The Review of Financial Studies · Vol. 39, No. 8 · 2026
Mortgage Design, Repayment Schedules, and Household Borrowing
How does the design of debt repayment schedules affect household borrowing? We exploit a Swedish policy reform that eliminated interest-only mortgages for loan-to-value ratios above 50%. We document substantial bunching at the threshold, leading to 5% lower borrowing. Wealthy borrowers drive the results, challenging credit constraints as the primary explanation. A life-cycle model shows that much of the effect comes from households experiencing ongoing flow disutility to amortization payments. Mortgage contracts with low initial payments substantially increase household borrowing and lifetime interest costs.
Setting
The 2016 Swedish amortization requirement
From June 2016, new Swedish mortgages with a loan-to-value ratio above 50% must repay at least 1% of the original loan each year, and those above 70% at least 2%. Below 50%, interest-only remains allowed. Swedish mortgages have linear repayment schedules, so the jump in the monthly payment at the threshold is entirely principal, not interest.
The rule applies to the whole loan, not the part above the threshold. A borrower just above 50% therefore faces a large implicit marginal amortization rate on the last percentage points of LTV. Once the balance falls below the threshold, a phone call to the bank turns amortization off. Seven of the eight large banks charge no fee, there is no new credit check, and no new appraisal.
Data come from Finansinspektionen's Mortgage Survey, 2011–2018, which covers all new mortgages issued by the eight largest banks on sampled days each autumn.
Evidence explorer
Re-estimate the bunching yourself
The counterfactual is the pre-reform LTV distribution, bin by bin. Choose the threshold, the years that form the counterfactual, and the limits of the excluded window. All four statistics update from the binned data.
Why the live numbers differ slightly from the paper
The paper bins the confidential microdata by rounding LTV to the nearest 0.5 (so the bin labelled 50 covers 49.75–50.25). The public file behind this page uses bins [a, a+0.5), which cannot be re-rounded. The bin [50, 50.5) mostly holds loans at exactly 50% (70% of them interest-only in 2016), but it also contains some loans just above the threshold. The live estimates are therefore close to, but not identical to, the published ones, and the live page reports no bootstrap standard errors. The published estimate, with its standard error, is always shown alongside. The page publishes shares only, no counts.
Method: the difference-in-bunching estimator
Let nj be the post-reform percent of loans in bin j and n̂j the percent in the same bin before the reform. With threshold T and window limits L and U:
B̂ = Σj=LT (nj − n̂j) M̂ = Σj>TU (nj − n̂j)
ΔLTV = B̂ / ĝ(T) where ĝ is the average counterfactual density over [L, T]
eα = (ΔLTV / T) / (α* − α0), α* = α0 + Δα + Δα · T / ΔLTV
The identifying assumption is that, absent the reform, the share of loans in each bin would have equalled its pre-reform share. This allows for round-number spikes at 50 that are stable over time, which a smooth polynomial counterfactual would attribute to the policy. M is negative when mass is missing above the threshold. The estimation sample is LTV 20–65 for the 50% threshold and 55–80 for the 70% threshold. At 50, α0 = 0 and Δα = 0.01. At 70, α0 = 0.01 and Δα = 0.01.
Year by year
Each panel shows one survey year. The pile at the threshold appears in 2016 and persists through 2018.
Supply-side checks
Mortgage rates show no discontinuity at the threshold, before or after the reform. The share of interest-only loans above 50% collapses after the reform, which is the policy itself.
Specification explorer
Every published estimate, side by side
Point estimates with 95% confidence intervals from bootstrapped standard errors (500 draws). Choose a statistic and a group, then select a row for details.
Model explorer
What kind of preference produces this pile?
A life-cycle model with housing, liquid savings, long-term mortgages, income risk and credit constraints, in the spirit of Campbell and Cocco (2003) and Cocco (2004). The model is solved in Julia. The charts below show its actual output.
Three versions of the household
Calibration and identification
Move the disutility parameter and read off the excess mass and missing mass the model implies. A flow disutility of Δk = 0.08 reproduces the empirical excess mass (model B = 7.53, data B = 7.47) with almost no hole. A one-off cost generates a large hole relative to the pile, which the data do not show.
The standard model does not bunch under alternative calibrations
Without a psychic cost the value function stays smooth at 50%, whatever the calibration. Each alternative is simulated under interest-only and under the Swedish requirement.
Model parameters
| Parameter | Symbol | Value | Source |
|---|---|---|---|
| Income process | |||
| Income persistence | ρ | 0.97 | Kovacs and Moran (2021) |
| Std. dev. of income shocks | σε | 0.180 | Kovacs and Moran (2021) |
| Std. dev. of initial income | σ0 | 0.410 | Kovacs and Moran (2021) |
| Share with zero initial assets | a0zero | 0.433 | Kovacs and Moran (2021) |
| Preferences | |||
| Time preference | β | 0.96 | Cocco (2004) |
| Risk aversion | γ | 5.0 | Cocco (2004) |
| Housing utility share | θ | 0.1 | Cocco (2004) |
| Disutility of renting | ζ | 0.03 | Leombroni et al. (2020) |
| Flow disutility of amortization | Δk | 0.08 | Calibrated to B = 7.47 |
| One-off disutility of amortization | Δn | 0.35 | Calibrated |
| Assets and mortgages | |||
| Real return on liquid asset | r | 0.018 | Swedish 3-month T-bill |
| Real return on housing | rH | 0.029 | Statistics Sweden |
| Mortgage interest rate | rM | 0.043 | Statistics Sweden |
| Maximum LTV | 1 − ψ | 0.85 | Swedish law |
| Refinancing cost (proportional / fixed) | f2, f3 | 5% / $3,000 | Federal Reserve Board |
| Transaction cost of moving | f1 | 0.05 | OECD (2011) |
| Demographics | |||
| Entry, retirement, death | t = 0, W, T | 22, 65, 120 | Attanasio et al. (2012) |
Implications
Why a kink matters beyond Sweden
Financial innovation and household debt
Start from a model economy with only fully amortizing mortgages and introduce interest-only loans. Mortgage debt rises by 33.4% when households have flow disutility, and by 9.1% without it. A notch would affect only borrowers near a threshold. A kink shifts everyone who amortizes, which is why the amplification is large.
Forced repayment and wealth accumulation
Bernstein and Koudijs (2024) find that each euro of mandatory amortization raises net wealth by almost a euro, even for households with liquid savings. Re-running their instrumental-variables design on simulated data, only the model with flow disutility matches both facts.
Cite
How to cite this paper
@article{backman2026mortgage,
author = {B{\"a}ckman, Claes and Moran, Patrick and van Santen, Peter},
title = {Mortgage Design, Repayment Schedules, and Household Borrowing},
journal = {The Review of Financial Studies},
year = {2026},
volume = {39},
number = {8},
pages = {2549--2594},
doi = {10.1093/rfs/hhaf115}
}
- Published article and Internet Appendix (Oxford University Press)
- Replication package (Harvard Dataverse, doi:10.7910/DVN/DZUGKD)
- Google Scholar