A property tax, put to the test

The Ministry of Finance published its draft on 23 December 2024: 1.45% of the generalised value a year on every dwelling its owner does not live in. Exempt would be owners who live there (permanent residence and actual use, which FURS could check by electricity use) and public rental housing. A let dwelling's tax would fall by 25% of the rent declared for income tax. A relative living there rent-free would not be exempt. The charge for the use of building land (NUSZ) would stay.

The consultation ran to 25 January 2025; the draft was withdrawn in July 2025, and on 1 June 2026 the new finance minister ruled a property tax out. The previous tax was annulled by the Constitutional Court in March 2014 (U-I-313/13), because the valuation models were not set by statute and an owner could not effectively challenge the value.

The simulation follows the draft with one simplification: 1.4% instead of 1.45%, and let dwellings in three variants. The draft's own revenue arithmetic reproduces: the ministry estimated about €600m (gross €2,340m, less 1,415 for homes their owners live in, 235 for public dwellings and 91 for the rental credit). The same rules on today's base give €735m; scaled to the base the ministry used, about €585m.

330,358 flats and houses, 36% of all, would pay, on average €2,383 a year. These are not only empty flats. There are 76,299 registered lets, 88,037 dwellings where a relative lives rent-free, 47,684 where somebody lives but nobody is registered, 19,417 holiday homes and 98,921 truly empty.

The largest single block is empty houses in the countryside: 67,849 houses at a median €1,518 a year, €124m in all. Many are derelict or in disputed estates and cannot respond to a tax.

How big the tax is, the same dwelling's rent says. A rent filed in the register is about 2.5% of the value a year in Slovenia, so 1.4% of the value means 53% of the rent in the median municipality (121 municipalities with at least 25 lettings). In Ljubljana 56%, in Maribor 46%, on the coast 69% to 85%, in Velenje 93%.

Filed rents run below asking rents; in Ljubljana asking rents are about 1.5 times as high, and against them the tax would take about 37%. With the draft's credit the burden is smaller, but it stays large.

What happens when owners respond? An empty dwelling's owner can let it, sell it, or register a family member there. The owner of a dwelling a relative lives in can give it to the relative. A landlord can sell. How likely each is depends on how much tax the dwelling pays against its rent.

Which way rents move is decided by one exemption: whether let dwellings pay. If every let pays, landlords sell to people who move in, the rental market shrinks, and Ljubljana's rents rise by +19% in five years (+10% to +29%). With the draft's credit the outcome is uncertain: −3% (−8% to +1%). If lets are exempt, empty dwellings are let and hidden lets registered: rents fall by −14% (−18% to −10%).

Prices fall a little, everywhere: in Ljubljana from −1.3% to −5.2%. Revenue shrinks as owners adapt: from €612m in the first year to €487m in the fifth if every let pays, and to €269m if lets are exempt, which is less than today's NUSZ (€290m). Pick the outcome and the region.

A sale is not a new home. When the tax makes a landlord sell, an owner only replaces the tenant. A new home is narrower: a dwelling that was empty or a holiday home and is then lived in.

If every let pays, 37,961 dwellings go up for sale in five years, about 55% more sales a year than today (13,890), but many of them were already lived in. New homes number 17,360. If lets are exempt there are fewer sales and more new homes: 26,271 (21,591 to 30,611), about 5,254 a year, as many as Slovenia builds in a year (5,126). The flow is largest at the start: 6,560 in year one, 3,729 in year five, as the responsive part of the empty stock runs out.

The largest single response is neither a sale nor a let but a deed of gift: about 40,828 dwellings pass in five years to the relative who lives in them, the same under all three designs. A gift to a child or parent is untaxed, and the dwelling becomes the recipient's home and is exempt. About 14,047 more avoid the tax by registering a residence.

If relatives living rent-free were exempt, the tax at its start would be about €217m smaller. Rents would barely change (+20%, −2%, −12%), because the gifts the exemption replaces never reached the market.

How large is this tax against what other countries do? Each foreign tax was applied to the same Slovenian dwellings and their generalised values. The range runs from €39m (Croatia) to €1,214m (the Netherlands, a tax on a deemed return). The 2024 draft is among the highest; taxes on empty homes alone raise a tenth as much.

A permanent 1.4% a year is worth 35% of a home's value at a 4% discount rate, and a buyer takes it off the price. The homes that would pay are worth €56bn of €160bn in all. If half to all of the tax is capitalised into prices, their value falls by €10–20bn at the announcement. An owner who avoids the tax loses nothing.

A 1.4% tax is as fair as its base is accurate. Take two flats that sold for €250,000. The first has a generalised value of €179,181, the second €370,683: the 10th and 90th percentiles of the ratio of value to price. The tax would be €2,509 and €5,190 a year. Only 65% of 2025's sales were within a quarter of their value.

A tax base's accuracy is measured against the international IAAO standard. The main measure is the COD, the mean deviation of the value-to-price ratio from its median: for homes it should be between 5 and 15, up to 20 for old or rural stock. On 14,753 sales from July 2024 to June 2026 the COD is 24.5, and 41% of homes are valued more than 20% from the median. Houses have 31.1, flats in blocks 20.9, homes built before 1945 30.7.

None of the 65 municipalities with at least 50 sales meets the standard: their COD runs from 18.3 to 39.1. Between municipalities the base is off by less: their median ratios run from 0.93 to 1.13. Even a perfect base would show a COD of about 9, because prices are themselves a little random.

The standard tests say the base is regressive: cheap homes valued too high. Ranked by price, the cheapest tenth has a ratio 1.43 times the median. That is mostly an artefact: a flat that happened to sell cheap shows a high ratio by construction. Ranked instead by an independent value (a model that sees neither the generalised value nor the dwelling's own price), the ratio runs from 0.96 to 1.06 across all ten deciles. The base is not unfair to the poorer; it is imprecise for everyone.

In buildings with at least two sales (2,325 buildings, 6,707 sales) we split the deviation three ways. 37% is the whole building mis-valued: the building is valued too high or too low. That can be corrected where there are sales. 44% is the difference between flats in the same building, which the register does not describe: the interior, a renovation never filed, the view, noise. The remaining 19% is price noise.

We tested corrections a statistician can build from data the state already holds, used the way an assessor would: calibrated on sales to June 2025 and tested on the 7,671 after. The best (a model on building data plus neighbouring sales) takes the COD from 24.5 to 22.1 and the share mis-taxed from 40% to 37%. A price per m² by zone is far worse: COD 38.0.

The best base these data allow removes about 2.4 of the 15.5 COD points between today's base and a perfect one. The correction removes the differences between municipalities and is worth making, but it does not reach the standard for houses or old stock.

At 2025 prices 43% of sold homes have a generalised value above their price: houses 51%, flats 39%. At the prices of the first half of 2026 (+9%) it is 30%. If the base were a fraction of the value there would be fewer: at 80% of the value 18%, at 70% 11%. A fraction does not repair differences between neighbours, but it cuts the number paying tax on more than the home is worth.

Since the base cannot be made accurate enough from data, fairness has to come from the design. All three proposals therefore share three rules.

The repaired base. Part two's correction on all 908,509 valued dwellings: €155bn in all instead of €160bn. Flats rise (median from €158,100 to €161,644), houses fall (from €147,000 to €136,672). The correction is a rule written into statute, which answers the Constitutional Court's objection.

A tax on 80% of the value, with a correspondingly higher rate so revenue stays the same. The share of payers taxed on more than the home is worth falls from a third to a sixth.

Appeal by evidence. The dwelling's own recent sale, or an appraisal, replaces the base. It is the only remedy for the single home's error. 141,693 dwellings (16%) have a sale of their own on record since 2016; the rest would need an appraisal.

A tax can have several aims: revenue, more homes on the market, lower rents, fairness. We set one condition every design must meet: Ljubljana's rents must fall in at least 90% of draws, and no region's median may rise. For selection the bar is stricter: even the worse outcome (the 90th percentile) must be −1.0% or lower.

We tested 257 designs, each with 40 draws. 209 meet the condition, 204 with the margin. The 2024 draft does not: Ljubljana's rents in year five +2.6% (−2.0% to +8.5%), falling in only 26% of draws.

What does each rule do? We compared pairs of designs that differ in that rule only, and took the mean change.

A rate that rises with years empty is the only rule that helps two aims at once: +38 €m and +3,673 new homes. Exempting lets is what turns rents down (−11.7% points). Tight markets only is the costliest rule in homes (−16,556), because most empty dwellings lie outside them. Exempting uninhabitable dwellings costs only revenue: a derelict house pays and never comes to market.

Each dot is one of the 257 designs. Designs that meet the condition are blue. There is no design that gives the most homes and the most revenue at once: the most homes come with little revenue, the most revenue with five times the payers. Pick what the vertical axis shows.

At a rate of 1.45% and a credit of 25% of the rent, rents do not fall reliably. At 1.5% a fall needs a credit of at least 50% of the rent (−9.1%) or an exemption (−13.9%); with 25% of the rent it stays at +3.2%. A rent is already taxed as income (22.5% of the rent after the standard deduction), so taxing it as income rather than as property is the fairer way.

From the designs that meet the condition we chose three with different emphases and re-ran them with 300 draws.

A is a vacancy tax. Only dwellings that are truly empty, by electricity and water use, and holiday homes pay, at 3.0% of market value. It brings the most homes to market (34,139 in five years) and lowers rents most (−16%). It touches one home in thirteen, and has a fifth of the 2024 draft's mis-taxed payers. It raises little, and less each year (€98m in year five), because it works: empty dwellings are let or sold.

B keeps the 2024 draft's breadth (312,273 payers) and its revenue (€427m in year five). It repairs rents with a credit of 50% of the rent and a rate rising with years empty. The draft's unfairness stays almost whole: 130,053 mis-taxed payers.

C is the 2024 draft without its two worst rules. Exempting lets turns rents down; exempting relatives closes the gift route and halves the payers. It is the middle on every axis: €290m (about NUSZ), 29,879 new homes, rents −14%, 71,275 mis-taxed payers.

This is not a technical decision. If the aim is housing, the answer is A: a vacancy tax checked by meters, not registrations, so it is hard to avoid. If the aim is revenue, it is B, at the price of a third of homes paying and nearly half the payers taxed on a wrong base. C is the 2024 draft with the fewest changes that reliably lowers rents.

All three need the same things, though: a base written into statute, a tax on a fraction of the value and an appeal by evidence. Without them, each would be open to the same objections as the tax the Constitutional Court annulled.

An owner is not a person. The register cannot link two dwellings to the same owner: every private owner carries the same marker. So we do not know how many households would pay, how many pay for several dwellings, or who pays by income. Every number is per dwelling.

Who lives where comes from the 2021 register, not from reality; hidden lets are calibrated on one municipality's electricity meters.

Behaviour is judgement. The response probabilities are calibrated on experience abroad and on ranges that seem plausible to us; that is why the band matters as much as the median. Revenue depends most on how well FURS could check who lives where.

No macroeconomics. Interest rates, wages, building, migration and tourism stay on the path without the tax. Short-term letting is not modelled, nor the building land the draft would also have taxed.

A vacancy tax assumes meter data may be used. Whether such a tax would survive a court challenge is outside this analysis. Revenue includes no administration cost.