative currency positioning becomes largely irrelevant.

The economic impact of the COVID-19 shutdown in the United Kingdom (GBP), Europe (EUR), Poland (PLN), Japan (JPY), or Norway (NOK) has been largely similar. A relative comparison of their respective currencies has little bearing, and so it has been in our portfolios. While there has been a small degree of variability, the impact has been small, and in line with predicted risk.

In our positioning of country/duration risk, we have favoured some regions (Europe over the United States; Poland over the United Kingdom; Australia over the United States, and pretty much anything over Japan). When governments have effectively moved to temporarily shut down their economies and most central banks around the world have cut interest rates close to zero and/or promised to buy huge amounts of government bonds issued by their governments, once again there has been limited variability versus predicted outcomes.

Sure, truly risk-free assets (there have really only been three as Japanese government bond yields were already on the floor: US Treasuries, UK Gilts and German Bunds) have outperformed, and there has been variability between some European government bonds. But once again, the outcome seen in portfolios lies well within expected market volatilities. The risk predictors have worked. Although there was variability of how duration was taken in respective markets, overall duration has been quite close to benchmark, and the outcome in portfolios has been largely as the risk models might have predicted.

WHERE RISK INDICATORS HAVE NOT WORKED, AND WHY?

Sector, spread and security selection have conspired to be a much more difficult space to predict outcomes versus the models. Market outcomes have been materially worse, but in part this variability versus prediction is entirely understandable ex post (another phrase used by economists, who having failed to predict the future, can now explain their forecast errors of the past). But where an industry (for example, autos, entertainment or tourism) is entirely shut down with an indeterminate end date, and a recovery rate is difficult to forecast the fact that these sectors have amplified risk from forecast data in portfolios should not be too surprising.

Mortgage repayments are of course uncertain when many workers are furloughed. Rent receipts on office space are uncertain when the return to the office is uncertain, and toll receipts are absent on motorways when no one is driving. But we do assume we will return to work, we will wish to move in a less socially distanced world, and to the degree that risk taken in these sectors seems to have been amplified in a negative skew of return from what was anticipated, it is in this area that we believe opportunities may exist.

We may debate the shape of economic recoveries: V-shaped (less likely as the shutdown becomes more sustained), U-shaped (more plausible especially given the scale of fiscal support made by governments to their economies) or L-shaped (one would have to be a true pessimist to assume that no recovery can be expected, and output loss will be permanent). But given these variable outcomes, it does not seem that the opportunities lie in guessing where interest rates may be (i.e., adjusting our duration stance) or which currencies might do better (at the momenta further imponderable). Rather, it seems that opportunities lie in the areas most affected over and above predicted forecasts.

So rather than berate the flaws of risk models, where “Radical Uncertainty” has exposed the tails of the forecast risks, the way things have turned out provides us clues as to where to look for recovery.


WHAT ARE THE RISKS?

All investments involve risks, including possible loss of principal. Diversification does not guarantee profits or eliminate the risk of investment losses. Stock prices fluctuate, sometimes rapidly and dramatically, due to factors affecting individual companies, particular industries or sectors, or general market conditions. Bond prices generally move in the opposite direction of interest rates. Thus, as the prices of bonds in an investment portfolio adjust to a rise in interest rates, the value of the portfolio may decline. Special risks are associated with foreign investing, including currency fluctuations, economic instability and political developments. Investments in emerging markets, of which frontier markets are a subset, involve heightened risks related to the same factors, in addition to those associated with these markets’ smaller size, lesser liquidity and lack of established legal, political, business and social frameworks to support securities markets. Because these frameworks are typically even less developed in frontier markets, as well as various factors including the increased potential for extreme price volatility, illiquidity, trade barriers and exchange controls, the risks associated with emerging markets are magnified in frontier markets.