Robert Tattersall, CFA, is co-founder of the Saxon family of mutual funds and the retired chief investment officer of Mackenzie Investments.
We all know that it is possible to drown in a river with an average depth of six inches – and so it is in the investment world, where investors are often surprised by the turbulence of short-term statistics that make up a benign long-term average. Many of the long-term averages that underpin our investment decisions are correctly computed, but they are also the centre point of a wide distribution. In statistical terms, the standard deviation around these averages is huge. As a result, investors are often surprised at how difficult it is to determine if a particular outcome is the result of skill and judgement, or simply the output of random noise.
These problems arise, for example, when assessing the benefit of diversifying across different asset classes to reduce portfolio volatility, or in deciding the appropriate time horizon to evaluate the value-added of an equity-portfolio manager. By chance, the current issue of the Financial Analysts Journal (Third Quarter 2018) has two articles that address these very topics.
In this article, I will review their findings on portfolio diversification and deal with value-added in the equity market at a later date.
The primary purpose of diversifying a portfolio is to reduce the volatility associated with exposure to a single asset class, no matter how attractive the expected return from that asset. In considering a candidate for this role, the first step is to establish the correlation of returns between pairs of asset classes. This measures the degree to which they move in harmony. The correlation can range from +1.0 when two assets move in lockstep, through 0, where the results are independent of each other, to -1.0, where they move in opposite directions in response to the same stimulus.
The ideal diversifying asset classes would have a negative correlation, but this rarely happens in financial markets. So investors are usually happy with a low positive correlation when adding a new asset class to an existing portfolio. A low correlation with mainstream investments is often part of the marketing pitch for new or exotic investment opportunities, such as hedge funds or private equity.
In their article, alarmingly titled “When diversification fails,” authors Sebastien Page and Robert Panariello start with a portfolio of U.S. stocks (the MSCI US Total Return Index) and look to add non-U.S. stocks in the form of the MSCI EAFE (Europe and the Far East) Index. Based on their chart, the long-term average correlation between these two indexes from 1970 through 2017 is about +0.65 – not a great diversifier, but a reasonable basis for expecting a small reduction in portfolio volatility. They then broke down the entire time frame into periods when the U.S. equity market was rallying and also in a sell-off mode before recalculating the correlation during these subperiods.
What they discovered is disappointing: when the U.S. equity market was in a strong upswing, the correlation with non-U.S. markets fell dramatically to as low as -0.17. In other words, the portfolio-dampening effect of diversification worked just fine when you least needed it – when the market was in an upswing. When the U.S. market suffered a sharp sell-off, the opposite occurred: correlations turned positive to the tune of +0.87. This analysis confirms what experienced market traders know intuitively: “In a financial crisis, all correlations turn to +1.0”.
The disconnect in the message from the long-term average correlation and the actual experience at times of market stress was not confined to non-U.S. equities. The authors look at most of the asset classes promoted as portfolio diversifiers: hedge funds, real estate, high-yield bonds and emerging-market bonds. All of them exhibit the same characteristic: correlations are low during market rallies when diversification acts as a drag on the portfolio, and high when the market tanks and the portfolio would benefit from asset classes marching to a different drummer.
The only exception to this phenomenon appears to be equities and long-term government bonds, where the real-time correlation works even better than the long-term average suggests. In bad times for the stock market, bond returns decouple because of a flight to quality, while they are positively correlated with stocks in good times. The authors point out, however, that the stock-bond correlation is difficult to estimate and the current very-low interest-rate environment has pushed both asset classes to a level where a single event could be negative for both stocks and bonds.
Although the research was conducted from the viewpoint of an investor in a U.S. equity portfolio, it is reasonable to assume that the same conclusion would apply in the Canadian financial market: correlations based on long-term averages are not helpful when markets are under stress. As the authors point out, “full-sample correlation is an average of extremes … during market crises, diversification across risk assets almost completely disappears. Moreover, diversification seems to work remarkably well when investors do not need it – during market rallies.”
What lessons are there for the individual investor in this article packed with statistics? The authors offer three suggestions.
First, and obviously, do not rely on full-sample correlations in portfolio construction. If you are presented with a new asset class that will allegedly reduce overall portfolio volatility, ask how the correlations stack up during times of market stress.
Second, do not be seduced by the track record of negative correlations between stocks and government bonds at a time when bond yields are at less than 3 per cent. Shocks to interest rates or inflation can turn this correlation positive.
Finally, they suggest that investors look beyond diversification to manage portfolio risk. Selling equity put options on a portion of the portfolio or simply rebalancing the asset mix over time directly addresses the downside risk by reducing exposure to large losses when volatility is high.
They conclude by pointing out that they are not arguing against diversification across traditional asset classes, just that the traditional measures understate the exposure to loss in times of market stress.