Refinance kept falling, and the Fed decides Wednesday

The week in rates

Sources: Freddie Mac PMMS national averages, Sep 10 2026, and the 10 year Treasury constant maturity via FRED, series DGS10, Sep 9 2026, the most recent value published. These are published national benchmark and average rates from the sources named above. They are not an offer, not a quote, and not a rate anyone is being offered by anyone, including me.

The 30 and 15 year figures describe conventional conforming purchase loans at 80 percent loan to value for a borrower with excellent credit, because that is what the Freddie Mac survey measures. That says nothing about what any individual qualifies for, and it is a particularly poor comparison for anybody with credit repair ahead of them, a small down payment, or a loan above the conforming limit.

The 10 year finished at 4.83 against 4.79 the same day a week earlier, a move of 4 basis points, and a basis point is one hundredth of a percentage point. The path matters more than the size. It fell 2 basis points to 4.77 on Sep 3, then climbed in every session after, 4.78 on Sep 4, 4.80 on Sep 8 and 4.83 on Sep 9, with Monday Sep 7 missing because the bond market was closed for Labor Day. So the week's range was 6 basis points wide even though the week over week move was 4. Then look at what those 4 basis points crossed. A week ago 4.79 was the highest reading since January 2025. Four basis points later, 4.83 is the highest since October 31 2023, which moved the comparison back more than a year on a move most people would not notice. That is what a thin trading range looks like from the inside. One more piece of shape: the benchmark and the survey moved together this week, 4 basis points against 5, where the week before the benchmark moved 13 against the survey's 5.

The benchmark behind second liens

Deliberately no average home equity loan or HELOC rate here. Second liens are portfolio products priced by each lender, so published averages for them disagree by close to a full point depending on the credit profile assumed, and quoting one would imply a precision that does not exist. Prime is different: it is a single published benchmark, the same number at every institution. A HELOC is priced at prime plus the lender's margin, so prime tells you which way that product moves when the Fed moves, and a fixed home equity loan does not move with it at all. That distinction is worth more than usual this week, because the Fed announces on Wednesday and prime conventionally follows the target range within a day or two. If the committee raises, most variable rate HELOCs reprice at their next billing cycle, though lines carrying rate floors, periodic caps, or a fixed rate conversion option may not move in step. A fixed second lien does not move at all. The margin is still the part that varies by lender, and the only honest way to know a client's is to price the file.

Sources: Bank prime loan rate via FRED, series DPRIME, Sep 9 2026. These are published national benchmark and average rates from the sources named above. They are not an offer, not a quote, and not a rate anyone is being offered by anyone, including me.

What applications did

Total application volume was down 2.7 percent for the week on a seasonally adjusted basis, and down 4 percent unadjusted. Purchase applications were down 0.2 percent for the week seasonally adjusted, with the unadjusted index down 3 percent for the week and 4 percent above the same week in 2025. Do not lift that 4 percent into a client email without reading the Labor Day section further down, because the week it is measured against was a holiday shortened one and the comparison is flattered by the calendar. Refinance applications were down 6 percent for the week and 25 percent below the same week a year ago, and that 25 percent is measured against the same holiday shortened week, so if anything it understates the decline. The release also showed the refinance share of all applications falling to 40.9 percent from 41.8. The adjustable share rose again to 8.5 percent, and the MBA put the average contract rate on a 30 year conforming loan at 6.85 percent for the week, its highest since June 2025.

Source: MBA Weekly Applications Survey, week ending Sep 4 2026, released Sep 9 2026, week over week figures seasonally adjusted unless stated otherwise, as publicly reported. Any rate or share named in this section is a published national average of applications already taken. These are published national benchmark and average rates from the sources named above. They are not an offer, not a quote, and not a rate anyone is being offered by anyone, including me.

The borrower picture

These numbers move slowly, but they decide which conversations are worth having this week. Each carries its own date because they refresh on different schedules.

Every figure above is a published national aggregate from the source named beside it. Any rate shown here, including the credit card average, is a national average and not an offer, a quote, or a rate anyone is being offered.

If you are the borrower reading this

If you are floating a rate and closing in the next several weeks, this is the week your loan officer earns the relationship, and you should not wait for them to call. Send one message today and ask two things: what it would cost to lock today, and what an extension would cost in dollars if your closing slipped. Ask before tomorrow morning rather than after, because the inflation report comes out at 8:30 Eastern and the Fed announces Wednesday, and pricing can move on both. Nobody can tell you which way, and anyone who says otherwise is guessing at your expense. If your file is not complete, or you are not under contract yet, the honest answer may be that pricing cannot be quoted on your loan at all, and a loan officer who tells you that plainly is being straight with you. If you are shopping rather than buying, ignore the national average and find out what a payment looks like on your own inputs, which is what the calculators on this site do. Separately, one thing costs nothing and is worth doing this month whatever rates do. Get a second quote on your homeowners insurance. Homeowners who shopped and switched carriers cut their premiums 6.6 percent over the past year, and got better coverage while they were at it, while homeowners who stayed with the same carrier were charged 10.4 percent more. Read that as the gap between two groups rather than as a promise to you, because the people who shop are not a random sample of homeowners and your own result depends on your home, your claims history, and what carriers will write in your area. Two things to watch when you do it. Check that a lower premium is not simply a higher deductible, which is not a saving, it is you agreeing to pay more when something actually happens. And take coverage questions to an insurance agent rather than to me, because I am not one. And if you own your home and are carrying credit card debt, the question worth asking is not what your mortgage rate is. It is what your borrowing costs across everything, the mortgage and the cards together. Put your real balances into the blended rate calculator and you will usually find the mortgage is the cheap money and the cards are the problem. That does not automatically mean you should borrow against the house. The answer is often a smaller second loan, and sometimes it is nothing at all.

What this means for your week

The rest of this is written for the loan officers reading. If you are a homeowner or a buyer, stay anyway: this is what a good conversation with one should sound like from the other side of the desk.

Call every floating file today, before CPI tomorrow and the Fed on Wednesday

That is the whole instruction and the rest of this section is why it holds. Start with the thing worth knowing that nobody will tell your client: a move of 4 basis points this week put the 10 year at its highest since October 31 2023, which means there is almost no prior trading range above here until you are back in 2023. That does not predict anything and you should not use it as a prediction. It tells you the benchmark is at the top of everything recent, which is a poor moment to let a borrower drift. Here is the state of play going in, and it is short on purpose because you have already read the Fed coverage this week. The federal funds target range, which is the band the committee sets for the rate banks charge each other overnight, is 3.50 to 3.75 percent. The committee held it there in July by a vote of nine to three, so this is a genuinely divided committee rather than one being reported as divided. The August jobs report on September 4 came in at 162,000 payrolls against a consensus near 53,000, with unemployment holding at 4.1 percent and average hourly earnings up 3.1 percent over the year, and the benchmark kept climbing in the sessions that followed. Market implied odds of an increase have run somewhere between roughly 48 and 66 percent across CME FedWatch, Kalshi and Polymarket since the Jackson Hole keynote on August 28, which is to say the market does not know either. Do not quote an odds number to a client. It will be wrong by the time they repeat it to their spouse. So the instruction is one line and it has not changed: every floating file gets a call today, before tomorrow's 8:30, and the ones still floating on Monday get a second call before Wednesday afternoon. Locking protects today's pricing for a set number of days, floating is a bet on data none of us have seen, and you should not pretend to know which one wins. Some clients will float anyway, which is a legitimate answer as long as they chose it on purpose. What you cannot survive is a borrower who finds out Friday afternoon that nobody called them Thursday. Every rate and range named in this block is a published national figure from the sources dated above. None of them is an offer, a quote, or a rate anyone is being offered.

Next week's application numbers are likely to look like a collapse, and they will not be one

Two calendar quirks are sitting in the applications data back to back, and they point in opposite directions. Start with the one already published. The unadjusted purchase index came in 4 percent above the same week in 2025, which reads like buyers showing up in a week when the average rate reached its highest since June 2025. Check the calendar before you repeat it. The week it is compared against ended September 5 2025, and Labor Day 2025 fell on Monday September 1, inside that week. This year's week ended September 4, and Labor Day 2026 fell on Monday September 7, outside it. So this year's week had five business days and the year ago week had four, and some of that 4 percent is the extra day rather than extra buyers. Now be careful about what you reach for to correct it, because this is where the argument usually goes wrong. The minus 0.2 percent seasonally adjusted purchase figure is a week over week number measured against the week ending August 28. It does not fix the year over year comparison and you should not offer it as the corrected version. The survey does not publish a business day adjusted year over year purchase figure at all. The honest position is that the plus 4 percent is flattered by the calendar and nobody can tell you by exactly how much. You can, though, say something about the size of it. A five business day week beat a four business day week by 4 percent, while the available days alone were up about a quarter. Applications are not proportional to business days, so do not turn that into a number and do not put a figure on it in writing. But the direction is not really ambiguous: underlying purchase demand is more likely running a little below last year than above it, which is the opposite of what the headline says. The same caution cuts the other way on refinance, and you have to say so or you are just spinning. That minus 25 percent year over year figure is measured against the same four day week, so if anything it understates the decline. Do not use the calendar to explain away only the number you dislike. Now the part you can get ahead of. Next week's survey, covering the week ending September 11 and released Wednesday September 16, is the week that contains this year's Labor Day. It will have four business days against last year's five, so the unadjusted numbers should be expected to look poor for that reason alone before you count anything real. So send the note now. Two sentences to your agents and your builder contacts: next week's mortgage application headlines are likely to show a sharp drop in the unadjusted numbers, a large part of it is the Labor Day holiday falling inside the survey week, and the number to watch is the seasonally adjusted purchase index rather than the headline. One honest limit before you send it. The MBA's own seasonal adjustment is designed to account for holiday weeks, so what you are cautioning about is the unadjusted figures and the year over year comparison, not a claim that their adjustment is broken. Say unadjusted when you send it and you cannot be wrong.

The insurance line is the only part of the payment where the increase is slowing

ICE published its September Mortgage Monitor this morning and the subject is property insurance, which is the one input to a housing payment currently moving in the borrower's favor. Nothing actually got cheaper, so do not tell anyone it did. What changed is the direction. Insurance costs are up 8.7 percent over the year, which is down from 11.4 percent at the start of this year and down from a peak of 15.1 percent at the end of 2024, and the gain in the second quarter was 1.8 percent, the smallest since ICE began tracking it. Insurance now accounts for 9.6 percent of the average monthly mortgage payment nationally, and the spread across the country is enormous, from 24.3 percent of the payment in New Orleans to 4.3 percent in San Jose. Here is the finding that turns this into a phone call. Homeowners who shopped and switched carriers cut their premiums by 6.6 percent, a record, and they did it while raising their coverage limits 7.3 percent and lowering their deductibles 1.4 percent, so they bought more protection for less money. Homeowners who stayed with their carrier were charged 10.4 percent more. Same year, same market, opposite outcomes. Now say the honest thing about that comparison before you repeat it to anybody, because it is where this argument usually gets oversold. Homeowners who shop are not a random sample of homeowners. They are disproportionately the ones who just got hit with a large renewal increase, who live in markets with real competition, and whose risk profile a new carrier will actually take. So treat those two numbers as the size of the gap between two groups, not as what any one client is going to get. Now size the gap, because the arithmetic is not what most people assume. Falling 6.6 percent is measured against the switcher's own premium last year, but your client's real alternative is not last year, it is staying, and stayers were charged 10.4 percent more. The gap between the two outcomes is about 15 percent. Then remember that 15 percent applies to the insurance line and not to the whole payment. Where insurance is 9.6 percent of the payment, the gap works out to roughly one and a half percent of the payment. Where it is 24.3 percent, it is about three and seven tenths of one percent. Where it is 4.3 percent, it is about seven tenths. So this is a real lever in a high premium market and a rounding error in a low one, and in neither case is it a rescue. It belongs on the file sitting a hair outside guidelines and it will do nothing for the file sitting two points outside. Now split the work, because live files and closed clients are two different jobs and running them together is why this angle usually dies on a Monday. On live files, run the revised premium through the affordability calculator with the client on the phone, which reports which ratio is actually binding, so they watch it move themselves rather than taking your word for it. For closed clients it is an email and not a call list, because the call has no revenue event attached and it will lose every time to a lock conversation. Sort that list by the escrowed homeowners insurance line from their closing file rather than by state. That is the number already sitting in every file, it sorts by real dollars at risk instead of by a proxy, and it beats sorting by state because the 24.3 percent figure is New Orleans specifically and not Louisiana generally. One timing point that decides whether the message is useful at all: insurance is genuinely shoppable in roughly the thirty to sixty days before the policy anniversary, and you cannot see anniversary dates for clients you closed three years ago, so write the email to be useful whenever it is opened rather than as an act now message. Before any of this touches your database, clear the list and the wording through your compliance desk. That book is your company's data and not yours, and outbound contact to past clients runs into do not call and telephone consumer protection rules that your shop, not this page, has to sign off on. Three more honest limits. A quoted premium is not a premium until there is a binder, which is the insurer's written confirmation that coverage is actually in force, so nothing goes in the file on a verbal. You are not an insurance agent and should not be advising on coverage levels or carriers, so point them at their own agent or an independent one and stay out of the middle. And a premium that fell only because the deductible went up is not a saving, it is a transfer of risk onto the client, which is worth saying out loud given that the homeowners in this data did the opposite.

The refinance business is down 25 percent, and the part still working is not about their rate

This is the standing play on this page and you have read a version of it here before. The reason to run it again this week is a date, and I will get to it. Refinance applications are 25 percent below a year ago and the refinance share fell to 40.9 percent of all applications. Note that the year ago week is the same four day Labor Day week from the section above, so if anything that 25 percent understates the decline. The wrong conclusion to draw is that there is no refinance business left. Look at what remains. Refinance is still four in every ten applications at a time when the MBA put the average contract rate on a 30 year conforming loan at 6.85 percent. The survey counts applications and not reasons, so it does not tell you why any of them applied and I am not going to pretend otherwise. But it is hard to build a rate and term case, meaning a refinance done only to lower the rate or shorten the term, when two thirds of every outstanding mortgage in the country carries a rate under 5 percent. What is left is people who need money. The table above says where it is going: ICE counts second liens carrying 54 percent of all equity extraction, HELOC balances stand at 459 billion dollars, and households carry 1.26 trillion dollars of credit card balances, with the accounts that actually revolve paying an average of 20.94 percent in the Federal Reserve's May reading. So the client worth calling is not the one with a high mortgage rate. It is the one with a low mortgage rate and expensive debt attached to everything else, and the entire skill is in not wrecking the first mortgage to solve the second problem. Do it in this order and it stays honest. Run the blended rate calculator first, with their real balances, so you and they can both see what their combined cost of borrowing actually is across the mortgage and the cards together. That single number does more work than any pitch, because it usually shows the mortgage is the cheap money and the cards are the problem, which tells you immediately that touching the first lien is the wrong move. Then take the second lien path and run it through the debt consolidation calculator so the real trade is visible: a lower monthly cost, against a balance that may now be stretched across a much longer term and secured by the house instead of unsecured. Say both halves, and say the house part especially, because a card balance cannot take the home and a second lien can. Now the date, and plan both branches before Wednesday rather than after. If the committee holds, nothing reprices. Your borrowers who floated and did not lock got a reprieve rather than a vindication, and you should be the one who tells them which of the two it was, because that is the conversation that gets them to lock next time. These second lien conversations lose their urgency but not their logic. If the committee raises, prime conventionally moves with the target range within a day or two, and most variable rate HELOCs reprice at their next billing cycle, with the exceptions noted higher up this page. That makes the fixed second lien conversation genuinely time sensitive for exactly the clients described above, and the note to send is one sentence: your line is likely to get more expensive at your next statement, here is what a fixed alternative would involve, and let us price your actual file rather than guess at it. Most of these conversations should end in a second lien rather than a refinance, and a fair number should end in neither. Both of those are the right answer sometimes, and you should be the one who says so even when the bigger loan pays better. Every rate named in this block is a published national benchmark or average from the sources dated above. None of them is an offer, a quote, or a rate anyone is being offered.

The week ahead

Two releases set up the next month of pricing and they are five days apart. The consumer price index for August lands Friday September 11 at 8:30 Eastern from the Bureau of Labor Statistics, the last major inflation reading the committee sees before it votes. The Federal Open Market Committee then meets Tuesday and Wednesday, September 15 and 16, and announces at 2 o'clock Eastern on Wednesday, with a Summary of Economic Projections attached that will tell you more about October and December than the decision itself tells you about September. One more release lands in between and it is the one from the Labor Day section above: the MBA applications survey comes out Wednesday September 16 covering the week ending September 11, the week containing this year's holiday, so the unadjusted figures should be expected to fall hard for calendar reasons before anything real is counted. Check live pricing before your calls on both mornings, and do not carry Thursday's assumption into Wednesday.

Where the numbers come from

Rates are the Freddie Mac Primary Mortgage Market Survey national average, published Sep 10 2026, and the 10 year Treasury constant maturity via FRED, series DGS10, Sep 9 2026, which is a daily yield rather than a closing price. The claim that 4.83 is the highest since Oct 31 2023, and that 6.76 is the highest weekly survey reading since June 2025, were both checked this week against the FRED series back to 2021. The prime rate is the bank prime loan rate via FRED, series DPRIME, Sep 9 2026. Application figures, the refinance and adjustable shares and the average contract rate are the Mortgage Bankers Association weekly survey for the week ending Sep 4 2026, released Sep 9 2026, as publicly reported. The Labor Day comparison is a calendar fact checked against the 2025 and 2026 holiday dates, Sep 1 2025 and Sep 7 2026, and is offered as a caution about the unadjusted figures and the year over year comparison rather than as any claim about the MBA's own seasonal adjustment. Employment figures are the Bureau of Labor Statistics Employment Situation for August 2026, released Sep 4 2026. The July FOMC vote and the current target range are from the Federal Reserve's published July 28 and 29 2026 materials. The Warsh keynote was delivered at the Jackson Hole Economic Policy Symposium on Aug 28 2026 and is published by the Federal Reserve Board. Odds of a rate increase are market implied pricing from CME FedWatch, Kalshi and Polymarket as publicly reported between Aug 28 and Sep 9 2026, they disagree with each other, and they change daily. Property insurance figures are the ICE Mortgage Monitor for September 2026, released Sep 10 2026. Household figures are from the New York Fed, the Federal Reserve, FHFA, ICE and Fidelity, each dated in the table above. Release dates and times in the week ahead are from the Bureau of Labor Statistics and Federal Reserve calendars, checked this week. This brief is educational market commentary. It is not an offer of credit, not a quote, not a commitment to lend, and not financial, tax, or legal advice. Figures are national aggregates and say nothing about what any individual borrower qualifies for.