Asset Class Correlations are Bunk

[craigr]

Ok everyone, here is my opinion on asset class correlations often touted in investment literature. What say you?

https://web.archive.org/web/20160324133 ... -all-bunk/

-- Craig

[LifestyleFreedom]

I learned investing from the point of view of modern portfolio theory and the efficient market theory.  The basis for this theory is that a portfolio of individually risky assets is less risky (as measured by volatility) when the individual assets have a less than perfect correlation with each other.  I still believe this theory is "largely" true and has a place in my portfolio.  But correlations do change at times for various reasons.  Back in the 1990s, for example, the correlation between U.S. and large-cap international stocks was around a third.  Since 2000, the correlation is closer to unity.

Why?  The conventional wisdom is that a lot of investors (professional and individual) learned about non-correlated asset classes and bought baskets of non-correlated assets to be diversified.  When a large amount of money gets invested in a certain style, that style often stops working.  If this is true and a large amount of money gets invested in the permanent portfolio, this style will also stop working.  Once a style stops working, investors become discouraged and abandon the style, so it starts working again because only the true believers are left.

Nearly all assets became correlated in late 2008 and early 2009 because most investors rushed for the exits in a panic.  They sold at any price.  What is never discussed in the financial press, however, is who were on the other side of those transactions.  The investors who knew what they were doing (or were incredibly lucky) ended up making a lot of money by being greedy when everyone else was fearful.  So it's possible for some investors to make good money when the market crashes.

Another example of non-correlated assets becoming highly correlated when markets panic is the story of Long-term Capital Management.  During normal markets, the assets (foreign bonds) owned by LTCM were non-correlated.  But when the Asian Contagion hit in 1998, investors decided to sell first and ask questions later.  The market plunged and LTCM, which was highly leveraged, had to be bailed out (read When Genius Failed for the story).

So I believe asset class correlations are relevant, but are not the end-all/be-all for a portfolio.  (I feel the same way about the permanent portfolio -- it has a place in my portfolio, but I will not make it my entire portfolio.)  I will continue to use a mixture of investing styles because I don't trust any particular one to work all the time.  I don't have any hard data on when various investing styles might fail, but I feel in my gut that it's better to use more than one style just in case the next financial panic is of a type the world hasn't experienced before (and there is no guarantee that the styles I use would survive such a panic, but that's the way I'm going to bet).

[moda0306]

Clive,

The reason any 2 PP assets may or may not correlate can be a simple explanation of the fact that you can have deflation (or less-than expected inflation) at the same time as prosperity, or you can have it at the same time as a recession.  LTT's performed wonderfully from 1982-2008, but that doesn't mean they were under the same prosperity vs recession vector.

Further, I have said before that due to the very "recessionary nature" of gold and LTT's, you can have deflationary recession and have gold go up (2008) and you can have inflationary recession, but yield-spread hikes will keep your LTT's from getting absolutely lambasted (1970's).  Also you can have disinflationary prosperity and have gold go down (1982-1999).

[Tortoise]

Great article, Craig! I really enjoyed it. Your analogy between economic cycles and the seasons is a great way to illustrate the difference between "random" price correlations on the one hand and causation on the other.

The mainstream ivory-tower theory of asset prices as independent, memoryless, and completely random processes with mysteriously time-varying correlation coefficients between them always struck me as a bit hard to swallow. Reality just isn't that neat and tidy. You mentioned Mandelbrot and Taleb in a comment below your article; they helped me, too, to understand why the dominant paradigm of clean-shaven and well-behaved means, standard deviations, and correlations in financial mathematics is mostly bunk.

[AdamA]

Great article, Craig! I really enjoyed it. Your analogy between economic cycles and the seasons is a great way to illustrate the difference between "random" price correlations on the one hand and causation on the other.

I agree, but I don't think it necessarily means that asset correlations are "bunk," because there are some asset classes that do have strong correlations (like those in the PP). 

The problem arises when investors look back at assets or sectors that have correlated in the past and assume that they will correlate in the future, when there is no real economic reason for them to do so. 

I would argue that one could run into the same problem trying to use a theory that made economic sense, but had no numbers to back it up. 

Having said that, I'm hard pressed to name any other group of assets that correlate as reliably well as those in the PP. 

[Roy]

As is often the case, things are both simpler and more complicated.  

It is always the overall portfolio that matters most.   Many times equities decline together—which is the problem that underlies the "bunk" accusation.  The popular business of having multiple equity asset classes ("diversification"), in the belief that those correlations will float a portfolio, is indeed a flawed way to view things.  Among themselves, the equities have not had the same correlative portfolio value as the addition of a significant amount of high-quality fixed income, no matter how cutely-divided the equity pie charts look.  There are exceptional years where some equities rise when most decline, but still.  

Looking at conventional portfolios, the most significant correlations have been between any Equities and Short or Intermediate Treasuries.  Over the last 40 years, any combination of equities totaling say, 30-40% , when coupled with plain 'ole Intermediate Treasuries, have done quite well in weathering the varied storms—including 2001-2002 and 2008.  One can quibble about real returns in the hyper-inflationary years (where the PP had higher returns than even the best conventional ports), but the fact remains a conservative approach using high-quality fixed income did fine.  In that basic way, correlations are not bunk and matter greatly.

I don't think the four economic climates tells us all we need to know about the PP or anything else.  As we have discussed repeatedly, Gold, for example, can have impact (panic and monetary) beyond just its stated inflation environment protection, yet sometimes does not confer much inflation protection at all—even across decades.  It has recently done well in DEflation.  And if you have 25% of it, that's a lot of variance potential that is not as neatly defined as assumed.  

Of course, the combination that makes the PP a portfolio is untypical.  But, the PP also works for sound economic reasons pertaining to risk and expected return.  This basic economic fact gets buried because most folks get hung up on its volatile, parts, its symmetry, etc.  That is, the two Treasury types in the PP have varying degrees of expected return for risk taken.  Obviously, the equities  have this in the greatest degree.  That's three asset classes of the PP behaving—as expected—in a basic way. (Backtesting simply demonstrates that economic base, as it does with anything else.)   In that fundamental sense (risk/return), the PP is similar to other concepts.  Put that all together with a low Beta exposure, and you get a lot of the explanatory value of the PP without needing to shoehorn it into only economic climates—even if it often can be reconciled that way. The Gold is what differentiates it, powerfully, but don't look past the plain economic underpinnings of risk and expected return that drive three-fourths of its components.

[craigr]

Thanks for the comments everyone.

[AgAuMoney]

Ok everyone, here is my opinion on asset class correlations often touted in investment literature. What say you?

https://web.archive.org/web/20160324133 ... -all-bunk/

I think your "very much PP related" articles should point people here for discussion. 

Overall I agree with your thesis and generally well written article.  I do disagree on a couple of minor points.

First point is ignoring crisis, and that leads to point two, using 2008 as an example of changing seasons and (dis)proving correlation.

Crisis might be thought of as a "black swan" event, or a non-seasonal disruption.  If talking about the weather in spring or summer, we'd talk about the freak hailstorm that dropped temperatures 50-60F in an afternoon and buried the town in hailstones.

2008 was such a financial system crisis, not of cash but of financial instruments.  Every financial instrument and asset was sold in an attempt to raise cash, dropping the price of all those things until the market could clear.  For a time, correlation was essentially 1 between stocks, treasuries, and gold and was positive with essentially all other assets.  Not because of the economic climate, but because of deleveraging and fear of future deleveraging.

And now with this perspective we know that we can look forward to the freak storm that froze wooly mammoths with flowers in their mouths and kept them on ice for hundreds of years.  What would that freak storm be in the financial/investing world?

Perhaps the collapse of the world's reserve currency.  That could precipitate a crisis when holding that currency is just insane, and asset prices normally priced in terms of that currency are in total disarray.  Then what happens to historical correlations?  Just like any crisis, it goes out the window.  If cash isn't the thing in that crisis, what will be the thing?  Maybe gold?  Maybe food or land?  Maybe the Yuan or the Canadian or Australian dollar?

Hopefully we'll never find out.

[AgAuMoney]

there are some asset classes that do have strong correlations (like those in the PP). 
...
Having said that, I'm hard pressed to name any other group of assets that correlate as reliably well as those in the PP. 

I'm a bit confused by your phrasing.  The assets in the PP do not correlate with each other.  They do respond to different economic conditions (or 'seasons').  Perhaps that is what you meant?

Correlation is always measured between two things as a value between 0 and 1, and the direction of correlation is given with the sign (+/-) of that value.  Strong correlation is +/-1.  Perfect positive correlation is +1 and perfect negative (inverse) correlation is -1.  No correlation is 0.

The assets in the PP are all historically very, very close to 0 with respect to each of the other assets.

Ideally they would be all 0's:

cashtbondstockgold
cash-000
tbond0-00
stock00-0
gold000-


That's the ideal, but in reality they sometimes, rarely but sometimes do move in unison.  Historically that is very rare, 2008 being one of those rare exceptions.

That lack of correlation is one of the strengths of the PP.  The other primary strength being the way the assets do respond to the 'seasons' of the economy.

[AdamA]

I'm a bit confused by your phrasing.  The assets in the PP do not correlate with each other.  They do respond to different economic conditions (or 'seasons').  Perhaps that is what you meant?

Yes, it's poorly phrased.