ARTICLE / AVM HONG KONG

AVM in Hong Kong: a complete guide for brokers, banks, and owners.

An Automated Valuation Model is a statistical system that prices property from data. This is how AVMs work in Hong Kong, what accuracy is realistic to expect, and how the regulatory context shapes their use.

By Ippei Folliot, founder, Quant Property Valuation. Published 26/04/2026, updated 03/05/2026.

01 / DEFINITION

What an AVM actually is.

An Automated Valuation Model (AVM) is a software system that estimates the market value of a property without a physical inspection. It takes inputs (address, size, building age, district, recent comparable sales, market conditions) and returns a price estimate, ideally with a confidence range.

AVMs are not new. The largest AVM globally is Zillow's Zestimate in the US, with hundreds of millions of valuations live at any time. CoreLogic, Quantarium, and HouseCanary supply AVM data to most major US banks. The UK uses Hometrack and Rightmove's RPI. Australia uses CoreLogic's RP Data.

In Hong Kong, AVM use is more concentrated. The major lenders (such as HSBC, Hang Seng, Bank of China, Standard Chartered, Citi) operate internal AVMs informed by Land Registry transaction data, sourced from a small set of vetted vendors. Independent AVMs serving brokers, family offices, and property owners directly remain less common, which is the gap QPV is built into.

02 / HOW IT WORKS

The five-input recipe.

Every credible AVM operates on a similar input recipe, regardless of vendor:

  1. Comparable transactions. Recent sales of similar units in the same building, neighbouring blocks, or matched estates. Most HK AVMs draw from the past 6 to 12 months.
  2. Property characteristics. Saleable area, year of completion, floor, view, orientation. In HK, club facilities and clubhouse age also matter for estate-level pricing.
  3. Location. District, MTR proximity, school net, harbour view. HK's 35+ districts each carry distinct premiums.
  4. Market context. Hang Seng Property Index, HIBOR, mortgage rates, transaction volume. AVMs that ignore market direction overprice in falling markets and underprice in rising ones.
  5. Property condition. Renovation status, photo-derived condition score where available. This is where modern AVMs add value over Land Registry-only models.

The AVM weights these inputs (the weights are the "model"), computes a point estimate, and surrounds it with a confidence interval. The interval is typically wider for older buildings, sparse-comparable estates, or volatile market periods.

03 / ACCURACY

How to judge an AVM's accuracy: PPE10.

The standard AVM accuracy metric is PPE10, the Percentage Predicted Error within 10%. PPE10 of 80% means 80% of the model's valuations landed within 10% of the eventual sale price.

Industry benchmarks:

Any AVM that claims "95% accurate" without specifying the metric or the sample is overstating. Accuracy varies by district, building, market condition, and time-since-comparable. A responsible AVM publishes its accuracy and the conditions under which it degrades.

04 / REGULATION

HKMA, Basel III, and the regulatory frame.

HK banks operate AVMs under the Hong Kong Monetary Authority (HKMA) supervisory framework, with Basel III capital adequacy requirements as the binding constraint on mortgage-book risk. Two implications matter.

First: bank AVMs must be auditable. A bank using an AVM for mortgage origination must be able to show HKMA inspectors how the valuation was computed, what data was used, and what governance sits over the model. This is why HK banks buy from a small number of vetted vendors rather than building in-house from scratch.

Second: AVMs are not a substitute for chartered surveyor valuations on high-LTV deals. The HKMA Supervisory Policy Manual sets the framework, and individual lenders set internal triggers within it for when a physical valuation by a HKIS-registered surveyor is required. The AVM is the first-pass screen; the surveyor is the second-pass for risk teams. See the AVM vs surveyor decision framework for when each is required.

For an independent AVM serving brokers, family offices and owners, the regulatory frame is different from a regulated valuation: the QPV output is explicitly a model estimate with a confidence range, not a binding valuation, and is positioned as a second-opinion tool alongside, not in place of, a HKIS-registered surveyor where one is required.

05 / USE CASES

Who actually uses AVMs in Hong Kong.

Banks (mortgage origination): as a first-pass screen for loan applications. Higher loan-to-value applications trigger a chartered surveyor follow-up.

Mortgage brokers: as a second opinion when a bank valuation comes in below the agreed transaction price, threatening the deal. Mortgage brokers often shop multiple banks; an independent AVM helps them argue for a higher valuation at a competitor lender.

Estate agents: typically use agency-internal pricing tools (CentaEstimate, MidlandPricer) rather than third-party AVMs. Independent shop-front agents may use 28Hse comparable transaction data manually.

Family offices and HNWI buyer's agents: for portfolio monitoring and acquisition due diligence. AVMs scale better than commissioning surveyor reports across a 20-property portfolio.

Property owners: to verify a bank valuation before closing, or to set a realistic listing price. The most common search query is some variant of "is the bank valuation low?" at the moment a deal is at risk.

06 / QPV

How QPV approaches HK AVM.

QPV operates as an independent AVM for Hong Kong residential property. The methodology is published in full at /methodology: standardised inputs, explainable outputs, confidence ranges, full audit trail. Unlike bank AVMs (internal black boxes) and unlike consumer estimator tools (point estimates with no confidence band), QPV outputs a price range with stated uncertainty and the underlying comparables.

Coverage spans 35+ HK districts, drawing from Land Registry transaction data 2024 onwards and AI-derived condition scores from public listing photos. The model is HK-native: trained exclusively on HK transactions, calibrated for HK building typologies (private estate, HOS, public estate, walk-up, single block), and aligned with the HKMA regulatory frame.

For brokers facing a low bank valuation, for owners verifying a bank's number before closing, or for family offices monitoring a portfolio, a QPV report provides a defensible second opinion with the audit trail to back it up. The full QPV product walks through what a sample report looks like end to end.

QUESTIONS

What people ask about HK AVMs.

What is an AVM in Hong Kong?

An Automated Valuation Model (AVM) is a statistical or machine-learning system that estimates the market value of a property using comparable transactions, building characteristics, and location data. In Hong Kong, AVMs are used by banks for mortgage origination, by brokers for listing pricing, and by property owners for second-opinion valuations.

Are AVMs accurate enough for mortgage valuations?

Mature AVMs in mature markets achieve PPE10 of 80-90%. HKMA-supervised banks use AVMs as a first-pass screen, often confirmed by a chartered surveyor for loan approvals above certain LTV thresholds. AVM output should always be treated as a model estimate, not a binding valuation.

How does an AVM differ from a bank valuation?

An AVM is a model-driven point or range estimate computed from data. A bank valuation is the lender's internal valuation, which may be informed by an AVM but is also subject to risk-management adjustments. The bank valuation is what the lender uses to size the mortgage; the AVM is one of several inputs.

How does QPV compare to bank AVMs in Hong Kong?

QPV publishes its methodology, confidence ranges, and audit trail. Bank AVMs are typically internal black boxes with point estimates only. QPV is a second-opinion tool independent of any specific lender.

Disclaimer. This article is general information about Automated Valuation Models in Hong Kong. It is not financial, legal, mortgage, or tax advice. For mortgage decisions consult a licensed mortgage broker or your bank. For valuations used in legal proceedings, court matters, IRD stamp-duty matters, or sworn evidence, instruct a HKIS-registered chartered surveyor. QPV outputs are model estimates with confidence ranges, not binding valuations. The names of specific banks and tools are referenced for illustrative context only and do not imply endorsement, partnership, or accuracy of any third-party output.
NEXT

See QPV's methodology in detail.

Walk through the inputs, weights, and confidence framework, or read the AVM-vs-surveyor decision framework next.