Skip to content

The Myth of Diversification Reconsidered: summary and key takeaways

The paper argues that earlier studies measured downside correlation the wrong way, because they only look at periods when both assets fall. Conditioning on one growth asset instead changes the picture for low-correlation pairs, especially stocks and bonds. Treasury bonds come out as the most useful complement to equities.

Kinlaw, W.; Kritzman, M.; Page, S. et al. · 2021 · The Journal of Portfolio ManagementSummary updated 20 November 2026
Published2021Journal article
Asset focusMulti-assetDiversification / correlation
TypePortfolio MgmtStock-Bond Correlation
Reading time4 minthis summary

The Myth of Diversification Reconsidered at a glance

The question
How should investors measure whether an asset really diversifies their main growth engine in down markets versus up markets?
Dataset & period
Six asset classes: US stocks, Treasuries, US corporate bonds, commodities, foreign developed stocks and emerging market stocks. Monthly returns from January 1976 to December 2019; emerging markets start in January 1988. A simulated example uses 500 returns for two assets.
Method
The authors compute upside and downside correlations for a given threshold, measured in standard deviations from the mean. They compare the observed values with those expected from a bivariate normal distribution, so that mechanical effects of splitting the sample are removed. They contrast conditioning on both assets with conditioning on one asset only, and summarise each pair with an average excess downside minus excess upside correlation. They then run a full-scale optimisation with a loss-averse utility function and compare it with mean-variance.

Key findings of The Myth of Diversification Reconsidered

  • Conditional correlations fall by construction as the threshold moves further out, even for a normal distribution. In the simulated example, a full-sample correlation of 0.50 becomes 0.33 when conditioning on one asset at a zero threshold.
  • Conditioning on both assets gave 0.27 in the same example. The double-condition approach shifts the reading by design and can mislead.
  • For US and foreign developed equities, the full-sample correlation is 0.66, and the asymmetry summary metric is 0.32. Both conditioning methods agree here, since most observations sit in the two same-direction quadrants.
  • For US equities and US corporate bonds, the methods disagree. Conditioning on both assets implies unusually high downside correlation. Conditioning on US equities alone shows some downside decoupling.
  • Using US equities as the conditioning asset, emerging market equities, foreign developed equities and commodities show asymmetry that works against the investor. Treasury and corporate bonds show asymmetry that works in the investor's favour.
  • Treasury bonds were the one asset that helped in every pairing when equities or other assets were the conditioning asset, although the help for commodities was negligible.

Robustness, as reported in the paper

  • The authors state that the full-scale optimisation results are robust to reasonable changes in the utility parameters.
  • Asymmetry values are only estimated at thresholds that leave enough observations in the tail (a minimum count is required before a threshold is used).
  • Full-scale optimisation is run on a simulated 1,000 years of returns built by resampling monthly observations from the 1988 to 2019 data.

Caveats and limits

  • The paper notes that interest rates were near zero when it was written, so Treasury bonds may not rally during equity sell-offs as much as they did historically.
  • Single conditioning means two coefficients per pair, so investors must pick which assets act as the main growth engines.
  • The portfolio optimisation uses illustrative expected returns, not forecasts, and the paper reports no transaction costs or out-of-sample test of the optimised portfolios (inferred).
  • The asymmetry analysis covers a single historical sample, so the results depend on the regimes inside that period (inferred).

Items marked (inferred) are our reading of the design, not statements by the authors.

Why it matters

Downside correlation estimates that condition on both assets can overstate how badly a low-correlation asset fails in a sell-off. Practitioners who stress-test diversification can use the single-condition method and compare against what a normal distribution would imply. The results also show why Treasury bonds were treated as the main downside complement to equities in this sample.

Questions about this paper

What is wrong with exceedance correlation as used in earlier studies?

It builds the downside estimate only from periods when both assets fall, which leaves out times when one asset's gains offset the other's losses.

Which asset class gave the most favourable correlation profile with equities?

Treasury bonds. Conditioned on any other asset class they showed favourable asymmetry, with negligible benefit for commodities.

Does the paper offer a way to build portfolios that account for asymmetry?

Yes. It uses full-scale optimisation with a loss-averse utility function and reports a smaller average tail loss than the mean-variance portfolio at the same expected return of 7%.

Where can I read the original paper?

The original is at its publisher: https://doi.org/10.3905/jpm.2021.1.273. We link to the paper and do not host the PDF.

Follow new quant research

Download MacroFinance market data as CSV or Excel, or pull it by API, and test the ideas in these papers yourself. New summaries are added as we review them.

Create a free account
Summaries are written by MacroFinance from the published papers, in our own words, for information only. They are not investment advice and may omit detail. Papers are not hosted here: each page links to the original at its publisher.