The quantitative AVM methodology behind every QPV valuation.
Built for Hong Kong property: standardised inputs, explainable outputs, a number you can audit.
Six standardised inputs, one repeatable method, every valuation. Here is what QPV measures, how it weights each factor, how it reports uncertainty, and how it proves the same number again six months later.
How QPV values a Hong Kong property.
QPV values a Hong Kong property by decomposing it into six standardised inputs, comparable transactions, building age, floor level, unit size, unit condition, and transaction recency, then weighting each one with figures derived from Hong Kong transaction data. The model selects the closest comparable sales by similarity score, gives recent sales more weight than older ones, and returns a point estimate inside a confidence range with a full audit trail. Feed the same inputs in six months later and the same output comes back within tolerance. That is the whole method in one paragraph. The rest of this page explains each part and why it is built this way.
The approach is quantitative, not a surveyor's freeform judgement and not a black-box AI score. It borrows its discipline from bank market risk management, where every asset is priced from its drivers, weighted by defined rules, and recorded so the number can be reviewed rather than trusted on faith. For a fuller primer on automated valuation models in this market, read the guide to AVMs in Hong Kong.
The V1 dataset behind the model covers 7,096 Hong Kong residential transactions across 35 plus districts, with 914 properties valued end to end and AI-assisted condition scoring from public listing photos, spanning Hong Kong Island, Kowloon, and the New Territories. Coverage grows as new transactions settle.
A model, not intuition. Explainable, not a black box.
Most property valuation in Hong Kong sits at one of two extremes. At one end is the surveyor: experienced, but slow, and reliant on judgement that is hard to reproduce or review. At the other end is the black-box AI score: fast, but unable to tell you why it landed where it did. QPV is built to take the discipline of the first and the speed of the second without inheriting their weaknesses.
The reference point is market risk management. A bank risk desk values thousands of financial instruments every day by decomposing each one into its drivers, applying defined weights, computing a price with a confidence interval, and recording the audit trail. Property is priced the same way: from comparable transactions, with explainable drivers, under a requirement to be reproducible. QPV applies the risk-desk logic to Hong Kong residential property so that a valuation can be defended line by line, not asserted.
This is the opposite of an opaque model. A surveyor and an AVM both produce a number; the difference QPV draws is whether the number arrives with its reasoning attached. For a direct comparison of the two approaches and when each is the right tool, see AVM versus surveyor valuation in Hong Kong.
Standardised inputs, not surveyor intuition.
Every QPV valuation is decomposed into the same six factors every time. No freeform adjustments. No 'sunset premium' or 'harbour view plus 8 percent' invented on the spot. Every factor has a defined weight derived from Hong Kong transaction data.
Comparable transactions
Recent sales of similar units in the same building or neighbouring blocks. Top 30 selected by similarity score.
Building age
Year of completion. Older buildings adjusted against renovation status and structural depreciation.
Floor level
High floor, mid floor, low floor. Premium and discount weightings derived from transaction data per district.
Unit size
Saleable area in square feet. Non-linear pricing curve calibrated per district.
Unit condition
AI-scored from public listing photos. Six-band scale, fully overridable with audit trail.
Transaction recency
Time decay on comparable weighting. Recent sales weight more heavily than older ones.
Two of these factors carry most of the explanatory power in Hong Kong: floor level and district. A high floor in a harbour-facing tower can trade at a large premium to a low floor in the same block, and the gap is not constant, it is calibrated per district from the data. The same is true across neighbourhoods, where price per square foot varies enough that a single citywide model would misprice most units. The way these premiums move from one area to the next is set out in the Hong Kong property valuation by district breakdown.
Risk teams get a repeatable input signature. Banks get a methodology that can be reviewed, not a narrative. Reviewed six months apart, the same inputs give the same output within tolerance.
How the comparables are chosen.
The single largest driver of any property valuation is the set of comparable transactions it leans on. Choose the wrong comparables and every downstream number is wrong, however precise it looks. QPV ranks candidate sales by a similarity score that blends building, location, size, floor, and age, then takes the closest set and weights them. Sales in the same building or neighbouring blocks count more than distant ones, and recent sales count more than older ones through an explicit time decay.
This is also why two honest valuers can disagree. If they start from different comparable sets, they reach different numbers, and neither is lying. The dispersion is structural, and it is the reason a defensible method has to make its comparable selection visible rather than hidden. We unpack that effect in why three surveyors give three valuations.
When comparable data is thin, for an unusual unit, a single block, a village house, or a property with a large renovation gap, QPV does not hide the weakness. It widens the confidence range to reflect it. Sparse data should make a valuation less certain, and the output should say so.
Every valuation shows its drivers.
A QPV valuation is not a number. It is a number with its reasoning visible: which comparable transactions were selected, how they were weighted, which features moved the estimate up or down, and by how much. Open any valuation and the audit trail opens with it.
This matters for three groups. Bank risk committees approving loans need to see the basis, not a headline figure. Regulators reviewing portfolio exposure need to trace it. Sellers deserve to understand why their property is valued at HK$11.2M and not HK$11.8M. In every case the answer is the same: the drivers are on the page, so the number can be challenged and defended rather than taken on trust.
Uncertainty is information.
Every QPV output is a band, not a single figure. Take a worked example: a 560 square foot Wan Chai apartment valued at HK$10.65M with a plus or minus 15 percent interval, which places the likely market value between HK$9.10M and HK$12.20M for typical similar transactions. The point estimate sits at the 69th percentile against 234 historical Wan Chai transactions, where the district median runs around HK$16,074 per square foot.
The point estimate inside its band. A narrow band is a confident valuation, a wide one is a warning.
The width of that band is itself a signal. It widens when comparable data is sparse, it ages with market volatility, and it tightens when fresh transactions stack up in a district. A narrow band is a confident valuation; a wide band is the model telling you to be careful. Banks underwrite against the low end of the range, sellers price against the high end, and regulators see both. Where a mortgage sits against that value is what decides whether an owner is in negative equity. The way QPV reports and reads that confidence is covered in detail in AVM confidence in Hong Kong.
The figures above are an illustrative, anonymised worked example, not a formal or regulated valuation. See Terms.
How accuracy is measured: PPE10, MdAPE, and honest reporting.
The standard accuracy metric for an automated valuation model is PPE10, the share of valuations that land within 10 percent of the eventual sale price. A related metric, MdAPE, reports the median absolute percentage error. Mature AVMs in mature Western markets achieve PPE10 of 80 to 90 percent. QPV targets 70 percent or more on its early-stage dataset, a deliberately conservative number for a prototype working from an early-stage Hong Kong transaction set, with a target of 85 percent or more within 12 months as Land Registry integration deepens.
The honest part matters as much as the number. Many providers quote a single accuracy figure without the sample, the date, or the property types it excludes. QPV reports accuracy as a target while the V1 benchmark is still being finalised, rather than publishing a flattering figure that would not survive scrutiny. The full explanation of these metrics, what they include, and where Hong Kong AVMs genuinely sit is in AVM accuracy explained, MdAPE, PPE10 and the Hong Kong gap. For the plain-English version aimed at owners and buyers, see how accurate online property valuations in Hong Kong really are.
Reproducible six months later.
Every valuation records its inputs, weights, comparables, photo-condition scores, market-condition flags, and output. The record is stored immutably. Six months later, a different team can replay the same inputs and get the same output within tolerance. This is a hard requirement for bank-grade valuation, not a nice-to-have.
For institutional valuation use cases, the audit trail is not optional. For mortgage origination, it is the difference between a loan approved and a loan denied. For sellers, it is proof the number was not negotiated into existence. A valuation that cannot be reproduced is an opinion; a valuation that can is evidence.
Built for Hong Kong, not retrofitted.
QPV is calibrated on Hong Kong transaction data from 2024 onwards, sourced from Land Registry instruments and normalised through QPV's data pipeline. Building stock is classified by Hong Kong typology: public estate, Home Ownership Scheme, private estate, walk-up, and single block. Regulatory context aligns with the HKMA, the HKIS, and Basel III as implemented in Hong Kong from 1 January 2025. A global AVM retrofitted to Hong Kong would miss the district-level nuances that drive a large share of price variation in this market.
This local grounding is what connects the method to the decisions people actually make. When a bank values a flat below the agreed purchase price, the gap is set by exactly these district and floor calibrations, a situation we walk through in when the bank valuation comes in low. The HKMA loan-to-value ceilings that turn that gap into a financing problem are explained in the HKMA loan-to-value and mortgage valuation guide. Coverage spans 35 plus Hong Kong districts and grows daily as new transactions settle.
Who relies on the method, and for what.
The same valuation serves different decisions depending on who is reading it. Bank risk and credit teams use the low end of the range and the audit trail for mortgage origination and portfolio review, where the loan-to-value ceiling depends directly on a defensible valuation. Regulators and auditors use the reproducibility to trace exposure across a book. Sellers and buyers use the point estimate and range to price realistically and to understand a bank valuation that disagrees with the agreed price. Brokers and agents use it as a fast, neutral second opinion before a deal is papered.
What ties these together is that each group needs a number it can stand behind. A surveyor's report carries professional weight on unusual properties and where the law requires it; a quantitative model carries consistency and speed across the thousands of standard transactions that make up most of the market. QPV is built for the second job and is explicit about the first.
What people ask.
How does QPV value a Hong Kong property?
QPV decomposes every property into six standardised inputs: comparable transactions, building age, floor level, unit size, unit condition, and transaction recency. Each input carries a weight derived from Hong Kong transaction data. The model selects the closest comparable sales by similarity score, weights recent sales more heavily, and returns a point estimate inside a confidence range with a full audit trail. The same inputs reproduce the same output six months later.
Why quantitative methods instead of a surveyor?
Market risk teams in banks value thousands of financial assets every day using a consistent logic: decompose every asset into its drivers, apply defined weights, compute a price with a confidence interval, record the audit trail. QPV applies that logic to Hong Kong property. A surveyor adds judgement on unusual properties, but a quantitative model gives consistency, speed, and a reviewable trail across thousands of standard units. The two are compared in full in AVM versus surveyor.
How does QPV measure accuracy?
QPV uses PPE10, the share of valuations within 10 percent of the eventual sale price. QPV targets 70 percent or more on its early-stage dataset, against 80 to 90 percent for mature Western AVMs. The 70 percent target is intentionally conservative for a prototype. With full Land Registry integration, QPV targets 85 percent or more within 12 months.
What data does QPV use?
Hong Kong Land Registry transaction records from 2024 onwards, sourced from registered Memorial instruments and normalised through QPV's data pipeline. Building-level and district-level metadata from QPV's own platform layer. AI condition scoring from public listing photos. QPV uses no non-public data sources. Every valuation can be reproduced from public inputs plus the QPV model.
Why does QPV give a range instead of a single number?
A single number hides how confident the model is. Every QPV output is a band with a point estimate. The band widens when comparable data is sparse and tightens when fresh transactions stack up. Banks underwrite against the low end, sellers price against the high end, regulators see both. How QPV reports that confidence is covered in AVM confidence in Hong Kong.
Is a QPV valuation a formal or regulated valuation?
No. QPV outputs are quantitative estimates with confidence ranges, not formal or regulated valuations, and not a substitute for a HKIS chartered surveyor report where one is legally required. Worked examples on the site are illustrative and anonymised. See Terms.
Can a QPV valuation be reproduced and audited?
Yes. Every valuation records its inputs, weights, selected comparables, photo-condition scores, market-condition flags, and output. The record is stored immutably, so a different team can replay the same inputs six months later and reach the same output within tolerance. That audit trail is a hard requirement for bank-grade and institutional valuation use cases.
See the methodology in action.
Walk through a sample valuation report or speak with the team about your portfolio.
Learn more: About QPV or read the full QPV article library.