Is a 7% Mortgage High? What Fifty-Five Years of Data Can and Cannot Tell You

On September 17, 2026, Freddie Mac published a 30-year fixed mortgage rate of 6.95%.¹ Measured against every weekly observation the survey has produced since April 2, 1971, that rate sits at the 45.5th percentile. About 45.5% of the 2,895 published weeks were at or below it, and the median week of the past 55 years was 7.21%.

Now change one thing. Start the sample in 2000 instead of 1971, and the same 6.95% is the 91.1st percentile. Start it in 2009 and it is the 96.8th. Nothing about the rate changed. Nothing about the data changed. The only thing that moved was the first row included in the calculation, and the answer moved 51 percentile points.

Both numbers are correct. They answer different questions, and most arguments about whether today’s mortgage is expensive are really arguments about which question is the right one. The same percentile gets cited to tell buyers they are complaining about a normal rate and to tell them the market has broken, so it is worth knowing what the statistic can carry.

Where 6.95% sits in the full record

Freddie Mac’s Primary Mortgage Market Survey is the longest continuously published series of U.S. mortgage rates, running weekly since April 1971.¹ Across all 2,895 observations through September 17, 2026:

The series peaks at 18.63% in the week of October 9, 1981 and bottoms at 2.65% in the week of January 7, 2021. Someone whose sense of a normal mortgage formed between 2012 and 2021 anchored on a decade whose median week was 3.83%, the 9th percentile of the full record, and 98% of whose weeks fall in its bottom fifth.

A reasonable first suspicion is that weekly data is doing something strange, since weekly observations cluster during long flat stretches. It is not. Averaging the same data to monthly puts 6.95% at the 45.2nd percentile, and averaging to annual puts it at the 46.4th. The sampling frequency is worth about one percentile point.

One caveat belongs here. Freddie Mac replaced its survey of roughly 125 lenders with application data from its Loan Product Advisor system on November 17, 2022. Freddie measured the difference between the two approaches for the 30-year at 0.066 percentage points over 2005 to 2021, with a standard deviation of 0.074 points, and backcast the new methodology to January 2005, so there is no raw splice at the changeover.² Two details matter more than that 6.6 basis points. The current series measures rates at application rather than at origination, and the companion series for discount points and origination fees was discontinued, so any comparison of all-in borrowing cost across 2022 is comparing two different things.

Inflation-adjusting moves today between the 39th and 67th percentile

The obvious objection to that percentile table is that it spends its first decade inside the Great Inflation. A 12% mortgage against 13% inflation is a different contract from a 12% mortgage against 2% inflation, and a median computed across both is measuring something blurry. Fisher’s relationship says a nominal rate is approximately a real rate plus expected inflation, so the comparison people want is on the real side.

That objection is worth taking seriously, and it fails. Subtracting trailing CPI-U inflation from each week’s published rate moves today’s position around, and never near the top of the distribution:

The range across five deflators runs from the 39th to the 67th percentile. That spread is real and it is worth reporting in full rather than quoting the most convenient end of it. The three-year and core readings sit higher because the 2022 and 2023 inflation spike is still inside a three-year window, and because core CPI ran 2.4% in August 2026 against 3.4% headline, an unusual configuration in which food and energy account for the gap.³ Any real mortgage rate you see quoted should say which inflation measure produced it.

What none of the five readings support is the claim that today’s mortgage is historically extraordinary. Deflating does not rescue that reading.

The 1970s were a cheaper loan in real terms than today

What inflation adjustment does change is which decade looks painful. The cleanest way to see it is to ask what each vintage of borrower paid in the end: the coupon at origination minus the inflation that was realized over the following ten years.

Two familiar arguments come out of this in opposite condition. A borrower who signed for 9.06% in June 1974 paid 1.28% a year after the inflation that followed, which is cheaper in real terms than anything available in 2026. So “my parents managed a 9% mortgage” is a weak argument: their loan was a bargain that today’s buyer cannot get. A borrower who signed for 18.63% in October 1981 paid 14.69% a year in real terms over the following decade. So “people paid 18% in 1981” describes something genuinely brutal.

That second figure needs a caveat, and the caveat is itself the interesting part. It measures the cost of holding that coupon for ten years, and the 1981 cohort did not hold it. Rates fell through the 1980s, and borrowers refinanced. The 1970s cohort had no such option worth taking, because rates rose after they signed, so they did hold their cheap real loans to term. American fixed-rate borrowers refinance when rates fall and sit still when rates rise, and that asymmetry is the prepayment option that mortgage investors price for.⁴ It is also why a 30-year mortgage is compared against the 10-year Treasury rather than the 30-year: the option shortens its effective life.

The theory behind the 1970s result is older than the data. Poterba showed in 1984 that inflation cuts the real cost of owning because nominal mortgage interest is deductible while the inflationary part of a house’s appreciation is never taxed. His simulations attribute as much as a 30% rise in real house prices to the increase in expected inflation over the 1970s.⁵ Anyone comparing a 7% mortgage in 1978 to a 7% mortgage in 2026 is comparing two different economic objects.

This is the same mechanism that makes a fixed-rate mortgage a partial inflation hedge, and it works only on the unexpected part. We cover where that framing holds and where it breaks in Is a 30-Year Fixed Mortgage an Inflation Hedge?

Try it: the mortgage rate percentile explorer

The explorer below runs the same calculation on the full Freddie Mac series in your browser. Drag the start year and the distribution redraws against the window you selected; the toggle switches between the published rate and the rate less trailing twelve-month CPI-U. The opening default is the whole record since 1971.

Dragging from 1971 to 2000 takes the reading from the 45.5th percentile to the 91.1st, and the window median from 7.21% to 5.22%. Dragging to 2009 reaches the 96.8th. Every one of those percentiles is arithmetically correct on the data it describes, which is the problem worth understanding.

Why the sample start changes the answer by 51 points

A percentile drawn from 2,895 observations sounds like a well-measured statistic. It would be, if those observations were 2,895 independent draws from a stable process. They are nothing like that.

The simplest way to see it needs no model at all. Over 55 years, the weekly series crosses its own 7.21% median just 24 times, in 30 runs. Two of those runs last more than two decades each:

• 1,162 consecutive weeks above the median, from April 1971 to July 1993. That is 22.3 years without a single week below 7.21%.
• 1,153 consecutive weeks below the median, from July 2001 to August 2023. That is 22.2 years without a single week above it.

The lag-1 weekly autocorrelation is 0.9994, and the autocorrelation function does not cross zero until a lag of 879 weeks, about 17 years. A standard effective-sample-size calculation on this series returns roughly four. That calculation assumes a stationarity the series plainly lacks, so the number should not be quoted as if it were precise, but the direction it points is right: the record contains a handful of long episodes, not thousands of independent observations.

Seen that way, the 7.21% median stops looking like a central tendency. It is close to the midpoint between one 22-year regime that sat above it and another 22-year regime that sat below it, which makes it an artifact of where a rising-then-falling arc happened to be cut. The start-date table falls out of the same fact:

There is published support for treating the series this way. Bauer and Rudebusch show that models of the Treasury curve which assume a constant long-run mean misread yield dynamics, because trend inflation and the equilibrium real rate both shift over time.⁶ A 30-year mortgage is the long Treasury yield plus a spread for prepayment and credit risk and intermediation cost,⁷ so a shifting-endpoint Treasury process implies a shifting-endpoint mortgage rate. There is no fixed level for the series to average around, which is why the average of a 55-year sample is not an equilibrium.

Two questions with two right answers

The 45.5th percentile and the 91.1st are answers to different questions, and each is the correct answer to its own.

Is 6.95% unusual for an American 30-year mortgage, across the whole period anyone has measured? No. It is close to the middle, in nominal terms and after inflation. If the claim on the table is that a 7% mortgage is some kind of historical aberration, the full record refutes it, and 3% to 4% mortgages are the aberration instead.

Is 6.95% unusual for anyone currently shopping for a house? Yes, severely. Against the quarter-century that produced every expectation a living buyer or seller holds, today sits at the 91st percentile. A household comparing today to the only market they have ever transacted in is not committing an error of arithmetic.

Neither number is a forecast, which is the third thing a percentile cannot do. Nothing pulls a mortgage rate toward the empirical median of a past sample. Where rates settle depends on inflation expectations, the equilibrium real rate, fiscal supply, global saving, term premiums and mortgage spreads. The New York Fed’s estimate of the natural rate of interest stood near 1.65% in the second quarter of 2026,⁸ and that estimate is itself model-bound and revised often. On September 16, 2026 the FOMC raised its target range to 3.75% to 4.00%, the first increase since 2023,⁹ while the 10-year Treasury closed the following day at 4.94%¹⁰ and the mortgage rate printed 201 basis points above it.

What a middle-of-the-distribution rate still costs

None of this says a 7% mortgage is cheap, and the arithmetic of the payment is where the percentile stops being relevant.

Principal and interest only, 30-year term. Moving from 3.00% to 6.95% cuts what a fixed $2,500 monthly budget can borrow by 36.3%. For the payment to stay flat, the price would have to fall by the same 36.3%, and prices did not do that.

Against the current market: the median existing home sold for $429,100 in August 2026,¹¹ and real median household income was $87,460 for calendar 2025.¹² With 20% down at 6.95%, principal and interest alone come to $2,272 a month, or 31.2% of that income, before property taxes, insurance and maintenance. The same house at 3.00% would be 19.9%. The price is 4.91 times income.

So a mortgage rate near its 55-year median coexists with unusually poor affordability, because affordability is a function of price, rate and income together, and only one of the three is middling. We have treated the pieces of that elsewhere rather than repeat them here:

• How Much of a Ten-Year Treasury Yield Is Risk Compensation? for why Fed cuts did not lower mortgage rates, and for the lock-in effect on existing owners.
• Why a 50-Year Mortgage Won’t Fix Housing Affordability for why lowering the payment without adding supply capitalizes into the price.
• How Much House Can High Earners Really Afford? for the household-level version, where three standard frameworks disagree by more than a million dollars.

For your own balance sheet, what decides the question is your after-tax borrowing cost against what you can earn on low-risk assets, how much liquidity you need, and what else the capital could do. A rate can sit at the 45th percentile of the historical record and still be the wrong rate for you to sign.

Bottom Line

A 6.95% mortgage on September 17, 2026 sat near the middle of the Freddie Mac record since 1971, in nominal terms and after inflation, and near the top of the record since 2000. The full-sample percentile is a fair correction to anyone who calls 7% unprecedented, and it is not evidence that rates will fall back toward 7.21%, that the 2010s were an anomaly the market owes a return from, or that buying a house today is as manageable as the rate alone suggests. The series holds a few long regimes rather than thousands of independent draws, so any percentile computed from it should be quoted with the window that produced it.

Key Takeaways

• Today’s rate is mid-distribution since 1971 and near the top since 2000. On September 17, 2026, 6.95% was the 45.5th percentile of 2,895 weekly observations back to April 1971, and the 91.1st percentile of the 1,394 weeks since 2000. Quote the window whenever you quote the number.
• Inflation adjustment does not change that conclusion. Across five trailing-CPI deflators, today’s real rate lands between the 39th and 67th percentile of the 1971 to 2026 record. The Great Inflation is not what makes the median 7.21%.
• The 1970s were a cheaper loan in real terms than 2026. Loans originated 1971 to 1979 carried a median 8.89% coupon and cost a median 2.00% a year after the inflation that followed. Loans originated in the 1980s cost 8.70%. The decade with the scary nominal rates and the decade with the expensive loans are not the same decade.
• 2,895 weekly observations are a handful of long episodes. The series crosses its own median 24 times in 55 years, including one 22.3-year run above it and one 22.2-year run below it. Lag-1 autocorrelation is 0.9994. The 55-year median is where a long arc got cut, and nothing pulls rates back toward it.
• A middling rate and poor affordability are both true at once. Moving from 3.00% to 6.95% cuts what a $2,500 monthly payment can borrow by 36.3%. At August 2026’s $429,100 median existing-home price and 2025’s $87,460 real median household income, 20% down at 6.95% puts principal and interest alone at 31.2% of income.

Frequently Asked Questions

Is a 7% mortgage rate high?

It depends entirely on the comparison period, and the difference is large enough to reverse the answer. Against every week Freddie Mac has published since April 1971, 6.95% on September 17, 2026 was the 45.5th percentile, slightly below the 7.21% median. Against the weeks since 2000, it was the 91.1st percentile. Both figures come from the same series and the same rate.

What is the median 30-year mortgage rate since 1971?

7.21%, across 2,895 weekly observations from April 2, 1971 through September 17, 2026. The mean is 7.68%, the peak is 18.63% in October 1981 and the trough is 2.65% in January 2021. Treat the median as a description of that sample rather than as a level rates tend toward: the series spent 22 straight years above it and then 22 straight years below it.

Were 3% mortgage rates normal?

No. Rates at or below 4% account for 12.7% of all published weeks since 1971, and rates at or below 3% for 1.9%. Every one of those weeks falls between 2011 and 2022. They reflected low inflation expectations, a low equilibrium real rate, and central bank purchases of Treasury and agency mortgage securities, and researchers still disagree about how much of the decline to attribute to the purchases as against the underlying market conditions.

Will mortgage rates go back to 3%?

Nothing in this analysis supports a forecast in either direction, and a historical percentile is not one. Future rates depend on inflation expectations, the equilibrium real rate, fiscal borrowing, global saving and investment, term premiums and mortgage spreads. Buying a house on the assumption that a refinance will rescue the payment treats a valuable option as a certainty. A purchase should work at the rate you sign, with a refinance counted as upside.

Why did my mortgage rate not fall when the Fed cut rates?

The Fed sets an overnight rate. A 30-year mortgage prices off long-term yields plus a spread for prepayment risk, credit risk and the cost of intermediation. On September 17, 2026 the effective federal funds rate was 3.88%, the 10-year Treasury was 4.94%, and the mortgage rate was 6.95%, or 201 basis points above the 10-year. Our guide on the yield curve, term premium and r-star works through the mechanism.

Was buying at 18% in 1981 harder than buying at 7% today?

On the financing alone, yes, and by a wide margin: a $400,000 loan costs $2,648 a month at 6.95% and $6,234 at 18.63%, and the 1981 coupon cost 14.69% a year after the inflation that followed. The comparison still does not settle affordability, because it holds the price fixed when the price is the variable that moved most. Price relative to income, the size of the required down payment, and the rate all enter, and only the rate is middling by historical standards.

Does the 2022 methodology change make the older data unusable?

No, though it makes the series something other than one homogeneous instrument across 55 years. Freddie Mac moved from a lender survey to Loan Product Advisor application data on November 17, 2022, measured the 30-year difference at 0.066 percentage points over 2005 to 2021, and backcast the new method to January 2005. The larger discontinuities are that rates are now measured at application rather than origination, and that the discount points and origination fees series was discontinued, which affects any pre- and post-2022 comparison of all-in cost more than 6.6 basis points does.

Related Guides

• How Much of a Ten-Year Treasury Yield Is Risk Compensation? for the decomposition behind the mortgage rate, and the lock-in effect on existing owners.
• Is a 30-Year Fixed Mortgage an Inflation Hedge? for why the 1970s cohort came out ahead, and when the framing breaks down.
• Why a 50-Year Mortgage Won’t Fix Housing Affordability for what happens when a policy lowers the payment without adding supply.
• Does the Fed Really Set Interest Rates? for why the overnight rate and the mortgage rate move separately.
• A 9% Mortgage Does Not Lose to 10% Stocks for what a rate at this level implies about paying the loan down versus investing.
• Did Buying This House Beat Renting? for a 24-year worked example of the ownership decision at an earlier point in this same distribution.

Sources

1. Freddie Mac, “Primary Mortgage Market Survey,” 30-year fixed-rate mortgage average, consolidated weekly history (PMMS_history.csv), April 2, 1971 through September 17, 2026. All percentile, median, run-length and autocorrelation figures above are computed from that file. freddiemac.com
2. Sam Khater, Len Kiefer and Mihwa Kim, “Freddie Mac’s Newly Enhanced Mortgage Rate Survey Explained,” Freddie Mac Economic and Housing Research Note, November 2022. Source of the November 17, 2022 transition date, the 0.066 percentage point estimated difference for the 30-year, the backcast to January 2005, and the discontinuation of the points and fees series. freddiemac.com
3. U.S. Bureau of Labor Statistics, “Consumer Price Index,” news release of September 11, 2026 (August 2026 reference month: headline CPI-U 3.4%, core 2.4% over twelve months), and series CUUR0000SA0 and CUUR0000SA0L1E from the BLS public API for the historical deflators. bls.gov
4. Gene Amromin, Neil Bhutta and Benjamin J. Keys, “Refinancing, Monetary Policy, and the Credit Cycle,” Annual Review of Financial Economics 12 (2020), pp. 67–93. A review article; cited here for the intensive and extensive margins of refinancing and for the share of U.S. originations that are 30-year fixed-rate loans. nber.org
5. James M. Poterba, “Tax Subsidies to Owner-Occupied Housing: An Asset-Market Approach,” Quarterly Journal of Economics 99(4), November 1984, pp. 729–752; working paper version NBER 553 (1980). Source of the estimate that 1970s expected inflation could account for as much as a 30% increase in real house prices. nber.org
6. Michael D. Bauer and Glenn D. Rudebusch, “Interest Rates under Falling Stars,” American Economic Review 110(5), May 2020, pp. 1316–54. Shows that accounting for time variation in trend inflation and the equilibrium real rate is necessary to model Treasury yield dynamics. aeaweb.org
7. Nina Boyarchenko, Andreas Fuster and David O. Lucca, “Understanding Mortgage Spreads,” Federal Reserve Bank of New York Staff Report No. 674, May 2014, revised June 2018; published in Review of Financial Studies 32(10), 2019, pp. 3799–3850. Attributes the cross-sectional pattern in agency mortgage spreads to prepayment risk and the time-series variation to a non-prepayment factor co-moving with securities supply and credit risk. newyorkfed.org
8. Federal Reserve Bank of New York, “Measuring the Natural Rate of Interest,” Holston-Laubach-Williams model estimates; reading of approximately 1.65% for 2026 Q2. newyorkfed.org
9. Board of Governors of the Federal Reserve System, FOMC statement of September 16, 2026, raising the target range for the federal funds rate to 3.75% to 4.00%. Effective federal funds rate of 3.88% from the Federal Reserve Bank of New York. federalreserve.gov
10. U.S. Department of the Treasury, Daily Treasury Par Yield Curve Rates, 2026. The 10-year constant maturity closed at 4.94% on September 17, 2026, and the 10-year TIPS real yield at 2.61%. home.treasury.gov
11. National Association of Realtors, “Existing-Home Sales,” August 2026, released September 10, 2026. Median existing-home price $429,100. nar.realtor
12. U.S. Census Bureau, “Income in the United States: 2025,” Report P60-289, released September 15, 2026. Real median household income of $87,460 for calendar year 2025. census.gov

Author disclosure

No business relationship with Freddie Mac, the National Association of Realtors, or any lender. Every percentile, median, run length, autocorrelation and real-rate figure above was computed from the Freddie Mac PMMS weekly file and BLS CPI-U series by a script published with the site, and the explorer recomputes them in your browser from the same data, so the two agree by construction. Real rates here subtract realized trailing inflation, which is a proxy for the expected inflation a borrower actually prices against and not a measurement of it; the ex-post vintage figures subtract inflation that was unknowable at origination. The effective sample size of roughly four assumes a stationarity the series does not have and is offered as an illustration rather than an estimate. Rates quoted as of September 17, 2026 change weekly. Nothing here is investment advice.