What percentage of foreclosures are actually redeemed?

Almost every article about foreclosure describes the process. Very few say what happens at the end of it, and the reason is that the answer is hard to measure.

Redemption leaves no positive record. There is no document filed that says "the owner redeemed". There is only the continued absence of a transfer. To know how often it happens you have to follow each window past its expiry and re-check the record — which is sustained work rather than a one-time extract, and it is why the question usually goes unanswered.

This page publishes what we have measured. Every figure carries its sample size, and the method is set out at the bottom so it can be checked.

The headline figure#

Across 326 resolved Minnesota redemption windows tracked from the sheriff's sale to resolution:

Restricting to outcomes confirmed outright rather than inferred — see the methodology note below — the rate is 37.5% of 285 windows.

Both are published because the difference is material and because publishing only the flattering one would be the kind of thing this page exists to avoid.

These are Minnesota figures. They should not be read as a national rate. About half of US states grant no post-sale redemption at all, and where it exists the period ranges from sixty days to two years. A rate measured under Minnesota's six-month window does not transfer.

Redemption by equity position#

This is the strongest signal in the data, and it inverts what most buyers assume.

The gradient is monotonic across all three bands.

The mechanism is straightforward. A lender bidding under half the assessed value leaves a large equity cushion. The owner has something worth saving and something to borrow against — a refinance, a family loan, a sale of the property itself. An owner whose debt approaches or exceeds the property's value has nothing to redeem for.

The practical inversion: the properties that look like the best deals at auction — a low bid against a high assessed value — are the ones most likely to be redeemed out from under the buyer. More than half of them were.

This cut is only available for Hennepin County, because the winning bid amount is published in the Hennepin sheriff's payload and not in the equivalent feeds from Dakota or Washington. Properties with no published bid are excluded from the band rather than pooled into a fourth category, because "no bid data" would appear on a page as though missing information predicted an outcome.

Redemption by homestead status#

A fourteen-point difference, and homestead status is published on the county assessor record — so it is knowable before a sale, at no cost.

Four county vocabularies had to be normalised to produce this. Dakota records FULL HOMESTEAD / NON HOMESTEAD / DISABLED VET HOMESTEAD / FRACTIONAL, Ramsey uses Y / N / P, Hennepin and Washington use Yes / No. Fractional and partial homesteads are held as a third category rather than folded into either side, because a fractional homestead is a genuinely different situation.

One coverage caveat. Anoka records a null rather than N for not-homesteaded — enumerated across all 140,279 Anoka parcels: Y on 105,887, null on 34,392, and no third value. A null elsewhere may mean unknown, so Anoka contributes fewer homestead rows than it has.

Redemption by county#

County differences here should be read cautiously. They partly reflect genuine differences in local conditions and partly reflect how completely each county's windows have been checked. A county checked less thoroughly looks like a county where less happens. Counties with fewer than twenty confirmed windows are not published for that reason.

How long resolution takes#

The statutory redemption period is six months in the ordinary Minnesota case. Actual resolution takes considerably longer.

Measured across 1,336 tracked Minnesota mortgage foreclosure windows:

Two observations.

Half of all windows were still unresolved at eighteen months, against a six-month statutory period. Postponements, bankruptcy filings, loss mitigation negotiations and litigation all extend the clock. Capital committed at a sheriff's sale sits far longer than the statute implies.

Owner sales cluster early and then stop. 15.0% of windows saw the owner sell during the redemption period by one year — and it is still 15.0% at eighteen months. Owners who sell during the window do it early or not at all.

Adding that 15.0% to the 33.4% redemption rate means nearly half of these situations do not end with the owner losing everything, which is not the impression most foreclosure content gives.

The two statutory tracks, and why they must be separated#

This section is a correction to our own earlier figures and is included because the error is instructive.

The timing table above was previously computed over 1,671 windows rather than 1,336. The extra 335 were tax forfeiture windows under Minn. Stat. ch. 281 — a three-year clock from judgment — pooled with ch. 580/582 mortgage redemption windows on a six-month clock.

Every one of those forfeiture windows was censored and always would be, because the outcome checker detects mortgage foreclosure outcomes and a forfeiture window cannot reach a foreclosure sale at all. They entered the risk set, sat in it for an average of 921 days, never resolved, and depressed the hazard at every horizon.

Removing them moved "reached a foreclosure sale within one year" from 18.8% to 33.0%.

Two different statutes, two different clocks, one estimate. The corrected figure is the one published above.

What we expected and got wrong#

Publishing the misses alongside the findings is the point of a research page. Three predictions failed against the data, and each failure was informative.

We expected the redemption rate to be lower. The working assumption before measuring was that redemption was rare — a right that exists on paper and is seldom exercised. A third is not rare. It means any analysis treating a sheriff's sale as an acquisition is wrong a third of the time, and that is a material error rather than a rounding one.

We expected homestead status to be the dominant predictor. It is a real effect — 38.6% against 24.6% — but a survival model fitted on this data found the homestead term far weaker than the raw rates suggested once county was controlled for, and its apparent strength was partly an artefact of the tax-forfeiture rows described above. Once those were removed, homestead was no longer statistically significant at conventional thresholds.

We expected bid-to-value to be weak. It is the strongest cut in the rate table, moving redemption from 20.0% to 58.1% across three bands. But the same survival model found it contributed essentially nothing to ranking individual windows once county was accounted for.

That last disagreement is unresolved and worth stating plainly. A variable that separates strongly as a marginal rate and does nothing in a hazard model is telling you something, and we do not yet know what. Either it is collinear with something already in the model, or the gradient is a subset effect that disappears within counties. We publish the rate table because the rate is observed and the model is a fit — but a reader should know the two methods disagree.

Why a rate is published rather than a model#

A per-property prediction would be more marketable than a rate table. It is not published, and the reason is a measurement.

A survival model built on this data reached a concordance index of about 0.80 before the statutory tracks were separated, and about 0.66 after. Roughly two-thirds of the pre-correction score came from a single binary covariate. Against that, an observed rate with a stated sample size answers the same question and survives being questioned.

The same thing happened with valuation: an automated model trained three times on growing data lost to a one-line county-year median every time, and the gap widened as data grew.

A calibrated rate is a prediction. "58.1% of 31 comparable windows were redeemed" tells an investor what they need and shows its own working. An unexplained score of 0.73 does not, and cannot be checked by the person relying on it.

Models will be published here when one demonstrably beats the rate it would replace, and the comparison will be shown.

Methodology#

Source. Sheriff's sale notices from county sheriff feeds and published legal notices, joined to a parcel spine built from county assessor and GIS records across 59 Minnesota counties.

Window construction. A redemption window opens at the sheriff's sale date and expires at the statutory period for that property — six months in the ordinary case, twelve for certain agricultural property and older mortgages, five weeks where formally declared abandoned.

Outcome detection. Two independent methods.

Ownership re-check. After expiry, the current owner of record is read from the county's live parcel service. An owner matching a lender, servicer or government pattern indicates the property was lost. The pattern list is maintained against observed owner strings rather than assumed — it initially missed every abbreviated form of Fannie Mae and every spelling of the Secretary of Veterans Affairs, which cost 56 outcomes until the recorded owner strings were read back and the list retuned.

Recorded sale matching. Sales from the Minnesota electronic Certificate of Real Estate Value are matched to windows by parcel. An arms-length sale inside the window by the pre-foreclosure owner indicates the owner sold; a sale by a lender indicates the property was lost.

Classification. An outcome is confirmed when a recorded document establishes it directly. It is inferred when the classification rests on a judgement — a repeat corporate seller treated as a certificate holder, a tax-forfeiture owner name, a sale well below assessed value. Of 326 resolved windows, 41 are inferred. Both figures are published.

Deduplication. Superseded and duplicate window records are excluded. This matters more than it sounds: 762 duplicate tracker rows were retired in a single cleanup pass, and a published rate computed without that filter was counting retired duplicates. Washington County read 93 resolved windows against a true 66 until the filter was added.

Sample floors. No cut is published below twenty confirmed windows, and no buyer-type or bid-band cut below fifteen.

What this cannot answer. When a redemption happens. Redemption is inferred from the absence of a post-expiry deed, and an absence has no date. Time to foreclosure sale is recoverable and is what the timing table measures; time to redemption is not.

Known limitations#

Minnesota only. The method transfers; the numbers do not.

Coverage varies by county. Some counties are checked more completely than others, which affects county-level comparisons more than statewide ones.

The bid-to-value cut is Hennepin only, because only Hennepin publishes the winning bid amount.

Inferred outcomes are a judgement. 41 of 326. Both the inclusive and confirmed-only figures are published so the effect is visible.

Sample sizes are modest in several cuts. A 31-window band is a real finding and not a precise one.

Citation#

These figures may be cited freely. Please include the sample size — a rate without its count is not a finding, and it is the discipline this whole dataset is built on.

Suggested form: Govire, observed Minnesota redemption outcomes, n=326 resolved windows, August 2026.

If you are a journalist, researcher or agency working on housing distress and need a cut that is not published here, the underlying windows exist and the question is usually answerable. Access for research and reporting use is free.