Property data and risk glossary

Asking price vs transaction price

An asking price is the figure a seller advertises; a transaction price is the figure written into the purchase contract and, eventually, into a transaction index. They differ in level, in timing and in what they say about the market. Asking prices are the seller’s opening position, move the day an advert is edited and are observable for the whole advertised stock. Transaction prices are the outcome of negotiation, reach the statistical office only after the transfer is registered, and are usually published as an index rather than as a level.

The two series do not measure the same basket, so their gap is not a discount figure. In Slovakia, for example, the central bank’s asking-price series rose 9.5 percent year on year in the second quarter of 2026 while Eurostat’s transaction-based house price index for the same country rose 14.4 percent in the first quarter: different quarters, different weights, both correct. A model that needs a level works from asking prices; a model that needs revision discipline anchors on the transaction index.

Bytero Live is an asking-price index and says so on the page; the index methodology guide explains what that allows and what it does not.

Sources: 6, 7

Automated valuation model (AVM)

An automated valuation model is a statistical model that estimates the market value of a property from recorded data about it and about comparable transactions or offers, without a physical inspection. Inputs are typically location, floor area, layout, building type, age and condition, plus the prices of nearby properties; the output is a value, a confidence band and the evidence the estimate rests on. The model families in use range from comparable-based grids and hedonic regressions to gradient-boosted trees.

In European lending the concept has a regulatory anchor. The Capital Requirements Regulation requires banks to monitor the value of residential collateral at least once every three years and commercial collateral at least once a year, and allows statistical methods for that monitoring; the EBA guidelines on loan origination, applying from 30 June 2021, set governance expectations for advanced statistical models used in valuation. An AVM is therefore not a substitute for every appraisal but a defined tool with a defined place.

Bytero Engine is an AVM of this kind; the AVM guide walks through the inputs, model families and error metrics.

Sources: 1, 2

Base period

The base period of an index is the period whose value is fixed, usually at 100, so that every other period is read relative to it. It is a presentational choice, not a statement about the market: rebasing an index from 2010 = 100 to 2015 = 100 changes every number and none of the growth rates. Eurostat’s house price index uses 2015 as its reference period, while national offices may keep an older base for their own series; the Slovak statistical office publishes its national index of realized dwelling prices on a 2010 = 100 base and Eurostat republishes the same series on 2015 = 100, where it stood at 222.7 in the first quarter of 2026.

Two cautions follow. An index says nothing about the price level in currency per square meter, only about change since the base. And comparing two indices requires them to share a base or to be rebased to one, otherwise two readings describe different starting points, not different markets.

Bytero Live sidesteps the question by publishing levels in euros per square meter rather than a 100-based index; see Bytero Live.

Sources: 3, 7, 8

Cadastre (list vlastníctva, katastr nemovitostí, księga wieczysta, tulajdoni lap)

A cadastre is the public register of land parcels, buildings and the rights attached to them. In Slovakia the ownership record is the list vlastníctva (LV), in Czechia the katastr nemovitostí, in Poland the księga wieczysta (the land and mortgage register kept by the courts) and in Hungary the tulajdoni lap (title sheet). Each records the owner, the parcel and building identifiers, the legal floor area of a flat and its co-ownership share of the common parts, and encumbrances such as mortgages and easements.

What a cadastre does not record matters as much. It rarely holds a usable postal address for a flat, never the price paid in a form the public can read, and nothing about the size of a house beyond its footprint. Under section 68 of the Slovak Cadastral Act the collection of deeds that contains the purchase contract, and therefore the price, is open only to the owner, other authorized persons and their successors. The cadastre is the spine to which prices, valuations and risk are attached, not a source of them.

Bytero Recover derives the floor area, address and property type that the register leaves implicit; see Recover.

Sources: 21, 22, 23

Chained Laspeyres index

A Laspeyres index measures price change between two periods using the quantities, or weights, of the earlier period, so that the answer reflects what was bought or advertised then, not now. A chained Laspeyres index applies that comparison one step at a time, period to period, and multiplies the steps together, so that the weights are refreshed at every link instead of being frozen at the base. Eurostat’s house price index is an annually chain-linked Laspeyres-type index with weights representative of the previous year and a 2015 = 100 reference period.

The virtue of chaining is that a change in what is advertised or transacted does not masquerade as a change in price. The cost is that each link carries its own sampling error, and an error in one link is carried forward into every later level. That is why a chained index built on a thin sample needs a gate on each step, and why a visible break is more honest than a join across two different baskets.

Bytero Live chains monthly cell medians of price per square meter with the previous month’s cell counts as weights; the methodology guide shows the arithmetic.

Sources: 3

Collateral monitoring

Collateral monitoring is the lender’s ongoing check that the property securing a loan is still worth what the loan file assumes. Article 208 of the Capital Requirements Regulation sets the floor: the value of immovable property must be monitored frequently, at least once a year for commercial property and at least once every three years for residential property, and more often when market conditions change materially. The same article allows statistical methods for the monitoring and requires a review by an independent valuer when the monitoring shows that the value may have declined materially relative to general market prices.

In practice monitoring is a portfolio task, not a file-by-file one: a bank re-values a book of tens of thousands of properties from an index or a model, flags the ones whose loan-to-value has moved past a threshold and sends only those for individual review. The quality of the monitoring is therefore the quality of the model and of the index it runs on, and both should be documented to the standard the EBA guidelines expect.

This is the batch use of an AVM; Bytero Engine re-values a book on a schedule and returns the evidence behind each value.

Sources: 1, 2

Comparable (comp)

A comparable, or comp, is a recently sold or currently advertised property that resembles the one being valued closely enough for its price to inform the valuation. Resemblance is judged on location, property type, floor area, layout, condition, age and the date of the price; the closer the match, the smaller the adjustment a valuer must make. The comparable method is the oldest valuation approach and the one that appraisers, courts and lenders find easiest to audit, because every step is a visible price and a visible adjustment.

Its weakness is scarcity. In a thin market the nearest comparable may be a year old or a kilometer away, and the adjustments then carry most of the value. Automated methods answer that by drawing on hundreds of weaker comparables and letting a model weight them, at the cost of the audit trail unless the model reports which comparables it used. A comparable drawn from an asking price also needs the gap between asking and transaction prices taken into account before it is treated as evidence of value.

Bytero Engine returns the comparables behind each valuation together with the input data and model version; see Engine.

Depth-damage curve

A depth-damage curve, or vulnerability function, converts a flood depth at a building into the fraction of the building’s value that is lost. It is the step that turns a hazard map into money: a few decimetres of water on the ground floor of a masonry house destroys a small share of the reinstatement value, two meters a large one. Curves are defined per occupancy class (residential, commercial, industrial) and per construction type, and they rise steeply in the first meter, where floors, plaster, electrics and contents sit, then flatten.

The most used open set is the global flood depth-damage database published by the Joint Research Centre in 2017, which gives a curve per continent and a maximum damage value per country derived from construction-cost data, with guidance on adjusting for urban or rural settings and for building materials. Curves calibrated on a carrier’s own claims are better still, but they are rarely public, and a curve transplanted from another climate or building stock is a common source of error.

Bytero Shield applies damage curves calibrated to real premiums; the expected annual loss guide shows a curve in use.

Sources: 15

Exceedance probability

The exceedance probability of a hazard level is the chance that it is equalled or exceeded in a given year. It is the reciprocal of the return period: a 100-year flood has an annual exceedance probability of 1 percent, a 20-year flood of 5 percent. Hazard maps are usually published for a small set of exceedance probabilities; the JRC river flood maps for Europe, for instance, give water depth for return periods from 10 to 500 years, which is annual exceedance probabilities from 10 percent down to 0.2 percent.

Two habits keep the concept honest. Express it over a horizon that matters to the decision: a 1 percent annual probability is a 26 percent probability over a 30-year mortgage, because one minus 0.99 to the power of 30 is 0.26. And remember that it is an estimate from a model or a record, not a law of nature: the probability assigned to a river reach changes when the hydrology is remodelled or when a new flood joins the record.

Bytero Shield reports modeled depth by return period for the address; see Shield.

Sources: 14

Expected annual loss (AAL)

Expected annual loss, also written AAL for average annual loss, is the loss a property would suffer per year averaged over a very long run of years, including the many years with no loss at all. It is computed by integrating the loss at each hazard level against the probability of that level: in the simplest form, the sum over return periods of the exceedance probability times the damage at that depth. A house that loses 40,000 euros in a 1-in-100-year flood and nothing otherwise has an expected annual loss of about 400 euros from that scenario alone.

AAL is the number that makes hazards comparable: flood, windstorm, hail and earthquake each collapse into euros per year, which can be summed per property and per portfolio. It is also the pure-risk core of a premium. Its limits are the limits of its inputs, above all the depth-damage curve and the value at risk, and it says nothing about the variance of a single year, which is what the probable maximum loss is for.

Bytero Shield returns an AAL and a probable maximum loss per address and per hazard; the AAL guide has a worked example.

Fluvial vs pluvial flooding

Fluvial flooding is a river leaving its channel; pluvial flooding is rain that cannot drain away fast enough and pools or runs across the surface, often far from any river. The two need different data. Fluvial risk is mapped by water authorities under the EU Floods Directive from hydraulic models of the river network: the Slovak hazard maps carry five return periods from Q5 to Q1000, the Czech ones four from Q5 to Q500. Pluvial risk is absent from most of those maps and has to be modeled from terrain, rainfall intensity and drainage, typically with a two-dimensional surface model over a fine grid.

The distinction matters for underwriting because a property outside every river zone can still flood from the street, and because the two perils accumulate differently: a fluvial event hits a valley, a pluvial event hits a city district. A risk score that reads only the river layer under-prices the house on the hill with a blocked culvert below it.

Bytero Shield combines the official river maps with a two-dimensional pluvial model at 5 m resolution in cities; see the address-level flood risk guide.

Sources: 13, 17, 18

Gradient boosting

Gradient boosting is a machine-learning method that builds a strong predictor out of many weak ones, usually shallow decision trees, by adding them one at a time and fitting each new tree to the errors the previous ones left behind. The result is a model that captures interactions a linear regression misses, such as the way a top floor adds value in a building with a lift and subtracts it in one without. XGBoost, described by Chen and Guestrin in 2016, is the implementation most valuation teams use because it is fast, handles missing values and exposes feature importance.

For property valuation the method’s strengths are accuracy on tabular data and tolerance of messy inputs; its weaknesses are extrapolation, where it flattens outside the range it has seen, and explanation, which needs additional tools such as SHAP values. It also needs retraining as the market moves, because a boosted model learns a snapshot of prices, not a rule for how they change.

Bytero Engine is an XGBoost model retrained weekly; see Engine.

Sources: 5

Gross rental yield

Gross rental yield is annual rent divided by property value, before any cost. A flat that lets for 800 euros a month against a value of 200,000 euros yields 9,600 divided by 200,000, or 4.8 percent. It is the first filter an investor applies and the number most often quoted in market commentary, precisely because it needs only two inputs and no assumptions about tax, vacancy or repairs.

Its usefulness ends where those assumptions begin. Two flats with the same gross yield can differ by a third in what the owner keeps once service charges, maintenance, insurance, agency fees, vacancy and income tax are deducted. Gross yield also depends on which value sits in the denominator: an asking price, a transaction price and a model estimate give three different yields for the same rent. When a yield is computed from a predicted rent against a predicted value, both estimates carry error and the ratio carries both, so the yield should be shown with its inputs.

Bytero Engine reports a yield from its rent estimate and its value estimate for the same property, with each input shown; see Engine.

Hazard zone (EU Floods Directive)

Directive 2007/60/EC obliges each member state to map flood hazard and flood risk for the areas it identifies as being at potentially significant risk, and to redo the exercise every six years. The hazard maps must show flood extent, water depth and, where appropriate, velocity for three scenarios: floods with a low probability or extreme events, floods with a medium probability whose likely return period is 100 years or more, and, where appropriate, floods with a high probability. Each state chooses its own return periods within that frame, which is why one country publishes Q5 to Q1000 and a neighbor Q5 to Q500.

A hazard zone is therefore membership in a mapped extent for one scenario, not a probability for the address and not a depth. The 2024 Slovak preliminary assessment, for instance, lists 202 areas of potentially significant flood risk; a reach outside that register has no official map at all, which is not the same as no risk.

Bytero Shield reads zone membership and modeled depth separately and reports both; the flood risk guide explains why.

Sources: 13, 16

Hedonic model

A hedonic model treats a property as a bundle of characteristics and estimates a price for each one: so many euros per square meter, a premium for a balcony, a discount per floor without a lift, a location effect per district. The price of any property is then the sum, or in log form the product, of those implicit prices. Fitted as a regression on a large sample, the model both values properties it has not seen and separates price change from quality change, which is why statistical offices use hedonic methods for house price indices; Eurostat’s 2013 handbook on residential property price indices sets out the standard forms.

The model’s strength is transparency: each coefficient is a readable statement about the market. Its weakness is rigidity, since a linear form struggles with interactions and with non-linear effects such as the value of the first ten square meters against the last ten. Gradient-boosted models trade that transparency for fit, and many production systems run both, one to value and one to explain.

Bytero builds hedonic-style features into a boosted model; the AVM guide compares the families.

Sources: 4

Hit rate

Hit rate is the share of automated valuations that fall within a stated band of the reference price, most often plus or minus 10 percent, and is written PPE10 (percentage of predictions within 10 percent) or PPE20 for the wider band. If 100 flats are valued and 68 of the estimates land between 90 and 110 percent of the price later paid, the hit rate at 10 percent is 68 percent. Lenders and auditors like the measure because it answers a plain question, how often is the model close enough, and because it can be reported per segment: by city, by property type, by price band.

Hit rate should always travel with the median absolute percentage error and with the size of the test set, because a high hit rate on a filtered sample is easy to manufacture. It should be measured against transaction prices where they exist and disclosed as measured against asking prices where they do not, since the two references are not interchangeable and the second flatters the model.

The AVM guide sets out how hit rate and MdAPE are read together and what to ask a vendor for.

Indexation of the sum insured

Indexation is the annual adjustment of the sum insured, and usually of the premium, by a published index so that the cover keeps pace with construction costs or property prices. Without it a sum set at inception drifts below the reinstatement cost within a few years of cost inflation, and the policyholder walks into underinsurance without any change on their side. Carriers write the mechanism into their terms: one Slovak wording indexes flats by the central bank’s residential price index, buildings by the statistical office’s construction cost indices and contents by consumer prices, gives ten weeks’ notice before renewal, and makes uninterrupted indexation a condition of its guarantee against underinsurance.

The practical questions for a book are which index each policy is tied to, whether indexation was ever refused or interrupted, and whether the starting sum was right in the first place, because an index preserves an error as faithfully as it preserves a correct value. Re-indexing a book against a fresh rebuild cost per property is the only way to find the policies where the chain broke.

Bytero Single Input Quote re-indexes up to 3,000 addresses per call from current rebuild cost; see SIQ.

Sources: 11

LiDAR and digital surface model

LiDAR (light detection and ranging) is airborne laser scanning: an aircraft sweeps the ground with laser pulses and records the return time of each, producing a point cloud of the terrain and everything standing on it. Two products are derived from the cloud. A digital terrain model (DTM) keeps the bare earth, which is what flood models run on; a digital surface model (DSM) keeps the tops of buildings, trees and bridges. Subtracting the one from the other gives the height of every object, and clipping that height to a building footprint gives storeys and enclosed volume.

National surveys now make this routine. The Slovak mapping authority scanned the whole country between 2017 and 2022 and publishes the 1 m DMR 5.0 terrain model and the DMP 1.0 surface model free of charge, with last-return point densities between 15 and 52 points per square meter. For rebuild cost that turns a declared footprint times a guessed storey count into a measurement.

Bytero Recover measures footprint, storeys, height and volume this way, verified against laser measurements of 1,473 houses; the LiDAR rebuild cost guide covers where the method breaks.

Sources: 19, 20

Loan-to-value (LTV)

Loan-to-value is the loan amount divided by the value of the property securing it. A 160,000 euro mortgage on a 200,000 euro flat has an LTV of 80 percent. At origination it measures the borrower’s equity cushion and drives both the lender’s capital requirement and the rate offered; over the life of the loan it moves with every repayment and with every change in the property’s value, which is why it must be re-computed rather than filed.

Macroprudential authorities cap it. The Czech National Bank sets an upper limit of 80 percent for mortgage loans, 90 percent for applicants under 36 buying owner-occupied housing, with a 5 percent volume exemption for loans above the cap, under its provision of general nature of 25 November 2021 as amended. The denominator is the contested part: an asking price, a transaction price and a model value give three LTVs for the same loan, and a lender that monitors on a stale value is reporting a stale ratio.

Bytero Engine supplies the value side at origination and for LTV monitoring; see Engine.

Sources: 9

Market value

Market value is the estimated amount for which a property should exchange on the valuation date between a willing buyer and a willing seller in an arm’s-length transaction after proper marketing, where each party has acted knowledgeably, prudently and without compulsion. That wording, from Article 4 of the Capital Requirements Regulation, matches the international and European valuation standards and is the value that lenders, courts and most insurers of flats mean when they say value.

Three things it is not. It is not the asking price, which is one seller’s opening position. It is not the rebuild cost, which is what a building would cost to construct again and can sit far above or below market value depending on the land beneath it. And it is not a single number in nature: it is an estimate, and every method that produces it, appraisal, comparables, hedonic regression or boosted model, carries an error band that should be reported with it.

Bytero Engine estimates market value with a confidence band and the comparables and inputs behind it; the rebuild cost vs market value guide sets the two values side by side.

Sources: 1

Median absolute percentage error

Median absolute percentage error, MdAPE, is the middle value of the absolute percentage errors of a set of valuations against their reference prices. If three estimates miss by 4, 7 and 25 percent, the MdAPE is 7 percent. It is the standard accuracy statistic for automated valuation models because it is in units everyone reads, percent of price, and because the median ignores the handful of wild misses that would dominate a mean.

That robustness is also its blind spot: a model with a MdAPE of 6 percent can still be 40 percent wrong on one property in twenty, and the lender holding that property does not care about the median. MdAPE should therefore be read with the hit rate, with the 90th percentile of error and with the size and selection of the test set. It should be measured out of sample and out of time, on transactions the model never saw and that happened after it was trained, and reported per segment, since a national figure hides the thin markets where the model is weakest.

The AVM guide explains how these metrics are measured and what a buyer should ask for.

Mix adjustment

Mix adjustment is the correction that stops a change in what is being sold or advertised from being read as a change in price. If large flats in the centre make up 20 percent of the offers one month and 30 percent the next, the raw average price per square meter rises even if every individual flat is priced exactly as before. A mix-adjusted index holds composition constant: it divides the market into cells, type by region by layout for instance, measures the price change inside each cell and combines the cells with fixed or lagged weights.

Official house price indices do this by design; Eurostat’s HPI weights each year by the previous year’s household expenditure on transacted dwellings, so the basket is refreshed annually and fixed within the year. The limit of the method is the cell: a cell with five properties gives a median that moves on noise, so a mix-adjusted index needs a floor on cell size and a rule for cells that are empty in one month.

Bytero Live is mix-adjusted on type, region and layout cells with the previous month’s counts as weights; see Bytero Live.

Sources: 3

Net yield

Net yield is annual rent minus the owner’s running costs, divided by the property value. Continuing the gross example, a flat that lets for 9,600 euros a year against a 200,000 euro value, with 2,400 euros a year of service charges, insurance, maintenance reserve and management, yields 7,200 divided by 200,000, or 3.6 percent, against a gross yield of 4.8. Add a month of vacancy every two years and the figure falls again.

The definition is only as good as its cost list, and lists differ: some include income tax and social or health contributions on rental income, which vary by jurisdiction and by the legal form of the owner; some deduct mortgage interest, which turns a property yield into a leveraged return on equity. A comparison between two net yields is meaningful only when the two lists match. For portfolio work the honest presentation is a gross yield, an itemised cost stack and the net yield that follows, so that the reader can swap in their own assumptions.

Price per square meter

Price per square meter is the price of a property divided by its floor area. It is the unit that makes properties of different sizes comparable and the unit in which most European residential statistics are published: the Slovak central bank’s asking-price series put the national average at 3,041 euros per square meter in the second quarter of 2026. Almost every valuation, index and piece of market commentary is built on it.

The denominator is where the trouble lives. Floor area can be the usable area, the total area including balconies and cellars, the area on the cadastral sheet or the area in the advert, and the same flat can carry four figures that differ by 10 percent or more. Houses are worse: a price per square meter of a house depends on whether the garage and the attic are counted and says nothing about the plot. A statistic that does not state its area definition cannot be compared with one that uses another.

Bytero Live is computed on price per square meter inside cells of the same type and layout, which limits how far area definitions can distort it; see Bytero Live.

Sources: 6

Rebuild (reinstatement) cost

Rebuild cost, or reinstatement cost, is what it would cost to construct the building again on the same site, including demolition, design fees and tax where they apply, and excluding the land. It is the value that property insurance for buildings is written on, because an insurer that pays a claim is paying for construction, not for a purchase. Slovak decree 492/2004 on the general value of property defines the corresponding východisková hodnota as the expert estimate of the value at which the building could be acquired by construction at the time of valuation, and the technická hodnota as that figure reduced for wear.

Rebuild cost and market value move for different reasons. Market value carries the land and the location; rebuild cost carries construction prices and the building’s size. In a city centre the market value of an old flat can be several times its rebuild cost; in a village the rebuild cost of a large house can exceed anything a buyer would pay. Insuring at the wrong one of the two is the commonest route into underinsurance.

Bytero Recover computes rebuild cost from measured building volume; the rebuild cost vs market value guide has a worked example.

Sources: 10

Return period

A return period is the average interval between events of a given size or larger, in years. A 100-year flood, written Q100, is a discharge that is equalled or exceeded on average once in 100 years, which is the same statement as an annual exceedance probability of 1 percent. It is an average over a very long run, not a schedule: two Q100 floods can come in consecutive years, and the chance of at least one in any 30-year period is about 26 percent.

The EU Floods Directive frames its three scenarios in these terms, with the medium-probability scenario defined as a likely return period of 100 years or more, and leaves the exact periods to each state; Slovak maps carry Q5, Q10, Q50, Q100 and Q1000, Czech maps Q5, Q20, Q100 and Q500. Return periods are also model outputs, re-estimated when hydrology is remodelled or when a new flood joins the record, so a Q100 line drawn in 2013 and one drawn in 2025 for the same reach are not the same line.

Bytero Shield reports modeled water depth at the address by return period; see Shield.

Sources: 13, 17, 18

Split-half standard error

Split-half standard error is a way of measuring how much of a statistic is signal and how much is sampling noise without assuming a distribution. The sample is divided into two halves by a fixed rule, the statistic is computed on each half independently, and the disagreement between the halves is the evidence: if the market really moved, both halves see the move; if the step is noise, they disagree. Repeating the split several times with different fixed rules and taking the spread gives a standard error in the statistic’s own units.

For a chained price index the statistic is each month-over-month step, and the method answers the question a thin sample raises, is this step real, before the step is published. The split must be deterministic: a random split re-rolls every night and lets borderline steps flap in and out of a public series. What the method cannot see is a bias shared by both halves.

Bytero Live measures every step over twelve fixed splits and withholds any step whose standard error exceeds 1.0 percentage point, breaking the chart rather than joining two baskets; the methodology guide gives the calibration.

Sources: 24

Sum insured

The sum insured is the maximum the insurer will pay under a property policy and the base on which the premium is computed. For a building it should equal the rebuild cost; for a flat, carriers differ, and their terms name either the rebuild value, the market value or the higher of the two. It is set at inception from the policyholder’s declared inputs or from the carrier’s own calculator, and then indexed, or not, each year.

Two consequences follow from getting it wrong. If the sum is below the insurable value, a proportional clause reduces every partial claim, not only a total loss. And the sum caps the payout even where the policy promises market value: one Slovak wording pays the general value of a flat after ordered demolition but no more than 115 percent of the sum insured. The sum insured is therefore not a guess at how much could be lost; it is the denominator of every claim, and the number on a book that most deserves re-checking against a fresh value.

Bytero Single Input Quote returns a sum insured computed by the statutory method, each coefficient traceable, alongside the market value; see SIQ.

Sources: 11

Technical premium

The technical premium is the price an insurer needs for a risk before commercial adjustment. Its core is the expected annual loss, the pure premium; on top of that sit a loading for the volatility of the loss, which is the capital the insurer must hold against a bad year, the expenses of acquisition and administration, reinsurance cost and a profit margin. A property with an expected annual loss of 400 euros from flood therefore needs a technical premium for that peril well above 400 euros, and the market premium sits above or below the technical one depending on competition.

The technical premium is the reference against which pricing decisions are judged: a tariff that charges less than it for a segment is buying market share with capital, and a tariff that charges much more is inviting a competitor to select the good risks out of the book. Comparing a carrier’s actual premiums with a technical premium built from public hazard data is the quickest way to find where a tariff is mispriced by segment.

Bytero Shield’s loss estimates are calibrated on 993 properties and the tariffs of ten carriers; see Shield.

Underinsurance and proportional reduction

Underinsurance is the state of a policy whose sum insured is below the insurable value of the property. Proportional reduction, or average, is what most building wordings do about it: the claim is reduced in the ratio of the sum insured to the insurable value, and that ratio is applied to every partial claim, not only to a total loss. A building that costs 250,000 euros to rebuild, insured for 150,000, suffers a 10,000 euro flood loss and receives 6,000. The Czech Civil Code writes the rule into section 2854; in Slovakia it is contractual, and one carrier’s wording adds a tolerance band that waives the reduction where the shortfall is under 15 percent, plus a guarantee where the sum was set by the carrier’s own calculator from true inputs and indexed without interruption.

The common path into underinsurance is not fraud but drift: a sum set correctly years ago and never re-indexed, or a flat insured at rebuild cost under a wording that measures a total loss against market value.

The rebuild cost vs market value guide walks through the arithmetic and the carrier clauses.

Sources: 11, 12