There are some strange correlations out there. For example, a recent article in the Financial Times cited data showing a correlation between share price and CEO manners. Those CEOs who said “please” and “thank you” more often saw their companies enjoy a higher share price. Another study done several years ago by a firm called InsideSales and cited in The Wall Street Journal found that over a nine-year period the average value of a sales deal closed during a new moon was twice that of a deal closed during a half moon and 43% higher than one closed during a full moon.
Big data and artificial intelligence have the ability to unearth ever deeper levels of correlation, but whether these relationships will have investment value remains to be seen. This is interesting to financial advisors and investors because, for those looking for exposure in alternative asset classes, the conversation inevitably anchors around a single data point: correlation.
In measuring correlation, analysts use a mathematical formula to understand return differences in a series of numbers. These return differences are combined with a measure of risk to arrive at single point number that defines the relationship between the two series. Two assets classes that are perfectly correlated have a value of “1”; if the asset classes are a perfect inverse correlation, the value is “-1”. No definable relationship between the asset classes is expressed as “0”.
Asset class relationships can exist anywhere along this spectrum, but note that these results say nothing about where the relationship is coming from, just that it is strong, weak, or non-existent. In fact, correlation provides no insight for end-result performance expectations. This is its first significant flaw.
Nonetheless, investors continue to seek out and invest in “non-correlated” assets, using correlation as a standalone proxy for diversification and risk management. With markets near all-time highs and everything that is going on in the world, there’s a natural desire to want to take some risk off the table, often through an investment in what are mathematically shown to be non-correlated or inversely correlated assets.
Hedging techniques—options contracts, futures, and short or swap exposure, for example—can be used to provide an exposure of -1 correlation. But these can be expensive, and what happens should the market continue to rise? The money spent on hedges is thrown out the window. Institutionally, the cost of carry can be whittled down through economies of scale, but for advisors and their clients, a more appropriate approach would be a portfolio-based hedge that provides returns, yet exhibits natural limitation of downside market participation through the diversification of risk premia. With a portfolio hedge, correlation becomes just one factor to consider, and the primary concern then shifts to another statistical measure of risk exposure: beta.
Beta is a measure of volatility relative to the market, with the market often defined as the S&P 500. A beta of less than 1 means a stock is in theory less volatile than the market; a beta of more than 1 means it’s more volatile. Unlike correlation, beta isn’t limited to +/- 1 so it can provide a more nuanced understanding of the relationship of the equity or asset class to the benchmark.
ETF investors know beta as the measure that is used for factor identification. For example, in a fund that uses a “quality” screen factor in its index, an equity with a higher beta relative to the quality metrics being used would result in that equity being included as a constituent in the fund; those with lower betas would be excluded. Beta is also the measure that is used in the Fama-French factor model, which provides for a market beta, size beta, value beta, etc. “Sensitivity” to the factor is more important going forward compared to the historic “relationship” that correlation would represent.
This brings us to a second flaw in depending on correlation alone for portfolio (or index) construction: correlation does not imply causation; rather, it identifies a historical relationship that may or may not persist and that may or may not have predictive value.
While they may think in terms of non-correlated assets, what investors in alternatives and alternative ETFs generally have in mind is replicating a hedge fund type of exposure, with the goal of generating mostly positive returns and avoiding substantial drawdowns. Multi-strategy funds and ETFs are one vehicle for this. These funds seek to navigate the various risk premia in the market, including systematic equity risk, providing a hedge fund-of-funds type of exposure. They may do this in part by incorporating both beta and correlation in the strategy or, in the case of an ETF, in the underlying rules for the index.
This is not to say that correlations can’t be useful. Consider an investment approach that employs multiple types of risk; one may be general market risk, another might be volatility, and yet another a long/short strategy. If many of the other risks do not provide return, or maybe even counteract each other, portfolio return would be a function of general market exposure – it would show high correlation to the market, as well as low beta. If the other return sources kicked in, the correlation to the broad market would naturally go down. Going forward, the market may provide negative returns, and the other strategies may counteract and provide positive returns, which could look like negative correlation, without changing the strategy—the goal of multi-strategy exposure.
The efficacy of this multi-strategy approach is evident during drawdown periods. While there have been no major bear markets in the nine plus years since the first multi-strategy ETF entered the market, there have been six market sell-offs of at least 5 percent since that fund was first introduced. (Full disclosure: my company, IndexIQ, was the sponsor of that ETF, the IQ Hedge Multi-Strategy Tracker ETF, ticker symbol QAI). In those instances, QAI experienced drawdowns that were 41.5 percent of the drawdown experienced by the S&P 500. Overall volatility was lower as well.
Investors love numbers, and new correlations surface all the time. Some will be useful, others not. (Wet sidewalks probably don’t cause rain, for example.). On the evidence to date, measuring correlation among asset classes alone is not sufficient for alternative asset investors to consistently achieve their risk management and diversification goals. Adding beta significantly improves the chances for success.
Dan Petersen is product manager at ETF provider IndexIQ.