Skip to content

When Diversification Fails: summary and key takeaways

The paper measures how correlations between US stocks and other assets change in extreme months. Across most risk assets, correlations are much higher in stock selloffs than in rallies. Treasuries are the main exception, though the paper warns the stock-bond link can turn positive.

Page, S.; Panariello, R.A. · 2018 · Financial Analysts JournalSummary updated 20 November 2026
Published2018Journal article
Asset focusMulti-assetDiversification / correlation
TypePortfolio MgmtStock-Bond Correlation
Reading time4 minthis summary

When Diversification Fails at a glance

The question
How do correlations between stocks and other asset classes, hedge funds, private assets, risk factors and bonds behave in the tails of the return distribution?
Dataset & period
Monthly data. US vs non-US stocks (MSCI US and MSCI EAFE total return): January 1970 to June 2017. US stocks vs long Treasuries (Barclays Capital Long U.S. Treasury Index): January 1973–June 2017. Other asset classes, hedge fund styles and risk factors use start dates set by data availability, with data through June 2017. Direct real estate and private equity use quarterly data. Tail correlations are reported at the 1st and 99th percentiles.
Method
The authors condition on the return of one asset only and compute the correlation with a second asset inside the selloff or rally subsample. This differs from double conditioning, which picks months where both assets move. They add observations from the rest of the distribution with exponential weights to reduce noise from few tail points. They compare each empirical profile with one simulated from a bivariate normal with the same means, volatilities and full-sample correlation.

Key findings of When Diversification Fails

  • For US vs non-US stocks, the conditional correlation fell to –17% in the strongest US rallies but rose to +87% in the worst US selloffs, so international diversification helped mainly on the upside.
  • Left-tail correlations with US stocks were much higher than right-tail correlations across the risk assets studied: real estate, corporate, high-yield, MBS and EM bonds, hedge funds, EM stocks, and style and size splits.
  • All hedge fund styles shown, including market-neutral, had higher left-tail than right-tail correlation with US stocks. The paper cites another study that found an average correlation jump of +33% for hedge fund strategies in financial crises.
  • Private assets show higher left-tail equity exposure than reported returns imply, because reported quarterly returns are smoothed. Rolling annual returns are less affected by smoothing.
  • Some risk factors held up better than asset classes. Currency and cross-asset momentum had much lower left-tail than right-tail correlation with stocks, while size and currency carry did not diversify well in selloffs.
  • Treasuries decouple from stocks in stock selloffs and correlate positively in rallies. The profile reverses when conditioning on bond returns, and the paper notes the 12-month stock-bond correlation was mostly positive in the 1970s and 1980s and that the 2013 taper tantrum hit both together.

Robustness, as reported in the paper

  • The data-augmentation estimates were reported to be similar in size and direction to conventional conditional correlations, but less noisy and less sensitive to outliers.
  • Empirical profiles are compared with simulated normal data given the same adjustment, and left-tail vs right-tail comparisons are used because conditioning bias largely cancels in that comparison.
  • Stock-Treasury results were reported as similar for the broad Treasury and intermediate Treasury indexes, though those results are not shown.
  • Private asset correlations are shown on both quarterly and rolling annual data to check the effect of return smoothing.

Caveats and limits

  • Conditional correlations are one measure of diversification. Conditional betas also reflect relative volatility, and the paper says it relies on prior studies to argue that offsetting volatility effects are unlikely.
  • The paper does not forecast left-tail events, so it does not say when correlations will rise.
  • Many results are shown only as charts and appendix material, so exact values for most asset pairs are not in the main text (inferred).
  • Start dates differ across assets, so profiles cover different sample periods and some tail estimates rest on few observations (inferred).

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

Why it matters

Full-sample correlations can overstate how much most risk assets diversify equity risk in a crash. The paper argues for tail-aware risk measures, scenario analysis and attention to the stock-bond correlation, which shocks to rates or inflation can turn positive.

Questions about this paper

Do all correlations go to one in a crisis?

The authors call that an oversimplification. Still, they find that diversification across risk assets almost completely disappears in market crises.

Do Treasuries still diversify stocks in selloffs?

In the sample, bonds decoupled from stocks during stock selloffs. The paper adds that shocks to rates or inflation can push the stock-bond correlation positive.

What tools does the paper suggest beyond diversification?

It points to tail-aware optimization, scenario analysis, tail-risk hedging, factors with short positions, defensive momentum and managed volatility.

Where can I read the original paper?

The original is at its publisher: https://doi.org/10.2469/faj.v74.n3.3. 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.