Asset Class Correlations: It’s All Bunk
June 13, 2011
Asset Class Correlations
Ever hear about asset class correlations? Well, it’s all bunk.
The idea of asset correlations is this: A 1.0 is perfect correlation. Meaning if Asset X moves up then Asset Y moves up in lockstep. A correlation of -1.0 is perfect negative movement. If Asset Y moves up then Asset Z moves down. Usually the range of correlations falls between the perfect positive correlation of 1.0 and perfect negative of -1.0.
You see this in investing books all the time. Stock asset classes X and Y have a correlation of 0.67 while bonds A and B have a correlation of 0.22, etc. You just throw them together and you get instant diversification! It looks so scientific being two decimal places and all.
Well, allow me to let you in on a secret: This information is worthless at best and dangerous at worst.
A World of Correlations
In the investing world many decide to build a portfolio using historical correlation data. They look at all of these assets over the years and assign a correlation number between them to try to figure out a relationship and how much of each you should own.
For instance over the past 30 or so years:
Total Stock Market and the Total Bond Market have a correlation of about 0.34. Meaning that some of the time are they moving upward at the same time but sometimes they aren’t.
Total Stock Market and Total International Stock Market have a correlation of 0.66. A stronger correlation indicating that if one is going up the other probably is, too.
Total Stock Market and Gold have a correlation of -0.27. Meaning that if the stock market is going up in value that gold is probably going down and vice versa.
Now you take a bunch of assets with different correlations and you are hoping that when one zigs the other zags. This diversification effect means over time that as one is gaining in value another is falling. During good markets you may give up some gains, but in down markets you have some assets that should do well to offset the losses.
Good theory, but normally used poorly in practice.
The reality is that asset class correlation data is irrelevant and can get you in trouble. In 2008 for instance many people held assets that supposedly had low correlation historically. Yet, when the markets crashed the correlations went up sharply and they all went down together. The diversification selected failed and large losses followed. Many claimed that “Diversification failed in 2008.”
Diversification didn’t fail. What failed are how portfolios were built to diversify the risks.
There are two primary failures with looking at asset class correlation alone:
1) Looking at correlation data without considering the underlying economy at the time covered.
2) Correlation data encapsulates big blocks of multi-decade returns which loses visibility into the specific events happening under the covers.
Let’s talk about these two things.
It’s Hot in Miami and New York – Who Cares?
What about the underlying economy? Why is this a big deal?
Let me explain using the weather. Below is a list of average temperatures in Fahrenheit in three cities:
I run this through my spreadsheet and these are the correlations I receive:
New York to Miami Correlation: 0.99
New York to Sydney Correlation: -0.98
What this tells me are that the temperatures of New York and Miami are highly correlated and the temperatures in New York and Sydney are not. When temps are going up in New York they are going up in Miami and when they are going down in New York they are going up in Sydney.
But why should this be? To many I see talking about asset class correlation the analysis stops here. They would simply say (if you built portfolios from cities) is:
“Just buy a little New York, a little Sydney to diversify and some Miami to boost returns.”
But when you ask a more fundamental question about why these cities are or are not correlated you may not get an answer. So you look further and think that perhaps it’s related to the months of the year. Hmmm…now that’s interesting.
Let’s add months to our chart:
Ok that adds some more context. Clearly you see the month of September, then October, then November, then December. What’s this? Why are the temps in New York and Miami falling but Sydney is going up? The months are all the same.
In Some Places, Santa Claus Wears a Bathing Suit
Let’s look deeper. Let’s ignore the months and look at the seasons instead. As you probably know, the Northern and Southern hemispheres have seasons that are reversed (When I was in New Zealand for instance, Santa Claus in December was shown to be in summer shorts carrying a surfboard and not wearing a big heavy coat):
Winter: December, January, February
Spring: March, April, May
Summer: June, July, August
Fall: September, October, November
Summer: December, January, February
Fall: March, April, May
Winter: June, July, August
Spring: September, October, November
With more context this correlation data now makes sense. The temperatures fall in New York and go up in Sydney because of the seasonal differences. It has nothing to do with the cities being correlated, the months of the year, etc. The temperature changes are due to the seasons and if I ran a correlation of temperature changes to seasons of the year you’d find that it is almost a perfect 1.0 match regardless of location on this planet.
What’s the Point?
Asset classes don’t move because of each other. Asset classes move because of the seasons in the economy.
Asset class correlations without an economic explanation why they move is dangerous. This is the problem of using asset class correlations alone to build a portfolio. It’s also why you should ignore tables of asset class correlation data you see in books, articles and other places. If they aren’t tying those assets to the economy then it’s all wasted ink.
Changing Seasons of the Economy
We come then to a core concept of the Permanent Portfolio. It is not built around this idea of asset class correlations. It is built around the idea of looking at the changing seasons of the economy:
Bonds don’t move up sharply in price because stocks moved down. Bonds move up because deflationary forces make it a good investment. Gold doesn’t move up because bond prices are falling. Gold moves up when people think inflation is becoming a threat. Stocks don’t move up because gold prices have come down. Stocks move up in price when people think the economy is going to be prosperous.
This is the difference between the Permanent Portfolio allocation and others. The assets chosen respond the best to the four conditions of the economy outlined above. Those assets are (for US Investors) Stocks, Long Term Treasury Bonds, Treasury Money Market Funds and Gold.
A Five Foot Deep Creek Can Still Kill You
So what about the second error I talked about above? The one where I said asset class correlation data masks some truly ugly details by bundling everything up over multiple decades?
Well it’s just something that averages do. They’re average. Taken as weather you may think the average January temperature above for New York is 30.7 degrees. However what you don’t see are those nasty years when it was in the low twenties with plenty of days in the single digits or lower.
It’s related to that old saw in statistics that a six foot tall man that can’t swim can still drown in a creek that is an average depth of five feet. What this average ignores is the creek is one foot deep in some places but 10 feet deep in others. It’s these extremes that can really burn investors and you don’t see them when you look at some correlation number that spans 40 years. The extremes are buried in the data and until you look you won’t see when the correlation of assets suddenly went to 1.0 perfect and huge losses incurred in certain years (witness 2008′s losses in what some thought were diversified investments).
Correlations Don’t Change Over Time
Incidentally, the changing of the economy is also why advocates of correlation data say “Correlations change over time.”
Actually, correlations don’t change. At least not when you look at the data from an economic cycle standpoint.
Stocks and bonds don’t suddenly become correlated out of the blue. What these folks are seeing is the economy shifting underneath that causes periods of over and underperformance for assets. The correlations though are not “changing.” The only thing changing is the time period they are analyzing and what the economy was doing and they are coming away with the wrong conclusions. Statistical tools are powerful when used correctly, but here they fall flat on their face because they are being mis-applied.
“Correlation does not equal causation” - Asset class correlations provide no explanation for cause and effect. Only tying the assets to the economy explains their price movements.
An Epiphany – At Least for Me
I took time to explain this because it was really an epiphany for me personally and answered many questions about how diversification can be made to work by applying economic understanding to the problem. In fact, I firmly believe it is a serious and grave error to not consider economic impacts on the asset classes you own and rely on correlation data only.
So ignore this asset class correlation stuff. It doesn’t answer the questions you need to have answered about investments and can get you into trouble by supplying you with a false sense of security. Instead, own assets that correlate to what the economy underneath is doing. This is where the power of diversification can really work for you.