Fixed or Variable
Two Canadian mortgages signed five years and three years ago, replayed against the rate path that actually happened — and what is left to decide on today
Ask a mortgage broker whether to take a fixed or a variable rate and you will get one of two answers, delivered with equal confidence. The first is that variable wins about ninety percent of the time, so take variable. The second is that nobody has a crystal ball, so take what lets you sleep.
Both are evasions. The first is a statistic from a study of the twentieth century. The second is true and useless. What neither does is tell you what the choice actually cost the people who made it, which is the one thing we can know for certain, because they made it and the rates then did what they did.
So this article does that instead. It takes two real signing dates — September 2021 and September 2023 — and replays both contracts against the Bank of Canada’s actual rate path, month by month, to the end of the term. Then it asks what, if anything, that tells you about the decision in front of you today.
Nothing here knows your income, your equity, your job security or your tolerance for a bad month. The calculations below are exact given their inputs, and the inputs are documented, but a mortgage is a decision about your own balance sheet and belongs with a broker or a fee-only planner who can see it. Every rate in this article is dated; rates move daily.
What the two rates actually are
They are not two prices for the same thing. They are two different contracts.
A variable rate is quoted as a discount off your lender’s prime rate — “prime minus 1.15”, say. Prime tracks the Bank of Canada’s overnight target almost mechanically: through the whole period in this article, at every one of the Big Six, prime was the overnight rate plus exactly 2.20, moved the day after each announcement, by the full amount, every time. (TD is the one wrinkle: its mortgage prime has sat 0.15 above regular prime since 2016, and its discounts are quoted off that, so the contract rate comes out comparable.) Your discount is locked for the term; the prime it is subtracted from is not.
A fixed rate is set at signing and priced off the five-year Government of Canada bond yield, not off the overnight rate at all. This matters more than most coverage admits: fixed and variable rates can and do move in opposite directions, because the bond market is pricing what it expects the Bank to do over five years while prime reflects only what the Bank has already done.
And the two carry different exit costs, which is where a surprising amount of the real money lives. Break a variable mortgage and the penalty is three months’ interest. Break a fixed one and it is the greater of three months’ interest and the interest rate differential — a number the big banks compute against their own posted rates, which sit well above the rates they actually lend at. We will come back to what that does to people.
Here is the path everything below is measured against.
The overnight rate went from 0.25% to 5.00% in ten increases over sixteen months, sat at 5.00% for eleven months, then came down in nine cuts over seventeen months to the 2.25% it has held since October 2025. Prime followed it from 2.45% to 7.20% and back to 4.45%. Very few Canadians holding a mortgage today had ever had to price a move of that size, in either direction.
Both look-backs are illustrative simulations, not a record of what any particular borrower was charged. They replay a $500,000 mortgage on a twenty-five-year amortization against the real prime path, and the assumptions behind them are deliberately simple:
- Monthly steps. Each month is charged the rate in force on the first of it. Real contracts adjust on the lender’s own effective date, and many people pay every two weeks rather than monthly.
- Semi-annual compounding throughout. That is the legal default for fixed-rate mortgages in Canada, but it is not universal for variable ones — TD, for instance, compounds its variable mortgages monthly. Applying one convention to every product slightly understates interest on the variable rows.
- Rates are the best broadly available insured rates archived for the signing month. An uninsured borrower paid roughly 25–30 basis points more on both sides, which moves the levels rather than the ranking.
- Nothing else is modelled — no fees, no prepayments, no penalties, no property tax, no insurance, no missed payments.
Everything scales linearly with the principal, so the shape of each result is robust even where the dollar figures are not precise to the dollar. Treat them as well-sourced illustrations of a mechanism, not as quotes.
Look-back one: signed September 2021
September 2021 was near the bottom. The best five-year fixed rate broadly available was 1.69%. The best five-year variable was prime minus 1.47, which against a prime of 2.45% meant 0.98% — seventy-one basis points cheaper on day one, and a smaller payment every month for as long as nothing changed.
Nothing changed for six months. Then everything did.
The arithmetic
| Contract | First payment | Highest payment | Interest charged | Owing at end of term |
|---|---|---|---|---|
| Five-year fixed at 1.69% | $2,043 | $2,043 | $38,677 | $416,084 |
| Variable, payment adjusts (prime − 1.47) | $1,879 | $3,064 | $87,613 | $432,823 |
| Variable, payment fixed at signing | $1,879 | $1,879 | $91,611 | $478,848 |
Fixed won, and it was not close: $48,935 less interest over the term than the adjustable variable, on a $500,000 mortgage. Scale it to your own balance — the gap is very close to ten percent of the principal.
It is worth being precise about why, because “rates went up” is not the whole story. The variable borrower’s rate spent the first six months below 1%, which is the best rate anyone in this article ever paid. They then spent roughly two years above 5%. The average across the term was 3.80% against a fixed 1.69%. The cheap start bought them about $2,500 of early advantage and cost them fifty thousand.
The third row is the one that hurt
Look again at the bottom row. That borrower’s payment never moved — $1,879, every month, for five years — and they ended the term owing $478,848 on a $500,000 mortgage. After five years of payments, they had repaid $21,152 of principal.
That row is the worst-case branch, and the assumption producing it deserves to be named: it lets the unpaid interest accumulate for the full term with the payment never changing and the lender never intervening. Many contracts do not permit that. RBC raises the payment rather than allow negative amortization at all; TD’s own materials describe increasing the payment or making a prepayment once the balance starts growing, and the big banks moved large numbers of clients out of negative amortization from late 2023 onward. So read $478,848 as what the arithmetic does if nothing intervenes — a ceiling on the damage, and the reason lenders intervene — rather than as the typical outcome.
This is the fixed-payment variable, and in 2021 it was the default at BMO, CIBC, RBC, TD, Desjardins and HSBC. About three-quarters of Canadian variable-rate mortgages worked this way. The payment stays flat and its internal split shifts toward interest as prime rises — until the payment no longer covers the interest at all. That threshold is the trigger rate, and past it the unpaid interest is added to the balance. In the simulation above it was crossed in January 2023.
This was not a fringe outcome. By October 2022, with variable rates around 5.1%, the Bank of Canada estimated that about half of all fixed-payment variable mortgages had reached their trigger rate — roughly 13% of every mortgage in the country. By November 2023 the Bank put it at up to 80%. In the third quarter of 2023, BMO, TD and CIBC reported that between 18% and 22% of their Canadian residential mortgage books — nearly $130 billion — were negatively amortizing.
So the honest summary of 2021 is worse than “variable lost”. Most variable borrowers did not experience a rising payment they could budget around. They experienced a payment that looked reassuringly stable while the debt behind it grew, and then a renewal.
What it looked like from inside
The regret is well documented, because people wrote it down as it happened. The highest-voted thread of the cycle, from October 2022, is titled “No good options left”; its author recounts being advised that rates could not possibly exceed 5%. The long tail is still arriving: a thread from April 2026 describes parents who cannot clear the deferred interest on a fixed-payment variable before their renewal.
But the most useful post is a quieter one, from a borrower who took 1.18% variable over 2.2% fixed in 2021 and wrote this September:
The maths was right, yours truly was wrong.
They had done the calculation correctly. Four quarter-point increases was the break-even; four seemed implausible; there were ten. And when they tried to use the escape hatch their lender had offered — the option to convert to a fixed rate mid-term — they found it was priced off posted rates, not the rates a new customer would have been quoted. That option appears in the sales conversation far more often than it appears usefully in a bad year.
There is a companion post worth reading beside it, from August 2021, in which someone observes that they have never met anyone unhappy with a variable rate in fifteen years. The top reply, written months before the first hike:
This is a sampling problem.
What locking in actually cost
The obvious response to all of this is that a 2021 variable borrower should have seen it coming and locked into a fixed rate partway through. Many tried. By September 2022, with prime at 5.45% and rising, brokers were reporting clients converting at around 4.7% — some as high as 5.39%, against a posted rate of 6.14% and a market-best five-year fixed of about 4.4%. Conversion is priced off the lender’s posted rate, so you rarely get the rate a new customer would be quoted.
Put that back through the simulation — variable from September 2021, converted to 4.70% fixed in September 2022, held to the end of the term:
| Contract | Interest charged | Owing at end of term |
|---|---|---|
| Five-year fixed at 1.69%, never touched | $38,677 | $416,084 |
| Variable, held all five years | $87,613 | $432,823 |
| Variable, locked into 4.70% in September 2022 | $92,629 | $435,018 |
Locking in was the worst of the three. It banked the damage already done, at a rate above the market, just before prime peaked and began an eighteen-month descent. This is the trap in the advice to “lock in when rates start rising”: by the time the increase is obvious enough to act on, it is largely priced into the fixed rate you are converting to. The escape hatch is most expensive at exactly the moment it feels most necessary.
Look-back two: signed September 2023
Now run it the other way. September 2023 was near the top: prime at 7.20%, the five-year Government of Canada bond at a sixteen-year high, and every headline insisting rates would stay higher for longer.
The best five-year fixed was 5.24%. The best three-year fixed was 5.94% — you paid a premium for the shorter commitment. The best variable was prime minus 1.25, or 5.95%, the most expensive of the three on day one.
The arithmetic
| Contract | First payment | Highest payment | Interest charged | Owing at end of term |
|---|---|---|---|---|
| Five-year fixed at 5.24% | $2,977 | $2,977 | $75,420 | $468,258 |
| Three-year fixed at 5.94% | $3,181 | $3,181 | $85,618 | $471,094 |
| Variable, payment adjusts (prime − 1.25) | $3,184 | $3,184 | $64,484 | $464,427 |
| Variable, payment fixed at signing | $3,184 | $3,184 | $64,072 | $449,442 |
Variable won, by $10,935 over three years. It started as the worst rate on the table and finished as much the best, because prime fell 275 basis points while it ran.
The bottom row is the trigger-rate story running in reverse, and it is the quiet winner of the whole table. A fixed payment against a falling prime does the opposite of what it did in 2021: the payment stays at $3,184 while the interest inside it shrinks, so more of each one goes to principal. That borrower pays slightly less interest than the adjustable-payment version and ends the term owing $449,442 — about $15,000 less than anyone else on the table. The contract feature that punished the 2021 cohort quietly rewarded this one. It is not a smarter product; it is the same product meeting a different rate path.
The row that is not finished
The five-year fixed line above stops at thirty-six payments because that is where the data stops, and that flatters it. That borrower is not free. They have two more years at 5.24% while today’s five-year fixed is 4.09% and today’s variable is 3.30%. They are paying roughly a point and a half over market on a shrinking balance, and their exit is governed by a penalty we will get to.
The three-year fixed holder, who looks worst in the table, renews this month. Whether that leaves them ahead over the full five years is a narrower question than it sounds, and worth doing rather than asserting.
Run both to September 2028 — the three-year holder renewing now at some rate, the five-year holder serving out their 5.24% — and the margin is this:
| They renew at | Total five-year interest, three-year path | vs. the $122,792 on the 5.24% |
|---|---|---|
| 3.30% (today’s variable) | $115,560 | $7,232 better |
| 3.94% (today’s three-year fixed) | $121,400 | $1,392 better |
| 4.00% | $121,948 | $844 better |
| 4.09% (today’s five-year fixed) | $122,769 | $22 better |
| 4.50% | $126,514 | $3,723 worse |
So the honest version is not “they finish ahead”. It is that the break-even renewal rate is about 4.09% — almost exactly today’s best five-year fixed — and the advantage at any plausible renewal runs from a few hundred to a few thousand dollars, small enough that a lender fee or a slightly worse quote erases it. The shorter term was not clearly the better bet. It was the bet that kept its options, and keeping those options turned out to be roughly free rather than profitable.
Why there is no 2023 regret genre
Search for people who took a fixed rate in 2023 and regret it, and you find almost nothing — which is odd, given how many of them signed near what turned out to be the peak.
Part of the explanation is visible in what those borrowers were discussing at the time. The threads from 2023 are full of people weighing two- and three-year terms against five, and being talked out of long commitments at high rates: one from September 2023 talks a borrower out of a 6.6% three-year using the forward curve; a widely-read one from April 2023 posts a bank’s renewal menu — variable 6.2%, two-year 5.55%, three-year 5.35%, five-year 4.19% — and the comments argue about the term, not the type. A cohort weighted toward shorter terms is one that has largely already renewed into something lower, which is what their posts now read like: relief rather than regret.
That is an observation about the discussion, though, not a measurement of the market — I could not pin a contemporaneous term-share statistic for 2023 originations. What regret does exist takes the form of “should I break my 5.09%?”, and the answer is usually that the penalty exceeds the saving — which brings us to the part of the decision that gets the least attention.
The penalty asymmetry
The two contracts have different exits, and the difference is not symmetric.
A variable mortgage costs three months’ interest to break. Call it $4,000 on a mid-sized mortgage, at almost any lender, in almost any rate environment.
A fixed mortgage costs the greater of three months’ interest and the interest rate differential. At a monoline lender, the IRD is computed against the rate they would offer a new borrower for your remaining term — a small number. At a big bank, it is computed against posted rates, which are advertised fictions maintained well above what the bank actually lends at. One documented case: a $300,000 five-year fixed at 3.19%, broken after a year, drew a major-bank penalty of about $16,800 against roughly $2,400 at a fair-penalty lender.
And the penalty grows as rates fall, which is the trap for the 2023 cohort. A borrower who took a three-year fixed at 4.30% in August 2023 on $500,000 would have faced about $5,400 on the three-month-interest basis. After TD cut its two-year posted rate by 195 basis points, the posted-rate IRD on that same mortgage came to more than $22,000.
So the real comparison is not two rates. It is two rates plus two option values. Fixed buys certainty of payment and sells you an expensive exit; variable sells you an uncertain payment and buys a cheap one. If there is any material chance you move, separate, refinance or sell mid-term, that cheap exit is worth real money, and it almost never appears in the rate comparison that sold you the mortgage.
The ninety-percent statistic
Now the claim that opens every fixed-versus-variable conversation.
It is real, it has an author, and it is older than most of the people quoting it. Moshe Milevsky, a finance professor at York University, published it in 2001: over 1950–2000, a Canadian borrower taking variable would have come out ahead 88.1% of the time. A 2008 update put it at 90.1% for 1950–2007.
Three things are almost always left out.
It is a study of a falling-rate half-century. The period runs from the post-war era through the inflation spike of the early 1980s — when the five-year fixed touched 21.75% — and down the long decline that followed. Variable won because rates spent most of the sample falling. That is a fact about history, not a property of variable-rate mortgages.
Milevsky’s own update qualified it. When he re-ran it against the deeply discounted fixed rates that had become normal, the figure dropped from 90.1% to 77.1%. The headline number assumes you are paying something close to the posted fixed rate, which a discounted borrower is not.
It is a statement about averages, not about you. You do not sign ninety mortgages. You sign one, for a defined term, starting on a specific Tuesday. Our two look-backs are a fair illustration: same country, same product, signing dates twenty-four months apart, opposite verdicts, and a five-figure gap either way.
Try it yourself
Everything above comes out of one small simulator. Here it is, with the inputs exposed — pick a signing date and a term and it replays that contract against the real rate path.
The calculator needs JavaScript. The two worked examples above are the same arithmetic with the inputs fixed.
Twelve signing dates are offered — the ones for which archived rate tables pin what was actually on the market that month. The calculator will not interpolate a contract nobody was offered.
Run all twelve and a pattern falls out that neither look-back alone shows. Fixed wins every signing date from June 2021 through June 2023, by margins that shrink steadily from $48,935 to $730. Variable wins every date from September 2023 onward. The crossover is not a rate level or a spread threshold — it is the month the Bank of Canada stopped raising. Which is another way of saying the choice was never really between two products. It was a bet on where you were in the cycle, and the people who got it right in either direction mostly did not know that was the bet they were placing.
Everything the calculator computes runs in your browser. The code and the rate data are in the site’s repository, and the same module generates the tables printed above, so the two cannot disagree.
So what about today?
Here is where things stand on 10 September 2026.
The overnight rate is 2.25%, held for the seventh consecutive meeting on September 2, unchanged since October 2025. Prime is 4.45%. The best five-year fixed is about 4.09%; the best five-year variable is about 3.30%, or prime minus 1.15. The three-year fixed sits at about 3.94%, slightly below the five-year — the market is charging you nothing for a shorter commitment, which is itself a signal.
That is a spread of roughly 0.79 points in variable’s favour, and it has been widening — the discount off prime deepened from about prime minus 0.66 in April to prime minus 1.15 in September, with prime itself standing still. Lenders competing on variable is not the same as rates falling.
What the forecasters say. The modal call for the end of 2026 is that the Bank does nothing at all: RBC, TD, BMO, CIBC and National Bank all have the overnight rate at 2.25% in December. Scotiabank is the hawkish outlier at 3.00%. Not one of the forecasts surveyed calls for a cut. For 2027 the dispersion opens up, between holding at 2.25% and rising to 3.25%.
The bond market is more hawkish still. As of early September, pricing implied a meaningful chance of a hike at the October meeting and a near-certainty of one by December. The Bank’s own September statement is two-sided: inflation around 3% on gasoline, core at 2.2% excluding it, new tariffs and counter-measures after trade talks broke down, and second-quarter GDP up 3.3% annualized. It gave no easing bias.
That the possibility of an increase is now being argued at all is the change worth noticing. On the day of the September hold, the mortgage broker Alex McFadyen published a video asking what usually precedes a Bank of Canada rate hike — a question that would have been eccentric a year ago, when every forecast pointed one way. Read it, and the bond pricing beside it, as evidence about the distribution rather than the direction: the case for variable no longer rests on cuts arriving, and the case against it no longer rests on them merely stalling.
So the variable borrower today is starting 0.79 points ahead, in an environment where the consensus is no cuts and the risk is priced toward increases. That is close to the opposite of September 2021, when the variable borrower started 0.71 points ahead with the overnight rate at its floor and only one direction available.
What people are actually doing. They are hedging. Among uninsured originations in the first quarter of 2026, half were fixed terms of under five years. The five-year fixed — the default for a generation — is down to about 15%. One lender found that 71% of applicants said they wanted a variable rate and 34% signed one; the gap between those two numbers is the sound of people remembering 2022.
The brokers are split, and honestly so. The rate-comparison side argues the spread makes variable the value play if you pay as though you were fixed. The brokers quoted in the press report that clients are choosing certainty anyway. Both camps have converged on the same compromise, which is the three-year fixed. One prominent broker’s own published arc captures the whole problem: in March, advising that a fixed rate below 4% was worth locking; by April, once fixed rates had risen into the 4s, advising people to wait and watch instead.
How to actually decide
The look-backs do not produce a winner. They produce a better set of questions than “which one wins”.
Start with the exit, not the rate. How likely is it that this mortgage ends early — a move, a sale, a separation, a refinance, a lump sum? If that chance is meaningful, the penalty difference is likely to dwarf the rate difference, and the answer leans variable regardless of the forecast. If the answer is genuinely “we are not moving”, the option has little value and you can decide on rate alone.
Then price the bad case, not the expected case. Take today’s variable at 3.30% and ask what your payment becomes if prime rises two points. If that number is uncomfortable, you have learned something a forecast cannot tell you. The 2021 cohort was not destroyed by a bad forecast; they were destroyed by having no margin when the forecast was wrong.
If you take variable, ask which kind. An adjustable-payment variable hurts immediately and visibly. A fixed-payment variable hides the same damage in the balance and hands it to you at renewal. The second is more comfortable and more dangerous when rates rise and — as the 2023 table shows — quietly better when they fall. Either way the difference is a line in the commitment letter, not a headline on the rate sheet. Ask which one you are being sold. Then, if your lender’s prepayment terms allow it, consider setting your payment as though you had taken the fixed rate: the extra goes to principal, it builds the buffer, and it converts the whole question into a smaller one.
Treat the term as a real variable. The entire market has moved to shorter fixed terms for a reason: a three-year term is a bet you only have to be roughly right about for three years, and it is currently cheaper than the five. Fixed-versus-variable is one axis. Term length is the other, and it is the one most conversations skip.
Do not let anyone tell you the statistic settles it. Variable won 88% of the time in a study of 1950–2000. It won in our 2023 look-back and lost badly in our 2021 one. What it does reliably is transfer interest-rate risk from the lender to you, in exchange for a discount. Whether that trade is good depends on the size of the discount — today, about eight-tenths of a point — and on whether you can survive being wrong.
That last part is not a financial question, and no calculator on this page can answer it.
Sources
Rates, forecasts and history were gathered on 10 September 2026 and are documented, source by source, in the research notes that accompany this article, including which figures come from archived snapshots and which are flagged as inferred.
Policy rates and prime. Bank of Canada, key interest rate and the 2 September 2026 decision; Valet series V39079, V80691311 and V80691335.
Rates on offer, then and now. Archived Ratehub and WOWA rate tables for September 2021 and September 2023; Canadian Mortgage Trends for contemporaneous lender pricing.
Trigger rates and negative amortization. Bank of Canada Staff Analytical Notes 2022-19 and 2023-19; bank Q3 2023 filings as reported by the Globe and Mail.
Renewals. Bank of Canada Financial Stability Report 2026.
Term and product shares. CMHC Residential Mortgage Industry Report, Q1 2026.
The 88% figure. Moshe Milevsky, Mortgage Financing: Floating Your Way to Prosperity (York University, 2001), and his 2008 update, as reported by the Globe and Mail.
Penalties. Robert McLister in the Globe and Mail on fair-penalty lenders; nesto, on posted-rate penalties after the cuts, January 2026.
Broker commentary. Alex McFadyen (Flow Mortgage Co), This Is What “ALWAYS” Happens BEFORE the Bank of Canada Raises Rates, 2 September 2026 — published the day of the Bank’s seventh consecutive hold, and a useful statement of the hike-risk case from the broker side. Worth watching with the usual caveat that the channel belongs to a mortgage brokerage.
Borrower accounts are quoted from public discussion threads without attributing them to named individuals.