Risk parity – the idea that risk, measured by volatility, should be allocated equally across the major asset classes – has been one of the major investment fads of the last 5 years.
The philosophical premise of risk parity is that correlation is more predictable than return.
The most common articulation of Risk parity works like this:
1. There are four macro regimes: high inflation, low inflation/deflation, high growth and low growth.
2. Different asset do well in each regime (this is the correlation point): nominal bonds do well in low inflation/deflation, commodities and index-linked bonds in high inflation, equities in low inflation and high growth.
3. If we cannot forecast which regime will prevail at any point in time, we should allocate equal amounts of risk across these assets, which diversify each other across regimes.
Portfolio construction in risk parity strategies often deploys leverage – typically of bond positions – in order to equalise the volatility of pure rates exposures with that of equities and commodities.
It doesn’t take much though to undermine this intellectual framework.
There are more than four seasons
The idea of “all weather” investing, assumes that we know all the seasons. And that we know the implication of each regime for all asset class returns. This is patently false. As we saw from 1998 to 2008, commodity returns can be high in a world of low inflation (which implies that the opposite is possible!); index-linked bonds have been one of the best performing assets of the last 20 years despite a secular decline in inflation; and nominal bonds in peripheral Europe behaved like risk assets in a deflationary shock (the Euro crisis).
Assets do not behave predictably and regimes are not easy to categorise. Take an extreme example. The main risk to an equity investor, on any time horizon, is not inflation or growth, but ownership of corporate cash flows. Sustained losses to equity investors occur because either insiders (managers), or governments, take the value. The principal-agent problem and property rights are much more fundamental determinants of the equity risk premium than volatility or correlation with cyclical factors.
The performance of equities during inflation also depends on your time horizon, and the cause of the inflation. Most thinking on this is biased by a sample of one (the 1970s), and cases of hyperinflation. (Hyperinflation is almost always a symptom of a failed state and often a suspension of property rights – usually bad news for all assets: owning agricultural land in Zimbabwe has hardly provided a hedge!).
Even in the case of the 1970s, the effect of inflation on asset returns is not straightforward. Someone buying the S&P500 in the 1970s and taking a thirty-year view, will have done extremely well. The macro chaos of the 1970s made equities very cheap, but did not permanently destroy earnings power.
A hidden assumption in all correlations and measures of volatility is embedded in the frequency of the data. A correlation which holds using daily, monthly, quarterly, or annual data, may have the opposite sign when we use 10-yr or 30yr frequencies. There is no a priori answer about which frequency to use, it depends on the time horizon, and the preferences and liabilities of the investor.
The predictability of asset returns
A common mistake in investing is to assume because value does not always predict returns, it never predicts return.
In a classic paper, John Cochrane, professor of finance at Chicago, argues that one of the most robust conclusions from decades of empirical work in finance is that yield provides the only predictable component returns. This is entirely intuitive. Take an extremely clear case: the yield-to-maturity on a 5-year treasury, is almost guaranteed to be the 5-year nominal return. Conchrane shows this is the case across all asset classes: dividend yields predict returns, not dividend growth, rental yields predict returns to property not rental growth, carry predicts currency return, not spot rate rate moves, and sovereign spreads predict returns not defaults.
Importantly, yield only predicts part of the return. From Cochrane’s purely empirical view, most of the rest of the variance is either cyclical, or unpredictable.
I would take a subtler view than this and argue that much of the time yields have little information, because assets are reasonably priced and other factors will dominate short-run returns, but periodically valuations are sufficiently extreme to dominate. A price earnings multiple of 30x or 4x for an equity index has huge information about prospective returns, so too does a negative real 10-year bond yield – such as prevailed earlier this year.
Far from being unpredictable, the biggest gains and losses in asset class returns appear to be very predictable.
Value affects correlation
To be clear, the periodic predictability of asset class returns is at odds with risk parity, because it implies that one should actively vary the amount of risk allocated to different asset classes, when valuation signals are sufficient to have information.
But can we also say anything about regimes and correlations? Despite the caveats expressed above about the variety of regimes and the unpredictability of correlations, I think certain correlations can be identified with a reasonable degree of confidence. Although it is extremely difficult to determine what regime will prevail in the distant future, understanding the current environment and the correlations likely to prevail may be more fruitful.
Risk-parity’s success and popularity has really been caused by a single regime-shift: a 20-year period were equities and bonds have delivered high returns, with bond returns highest during phases when equity returns are negative. A negative bond-equity correlation at short frequencies combined with high positive returns to both assets over the entire period creates very high measured Sharpe ratios, particularly if the bond positions are leverage.
Although the extent of these returns cannot be repeated, because yields in developed markets are now too low, the regime that produced the negative bond-equity correlation is likely to persist. Low inflation looks like an uncontentious assumption. After all, the downward trend in OECD inflation trend of the last 20-years (fig 2), has continued despite an oil price shock (2008) and huge central bank balance sheet expansion. It is true that periodic fears over monetary tightening, such as the recent tapering episode, can threaten risk assets. But these effects are likely to be temporary. Most importantly, the enduring bond equity correlation in a world of low inflation is likely to be negative, particularly in phases where there are sustained threats to growth and profitability. In other words, in the genuine tails of distribution of equity returns, it still seems reasonable to expect bonds to provide some protection.
But the degree of protection bonds provide is determined by value and the term premium. Ironically, as many are now rethinking the merits of “risk-parity”, the central correlation that created its track record, is now attractively priced. The long end of the yield curve is now paying elevated levels of term premium – which is precisely when it is cheap as a diversifier against cyclical growth risks (see fig 3).
Conclusions: sample-of-one investing
Investment fads which sound scientific have usually benefitted from something simple, and unrelated to the “science”. The spectacular Sharpe ratio of a long global equity and leveraged bond portfolio was caused by a one-off shift in regime – namely a structural decline in real interest rates and inflation. If real interest rates decline on a trend basis, bonds will deliver great returns. In an environment of low and declining inflation it is also to be expected that bonds and equities will have a negative correlation during cyclical shocks.
These returns are no longer repeatable for the simple reason that nominal interest rates can go from 6% to 0%, but it is virtually impossible for them to go from 0% to -6%. This simple observations reveal two key points: returns are predictable (cash and short dated bonds will deliver very low nominal returns and possibly negative real returns); and valuations affect diversification and correlation – a JGB yielding 0.6% provides virtually no protection against a Japanese recession or collapse in equity prices.
Importantly, similar returns could have been generated without index-linked bonds or commodities. And for most of this sample period, growth has been relatively stable (bar during recessions) and inflation extremely stable (albeit trending lower). The winning strategy was not a complex all-weather portfolio diversifying investors against a host of risks, but rather a combined equity and levered bond portfolio in a world of relative-stable trend growth, and declining real interest rates and inflation.
Correlation is not fixed or easily predictable and there are so many environments that the idea of creating an “all weather” investment strategy is a fallacy. Governments can default; commodity prices can collapse in the face of technological innovation; apparently fundamental correlations such as that between real interest rates and growth are not fixed.
This year has highlighted many of these points, and many are now questioning the merits of risk-parity and “all-weather” investing. It’s about time. The irony is that the core correlation which created this fad now looks attractive.