Four-year-olds are prone to asking, “Why?” Why does the sun set? Why does the car need gas? Why does the canoe float? Why did God create the animals? Why do I need to go to bed?
As we age, these questions persist but change in tone, losing their innocence. Why do terrorists attack? Why is the climate behaving this way? Why has the economy faltered? Why do World Cup referees miss the obvious?
The investment industry expends a great deal of energy trying to understand why certain conditions have occurred in the investment markets and how those conditions can help predict future activity in those markets. For the last 30+ years, finance academics and analysts have evaluated “correlation” to explain how action in one part of the investment markets causes action in another part. Correlation is a calculation intended to indicate that if “X” happens to the US bond market, then one can expect “Y” to happen with US equities, if “X” happens with US equities, then “Y” should happen with Non US equities, and so on. A correlation of +1.00 indicates that two asset classes move together, 0.00 indicates they move independently, and -1.00 that they move inversely.
This would be an aside of little interest to most people except for one critical piece of information: Almost all financial planning tools and almost all asset allocation modeling software assumes that one can predict the correlation between asset classes with a reasonably high degree of accuracy and that correlation is reasonably consistent over time. Advisors to pension plans, insurance companies, endowments, foundations, and private investors often use analytic tools that rely on predicting the correlation between asset classes to determine how best to strive to meet their objectives.
Yet, predicting correlation between asset classes in a complex and ever-evolving economy is not easy. Most applications use historic correlations between asset classes to estimate future correlations. But here at Syntrinsic, we have a problem with that. Our analysis finds that historic correlation between asset classes is not consistent, predictive, or otherwise reliable as a measure of future correlation. In short, the past does not predict the future.
Consider the example of US bonds and US equities (stocks). If one calculates the rolling 12-month correlation between these two asset classes over the past 25 years (1986 to 2010), one would expect to see some consistency of that correlation. However, the correlation ranges from +0.89 to -0.78. Given that the highest correlation possible is +1.00 and the lowest is -1.00, it would be hard to argue that there is a meaningful data point that can be used to describe the historic correlation between these assets. And yet, many analysts actually use historic correlation to estimate future correlation.
Some would argue that 12 month correlation is insufficient, and that the measure of correlation really requires longer sweeps of time. So we ran our analysis for 2, 3, 4, 5, and 10 years as well. While the result is smoother than the 1 year analysis, it is hardly reassuring for those who think of correlation as a reasonably fixed and predictive data point. The 10-year analysis—which is the smoothest—still ends up with a range from +0.53 to -0.23.
Most importantly, whether short-term or long-term data sets are being used, try to find a period when the backward looking calculation of correlation would have predicted the next 5-7-10 years. You won’t likely find one. For example, if one took a snapshot in December 1997, when 10-year correlation was at +0.53, and used that to predict correlation for 1998-2007, then one would have been terribly disappointed, as that next ten years had a correlation of -0.22.
For some, these observations may seem technical, obscure, and even irrelevant; however, deconstructing the use of correlation in financial planning ties into many major fiscal policies. Just consider the intense debates raging about the funding of municipal and corporate pension plans—much of that debate is linked to models that assume forward looking behavior based in significant part on historic correlations. If those correlations are not adequately predictive, then what?
This is a great time for asking “Why?” at every level of business, finance and investment. Why were certain investment models created in the first place? Why did those models persist through the boom years of 1981-1999? Why have those models failed to meet expectations in 2000-2010? And obviously, what are we planning to do about it?
Today’s four-year-olds are depending on our ability to address these questions with candor and rigor. It would be embarrassing if 30 years from now they were asking us why we kept using models that don’t work .
Even FIFA is considering adding some sort of instant replay to the World Cup. That just goes to show that anything is possible if one asks the right questions.