Why Stock-Bond Correlation Changes

Measure returns rather than chart levels

A correlation of equity-price levels with bond-yield levels does not answer how the two investments' returns co-move. Use defined return intervals and identify whether dividends, coupons, accrued interest, and reinvestment are included. Bond prices generally move inversely to yields under fixed-cash-flow conditions, so substituting yield changes for bond returns can even reverse the apparent sign of the relationship.

Match currency and hedging treatment. A domestic equity index compared with an unhedged foreign bond index includes currency effects that a local stock-versus-government-bond study may not intend. State the government maturity exposure as well: short bills and long bonds need not respond similarly. A chart labeled stock-bond correlation should name both actual series rather than allowing readers to assume a familiar but different benchmark.

Explain the growth-news channel conditionally

When lower growth reduces expected corporate cash flows and also leads markets to expect easier policy, equities can fall while high-quality bond prices rise. This mechanism can contribute to negative return correlation. It relies on the response of discount rates and on the type of bond; credit-sensitive debt may suffer from widening spreads even while government yields decline.

BIS researchers' The correlation of equity and bond returns discusses how the inflation environment and expected monetary response can alter this relationship. Attribute that explanation to the analysis and its setting, rather than calling it a universal law. A severe financing disruption, changes in term premiums, or shifting equity risk compensation can modify the response even if the original news concerns real activity.

Explain the inflation-news channel separately

An unexpected inflation increase can raise nominal yields and lower outstanding fixed-payment bond prices. It may also reduce equity value through higher required returns, pressure on margins, or a more restrictive expected policy path. Stocks and bonds can then decline together. Whether these effects dominate depends on pricing power, the shock's source, starting expectations, and the central bank response perceived by market participants.

This does not imply that every inflation release generates a positive correlation or that a simple inflation threshold determines the sign forever. Correlation summarizes the mixture of shocks and responses in the sample. A change in that mixture can alter the estimate without changing the fundamental mathematics of either asset. Regime descriptions should therefore remain hypotheses about mechanisms, not automatic assignments of future correlation values.

Quantify a synthetic portfolio sensitivity

Assume an illustrative portfolio with 60% equities and 40% bonds, annualized volatilities of 18% and 7%, and no leverage. Variance equals the sum of squared weighted volatilities plus twice the product of weights, volatilities, and correlation. With correlation negative 0.3, the synthetic portfolio volatility is approximately 10.31%. With correlation positive 0.5 and all other inputs unchanged, it rises to approximately 12.44%.

CFA Institute's Portfolio Risk and Return: Part I describes this dependence on weights, volatility, and covariance. The example changes only correlation to isolate the arithmetic; actual regimes can change all inputs together. A favorable correlation does not guarantee a low-volatility portfolio if individual volatilities increase sharply. Similarly, a lower estimated variance does not specify maximum loss or establish whether an investor can fund an adverse cash-flow event.

Distinguish average correlation from stress protection

For another synthetic comparison, hold the same weights and assume equities lose 20%. If bonds gain 5%, the simple one-period portfolio return is negative 10%. If bonds instead lose 8%, the portfolio return is negative 15.2%. These are illustrative stress combinations, not estimated probabilities. They answer a particular loss question more directly than a single correlation coefficient calculated over many ordinary days.

A negative full-sample correlation can coexist with damaging joint losses in selected intervals. Conversely, two assets can have positive average correlation while one still reduces portfolio variability because its own volatility is lower. Separate diversification, downside protection, and liability matching. None is fully characterized by the sign of a statistic alone, and the appropriate measurement depends on which research question is being asked.

Check windows, calendars, and estimation uncertainty

Short windows adapt quickly but contain fewer observations and can swing with a handful of unusual days. Long windows provide more observations but mix different monetary and inflation environments. Overlapping rolling windows also produce highly dependent adjacent estimates; a smooth line is not evidence of many independent confirmations. Compare several predeclared horizons rather than selecting the one that best fits a preferred narrative.

Non-synchronous market closes, holidays, stale prices, and different return frequencies can change observed co-movement. Align observations deliberately and explain missing-data treatment. Statistical uncertainty also grows when returns are serially dependent or unusually heavy-tailed. A displayed coefficient of negative 0.20 should not be interpreted as economically distinct from negative 0.18 without considering estimation uncertainty and the sensitivity to reasonable construction choices.

Preserve the scope of the empirical claim

Report the series names, total-return definitions, timestamps, sample, currency, and methodology beside the result. When discussing BIS evidence, retain the publication period and distinguish the authors' interpretation from an institutional prediction. A historical finding about US government bonds and equities does not automatically transfer to corporate bonds, another country, or a portfolio with an active currency overlay.

Conclude by describing what co-moved, which shocks plausibly explain it, and how the portfolio result changes under alternatives. Do not infer causality from a correlation plot or describe a bond allocation as guaranteed insurance. The defensible output is a measured historical relationship and a transparent sensitivity analysis, with enough context to understand when its diversification properties could weaken or differ from the sample average.

What to retain

• Correlation depends on the instruments, return definitions, currency, and sample.
• Growth and inflation news can create different stock-bond relationships.
• Historical correlation is not a guarantee of protection during a particular stress.

Source notes

1. BIS Quarterly Review: The correlation of equity and bond returns
2. CFA Institute: Portfolio Risk and Return: Part I

Related research