Equity Duration Is Not Where Investors Think: A Cross-Sectional Decomposition of Rate Beta, 2021–2026

00Thesis

“Technology is long-duration, so it sells off when rates rise” is the most repeated rule in macro-equity strategy. It is also, on careful measurement, mostly wrong. The error has three sources, and correcting all three reverses the conclusion: investors regress on nominal yields when the discount-rate variable is the real yield; they run univariate regressions that confound rate sensitivity with equity beta; and they use homoskedastic standard errors that overstate significance on fat-tailed daily returns. After decomposing nominal-yield changes into real and breakeven components, controlling for the market factor, and computing Newey-West (HAC) standard errors, the real-rate duration of US equities is concentrated in bond-proxy sectors — and Technology has essentially none.

01The regime: nominal, real, breakeven

The repricing of this cycle was a real-rate event. The 10-year real yield (TIPS) traveled from -1.08% to 1.9% — roughly a 300bp rise in the equity discount rate — while breakeven inflation (nominal minus real) round-tripped from 2.01% to 2.36% (Fig. 1). Decomposing the nominal move matters because the two components hit equities through different channels: real yields are the discount-rate shock; breakevens are the growth/inflation impulse. Every regression below is run on both.

02The stock–bond correlation is regime-dependent

The sign of the equity–rate relationship is itself the macro regime. Figure 2 plots the 63-day rolling correlation between daily changes in the nominal 10-year and the S&P 500’s return: mildly positive in 2021 (rates and stocks rose together on growth), sharply negative through 2022 (every yield uptick a valuation shock), and drifting back toward zero since. A single pooled correlation is the wrong summary statistic for a relationship this state-dependent — and at a 63-day window the estimate carries wide confidence bands, so “normalized to zero” should be read as within noise of zero, not precisely zero.

03Decomposing the cross-section of rate beta

For each sector I estimate the empirical rate beta four ways — univariate on Δnominal, Δreal, and Δbreakeven, and then the cleanest specification, the real-rate beta controlling for the market factor:

All $\beta$ are scaled to a +100bp move; t-statistics use Newey-West HAC standard errors (Bartlett kernel, 5 daily lags) to correct for the heteroskedasticity and serial correlation in daily returns. Table 1 is the core result. Three patterns jump out.

Table 1. Sector rate-beta decomposition, daily 2021-01 → 2026-01 (n≈1,265; XLRE/XLC shorter histories). Coefficients are % return per +100bp; Newey-West HAC t in parentheses (faded = |t|<2). Source: Federal Reserve Economic Data (FRED) — DGS10, DFII10, Yahoo Finance (daily adjusted close); author's calculations.

First, the real-yield ladder is far cleaner than the nominal one. On nominal yields, Energy (+3.6%) and Financials (+1.1%) carry positive betas — the apparent “sectors that benefit from rates.” On real yields, every sector loads negatively, including Energy (−1.6%) and Financials (−1.6%). The positive nominal betas were never a real-rate benefit. Second, the inflation channel explains the flip: Energy’s breakeven beta is +18.0% (t=10.1) and Financials’ +8.8% (t=5.5) — these are reflation trades, co-moving with breakevens, not discount-rate plays. Third, and most important, controlling for the market collapses the middle of the table. Once $r_{\text{mkt}}$ is included, Technology’s real-rate beta is +0.7% (t=1.8) — economically nil and the wrong sign — while Real Estate (−5.1%, t=−8.1) and Utilities (−4.2%, t=−7.6) retain large, highly significant duration. The duration ladder is real; it just lives in the bond proxies.

04Is Technology really long-duration?

Take the question head-on with a full five-factor model — real yield, breakeven, market, ΔVIX, and oil — for the cleanest contrast in the universe, Technology vs. Real Estate (Table 2). The result is stark. For Technology, the market factor has a beta of 1.4 (t=49.5) and the regression R² is 0.86: market risk explains 86% of XLK’s daily variance, and its conditional real-rate beta is +0.69% (t=1.9) — positive and insignificant. Real Estate, by contrast, keeps a real-rate beta of -5.31% (t=-8.7) even with all five controls. Technology’s reputation as the market’s premier “long-duration” asset is, in the data, a reputation for high equity beta. When rates rise in a risk-off move, Tech falls because it is high-beta, not because it is long-duration; the bond proxies fall because they are genuinely rate-sensitive.

Table 2. Five-factor model, daily 2021–2026, HAC t. β real/breakeven in % per +100bp; market β unitless; ΔVIX/oil per unit/return. Source: Federal Reserve Economic Data (FRED) — DGS10, DFII10, Yahoo Finance (daily adjusted close); author's calculations.

05Growth/value: a noisy duration proxy

If duration were a style-level phenomenon, growth-minus-value should load on real rates. It does not — robustly, across definitions, once measured correctly. The popular figure is the monthly IWF/IWD beta of -2.57 per +100bp, but it is insignificant (t=-1.6, n=60) and, more to the point, run at the wrong frequency — 60 monthly observations against 1,265 daily for the sectors. At daily frequency the growth-value real-rate beta is small and insignificant across all three definitions (Table 3), and after controlling for the market it flips positive. The earlier “growth is long-duration” signal was largely a frequency artifact compounded by basket composition: IWF blends genuine bond-proxy-free mega-cap quality with AI exposure, IWD blends Financials and Energy. Growth/value is not a clean duration factor.

Table 3. Growth-minus-value real-rate beta across definitions and frequencies, 2021–2026, HAC t. Source: Yahoo Finance (daily adjusted close), Federal Reserve Economic Data (FRED) — DGS10, DFII10; author's calculations.

06Robustness: is the ladder structural?

A five-year window is one regime. To separate structure from artifact, Figure 4 plots the rolling 252-day market-controlled real-rate beta for the key sectors back to 2011, and Table 4 splits the estimate into pre-inflation (2010–20), tightening (2021–22), and plateau (2023–26) sub-samples. Bond-proxy duration is structural: Utilities and Real Estate carry large negative real-rate betas in every sub-period and every rolling window. Technology’s is not — it is near zero pre-2020, briefly negative in the 2022 selloff, and positive in 2023–26. What looks like “Tech duration” is a transient of the 2022 regime, not a stable factor loading.

Table 4. Market-controlled real-rate beta by sub-period (% per +100bp). Source: Federal Reserve Economic Data (FRED) — DGS10, DFII10, Yahoo Finance (daily adjusted close); author's calculations.

Two honest caveats. The R²s in Table 1 are small — rates explain under 9% of even Real Estate’s daily return variance — so “rates as the master factor” overstates the share of variance rates command; the accurate claim is about cross-sectional ranking, not explanatory power. And these are tradeable-proxy betas (SPDR ETFs, with expense ratios and reconstitution), not pure-factor betas; the ranking is what survives, not the second decimal.

07Portfolio implications

The positioning follows, with appropriate humility about borderline t-stats. To make it concrete, Table 5 reports the mean return of three candidate rate hedges on the top-decile nominal-yield-up days of the sample (mean move +11bp). Shorting Technology — the consensus rate hedge — returns just +0.2% on those days, because Tech’s move is mostly market beta that a rate shock only partially drives. Shorting the bond proxies returns +0.59%, and the reflation-vs-bond-proxy spread +0.74% — a materially cleaner hedge per unit of rate move.

Table 5. Mean hedge return on the 127 worst rate-up days of 1266 in 2021–2026. Source: Yahoo Finance (daily adjusted close), Federal Reserve Economic Data (FRED) — DGS10, DFII10; author's calculations.

• →Hedging rate risk by underweighting Technology is mis-specified — most of that exposure is equity beta, and it is a weak, regime-unstable rate hedge. The cleaner hedge is the bond-proxy basket (Real Estate, Utilities), where the real-rate beta is large, significant, and structural.
• →But a short-Utilities/REITs position is not a pure rate hedge — it is also a quality/defensive underweight that will hurt in risk-off rallies when Treasuries and bond proxies both rally. Size it as a rate tilt, not a costless hedge.
• →Treat the growth-value duration trade as no signal, not a weak one: it is insignificant at daily frequency across every definition tested.

08Conclusion

After decomposing nominal yields into real and breakeven components, controlling for equity beta, and using HAC standard errors, the result is sharper than the consensus framing and opposite to it in one important place. Real-rate duration is concentrated in bond-proxy sectors — Real Estate and Utilities — where it is large, highly significant, and stable across regimes. Technology, the textbook long-duration asset, shows essentially no real-rate duration once equity beta is removed; its rate “sensitivity” is market beta. And the sectors that look like rate beneficiaries on nominal yields — Energy, Financials — are reflation trades loading on breakeven inflation, not real rates. Rate risk lives where the cash flows are genuinely bond-like, not where the market beta is highest.

AData & method