Skip to main content

Why Macro Betas Are Regime-Dependent (Unstable)

Updated Jun 24, 2026 at 2:35pm

Research Draft High 1,204 words

A "macro beta" is the measured sensitivity of an asset's return to a macro factor — how much a stock, sector, or portfolio moves per unit of move in rates, inflation surprises, the dollar, oil, or the broad equity market. The core tension is that this sensitivity is not a fixed property of the asset. It is a relationship that depends on the prevailing economic regime, and that relationship can weaken, strengthen, or flip sign when the regime changes. A beta estimated over a calm low-inflation decade can be actively misleading in a high-inflation tightening cycle. The number looks like a constant; it behaves like a variable.

How macro betas are formed and measured

A macro beta is the slope coefficient from regressing asset returns on a factor's returns or surprises:

> r_asset = α + β · (factor) + ε

In a single-factor (CAPM) setting the factor is the market; in multi-factor macro models the right-hand side stacks several factors (e.g. growth, inflation, real-rate, credit, dollar, commodity). Practitioners estimate β over a trailing window — a rolling 60-month or 1–3-year regression is conventional — which mechanically means the estimate is a backward-looking average over whatever regime dominated that window.

The instability shows up the moment you let β vary. The empirical literature uses three main techniques to model time-varying betas: rolling regressions, Kalman-filter / state-space estimation, and regime-switching (Markov-switching) and threshold-GARCH models that let β take different values in distinct states (commonly labelled bear / normal / bull, or expansion / recession). Studies across equity, commodity, and sustainable-stock markets consistently find that these time-varying-beta models fit the data better than constant-beta models, and that market beta "changes asymmetrically and nonlinearly" between regimes (Springer/Annals of Operations Research; MDPI Mathematics).

How regime-dependence shows up in practice

The cleanest, most-studied example is the stock–bond correlation, which is itself a macro beta (equities' sensitivity to the bond/rates factor). It is not merely noisy — it has switched sign across history:

  • It was positive on average through roughly 1970–1999 (one widely cited figure puts the US average near +0.35 over 1970–1999), the high-inflation era (AlphaArchitect; Financial Analysts Journal/QuantPedia).
  • It turned persistently negative from the late 1990s/early 2000s through the early 2020s (one estimate: about −0.3 over 2000–2022), the era in which "bonds hedge stocks" became received wisdom; sources variously date the onset to late 1997, 1998, or 2000, so no single start date is authoritative (AQR; AlphaArchitect; Russell Investments).
  • It flipped positive again in 2022, with the SPX–AGG correlation reportedly spiking to roughly +0.5 — the highest in that sample — as the Fed tightened aggressively and stocks and bonds fell together (AlphaArchitect; Morningstar).

The mechanism is well-identified: the sign of the correlation depends on what kind of shock dominates. Demand shocks and flight-to-quality episodes push the correlation negative (bonds rally as stocks fall); inflation surprises and monetary-policy shocks push it positive (both fall together). The common driver is shared sensitivity to unexpected inflation — the stock–bond correlation tends to be high when inflation and real short-rate uncertainty are high (NBER w15260; Journal of Banking & Finance, "One and a half centuries of evidence"). So the same asset has a different rates-beta depending on whether the economy is in a growth-shock regime or an inflation-shock regime.

The same logic applies to other macro betas: a sector's oil-beta differs in supply-shock vs demand-shock regimes; a stock's dollar-beta depends on whether the dollar is moving on risk-off flows or rate differentials; equity market beta itself rises in bear/high-volatility states. The practical takeaway for any factor-sensitivity model: a single static β is a regime-blended average, and the blend can be dominated by conditions that no longer hold.

Adoption, debate & evidence

That betas are time-varying is not contested — it is one of the better-established empirical facts in asset pricing, and the diversification-failure of bonds in 2022 made it front-page material for practitioners. The debate is over whether modelling the time-variation rescues asset-pricing theory.

The key skeptical result is Lewellen and Nagel (2006), "The Conditional CAPM Does Not Explain Asset-Pricing Anomalies." They argue that even though betas genuinely vary with the cycle, the covariance between beta and the expected risk premium is too small to explain anomalies like value and size — i.e., the conditional CAPM still fails. So there are two honest, separable claims: (1) macro betas are unstable and regime-dependent — strongly supported; (2) you can profitably exploit that instability with a conditioning model — much weaker, and contested. Folklore ("bonds always hedge stocks," "utilities are a stable rate play") overstates the stability of relationships that the data show are conditional. Measured: the relationships are real within a regime and unreliable across regimes, and regimes can persist for a decade, which makes out-of-sample estimation genuinely hard.

Strengths & limitations

Treating macro betas as regime-dependent is the more accurate model, and it explains real-world failures (the 60/40 drawdown of 2022) that static beta cannot. Its honest limitations:

  • Estimation lag. Rolling and even Kalman-filter betas update after a regime has begun; you learn the new beta partly by living through the losses.
  • Regime identification is itself uncertain. You rarely know in real time which regime you're in; misclassification feeds straight into the beta.
  • Overfitting. Allowing many states and thresholds fits history beautifully and forecasts poorly — the Lewellen–Nagel caution applies.
  • The #1 misuse: taking a beta (or a correlation, or a "hedge") estimated entirely within one regime and assuming it is a structural constant — e.g. sizing a portfolio's "bond hedge" on 2000–2020 data going into an inflation regime. That single error caused much of the diversification disappointment of 2022.

Sources

Disputes flagged: that betas are unstable/regime-dependent is well-established; whether conditioning models can be exploited profitably is contested (Lewellen–Nagel). Exact correlation figures (e.g. ≈+0.35 over 1970–1999, ≈−0.3 over 2000–2022, the ≈+0.5 spike in 2022) and regime start/end dates are sample-, window- and source-dependent — the negative-regime onset is variously dated to late 1997, 1998, or 2000 — and are quoted as reported by the cited sources, not as universal constants.