Portfolio Implications of Time-Varying Correlations

The empirical research demonstrates that using static long-term historical correlations between assets in portfolio management can lead to substantial underperformance as the large swings between asset return correlations have considerable implications for risk and, thus, portfolio construction (see here, here and here). Investors were reminded of the lessons history provided in 2022 when the correlation between stocks and bonds shifted to a strong positive correlation at the wrong time as both assets experienced double digit losses.

Jack Strauss and T.H. Williams, authors of the study “How to Diversify Differently: Time-Varying Correlations, Determinants, and Regimes,” published in the February 2025 issue of the Journal of Portfolio Management, investigated the performance of a dynamic regime-switching (RS) tactical asset allocation portfolio compared to a more traditional static 60/40 benchmark. They began by noting that prior research had identified four correlation regimes defined by rising/falling markets and positive/negative asset correlations:

· “Everyone-Wins (EW): Positive SBC with rising markets. High correlations during periods of positive returns allow investors to benefit from holding a diverse range of assets. Correlational risk, however, is priced in the market—investors demand a premium for holding assets with high systematic risk during periods of increased correlation, presenting nuanced diversification challenges during rising markets with positive correlations.

· Risk-On (RO): Negative SBC amid rising markets. Investors are inclined toward riskier assets because of optimistic conditions, growth, or general optimism, often driven by accommodating monetary policies. Stock–bond relationships are more positive in low-uncertainty phases, which often align with RO regimes.

· Flight-to-Safety (FTS): Negative SBC coupled with falling equity markets. In periods of market stress or uncertainty, investors reallocate their investments from riskier assets to more secure or higher-quality ones—bonds are often a refuge when equity markets decline. There is a similar effect when examining yield spreads in European sovereign debt, where they discover evidence of FTS behavior during market stress, with investors seeking refuge in higher-quality government bonds. Political uncertainties can amplify FTS tendencies.

· Nowhere-to-Hide (NTH): Markets experience negative returns, and SBC correlations are positive, limiting diversification opportunities. Such regimes are generally observed following a transition from an FTS regime. Correlated markets can induce shifts in stock–bond return correlations during inflationary risks, escalating investment risk. Higher market correlations exist during periods of intensified volatility and negative yields—opportunities for diversification recede markedly during financial turmoil. Investors typically reallocate their investments from riskier assets to more secure or higher-quality assets during periods of correlated market stress or uncertainty. This behavior results in a surge in demand for low-risk assets, such as government bonds, while causing a decrease in the value of riskier assets like equities.”

To identify breaks in correlation regimes they applied structural break tests and wavelet coherence (WC) methods to investigate the time-varying nature of correlations across 15 asset pairs. (WC measures the correlation between signals x and y in the time-frequency plane and is useful for analyzing nonstationary signals.)

Next, they examined the determinants of asset return correlations to evaluate dynamic lead/lag relationships between macroeconomic variables—inflation, expected inflation, leading indicators, and sentiment—and 15 correlational pairs. Then they constructed a real-time portfolio allocation based on changes in correlations—when a regime changes, the portfolio changes its allocation one month after to allow an implementable strategy. The EW regime received an allocation of 60/40 to the benchmark portfolio, an RO regime received an allocation of 80/20, the FTS regime received a 40/60 allocation to support this risk-off, negatively correlated environment, and the NTH regime received a 100% allocation to bonds.

Their data sample used month-end index and price data for six primary asset classes (US large stocks, US small stocks, developed market international stocks, REITS, gold, and Treasury bonds) from January 1982 through December 2023, resulting in 504 monthly observations. As a test of robustness they also studied the subperiod 2000-2023, and expanded the number of asset classes to include US mid-caps, high-yield bonds, and emerging market stocks. They also substituted a 50/50 mix of commodities and hedge funds for gold). Following is a summary of their key findings:

· There has been temporal instability in correlation between asset returns—asset return correlations exhibit substantial variability across time periods and horizon. For example, the one-year SBC correlation crossed zero (the mean) 18 times in the sample and experienced large swings, swinging from 0.23 before the financial crisis to −0.78 in 2008 during the crisis, then to over 0.40 in 2009, and back to near −0.80 in 2010. Before COVID-19, the SBC correlation was around −0.75; by 2020, it was around 0.60

· There were sharp changes in correlation for stocks/foreign, stocks/REITs, and bonds/gold. The stocks/foreign correlation was the most stable. However, it still ranged from 0.45 to 0.96 over the past decade—a change of more than 50%. Stocks/REITs exhibited significant swings since 2010, with the correlation dropping from 0.90 to nearly zero three times. The bond/gold correlation also showed shifts from 0.25 to over 0.55 since 2010.

· There were significant and multiple structural breaks in all 15 correlations.

· Although correlations may be volatile, correlational regimes are more stable.

· Over the 42-year sample, regimes changed 70 out of 504 months, resulting in a 13.9% turnover, or an average regime change every 5 years, 10 months.

· There has been significant bi-causality between their four macroeconomic determinants and asset return correlations, indicating extensive feedback relationships (e.g., inflation and sentiment both forecast and are forecasted by the stock/bond correlation (SBC) or gold/REIT correlation). Thus, macroeconomic factors lead return correlations and changes in correlations anticipate changes in macroeconomic conditions.

· Although correlations may be volatile, correlational regimes were more stable.

· The lagged macroeconomic variables predicted both changes in correlations and regimes defined by whether the stock and bond markets are rising or falling.

· The annualized returns of the individual EW, RO, FTS, and NTH regimes were similar, ranging from 8.51% to 9.01%. Because of lower risk, however, all four regimes exhibit higher annualized Sharpe and Sortino ratios (except for FTS, which had a Sharpe ratio of 0.62) compared to the 60/40 benchmark (Sharpe ratio of 0.63). EW, RO, and NTH regimes also had lower maximum drawdowns than the benchmark.

· The RS portfolio produced an annualized return of 10.11% and a Sharpe ratio of 0.76, compared to the 60/40 benchmark’s annualized return of 8.98% and Sharpe ratio of 0.63. The RS portfolio achieved a similar return to the 100/0 portfolio (10.30%) but with a 38% lower standard deviation (8.22% versus 13.25%). The maximum drawdown for the RS portfolio was only 17.91%, significantly lower than the 47.71% drawdown of the riskiest 100/0 portfolio and the 27.87% drawdown of the 60/40 portfolio.

· Their findings were robust to various tests.

Their findings led Strauss and Williams to conclude: “Major asset classes experience substantial changes in correlations over time, challenging the fundamental premise of buy-and-hold strategic asset allocation based on the assumption of time-invariant correlations.” They added: “The high persistence and stability of regimes can improve tactical allocation based on regime shifts when correlations are time varying…. Our results show that by exploiting RS to take advantage of time-varying correlations, investors can gain valuable insights into how to diversify differently.” They noted that while their RS portfolio did not outperform every time, it offered relatively consistent performance over different decades, and it outperformed the benchmarks on a risk-adjusted basis during the entire sample period.

Investor Takeaways

The research shows that because correlation regimes tend to be long lasting, and at least to some degree predictable, historically investors could have improved buy-and-hold strategies using a tactical approach and regime shifting their allocations. With that said, as Kevin Grogan and I showed in our book “Reducing the Risk of Black Swans” investors can also reduce downside risk without sacrificing expected returns by using the strategic approach of reducing the allocation to assets that have high exposure to economic cycle risk and replacing those allocations with assets that have no, or much lower, correlation to the economic cycle risk of stocks and the duration/inflation risk of bonds (for example, reinsurance strategies such as Stone Ridge’s SRRIX, long-short market neutral factor strategies such as AQR’s QSPRX, and senior, secured, and sponsored by private equity private credit interval funds such a Cliffwater’s CCLFX—I have significant allocations to those three funds). As Grogan and I showed in the book, investors can also lower their exposure to market beta (without lowering expected returns) by adding exposure to equities with exposure to factors which have historically provided premiums (value, momentum, and profitability/quality) but have had negative correlation to market beta)—as I have done personally in my portfolio.

Larry Swedroe is the author or co-author of 18 books on investing, including his latest Enrich Your Future.