The recent financial crisis unhinged many a conviction when traditional risk models revealed weaknesses and the concept of diversification, the cornerstone of portfolio theory, collapsed at the time it was most needed.
In the fourth quarter of 2008, correlations peaked as assets, previously thought to be unrelated, moved in similar directions. As correlations approached one – the point where correlated assets show identical behavior – Harry Markowitz’s concept of holding imperfectly correlated assets to limit portfolio risk without reducing returns seemed rebutted.
Average correlations between equity markets have increased over the last 20 years, as shown in a 2008 risklab study (r The study examined market behavior based on the weekly returns of major indices. Risklab analyzed how correlations behave over time, focusing on equity markets. More volatile than bonds, stocks are the main contributor to the overall volatility of a portfolio.
In addition, the study found evidence that correlations increase during market downturns, a reflection of dangerous herd behavior displayed during critical market periods. To verify this observation as a statistical basis, risklab divided the last 20 years into periods of “normal markets,” economic turmoil and sharp downturns.
The principal factor underlying rising correlations is globalization. Today, market information is disseminated much faster and boundaries between asset classes are becoming blurred. Diversification, part of the modern portfolio theory developed from Markowitz and his seminal paper
Portfolio Selection (1952) needs to be rethought to fit a computerized world if it is to remain valid and assist us to escape the maelstrom of depreciation in one particular asset class. Investors need to do justice to the stochastic nature of correlations. The question is how?
There was no reason for the coupling we saw in 2008 except trading behavior Prof. Rama Cont
By diversifying not only across markets, but also across investment strategies, says Rama Cont, associate professor at
Columbia University and director of its Center for Financial Engineering. “Portfolio managers need to acknowledge and factor in that correlations have a dynamic component due to feedback effects from investment strategies used. “There was no reason for the coupling we saw in 2008 except endogenous correlation generated by trading,” says Cont. “At the time, many market participants held large exposures to subprime securities, and when these started to lose value, they had to sell their most liquid assets, which happened to be stock holdings.”
To anticipate such scenarios in the future, risk management has to move to the next level, Cont argues. Correlations are not fixed parameters for which the only issue is to estimate them, but dynamic quantities which react by rising if liquidity drops rapidly. “Diversification results in coupling across markets, and actual correlations significantly differ from what most market participants would expect. The point is they are dynamically influenced by asset managers’ behavior in the market.”
If portfolio managers want to continue to use diversification in a globalized world of instant information, they have to internalize the twofold nature of correlations: Fundamental correlations can be considered stable and are a result of common economic factors. More importantly, dynamic correlations, the second component, are influenced by market participants’ trading strategies and are inherently unstable.
Dynamic correlations could be factored into risk management if traders diversified across investment strategies so as to take trading patterns and crowding effects into account. “Portfolio managers should monitor not only returns and volatilities, but also trading patterns in the market.” The approach is not entirely new, Cont says, as traders already consider their peers’ behavior within single asset classes.
Yet, the benefit of this information is lost on a portfolio level. A crowded trade might look momentarily profitable, but can turn out to carry significant tail risk, if, due to exogenous reasons, traders have to sell similar portfolios at the same time. “After the fall of Lehman Brothers, correlations rose and significantly dropped when the banks’ sell-off stopped. This confirms our model,” says Cont.
His diagnosis is not limited to the US financial markets. The 2000 Asian crisis created similarly unexpected correlations. Asian and Brazilian markets moved in similar directions, as emerging market funds started selling their most liquid assets, which happened to be Brazilian. “This created a strong coupling a posteriori, although it would have been almost impossible to detect these correlations a priori based on historical data.”
Indices observing various market strategies and depicting the crowding of a given trade – although difficult to establish – could act as early warning of a future crisis, says Cont. This would not invalidate Markowitz’s approach, but rather enhance it. “The portfolio theory provides an operational framework to make decisions based on correlations. Markowitz never suggested using historical data or sample covariance as a measure of risk,” Cont explains.
“By doing so, the risk measure has to be off the mark as the relevant figures – dynamic correlations and feedback effects due to traders’ behavior – are omitted.” If, in the next crisis, traders deleverage simultaneously without having established a way to identify dynamic correlations, Cont predicts correlations will approach one as in November 2008 – and result in the same levels of grief.