by Olivier d'Assier
To paraphrase Mark Twain, the two most important days in an investor’s life are the day he invests and the day he finds out he was right (or wrong). Every day in between seems to belong to Donald Trump. Welcome to the high-stakes world of binary geopolitics and the revolving door of “I love you, Me neither” meetings between Trump and pretty much any other heads of state.
Investing is forecasting, as we’ve said many times before. In the normal course of business, this means forecasting three things. First, which of the assets in my investment universe will go up or down based on my investment thesis. Second, what will other investors do1. And third, because it is impossible to have an informed view about every asset in the investment universe, what correlation regime should be used to estimate the co-movements of those other assets with the ones we do hold a strong view on? And therein lies the Trump rub.
As was pointed out in our research paper “The Correlation See-Saw”, cross-asset class correlation swings have been occurring at a much faster pace this year given the cluster of geopolitical risks ranging from Brexit concerns, to rising US interest rates, to North Korea, to European domestic politics (Italy, Germany, and now the UK), and a looming escalation in the currently ‘mild’ trade war between the US and pretty much the rest of the world (but especially China – the second largest economy). This means that equities and bonds can be negatively correlated one month, and not at all the next. Commodities and equities can (and have) move from a positive to a negative relationship just as easily. Currencies fluctuate between being positively correlated with equities one month, and with bonds the next. Getting that regime wrong on a stress test will basically invalidate your results, possibly making them directionally incorrect in the process. (Of course, getting the regime wrong in your actual portfolio can also be a big problem, meaning that stress-testing is even more important than usual these days.)
Take modelling a trade war scenario which my colleague Christoph Schon wrote about on Monday. In this example, the risk analysis conducted on April 6th 2018 and using the correlations from the previous 60 trading days as a guide, resulted in a cross-asset class diversification equal to 44% of the total hypothetical portfolio risk if all major risk factor types had been perfectly positively correlated2. Using the same correlations in a risk-off stress test where we shocked selected factors and used those correlations to estimate the co-movements on the rest, gave us the results discussed in his blog post. But what if the correlations change? What if volatility rises and correlations now mean that cross-asset class diversification only shaves off less-than 12% of total (perfectly correlated) portfolio risk, as it did throughout February of this year?
The matrix below shows the change in cross-asset class correlations, with the top/bottom 5 highlighted, between February 16 and April 6 2018 (in less than two months). As we can see the changes are significant, especially for the largest holdings in the portfolio3. Some correlations went from positive to negative. US Equities (30% weight) saw its correlation of +0.75 with International Treasuries (10% weight) and +0.64 with Global Inflation-Linked bonds (5% weight) on February 16 switch to -0.36 and -0.53 respectively by April 6th. Its correlation with Gold (3% weight) went from +0.69 to -0.35. The correlation between International Treasuries (10% weight) and the Japanese Yen (1%), although not a dominant relationship in the portfolio, went from +0.25 to -0.25.
Today, diversification is back to providing some 40%+ in risk-minimization benefit, but it has been on a downward path since November 2016 (see chart below) for Axioma’s global multi-asset class model portfolio.
Investors have been caught4 between the risk-on scenario of strong earnings growth and continued upbeat CEO projections, and the risk-off scenario of a global trade war as geopolitical events unfold; putting on a bullish trade one week only to reverse it with a flight-to-quality the next. This explains the directionless but still volatile markets we’ve seen since the January highs. The ‘question du jour’ on the Trump menu this week is an escalating trade war with just about everyone, but it isn’t being asked unilaterally, in a vacuum of other geopolitical risk events. It is being asked while multiple geopolitical events are occurring at the same time. You have Brexit and the troubled UK government. You have the NATO meeting and Trump’s confrontational stance with his security partners. You have the Trump-Putin summit, the post Trump-Kim Jung Un one. You have rising US interest rates and a flattening yield curve. Etc.
Investors are hoping the current Q2 earning season and CEOs can provide some direction and guidance as to which correlation regime to use when modelling the future5. Regardless, once they stop talking, there is another three months of just Trump talking, and we’re back to not knowing which correlation regime to use. So, until clarity (or sanity depending on how you feel about it) returns to geopolitics, instead of using a single correlation matrix calibrated over a fixed period of time for your stress test, perhaps developing two scenarios representing different correlation regimes and assigning probabilities to each one might be the more prudent (and scalable) way to go. The question then becomes, what are the appropriate probabilities to use? These days, this (also) seems like a Trump question – only he would know. Who said risk management was no art?
Investors have to do a lot more ‘stress testing,’ expert says
Friday, July 20th, 2018
Olivier d’Assier, head of applied research, APAC at Axioma, discusses what’s next for investors as they anticipate the outcomes of trade tensions between US and China. Watch here >
1. Successful investing involves anticipating the expectations of others.↩
2. Total hypothetical portfolio risk if all risk types were perfectly correlated would have been 11.51%, but cross-asset class correlations being less than 1.0 meant the portfolio benefitted from a total diversification of 5.01% (or 44% risk reduction) for a total portfolio (correlated) risk of 6.5%.↩
3. For the full portfolio composition details, see the Multi-Asset Class Risk Monitor report here↩
4. Some might even say they have been in denial…↩
5. Monitor the shape of the US yield curve for signs CEOs have succeeded in unnerving investors.↩
