When is diversification important?

There has been some minor buzz in the blogosphere over a study showing that the benefits of international equity diversification are falling (see comments from Paul Kedrosky and FT Alphaville). Moreover, I recently had a discussion with another investment professional about issues surrounding portfolio construction and diversification. His comment to me was a textbook one: “I would rather try and get… reduction in volatility through more diversification.”


Diversification when you need it the most
To someone as simple-minded as me, diversification during "normal" periods is nice and the higher correlations exhibited by global stock markets is a minor concern, but the most valuable form of diversification occurs during panics and crisis episodes - when virtually all asset classes go down. That's when you want something in your portfolio to save you.

Consider example of the crisis in 2008. During that and other periods of financial stress, the correlations of seemingly uncorrelated asset classes tend to converge to 1. Balanced fund portfolios, which were advertised as to be sufficiently diversified to withstand these kinds of shocks, were down 20-30%.

Max Darnell of First Quadrant addressed this diversification problem by explaining that you have to take a bet somewhere. Even though you may be diversifying your asset betas across asset classes, the fact that you are taking on beta generally means that you are making a bet on risk [emphasis mine]:

In short, diversification is not intended to be a tool for risk avoidance. Rather, it is meant to be used as though it were an acid that dissolves away impurities, i.e., uncompensated risk, leaving a pure risk that is more desirable principally because we are rewarded for holding it. The remaining risk will be risky. Otherwise, we wouldn’t be compensated for it.

Correlation isn’t causality
I wrote before that you shouldn’t confuse correlation with causality. Standard MPT style correlation analysis of asset returns are based on statistical correlations. Statistical correlations will tend to move around. Most critical to downside risk protection, don’t expect statistical correlations to hold up in accordance with historical experience during periods of extreme global volatility.

I prefer to look for diversification more analytically. For example, hard assets tend to perform well during periods of rising inflationary expectations, but default-free fixed income assets such as government bonds will perform poorly. Conversely, we can expect that during a credit crisis, which is associated with heightened default risk, prices of default-free government bonds with good credit will rise, while hard asset prices would be under pressure.


Alpha from beta?
Supposing you had a model that could identify macro-economic regimes, then a strategy of buying hard assets during periods of inflation; default-free assets during periods of deflation; and equities during periods of benign macro-economic risk would make good sense.

That would be creating alpha from diversification beta.