Bonds Are Not Rougher Than Stocks

1Introduction

Since late September, a specific alarm has run through rates and volatility commentary. The MOVE index, which measures expected volatility in Treasury yields, has climbed while VIX has stayed in the mid-teens. CrossVol reported MOVE at 107 with a MOVE-to-VIX ratio of about 7.0 in its October 4 note.¹ Fundstrat noted the gauges "diverging" with MOVE at 104.58 on September 24.² yieldcurve.pro put MOVE at 106.6 on September 29 with a ratio of 6.65, higher than on 89% of days since November 2002.³ The macro backdrop is not hard to find. The Federal Reserve raised the policy rate 25 bp on September 16 to a 3.75% to 4.00% range, its first hike since 2023.⁴ The 10-year Treasury yield closed at 5.19% on October 14.⁵

The implied reading is that the bond market is pricing more turbulence than the stock market, and that stocks have not caught up. This paper tests that reading with realized data, which is the only way to go back past 2002, and asks a second question the alarm usually skips: if bonds are rough, what does that change for the investor who holds both?

We do four things.

First, we put stock and bond volatility on the same scale: percentage price volatility, not basis points of yield. Second, we compute the ratio of the two back to 1962 and ask how unusual today is. Third, we test whether the ratio forecasts stock volatility or stock returns. Fourth, we move from volatility to the portfolio and measure what bonds contributed to a 60/40 portfolio and to sell-offs under different stock-bond correlation regimes.

The MOVE index is proprietary and its history is licensed, so we use it only as the published recent values quoted by the sources above. Every long-history exhibit uses a named public series, and the bond return series is built from them. Section 2 describes the construction and Appendix A lists every series and parameter.

One more note on prior work, which we take up in Section 7. AQR has argued that a positive stock-bond correlation is not a reason to hold more equity risk, and that the sign of the correlation is regime-dependent.⁶⁷ The yieldcurve.pro piece finds that VIX and MOVE describe stress but do not predict it.³ This paper agrees with both and adds what they do not test: a 1972-onward realized-volatility ratio, a bond-share-of-variance decomposition, and bond returns in stock sell-offs split by an ex-ante correlation reading.

2Putting both on one scale

2.1 Why the units matter

MOVE is quoted in basis points of yield volatility. VIX is quoted in percentage points of S&P 500 return volatility. Dividing one by the other yields a number such as 6.65. That number has no natural benchmark, because it compares the volatility of a rate with the volatility of a price. What an investor actually holds is the price of the bond, and the price moves by roughly modified duration times the yield change:

For a 10-year par bond, modified duration falls as the yield rises. We compute it exactly for a semiannual coupon bond (Exhibit 1). It is 9.25 years at a 1.5% yield, 8.58 at 3%, 7.72 at 5.2%, 6.8 at 8% and 5.1 at 15%.

So a 100 bp annual yield volatility means about 9.25% price volatility at a 1.5% yield, and about 7.7% at today's 5.2%. A basis point-based ratio therefore drifts upward as yields rise, even if the price risk relative to stocks stays put. Over 1990 to 2026 the correlation between the daily bp-to-VIX ratio and the yield level is +0.215. The price-based ratio's correlation with the yield level is -0.03.

There is a second effect. Yield volatility itself rises with the yield level, but less than one for one. Across 65 calendar years, the log of annual yield volatility in basis points regresses on the log of the average yield with a slope of 0.30 (standard error 0.11, R-squared 0.09). The relationship is loose, but it says that a higher yield level is only partly a higher-volatility regime in basis points.

2.2 The bond series

There is no free public daily total return index for 10-year Treasuries over this whole period that we could find, so we construct one from the FRED yield.¹⁰ Each day we reprice a 10-year par bond, semiannual, with a coupon equal to the previous day's constant-maturity yield (FRED DGS10), at the new yield with maturity shortened by 1/252 year, and add the coupon accrual. This is a total-return proxy for a bond that rolls at constant maturity. It ignores the roll between on-the-run issues and bid-offer costs.

We checked it against two independent sources. Against the annual 10-year Treasury bond return in Damodaran's historical returns table (1962 to 2024), the annual correlation is 0.988, the mean absolute difference is 1.0 percentage point, and the maximum difference is 6.7 points, in 1982.⁸ Against the monthly bond return in Shiller's data (1962 to 2023) the monthly correlation is 0.62 and annualized volatility is 7.9% against 6.7%. Shiller's series is built from yearly interest rate data and a different duration assumption, so we treat it only as a loose check. Annual returns in the years that matter for this paper are +39.5% in 1982, -7.2% in 1994, +20.5% in 2008 and -16.4% in 2022.

Stocks are the Fama-French market return (Mkt-RF plus RF) from the Kenneth French Data Library, which runs to August 31, 2026.⁹ Against Damodaran's annual S&P 500 returns the correlation is 0.985. The window for which we have both series therefore ends August 31. For the last six weeks we use VIX alone.

A caution on the early data. FRED's 10-year yield is rounded to the basis point, and in the 1960s yields moved by less than a basis point on about 30% of days. Realized volatility in that decade is biased low and coarse. We show results from 1962 and from 1971 or 1972, and none of the conclusions depends on the 1960s.

3Sixty years of the ratio

3.1 The level

Exhibit 3 shows the 63-day realized volatility of stocks and bonds. Bond volatility is lower than stock volatility nearly everywhere. The exceptions are 1980 to 1982 and 1985, when 10-year yields moved by a point or more in weeks, and the early 1990s.

Exhibit 4 plots the ratio. Its median is 0.46, the 10th percentile 0.24 and the 90th 0.83. It exceeds 1 on 503 days. By year, those days are 1980 (104), 1981 (168), 1982 (47), 1985 (88), 1986 (16), 1993 (8), 1994 (35), 1995 (31), 2011 (5) and 1971 (1). The peak is 1.46, on July 8, 1985. In every other year of the sample, stocks were more volatile than bonds over every 63-day window. The latest reading, at August 31, is 0.40, which sits at the 37th percentile.

The calendar-year view agrees. The annual ratio was 1.02 in 1980, 1.05 in 1981, 0.98 in 1985, 0.42 in 2022, 0.71 in 2023 and 0.40 for 2026 to date.

3.2 The recent window in both units

The published MOVE readings above are elevated. Realized yield volatility, which we can compute, tells a similar story: 21-day realized volatility of the daily change in the 10-year yield rose from 59 bp (annualized) on September 9 to 84.5 bp on October 14. Over 1962 to 2026 that is the 58th percentile, and over 1990 to 2026 the 55th. Realized bond price volatility over the same 21 days is 6.6%.

Against VIX, which closed at 16.87 on October 14, the realized ratios are:

Both sit within the ordinary range. We make no claim about the MOVE-to-VIX ratio itself, which includes an implied volatility premium we cannot reconstruct from public data. What the realized numbers say is that the underlying bond moves have been larger than a month ago, but not unusual against a one-month stock volatility reading, and not unusual in price terms. The gap between MOVE near 107 and realized yield volatility near 85 bp is a premium for expected future volatility, which is the part this paper does not test.

3.3 Episodes

Ten episodes help calibrate what high bond volatility looks like.

Two patterns stand out. The episodes with a ratio near 1 (1979 to 1985, 1994) are periods when bond volatility was high and stock volatility was moderate or low. In 1994, stock volatility for the calendar year was 9.4%, in the lowest fifth of the 65 years, and bonds lost 8.6% from February to December while stocks lost 3.8%. The ratio is high partly because the denominator is small. Second, when stocks crash (1987, 2008, 2020), the ratio falls to 0.2 to 0.3 because stock volatility explodes. A ratio near 1 describes a quiet equity market with a bond market that moves. A ratio near 0.2 describes a stock crisis.

This matters for the alarm. A high bond-to-stock ratio is, to a large extent, a statement about the denominator.

4Does the ratio forecast anything?

If bond volatility is a leading indicator for stock volatility, the ratio should predict stock volatility changes, controlling for the current level. We test that.

4.1 Design

For each trading day t from June 1972 to June 2026, we compute the ratio of 63-day realized bond volatility to 63-day realized stock volatility, and the following outcomes over t+1 to t+63: realized stock volatility, and excess stock return. We define "high" as the ratio being in the top decile of its expanding history (burn-in of 2,520 days), so that no look-ahead enters the threshold. Standard errors use a Newey-West HAC correction with 63 lags because the forward windows overlap. We also run a non-overlapping version.

4.2 Results

High-ratio days are days of low stock volatility. Current 63-day stock volatility averages 11.2% on high days and 15.4% on all others. Next-period stock volatility averages 12.2% and 15.4%. Stock volatility rises by 1.0 point after high-ratio days, against -0.07 points otherwise. On 63% of high days, stock volatility rises over the next 63 days, against 44% otherwise.

That looks like a signal until you control for the current level. Here is the regression of log forward stock volatility on log current stock volatility, with and without the log ratio.

None of the coefficients is distinguishable from zero. R-squared is 0.338 with or without the ratio. Replacing the ratio with log bond volatility on its own gives the same answer: bond volatility adds nothing beyond stock volatility itself.

For stock returns the picture is the same. After high-ratio days, the mean forward 63-day excess stock return is 0.76%, against 1.48% otherwise. The HAC t-statistic of the difference is -0.57. Since 1990 the difference flips sign and is +0.5 point (t = 0.41).

4.3 The implied-volatility version

The alarm is stated in terms of an implied bond volatility index, and the forward-looking measure on the stock side is VIX. We can run the equivalent test with the data we have: does realized bond volatility add to VIX when forecasting the next month of realized stock volatility? From January 1990 to July 2026 (9,114 daily observations), we regress the log of the next 21 days of realized stock volatility on log VIX and log 21-day realized bond volatility.

VIX already carries what is in bond volatility for this purpose. This is consistent with yieldcurve.pro's finding for the implied series over 24 years.³ Here it holds for realized bond volatility over 36 years.

4.4 What this does and does not show

It shows that the ratio does not carry information about the stock market beyond stock volatility. It does not show that bond volatility is unimportant. In 1994, 2013, 2022 and 2023 the bond market was where the news was. The test is narrower: a high reading of the ratio is not a reason to expect more stock volatility than the stock market's own volatility already implies.

5Bonds inside the portfolio

For an investor holding both, the first-order question is not whether bond volatility is high relative to stocks. It is what the bond position is contributing to portfolio risk, and whether it helps when stocks fall. Both depend on the stock-bond correlation more than on relative volatility.

5.1 The correlation regime

The stock-bond correlation has changed sign twice in this sample. Decade mean of the 252-day rolling correlation: 1960s +0.05, 1970s +0.25, 1980s +0.37, 1990s +0.32, 2000s -0.24, 2010s -0.42 and 2020s -0.06. At August 31, 2026 the trailing correlation is +0.30, the 74th percentile of the 1962 to 2026 distribution. Over the whole sample the correlation is positive on 57.8% of days.

This agrees with long-sample evidence. AQR's April 2026 note observes that the correlation was more often positive than negative from 1900 to 2000, so a positive sign is the historical norm.⁶ The 2000 to 2021 negative-correlation period is the period many portfolio managers learned from.

5.2 What bonds contribute to 60/40 risk

For a daily-rebalanced 60/40 portfolio, the bond contribution to variance is 0.4² σb² + 0.6 × 0.4 × cov(s, b), divided by portfolio variance. When correlation is negative, the second term is negative and the bond sleeve reduces risk. Using 252-day windows:

The latest reading is 12.1% (58th percentile, median 9.0%, peak 38.5% in 1981). Trailing 60/40 volatility is 8.7%.

The counterfactual is more useful than the level. Hold today's volatilities fixed and vary only the correlation: the bond share of 60/40 variance is -4.2% at a correlation of -0.4, 6.5% at 0, 12.1% at +0.3 and 16.5% at +0.6. The relative-volatility ratio, the thing the alarm is about, is held constant in this table. The correlation moves the answer by 20 percentage points.

For 2022, the same exercise at that year's volatilities (stocks 25.4%, bonds 10.6%): actual 60/40 volatility was 16.5% with a correlation of +0.17. It would have been 15.8% at zero correlation and 14.1% at -0.4. So even a favorable correlation would have cut 2022's portfolio volatility by about a seventh, not by half.

5.3 What bonds did in stock sell-offs

We take all 63-day windows (overlapping, daily start) in which the stock market lost at least 10%, and condition on the stock-bond correlation measured over the 252 days before the window started, so no look-ahead enters the sorting.

A regression of the bond return on a positive-correlation indicator, in these windows, gives -4.96 points (HAC standard error 1.24, t = -4.0). Across all windows, the sensitivity of the bond return to the stock return flips sign with the correlation regime: a slope of -0.16 (t = -4.7) when correlation is zero or negative and +0.15 when it is positive (the interaction is +0.32, t = 5.8). A positive correlation means that bonds fall with stocks, which is what the definition says. The point of the exercise is that the prior correlation, measured before the sell-off, was informative about it.

Two caveats limit how far to push this.

The first is the era overlap. Of the 913 sell-off windows, only 21 before 1990 had a negative prior correlation, and only 89 after 1990 had a positive one. Within the post-1990 sample the gap is still present (+5.8% in the negative-correlation windows against +2.3% in the positive ones), but the second figure rests on 89 overlapping windows from a few episodes. Excluding 1979 to 1982, the main result is unchanged (+5.6% against +0.7%).

The second is the number of independent events. Counting only windows that start at least 63 trading days apart, and taking the earliest qualifying window each time, leaves 39 windows. Of these, 20 had a zero or negative prior correlation (mean bond return +2.7%, bonds negative in 30%) and 19 had a positive one (mean -1.2%, bonds negative in 63%). The direction matches the full sample and the gap is smaller. The 39 windows still come from a small number of episodes. The HAC t-statistic above corrects the standard error for overlap but does not create independent episodes.

5.4 The years both asset classes lost

In 65 calendar years, stocks and bonds both lost in only three: 1969, 1994 and 2022. In those years, the full-year correlation was +0.29, +0.62 and +0.17. All three were years of rising yields: 1994 added 214 bp to the 10-year yield and 2022 added 236 bp. The three worst 60/40 years were 2008 (-20.1%), 1974 (-18.6%) and 2022 (-18.0%). Only 2022 had both legs down. In 2008 bonds returned +20.5% while stocks lost 41.6%. In 1974 stocks lost 31.5% and bonds gained 4.0%. In all, 14 of 65 calendar years were negative for 60/40.

The maximum drawdown of the 60/40 portfolio was -34.3% (March 9, 2009). The stock market's was -57.8% and our constructed bond series -27.1% (October 19, 2023). As of August 31, 2026, 60/40 sits 1.2% below its peak.

6The last six weeks

The 10-year yield averaged about 4.5% in the first days of June, reached 5.31% on October 5 and closed at 5.19% on October 14. The Fed's hike was on September 16. CNBC reports a deficit of about $2 trillion, a war with Iran that began February 28 and mortgage rates near 7.5%.⁵ These are the circumstances in which the bond market is moving. The measurement above says the moves are bigger than in the summer and not rare against history.

The published MOVE levels, 104.58 on September 24,² 107 on October 4¹ and 106.6 on September 29,³ show that expected yield volatility is elevated against the realized 85 bp. We cannot place those readings in a long history without the licensed series, so we do not.

What the data do say is the following. The portfolio question at today's correlation of +0.30 is more consequential than the volatility-ratio question. With the correlation at this level, bonds contribute 12% of 60/40 variance and the historical evidence in Section 5 says they have been weak protection in stock sell-offs. That is true at a ratio of 0.40 as well as at 0.80.

7Relation to prior work

AQR. In April 2026 AQR argued that a positive stock-bond correlation is a poor reason to add equity risk to a portfolio, and that the correlation sign has been historically mixed.⁶ The earlier Journal of Portfolio Management article shows that the correlation depends on the relative volatility of growth and inflation news and the correlation between them, with a model that explains about 70% of long-term variation.⁷ We do not model the macro drivers. Our contribution is on the market side: the volatility ratio has no forecasting content beyond stock volatility, the ratio of basis-point volatility to VIX contains a unit effect that rises with the yield level, and bond returns in stock sell-offs are different depending on the prior correlation sign.

yieldcurve.pro. The firm's note finds that VIX and MOVE describe stress and do not predict it, and reports a MOVE-to-VIX ratio of 6.65 that is above 89% of days since November 2002, and a "risk multiple" of 2.08 against a median of 2.50.³ We agree with the conclusion and extend it in three ways: realized volatility from 1972 rather than implied from 2002, which adds the 1980 to 1986 and 1993 to 1995 stretches when bond volatility did exceed stock volatility; the conversion of bp volatility into price terms; and the portfolio decomposition.

MOVE-to-VIX commentary. CrossVol and Fundstrat describe the gap as a divergence.¹² Our realized data do not contradict that the bond market is more volatile than a month ago. They say that, relative to stocks and in price terms, today's bond volatility is mid-range.

8Concluding thoughts

The idea that bonds have become rougher than stocks fails on the data. In price terms, bonds were more volatile than stocks over a 63-day window on about 3% of days since 1962. Almost all of those days fall in the Volcker disinflation and its aftermath (1980 to 1986) and the 1993 to 1995 bond rout, with five days in 2011 and one in 1971. Today's ratio is 0.40. In 21-day terms against VIX it is 0.39. The bp-based figure of 5.0 against VIX is also unremarkable in realized terms, and the larger number in the market commentary is mostly a unit and level effect plus an implied volatility premium we cannot test.

A high ratio is not information about stocks. It marks a quiet stock market. Once current stock volatility is in the regression, the ratio's coefficient is zero over 54 years and in every subsample we looked at.

The question that matters for a 60/40 holder is what bonds do when stocks fall. The answer in this sample depends on the sign of the stock-bond correlation before the sell-off. When it was zero or negative, bonds gained 5.6% on average in sell-off windows and lost money 13% of the time. When positive, bonds gained 0.6% and lost money 47% of the time. The correlation is positive now, at +0.30, so the more realistic framing of the current risk is that bonds are a smaller diversifier than in 2000 to 2021, not that bonds are a source of stock-market trouble.

Where this could be wrong. The bond series is constructed from yields, not an index, and the 1960s yields are coarse. Stock data end August 31, so the last six weeks are measured against VIX only. The sell-off evidence has 39 non-overlapping windows from a small number of episodes, and the negative-correlation regime is concentrated after 1990. Implied volatility has information we did not test, because the MOVE history is licensed. And a correlation that changes sign slowly says little about its next six months.

What would change the conclusion. A sustained ratio above 1, which has happened only in the Volcker-era and 1994 to 1995 stretches, would not be an equity warning on this evidence but would mean bonds are the larger risk in a 60/40 portfolio for as long as it lasted. The stock-bond correlation staying positive through a stock sell-off would confirm the regime reading. A sell-off in which bonds rallied at a positive prior correlation would count against it.

Notes

1. Data are as published at the retrieval date, October 15, 2026, and reflect the latest revision then available. FRED series (DGS10 and VIXCLS) run to October 14, 2026. Kenneth French Data Library stock returns end August 31, 2026, so every statistic that uses stocks and bonds together ends on that date.
2. The bond return series is constructed from the 10-year constant-maturity Treasury yield and is a proxy for a bond rolled at constant maturity. It ignores the roll between on-the-run issues and bid-offer costs. FRED reports the yield to the nearest basis point, and in the 1960s the yield was unchanged on about 30% of days, so realized volatility in that decade is biased low.
3. Volatility is the standard deviation of daily returns times the square root of 252. Windows are 63 trading days for the long history and 21 for the recent window and the VIX test. Percentiles use the full sample unless stated. The forecasting tests use percentiles of the expanding history with a burn-in of 2,520 days, so no later data enter the threshold.
4. The forecasting regressions overlap, so standard errors use a Newey-West correction with 63 lags (21 lags for the 21-day horizon). Sell-off windows overlap in the same way. Counts of windows are not counts of independent episodes.
5. The MOVE index is proprietary and its history is licensed. The only MOVE values in this paper are the published readings quoted from the sources cited, and none was recomputed.

Appendix A. Data series and construction

Constructed bond return. Each day we price a 10-year par bond with a semiannual coupon equal to the previous day’s yield, at the new yield and with 1/252 year less maturity, and add the coupon accrual. The 60/40 portfolio is rebalanced daily at weights of 0.6 for stocks and 0.4 for bonds. The bond share of portfolio variance is (0.4² σ_b² + 0.6 × 0.4 × cov(s, b)) divided by portfolio variance, over 252 days. The stock-bond correlation is the 252-day Pearson correlation of daily returns. Modified duration is computed numerically for a semiannual par bond with 10 years to maturity.

Checks. Against Damodaran’s annual 10-year Treasury return (1962 to 2024) the annual correlation is 0.988, the mean absolute difference is 1.0 percentage point and the largest difference is 6.7 points, in 1982. Against Shiller’s monthly series the correlation is 0.62. Our market series has an annual correlation of 0.985 with Damodaran’s S&P 500 return.

References

1. CrossVol Research, "Volatility is starting to talk, even", https://newsletter.crossvolresearch.com/p/volatility-is-starting-to-talk-even
2. Fundstrat, "Bond market and stock market volatility gauges are diverging", September 25, 2026, https://fundstratdirect.com/market-intelligence/first-to-market/2026/09/25/bond-market-and-stock-market-volatility-gauges-are-diverging
3. yieldcurve.pro, "VIX and MOVE describe market stress but do not predict it", https://www.yieldcurve.pro/blog/vix-and-move-describe-market-stress-but-do-not-predict-it
4. Federal Reserve Board, FOMC statement, September 16, 2026, https://www.federalreserve.gov/newsevents/pressreleases/monetary20260916a.htm; CNBC, https://www.cnbc.com/2026/09/16/fed-rate-decision-september-2026.html
5. CNBC, "Analysis: Why a $2 trillion deficit and 5% Treasury yields are becoming America's biggest economic challenge", October 15, 2026, https://www.cnbc.com/2026/10/15/treasury-yields-deficit-national-debt-ai.html
6. AQR, "A Positive Stock-Bond Correlation Is a Terrible Reason to Add More Equity Risk to Your Portfolio", AQR Perspectives, April 8, 2026, https://www.aqr.com/Insights/Perspectives/A-Positive-Stock-Bond-Correlation-Is-a-Terrible-Reason-to-Add-More-Equity-Risk-to-Your-Portfolio
7. Brixton, Brooks, Hecht, Ilmanen, Maloney and McQuinn, "A Changing Stock-Bond Correlation: Drivers and Implications", The Journal of Portfolio Management, listed by AQR February 28, 2023, https://www.aqr.com/Insights/Research/Journal-Article/A-Changing-Stock-Bond-Correlation
8. Damodaran, Historical Returns on Stocks, Bonds and Bills, https://pages.stern.nyu.edu/~adamodar/New_Home_Page/datafile/histretSP.html
9. Kenneth R. French Data Library, Fama/French research factors (daily), https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html
10. FRED, Federal Reserve Bank of St. Louis: DGS10, https://fred.stlouisfed.org/series/DGS10; VIXCLS, https://fred.stlouisfed.org/series/VIXCLS