Correlations show us how assets have moved relative to each other in the past. As multi-asset investors, one of our key objectives is to identify assets that improve diversification. To do this, we try to combine assets with low or even negative correlations. This sounds easy, but can be surprisingly difficult in reality.
Trying to make sense of a correlation matrix for a wide range of asset classes (see below) is a complicated task – too many numbers and too much detail can hide the bigger picture. Looking at individual correlations – between Japanese equities and emerging market debt (EMD), for instance – isn’t particularly important for a multi-asset investor; we need to see how an asset class behaves relative to all the other asset classes we hold (we therefore look at a full column of the matrix) and relative to a total portfolio.
There are newer quantitative methods however that promise more intuitive, graphical access to the data. We call them our correlation galaxies given that asset classes (the dots or 'stars') are clustered and arranged in a way to visually reflect the complex correlation matrix. If we just consider one asset class – we try to place it close to all asset classes that it has a high and positive correlation with, and far away from those assets where it has a low or even negative correlation.
The axes themselves have no particular meaning; we only need to worry about the distance between the stars. The methodology places all stars simultaneously to achieve a close fit in the distances between the dots and the correlations between each combination. In addition, the strongest positive links between assets are shown as dotted, dashed and solid lines.
The above chart has pretty much the same information as the large correlation matrix above but is so much clearer. We see various clusters (galaxies) of assets that are closely linked, as well as asset classes (intergalactic stars) that seem to have no strong links to the rest of the asset universe. There are many interesting conclusions we can draw from this chart, but I’ll just highlight a few points:
While all of the above is very intuitive, there are two main 'issues' here. First, there is a statistical one as the standard correlation assumes that returns are well behaved (follow a normal distribution) which we know isn't true. A possible solution for this problem is to use 'rank correlation measures' such as Kendall’s tau or Spearman’s rho.
Second, the idea that we can simply rely on some historic relationship to predict the future isn’t holding up well in the data. Rolling five-year numbers (below and covered by Daria) show that the correlation within the main asset class 'blocks' is consistently high and relatively stable. But there is huge variation between the main galaxies (between equities, bonds, etc) and some correlations swing from positive to negative and vice versa. So what reduces risk over some periods is adding to risk over another.
A time-lapse video of our analysis shows how the main galaxies (regional equities, bond markets, etc.) stay together, while their relative location and links between the blocks move around a lot. Worryingly, the relatively stable structure clearly breaks during the global financial crisis in 2008/2009 when diversification benefits that investors may have banked on previously didn't materialise. This should be a concern for anyone relying purely on correlations to make investment decisions.
What are the implications of this research? Some may argue that we can continue to use correlations as we are unlikely to see a sharp break in asset relationships over the short term, but that misses the point as it is exactly in these unexpected events (such as the 2008 financial crisis) when we need the benefits of diversification the most.
Our own approach is to combine those asset classes with stable correlations – as this helps to reduce the complexity. But we want to take into account that there may be no stable link between these clusters (which means that we do not rely on the correlations between the clusters, and this is where we deviate strongly from risk parity approaches). Instead, we spend much of our time and effort to understand the changing pattern of behaviours and look for robust solutions. For risk management purposes, we use scenarios and tail risk analysis to test portfolio outcomes when markets go rogue.
This blog comes with a health warning: Higher fee pots die younger. And the UK government is taking notice.
Emerging market assets have long been a source of both potential profit and peril for investors. 2017 saw an incredible streak of capital inflows into emerging market equities, bonds and currencies. Whilst returns are still characteristically volatile, this historically maverick asset class has become more mature and resilient than ever before, as was highlighted during February's market sell-off.