Should you wait for lower rates to refinance? What 35 years of Freddie Mac data show
Waiting costs money for certain and pays only if the rate falls far enough. Of the 1,844 weeks since January 4, 1991 that have a full six months of Freddie Mac releases after them, the 30-year average was at least half a point lower six months later in 22.5% and at least half a point higher in 17.1%. On $350,000 at a rate 0.75 points above a quote, each month of delay costs about $219 in interest. Every figure here is history through the October 29, 2026 release: a fixed record, not a live rate and not a forecast.
Worksheet steps (13)
- What a month of waiting costs before any rate moves
- The record: one file, 2,901 weekly rows, last row October 29, 2026
- After every week since 1991: where the 30-year rate stood 3, 6 and 12 months later
- How far the rate moved: medians, the middle 80% and the extremes
- A record that begins in a falling market: the same measures by era
- 1,818 windows, roughly 35 separate years
- Starting near 7%: what the high-rate weeks do and do not show
- Five largest twelve-month rises and falls, with dates
- Turning a rate move into dollars: the anchor loan after a six-month wait
- How small a fall repays the wait, set beside the rise that erases the gain
- Refinance now and again later: the second closing cost
- What the survey average cannot tell you about your own quote
- Using the record without a forecast: five numbers to work out first
What a month of waiting costs before any rate moves
Waiting is the one certain part of the decision, so price it first. The anchor loan on this page owes $350,000 with 336 months (28 years) left at a note rate of 7.99%, and the refinance quote is 7.24%, which is the Freddie Mac 30-year average in the October 29, 2026 release set equal to the quote for the arithmetic. It is an average of purchase applications, not an offer to anyone. The new loan keeps the same 336-month maturity, so only the rate changes and no term restart is hidden in the savings. The payment falls from $2,611 to $2,434, a drop of $177 a month.
That drop understates what a month of delay costs. In month one the old loan charges $2330.42 in interest and the new one $2111.67, so the interest gap is $218.75, which is the balance times the rate gap divided by 12. The payment gap is smaller because the same lower-rate payment sends more to principal: $322.55 against $280.85, a difference of $41.71. Principal repaid is equity, not cost, so the cost of staying on the old rate one more month is the interest gap. The payment gap tells you how much your budget changes; the interest gap tells you what the delay costs.
The table gives both numbers for three balances and four gaps, all against the same 7.24% quote. The interest column is exact to the cent before rounding. A $200,000 loan at a 0.50-point gap loses $83 a month to delay; a $500,000 loan at a 1.50-point gap loses $625.
| Balance | Your note rate | Gap to the quote | Payment falls by | Interest saved in month one |
|---|---|---|---|---|
| $200,000 | 7.74% | 0.50 points | $67 | $83 |
| $200,000 | 7.99% | 0.75 points | $101 | $125 |
| $200,000 | 8.24% | 1.00 points | $135 | $167 |
| $200,000 | 8.74% | 1.50 points | $205 | $250 |
| $350,000 | 7.74% | 0.50 points | $117 | $146 |
| $350,000 | 7.99% | 0.75 points | $177 | $219 |
| $350,000 | 8.24% | 1.00 points | $237 | $292 |
| $350,000 | 8.74% | 1.50 points | $359 | $438 |
| $500,000 | 7.74% | 0.50 points | $168 | $208 |
| $500,000 | 7.99% | 0.75 points | $253 | $313 |
| $500,000 | 8.24% | 1.00 points | $339 | $417 |
| $500,000 | 8.74% | 1.50 points | $513 | $625 |
The Fed decisions and mortgage rates record uses a different but compatible measure in its waiting example: on $350,000 with 26 years left, a quote of 6.90% against a 7.40% note saves $113.55 a month in payment, and three months of that is $341. The interest gap on the same loan is $145.83 a month, or $437.50 over three months. Both are right for what they measure. This page counts interest because it is the money that leaves and does not come back, so keep the two apart when comparing pages: the payment saving is a budget figure and the interest gap is the cost of delay.
Counted to the end of a seven-year stay, with the $7,000 closing cost paid on either date and the balance owed at month 84 included, refinancing today beats staying on the old loan by $11,545. If the quote were still 7.24% after the wait, delaying three, six or twelve months costs $715, $1,427 and $2,840 against refinancing today. The first piece of each is the interest gap collected over the waiting months ($656, $1,314 and $2,630). The rest is interest on the principal the old rate left unpaid, which is $125, $251 and $505 of extra balance at the end of the wait and keeps accruing at 7.24% until month 84. These totals ignore taxes, escrow and the time value of money, so they are the right size for a decision and not a statement of exact dollars.
The record: one file, 2,901 weekly rows, last row October 29, 2026
Freddie Mac publishes the full weekly history of its Primary Mortgage Market Survey as a CSV file. As downloaded on November 3, 2026, the file holds 2,901 rows of 30-year averages from April 2, 1971 to October 29, 2026. The release of October 29, 2026 was 7.24% and is the newest row. Every statistic below is computed from that one file, so the tables agree with each other and with the prose. Because the figures are a fixed history through that release, a later Thursday release will not change them; they are not a live rate, and the page does not update itself.
The look-ahead is counted in releases: three months is 13 weekly releases ahead (91 days), six months is 26 (182 days) and twelve months is 52 (364 days). The file has a few short or long weeks around holidays, and every window used here spans within six days of the nominal 91, 182 or 364. The change for a window is the later rate minus the starting rate, in hundredths of a point, so a fall of "0.50 or more" means the later average was at least 0.50 points below the starting one. The start weeks run from January 4, 1991. The last start with a full three-month look-ahead is July 30, 2026, with six months April 30, 2026 and with twelve months October 30, 2025, which gives 1,857, 1,844 and 1,818 start weeks.
The page starts in 1991 on purpose. The file reaches back to 1971, but the weeks around 1979 to 1982 carry moves far larger than anything after 1991. With the 1971 start the largest twelve-month rise in the file is +5.87 points from the week of April 6, 1979, and the largest fall is -4.59 from the week of February 12, 1982. The share of twelve-month windows with a fall of 0.50 or more is 31.2% on the full file against 34.0% from 1991, and the share with a rise of 0.50 or more is 27.4% against 21.8%. The 1991 start keeps those two extreme episodes from setting the tone.
Two details of the series matter when reading it across decades. First, the way the average is gathered changed on November 17, 2022, when Freddie Mac stopped surveying lenders and began computing the average from loan applications submitted to its Loan Product Advisor system (Freddie Mac, PMMS page, checked November 3, 2026). Freddie Mac's November 2022 research note tested the new method against the old one for 2005 through 2021 and reported: "On average, the difference between the traditional survey approach and the new approach for the 30-year FRM rate is about 0.066 percentage points from 2005 through 2021, with a standard deviation of about 0.074 percentage points." That is small beside the half-point and full-point moves counted below, but it is a seam in the series, and how the Freddie Mac survey is built covers the borrower profile the average describes. Second, through the release of November 10, 2022 the file also lists average points for the 30-year loan. Across the 1,663 weeks since 1991 with a points figure, the average ranged from 0.3 to 2.2 points and averaged 0.92. From November 17, 2022 the file has no points figure, so a rate quoted before that date came with an average cost that later rates do not state.
After every week since 1991: where the 30-year rate stood 3, 6 and 12 months later
The table is the whole record in three rows. Each row takes every starting week and asks how far the 30-year average had moved by a fixed number of releases later, then counts how many weeks fell into each band. The outer columns are cumulative: "fell 0.50 or more" includes the weeks that fell 1.00 or more. The middle column holds the weeks that stayed within 0.49 points either way, so the fell 0.50 column, the middle column and the rose 0.50 column add up to the start weeks.
| Look-ahead | Start weeks | Fell 1.00 or more | Fell 0.50 or more | Moved less than 0.50 either way | Rose 0.50 or more | Rose 1.00 or more |
|---|---|---|---|---|---|---|
| 3 months (13 releases) | 1,857 | 0.5% (9) | 9.7% (181) | 79.5% (1,477) | 10.7% (199) | 2.2% (41) |
| 6 months (26 releases) | 1,844 | 3.3% (60) | 22.5% (415) | 60.4% (1,113) | 17.1% (316) | 5.2% (96) |
| 12 months (52 releases) | 1,818 | 12.5% (227) | 34.0% (619) | 44.1% (802) | 21.8% (397) | 10.3% (188) |
At three months the average mostly stayed put: in 79.5% of weeks (1,477 of 1,857) the change was smaller than half a point in either direction. The tails were close to even and leaned the wrong way for a waiter: 199 weeks (10.7%) ended 0.50 or more higher and 181 (9.7%) ended 0.50 or more lower. A full-point fall by the end of three months happened in 9 weeks, 0.5%, against 41 weeks (2.2%) with a full-point rise.
At six months the tails fatten and tilt toward falls at the half-point mark: 415 weeks (22.5%) fell 0.50 or more and 316 (17.1%) rose that much. At the full-point mark the order flips: 96 weeks (5.2%) rose 1.00 or more and 60 (3.3%) fell that much. A reader hoping for a big drop should note that a full-point rise by the end of six months showed up more often than a full-point fall.
At twelve months falls lead at both marks: 619 weeks (34.0%) fell by 0.50 or more against 397 (21.8%) that rose by it, and 227 (12.5%) fell by 1.00 or more against 188 (10.3%) that rose by it. Even so, in 802 weeks (44.1%) the twelve-month change was under half a point either way, smaller than the 0.75-point gap on the anchor loan, and in 669 weeks (36.8%) the rate was higher a year on than at the start.
Counting only the end of each window matches a decision made on a chosen date. A different count is flattering and unusable: in 1,036 of the 1,818 twelve-month windows (57.0%) at least one of the next 52 releases was 0.50 or more below the starting rate, and in 333 (18.3%) at least one was 1.00 or more below. Those are counts of a low that existed in hindsight. A homeowner who waits does not know which release will be the low, and the rate on the day of closing is the one that sets the payment.
How far the rate moved: medians, the middle 80% and the extremes
Shares of weeks say how often, not how much. The second table gives the size of the typical move and of the extremes for the same start weeks. The median change was -0.07, -0.12 and -0.20 points at three, six and twelve months, a drift downward that the era table below explains. The spread matters more than the middle: the middle 80% of twelve-month changes ran from -1.05 to +1.01, a range of about two points, and at six months from -0.71 to +0.72.
| Look-ahead | Median change | Middle 80% of weeks | Largest fall (start week) | Largest rise (start week) |
|---|---|---|---|---|
| 3 months | -0.07 | -0.49 to +0.51 | -1.50 (October 16, 2008, from 6.46%) | +1.96 (August 4, 2022, from 4.99%) |
| 6 months | -0.12 | -0.71 to +0.72 | -1.78 (January 13, 1995, from 9.19%) | +2.76 (December 23, 2021, from 3.05%) |
| 12 months | -0.20 | -1.05 to +1.01 | -2.20 (January 6, 1995, from 9.22%) | +4.10 (November 10, 2021, from 2.98%) |
The largest rise is bigger than the largest fall at every horizon: +1.96 against -1.50 at three months, +2.76 against -1.78 at six and +4.10 against -2.20 at twelve. The worst rises began from low rates (2.98% in the week of November 10, 2021, 3.05% in the week of December 23, 2021), while the largest falls began from 9.22% in January 1995 and 6.46% in October 2008. A move of that size carries a dollar meaning only when applied to a particular balance, and the next sections do that for one loan; the percentage points in this table are the record, not a scenario for 2027.
Reading the middle 80% against the cost of waiting makes the trade plain without predicting anything. A month of delay on the anchor loan costs $219. A half-point change in the quote moves the payment on that loan by $115 a month down or $117 a month up, and that difference repeats in every one of the remaining months, while the delay cost is collected only while you wait. A move of that size is common by the end of six months, in either direction: 39.6% of six-month windows ended 0.50 or more away from the start, up or down. That is why the sections that follow look at the distribution of moves and not at the cost of waiting alone.
A record that begins in a falling market: the same measures by era
The 30-year average was 9.56% in the first week of 1991 and 2.65% in the week of January 7, 2021, its lowest in the file. A series that fell 6.91 points between January 1991 and January 2021 will show more falls than rises in a pooled count, and the negative medians above are partly that drift. Splitting by the year in which the start week fell shows how much the answer depends on the era.
| Start weeks in | 6 months: weeks | 6 months: fell | 6 months: rose | 6 months: median | 12 months: weeks | 12 months: fell | 12 months: rose | 12 months: median |
|---|---|---|---|---|---|---|---|---|
| 1991 to 1999 | 470 | 28.5% | 21.5% | -0.17 | 470 | 44.7% | 27.9% | -0.34 |
| 2000 to 2009 | 522 | 26.4% | 7.7% | -0.12 | 522 | 37.0% | 9.6% | -0.25 |
| 2010 to 2019 | 521 | 16.5% | 15.0% | -0.16 | 521 | 33.0% | 18.8% | -0.24 |
| 2020 to 2026 | 331 | 17.2% | 29.3% | +0.12 | 305 | 14.4% | 38.7% | +0.21 |
| Starts from November 17, 2022 (survey method change) | 181 | 19.9% | 19.9% | +0.02 | 155 | 22.6% | 14.2% | -0.09 |
For twelve-month windows, the share of start weeks followed by a fall of 0.50 or more shrinks across the eras, from 44.7% in the 1990s start weeks to 37.0% in the 2000s, 33.0% in the 2010s and 14.4% from 2020. The share followed by a rise of that size runs the other way, 27.9%, 9.6%, 18.8% and 38.7%. The median twelve-month change turned from -0.34 in the 1990s start weeks and -0.24 in the 2010s to +0.21 for starts since 2020. The 2020 to 2026 row includes the run from the 2.65% low in January 2021 to 7.08% in November 2022, and the rate of 7.79% on October 26, 2023, the highest weekly average since the week of October 20, 2000 (7.83%).
The last row isolates the weeks that start after the survey changed method. Of its 155 twelve-month windows, 35 (22.6%) fell by 0.50 or more and 22 (14.2%) rose by that much, with a median of -0.09. Of its 181 six-month windows, 36 (19.9%) fell by 0.50 or more and 36 (19.9%) rose by that much. Those start weeks cover about three years and their windows share most of their weeks with each other, so they describe one stretch of the market and not a rate of recurrence.
What the era table does and does not support: it shows that pooled falls reflect a long downtrend and that the same measure points the other way when the window starts after 2020. It does not support any statement about which of those regimes the next twelve months resemble. That is a forecast, and this page makes none.
1,818 windows, roughly 35 separate years
Consecutive start weeks have almost identical windows. The twelve-month window that starts one week later shares 51 of its 52 weeks with the one before, so 1,818 windows is not 1,818 independent observations. The record holds roughly 35 separate years of start weeks, and in practice the rate has moved in multi-year trends that make even those years less than independent. Every share in the earlier tables should be read with that in mind: the precision implied by one decimal place is not real.
The table shows the size of the problem. Taking every 52nd start week gives a sample of twelve-month windows with no overlap, and there are 52 ways to choose the first week. Across those 52 samples of 34 to 35 windows each, the share that fell by 0.50 or more ran from 25.7% to 40.0%, against 34.0% in the full count. The share that rose by 0.50 or more ran from 14.7% to 28.6%, against 21.8%. The share that fell by 1.00 or more ran from 8.6% to 20.0%. A choice with no bearing on the economy, which week of the year to start counting from, moves the answer by 14.3 percentage points.
| Look-ahead | All overlapping windows: fell 0.50+ | Non-overlapping samples: fell 0.50+ | All overlapping windows: rose 0.50+ | Non-overlapping samples: rose 0.50+ | Windows per non-overlapping sample |
|---|---|---|---|---|---|
| 3 months | 9.7% | 7.7% to 11.9% | 10.7% | 7.0% to 13.3% | 142 to 143 |
| 6 months | 22.5% | 18.3% to 28.2% | 17.1% | 14.1% to 19.7% | 70 to 71 |
| 12 months | 34.0% | 25.7% to 40.0% | 21.8% | 14.7% to 28.6% | 34 to 35 |
The same lumpiness shows in the calendar. Sorting twelve-month windows by the year of the start week, 10 of 35 start years had a majority of their weeks followed by a fall of 0.50 or more, 8 had a majority followed by a rise of that size, and 17 had neither. In two start years (2011 and 2019) every start week was followed by such a fall, and in one (2021) every start week was followed by such a rise. The record is a handful of long runs, with wide gaps between them, and an episode counts once however many weeks it covers.
A reasonable way to hold the numbers: the distribution is real as a description of 35 years and fuzzy as an input to any single decision. The tables support the statements that large moves in both directions are ordinary over six and twelve months, that a rise bigger than the rate gap has happened often enough to matter, and that the typical three-month change is small. They do not support a probability for the coming months.
Starting near 7%: what the high-rate weeks do and do not show
The 30-year average stood at 7.24% on October 29, 2026. Since 1991 it was at or above that level in 411 of 1,870 weeks (22.0%), but since January 1, 2000 in only 65 of 1,400. Grouping start weeks by the rate they began at is the closest the record gets to asking "what happened from here", so the table shows it. The third column carries the caveat: how many calendar years the weeks come from.
| 30-year rate in the start week | Start weeks | Start years | Fell 0.50 or more | Rose 0.50 or more | Median change |
|---|---|---|---|---|---|
| Under 4.00% | 363 | 12 | 14.3% | 42.7% | +0.37 |
| 4.00% to 4.99% | 288 | 12 | 45.8% | 3.5% | -0.41 |
| 5.00% to 5.99% | 207 | 10 | 18.4% | 34.3% | +0.02 |
| 6.00% to 6.99% | 437 | 18 | 24.0% | 19.7% | -0.11 |
| 7.00% to 7.99% | 304 | 15 | 33.6% | 24.7% | -0.20 |
| 7.00% to 7.99%, start weeks from 2001 on | 63 | 6 | 52.4% | 1.6% | -0.57 |
| 8.00% and above | 219 | 9 | 86.8% | 0.0% | -1.02 |
Start weeks in the 7.00% to 7.99% band, the band that holds today's rate, were followed twelve months later by a fall of 0.50 or more in 33.6% of 304 cases and a rise of that size in 24.7%. But 11 of the 15 start years in that band lie between 1992 and 2002, and the weeks at 8.00% and above, all from 1991 to 2000, fell by 0.50 or more 86.8% of the time and never rose by it.
The more recent subset is thin. Start weeks in the 7.00% to 7.99% band from 2001 on number 63, from six calendar years: 30 weeks in 2001, 7 in 2002, 2 in 2022, 17 in 2023, 6 in 2024 and 1 in 2025. Of them, 33 (52.4%) were followed by a fall of 0.50 or more and 1 by a rise of that size, with a median of -0.57. Six calendar years, two of them consecutive years at the start of the 2000s, are a few episodes and not a base rate, and the two weeks from 2022 were both followed by rises, the larger +0.71.
Neither subset says that a high rate tends to fall, and the other bands do not line up on one direction either. The weeks under 4.00% were followed by a rise of 0.50 or more in 42.7% of cases, the 5.00% to 5.99% weeks by such a rise in 34.3%, and the 4.00% to 4.99% weeks by a fall of that size in 45.8%. Where a band leans hard one way, it is because one or two episodes supply most of its weeks: the 8.00% and above band is made of start weeks from 1991 to 2000 and nothing else. The starting level describes each episode; it should not be read as a signal.
Five largest twelve-month rises and falls, with dates
The extremes show what the tails look like in rates and dates, which a share cannot. The table lists the five largest twelve-month rises and falls, selected so that no two windows in a list overlap. The windows are in points of the 30-year rate; the figures are not converted to dollars on today's loan, because a rise from a 2.98% start is not a scenario for a 7.24% start.
| Rank | Start week | Rate then | Release 52 weeks later | Change |
|---|---|---|---|---|
| Rise 1 | November 10, 2021 | 2.98% | November 10, 2022: 7.08% | +4.10 |
| Rise 2 | January 14, 1994 | 6.99% | January 13, 1995: 9.19% | +2.20 |
| Rise 3 | February 12, 1999 | 6.77% | February 11, 2000: 8.36% | +1.59 |
| Rise 4 | June 30, 2005 | 5.53% | June 29, 2006: 6.78% | +1.25 |
| Rise 5 | January 3, 2013 | 3.34% | January 2, 2014: 4.53% | +1.19 |
| Fall 1 | January 6, 1995 | 9.22% | January 5, 1996: 7.02% | -2.20 |
| Fall 2 | October 16, 2008 | 6.46% | October 15, 2009: 4.92% | -1.54 |
| Fall 3 | July 12, 1991 | 9.64% | July 10, 1992: 8.13% | -1.51 |
| Fall 4 | May 19, 2000 | 8.64% | May 18, 2001: 7.14% | -1.50 |
| Fall 5 | June 14, 2002 | 6.71% | June 13, 2003: 5.21% | -1.50 |
The largest rise, +4.10 points, ran from 2.98% in the week of November 10, 2021 to 7.08% in the week of November 10, 2022. The next largest, +2.20, began on January 14, 1994 at 6.99% and ended at 9.19%. After those two the rises are +1.59 points and smaller. The falls are more evenly sized: the largest, -2.20, began on January 6, 1995, and the next four are between -1.54 and -1.50.
For six-month windows the largest rises were +2.76 (3.05% to 5.81%, from the week of December 23, 2021), +1.75 (6.74% to 8.49%, from the week of October 22, 1993), +1.38 (6.77% to 8.15%, from the week of February 12, 1999). The largest falls were -1.78 (9.19% to 7.41%, from the week of January 13, 1995), -1.68 (6.46% to 4.78%, from the week of October 30, 2008), -1.41 (9.64% to 8.23%, from the week of July 12, 1991).
A rise equal to the rate gap erases the entire gain from refinancing; on the anchor loan that is 0.75 points. All five twelve-month rises in the table are larger than that, and the record as a whole had a twelve-month rise of 0.75 or more in 277 of 1,818 start weeks (15.2%). A fall of the same size in the other direction would add about $172 a month in saving on top of the $177 the first refinance already delivers.
Turning a rate move into dollars: the anchor loan after a six-month wait
The outcomes of waiting come in two kinds. If the quote falls, you refinance later at the lower rate and keep more of each payment. If it rises by as much as the gap, the refinance stops paying, you stay on the old loan, and the gain from refinancing today, $11,545 on these inputs, is gone. If it rises a little, the gain shrinks and may no longer cover the closing costs inside the stay.
The table prices five endpoints for a six-month wait. "Result if you refinance then" compares the new loan with staying on the old loan to month 84. "Best available" takes the larger of that and zero, because a homeowner would not refinance if it lost money. "Compared with refinancing today" subtracts the $11,545 the refinance earns if done now. The last column puts each row beside the share of the record's six-month windows that moved by at least that much; the unchanged row uses the band under 0.50 either way, because an exactly unchanged rate is rare.
| Quote six months later | Payment on the new loan | Result if you refinance then | Best available (refinance or stay) | Compared with refinancing today | History, six-month windows |
|---|---|---|---|---|---|
| 6.24% (down 1.00) | $2,210 | $32,792 | $32,792 | +$21,247 | 3.3% (60) fell 1.00+ |
| 6.74% (down 0.50) | $2,322 | $21,479 | $21,479 | +$9,934 | 22.5% (415) fell 0.50+ |
| 7.24% (unchanged) | $2,436 | $10,118 | $10,118 | -$1,427 | 60.4% (1,113) moved less than 0.50 |
| 7.74% (up 0.50) | $2,552 | -$1,285 | $0 | -$11,545 | 17.1% (316) rose 0.50+ |
| 8.24% (up 1.00) | $2,671 | -$12,724 | $0 | -$11,545 | 5.2% (96) rose 1.00+ |
Compared with refinancing today, a six-month wait leaves you $1,427 behind if the quote has not moved, $9,934 ahead if it is down 0.50 and $21,247 ahead if it is down 1.00. If it is up 0.50, the monthly saving shrinks to $60 and the $7,000 takes about 118 months to recover, past the 84-month stay, so the refinance is not made and the whole $11,545 is gone; at up 1.00 the new loan would cost more than the old one. A rise as small as 0.44 points is enough to make a refinance at the end of the wait lose money inside the stay. In the record, 22.5% of six-month windows ended 0.50 or more lower and 17.1% ended 0.50 or more higher. The sizes are similar in dollars, $9,934 gained against $11,545 lost, and the two frequencies differ by 5.4 percentage points, so the record does not show a lopsided bet in either direction. What it shows is a bet about 8 times the size of the $1,427 that six months of delay costs when the rate stands still.
A three-month or twelve-month wait changes the dollar columns a little and the logic not at all. Waiting three months and finding the quote down 0.50 leaves $11,117 ahead of refinancing today, and waiting twelve months with the same fall leaves $7,585 ahead; the longer wait gains less because fewer months remain in the seven-year stay to collect the lower payment. A rise of 0.50 or more leaves you $11,545 behind after either wait. Doing nothing about a rise is the "stay" branch, so the loss is the refinance not made, not a larger bill.
Neither the table nor the last column gives an expected value, and the page does not compute one. An expected value would turn the frequencies of 35 years of weeks into a probability for the coming six months, and the previous sections explain why the record cannot carry that weight.
A thin gap and a short stay: $500,000, 0.50 points, six months of waiting
Inputs: $500,000 owed, note rate 7.74%, 336 months left, closing costs $9,000 paid in cash, five-year stay (60 months). The quote is set equal to 7.24%, the Freddie Mac 30-year average for the week of October 29, 2026. The payment falls from $3,645 to $3,477, $168 a month, and the interest gap is $208 a month. A simple break-even is month 54, inside the stay, and counting payments, closing costs and the balance owed at month 60, refinancing today beats staying put by $3,592. Waiting six months with the quote unchanged costs $1,328 of that, 37% of the gain. If the quote is down 0.50 at the end of the wait, the refinance later is worth $13,500, which is $9,908 more than refinancing today. If it is up 0.11 or more, the closing costs no longer pay back inside five years and the gain is gone, $3,592 less than refinancing today; the break-even rise is about 0.10 points. In the record, 35.5% of six-month windows ended with a rise of 0.11 or more (28.0% with a rise of 0.25 or more), and 22.5% ended with a fall of 0.50 or more. The fall that repays the six-month wait is 0.06 points, and 55.3% of six-month windows had a fall that large. A thin gap and a short stay make the wait cheap in dollars and make the gain fragile: a rise of 0.25 points costs the whole $3,592.
How small a fall repays the wait, set beside the rise that erases the gain
Two numbers bracket a decision to wait. The first is the fall that makes up for the waiting months. Solving the anchor loan's arithmetic gives the fall at which a wait of three, six or twelve months ends exactly level with refinancing today: 0.03, 0.06 and 0.14 points. A lower payment is collected for most of the stay, so a small fall repays months of the delay. The second number is the rise that wipes out the gain, which equals the 0.75-point gap.
| Wait | Cost if the rate is unchanged | Fall that repays it | History: start weeks with a fall of at least that | History: start weeks with a rise of 0.75 or more |
|---|---|---|---|---|
| 3 months | $715 | 0.03 points | 53.2% (987) for 0.04+ | 4.4% (81) |
| 6 months | $1,427 | 0.06 points | 54.7% (1,009) for 0.07+ | 9.3% (172) |
| 12 months | $2,840 | 0.14 points | 54.0% (982) for 0.14+ | 15.2% (277) |
The break-even falls are small, and history had falls that large in 53.2% to 54.7% of start weeks across the three horizons. The fall required for twelve months is 0.14 points, and the quote moved down that much in 54.0% of twelve-month windows. These numbers show that the delay itself is cheap against the value of a modest fall. They are not the same as odds that waiting wins, for two reasons. A fall of 0.04 to 0.14 points is hard to tell apart from no change: 29.2% of three-month windows and 14.9% of twelve-month windows ended within 0.14 points of the start, up or down. And the same windows contain 9.3% of six-month starts and 15.2% of twelve-month starts with a rise of 0.75 or more, which ends the gain entirely instead of only delaying it.
The result moves with the gap. At a 0.50-point gap on the same balance, the fall that repays a six-month wait is 0.04 points, and the gain from refinancing today is $5,353. At a 1.50-point gap the fall that repays a twelve-month wait is 0.27 points, which 820 of 1,818 twelve-month windows (45.1%) reached, and refinancing today gains $30,175. The smaller the gap, the less there is to lose to a rise and the less a wait costs; the larger the gap, the more is at stake in each direction. Which side is better for any particular borrower depends on a tolerance for being wrong that no table sets.
Two cautions on how to use the rule. The required fall assumes a seven-year stay; a shorter stay leaves fewer months to collect a lower payment, so the required fall grows. And it assumes you can find the same closing costs and the same quote-to-average difference later, which the history cannot show.
Refinance now and again later: the second closing cost
Some borrowers frame the choice differently: refinance at today's quote and refinance again if rates fall. That route does not need a prediction, but it does need a second set of closing costs, and a second refinance has its own break-even. The table takes the anchor loan, refinanced today at 7.24%, and asks what a second refinance twelve months later would save for each size of fall since today. The balance twelve months in is $345,998, the closing costs are $7,000 again, and the new loan keeps the same maturity.
| Fall since today | Monthly payment saving on the second loan | Break-even month for the second closing | History: twelve-month windows with a fall of at least d |
|---|---|---|---|
| 0.50 points (to 6.74%) | $113 | Month 62 | 34.0% (619) |
| 0.75 points (to 6.49%) | $168 | Month 42 | 22.1% (402) |
| 1.00 points (to 6.24%) | $223 | Month 31 | 12.5% (227) |
| 1.50 points (to 5.74%) | $331 | Month 21 | 1.2% (21) |
A fall of 0.50 points saves $113 a month on the second loan and breaks even in month 62, which is about five years. A fall of 1.00 point breaks even in month 31. A fall of 1.50 points breaks even in month 21. The remaining stay in the example is 72 months at that point, so a 0.50-point fall clears the closing costs by about $1,116 and a 1.00-point fall by about $9,064, before counting the lower balance at exit. The last column shows how often the record had a fall that large within twelve months: 34.0% of start weeks for 0.50, 12.5% for 1.00 and 1.2% for 1.50.
The comparison with waiting is this. Refinancing now takes the first gain on the quote you can see: $11,545 on these inputs. It then keeps the option of a second refinance if the rate falls far enough for the second closing to pay, and costs a second $7,000 if it does. Waiting keeps the cash but takes the delay cost every month and forfeits the gain if the quote rises by the gap. Neither path is free, and the table puts a price on the second one. The guides on refinancing too often, how soon you can refinance again and the no-closing-cost refinance trade-off cover the rules and costs around a second loan, and rate lock fees and extensions show what a locked quote costs if closing slips.
What the survey average cannot tell you about your own quote
Every number above is about the Freddie Mac average, and the average is a description of one kind of loan. Freddie Mac's November 2022 note lists the filters: purchase applications, conventional and conforming, 30-year or 15-year fixed, one-unit owner-occupied homes in the lower 48 states and the District of Columbia, with a loan-to-value ratio from 75 to 80 and a credit score of 740 or above. Your quote can sit above or below that average on any given day, and a refinance is outside the survey's loan purpose by definition. The guide to how the Freddie Mac survey is built walks through the differences line by line.
This page does not measure how closely an individual lender's quote follows the weekly average. If your quote moved by only half of the average's move, or twice it, the dollar tables would change accordingly. Nor does it measure what lenders charged in points, because the file stops listing points after the release of November 10, 2022. The unit of the history is the average rate on applications in a week, which is why the moves are called "the 30-year average" throughout, not "your rate".
The history also leaves out anything about your file. A change in income, credit score or home value between now and the later date can change the rate you are offered whatever the average does. The guides on reading a refinance Loan Estimate and what the lock does deal with the quote in front of you. The weekly average is for context: the Fed decisions and mortgage rates record shows how a policy headline sat beside the same series in the last two years, and the rate tracker keeps the latest dated releases.
Using the record without a forecast: five numbers to work out first
The record cannot say whether to wait. It can organize the comparison, using five numbers from your own loan, and then you decide how much uncertainty to carry. The steps use the same arithmetic as the tables, so any result can be checked against them.
- Work out the interest gap per month: balance times the difference between your note rate and the quote, divided by 12. It is what each month of waiting costs while the quote is unchanged ($219 on the anchor loan).
- Work out the gain from refinancing today over the years you expect to stay: the payment saving over the months of the stay, plus the reduction in the balance owed at exit, minus closing costs. If that figure is not positive on today's quote, the question of waiting is about getting to positive, not about a saving you are giving up.
- Find the rise in the quote that takes the monthly saving to zero. It equals your rate gap. Compare it with the share of start weeks in the record with a rise that large over the wait you are considering (9.3% for 0.75 over six months, 17.1% for 0.50, 5.2% for 1.00).
- Find the fall that repays the wait, from the interest gap and the months left in the stay. On the anchor loan it is 0.06 points for a six-month wait, and a fall that small is hard to tell from no change: 18.1% of six-month windows ended within 0.14 points of the start, up or down. Treat it as a measure of how cheap the delay is, not a target.
- Price the alternative of refinancing now and again later, with a second closing cost and a second break-even month. A fall of 0.50 points paid back in about five years on the anchor loan; a fall of 1.00 point in about two and a half.
The calculator on this site runs the break-even month for a quote you have; the break-even guide explains what goes into it, and how much lower a rate needs to be gives the break-even month across balances and closing costs. The record on this page adds the one thing those tools leave out: how wide the range of rate moves has been over the waiting period.
What to have on the table before choosing to wait
- Your note rate, remaining balance and months left from the latest statement, so the interest gap is balance times rate difference divided by 12.
- A written quote or Loan Estimate with its date, and the closing costs on it, so the gain over your stay can be computed and not guessed.
- The years you expect to keep the home, as a month count, because every wait is paid for out of the months left in the stay.
- The Freddie Mac average for the week of your quote, from the dated release, to compare the quote with the average and not with a headline.
- The gain from refinancing today and the rise in the quote that would erase it, written side by side.
- The cost of a second closing if you refinance now and the rate later falls, with the break-even month for it.
Questions about waiting for lower mortgage rates
Will mortgage rates go down if I wait?
Nobody can say, and this page does not forecast. In the Freddie Mac record since January 1991, the 30-year average was at least 0.50 points lower six months later in 22.5% of weeks and at least 0.50 points higher in 17.1%. Twelve months later the figures were 34.0% lower and 21.8% higher. The record is dominated by a long downtrend and by a few long runs, and the era table shows the pattern reversed for starts after 2020.
How much does it cost to wait a month to refinance?
The monthly cost is the interest gap: balance times the difference between your note rate and the quote, divided by 12. On $350,000 with a 0.75-point gap it is $219 a month, and on $500,000 with a 0.50-point gap $208. The drop in your payment is smaller, $177 and $168 for those two loans, because the lower rate sends more of each payment to principal, but the payment drop is not the money the delay costs.
How often have rates fallen by half a point within a year?
In 619 of 1,818 weekly start weeks since January 4, 1991 (34.0%), the 30-year average was 0.50 or more lower 52 releases later. In 397 (21.8%) it was 0.50 or more higher. These windows overlap, so the real sample is closer to 35 separate years, and sampling every 52nd week gives a share anywhere from 25.7% to 40.0%.
If I refinance now and rates fall later, can I refinance again?
Yes, with a second set of closing costs and a second break-even month. On the anchor loan with $7,000 in closing costs, a fall of 0.50 points a year after the first refinance breaks even in month 62, a fall of 1.00 point in month 31. Your lender's rules and any recent-refinance timing limits are in your documents; the guide on how soon you can refinance again covers the common ones.
Do Federal Reserve rate cuts mean mortgage rates will fall?
Not reliably. A 30-year fixed rate follows long-term market rates and expectations, not the Fed's overnight target. The Fed decisions and mortgage rates record lines up seven target changes from September 2024 to September 2026 with the same Freddie Mac series. After the three 2024 cuts, which lowered the target by 1.00 point in total, the 30-year average was higher in January 2025 (7.04% on January 16) than in September 2024 (6.09% on September 19).
Is the Freddie Mac average the rate I would be offered?
No. It is an average of purchase applications from borrowers with a credit score of 740 or higher, a loan-to-value ratio of 75% to 80%, conforming loan size, one-unit owner-occupied homes. A refinance falls outside that selection, and points are not listed after November 2022. Use the average to date the market and compare quotes with each other. The guide to how the Freddie Mac survey is built explains the gap.
Sources and verification
- Freddie Mac, Primary Mortgage Market Survey weekly history file (PMMS_history.csv): Downloaded November 3, 2026: 2,901 weekly rows of 30-year fixed averages from April 2, 1971 to October 29, 2026, with an average points column through November 10, 2022. Source of every rate and every statistic on this page. Checked November 3, 2026.
- Freddie Mac, Primary Mortgage Market Survey: U.S. weekly averages as of 10/29/2026: 30-year fixed 7.24%, 15-year fixed 6.60%. Survey method: loan applications submitted to Loan Product Advisor since November 17, 2022; releases on Thursdays at 12 p.m. ET, or on the Wednesday when a U.S. holiday falls on a Thursday; fees and points no longer reported. Checked November 3, 2026.
- Freddie Mac, Newly Enhanced Mortgage Rate Survey Explained (research note, November 3, 2022): Selection filters for the average; the 0.066-point average difference between the old and new methods from 2005 through 2021, with a standard deviation of about 0.074 points. Checked November 3, 2026.
The statistics are computed on this site from Freddie Mac's PMMS_history.csv as downloaded on November 3, 2026, whose last row is the release of October 29, 2026. They are a fixed history through that date and not a live number. They describe the weekly average of 30-year fixed-rate purchase applications, not any lender's quote, and they are not a forecast of any future rate or a probability for the coming months. Overlapping windows are not independent: the record holds about 35 separate years. The series changed method on November 17, 2022, and lists no points after November 10, 2022. The dollar figures use illustration inputs stated beside each table: a $350,000 balance at 7.99% and a $500,000 balance at 7.74%, 336 months left, a quote set equal to the 7.24% average, closing costs of $7,000 and $9,000, and stays of 84 and 60 months; they ignore taxes, insurance, escrow and the time value of money, and are not quotes or advice. The page does not cover FHA, VA or USDA streamline rules, adjustable-rate loans, the 15-year series, or the Freddie Mac archive before 1991. Educational only, not a loan offer.