How PropGPT calculates
Every number PropGPT shows is either sourced data, with its source next to it, or a calculation described on this page. Calculations run in code, not in an AI model.
PropGPT match score
Seven factors compare a project with your Home Profile. Each factor scores 0 to 100%. The total is the weighted sum, from 0 to 100.
| Factor | Weight | Rule |
|---|---|---|
| Budget fit | 25% | Cheapest unit type with at least your bedroom count. At or below your target budget scores 100%; between target and maximum falls linearly to 50%; above your maximum scores 0%. |
| Location fit | 20% | Project in one of your preferred districts scores 100%; in a preferred region 70%; elsewhere 20%. |
| Layout fit | 15% | A unit type with at least your bedroom count and minimum floor area scores 100%; bedrooms only 50%; neither 0%. |
| Transport | 10% | Straight-line distance to the nearest MRT exit. Within your maximum scores 100%, falling linearly to 0% at twice your maximum. |
| Tenure | 10% | Matches your preference: 100%. You prefer freehold but the project is 99-year leasehold: 30%. 999-year leasehold counts as freehold. |
| Lifestyle | 10% | Share of your lifestyle preferences that the data can confirm (for example a school or park within 1 km). Preferences the data cannot check are left out. |
| Value indicator | 10% | Project median PSF compared with the median resale PSF within 1.5 km over the last 12 months. Premium of 10% or less scores 100%, falling linearly to 0% at a 40% premium. Needs at least 5 resale transactions. |
If a factor has no data, or you gave no preference for it, PropGPT shows it as unknown, leaves it out, and scales the other weights up so they add to 100%. With fewer than three known factors, PropGPT shows no score. Paid placement never changes the score.
Costs and loans
Monthly loan payment
Standard annuity formula: payment = P × r / (1 − (1 + r)−n), with r the monthly rate and n the number of months. Compare uses the starting price, a 75% loan, 3% interest and 30 years unless you change them.
Buyer's stamp duty (BSD), residential
| First S$180,000 | 1% |
| Next S$180,000 | 2% |
| Next S$640,000 | 3% |
| Next S$500,000 | 4% |
| Next S$1,500,000 | 5% |
| Remaining amount | 6% |
Source: www.iras.gov.sg, rates in force from 15 Feb 2023.
Additional buyer's stamp duty (ABSD), individuals
| Buyer | 1st home | 2nd | 3rd and later |
|---|---|---|---|
| Singapore citizen | 0% | 20% | 30% |
| Permanent resident | 5% | 30% | 35% |
| Foreigner | 60% | 60% | 60% |
Source: www.iras.gov.sg, rates in force from 27 Apr 2023.
Maximum loan under TDSR
Total debt payments may not exceed 55% of gross monthly income, tested at a 4% interest rate. PropGPT converts the remaining monthly amount into a 30-year loan amount. Banks make the final decision.
Source: www.mas.gov.sg, rates in force from 30 Sept 2022.
Acquisition cost
Price + BSD + ABSD + a legal fee estimate of S$3,000. The legal fee is a PropGPT assumption, not a quotation. Renovation comes from your profile; PropGPT does not estimate it yet.
Data freshness
Each figure records when PropGPT read it. A figure older than these limits shows “May be outdated”:
- Prices and availability: 30 days
- Sales progress: 45 days
- Fixed facts such as tenure and land size: 365 days
Source hierarchy
When two sources disagree, PropGPT keeps the value from the higher source and logs the conflict for review:
- Official government source
- Official developer source
- Licensed data provider
- Verified partner
- Public dataset
- Established secondary source
- Property portal or agent site
- User-provided
- AI-derived
A property portal or agent site is used only when no higher source publishes a value, and that value shows “from a portal or agent”. Each project is checked again every month, and a better source replaces it as soon as one is found.
Is it worth buying?
Each project page sets out what a property agent would weigh up. Every line gets a marker: in your favour, neutral, watch out, or not known yet when there is too little data. Each line gives a plain result and one sentence on what it means for you. “Show the calculation” on each line opens what it compares, the figures, how fair the comparison is (very fair, fair or a rough guide only) and what the check cannot see. The measurable lines add up to the financial score; PropGPT never makes a buy or avoid call. PropGPT never compares with HDB flats or landed homes. Comparisons use condos and apartments, or ECs for an EC, within 1.5 km (straight line), in the last 12 months, and only units in the project's size range (its smallest to largest unit, widened 20%). “Launches selling now” are the projects PropGPT lists as launched or in preview; their figures set the typical values, so the markers move with the market.
| Line | In your favour | Watch out | Not known when |
|---|---|---|---|
| Price vs similar resale nearby: this project's psf vs the median resale psf, each resale scaled to this project's lease with the SLA leasehold table, and to its building age at about 0.9% a year when both completion years are known (EdgeProp; a launch counts as new). The rate was measured from 10,793 private resales in 1,531 condos, Oct 2025 to Oct 2026, comparing condos of different ages in the same area, size and floor; where a completion year is missing, age is not adjusted and the line says so | Premium 5% percentage points or more below the usual premium (the middle of all launches selling now) | 5% percentage points or more above it | Fewer than 10 resales |
| Price vs other new launches nearby | 5% or more cheaper | 5% or more dearer | Fewer than 10 sales |
| Price vs launches in the region (ECs: all ECs) | 10% or more below the typical price | 10% or more above | Fewer than 5 other launches |
| Rental yield: median nearby rent per sq ft (URA contracts, last 4 quarters, size-range midpoint) × 12 ÷ this project's psf | 0.3% percentage points or more above the typical yield of launches | 0.3% percentage points or more below | Fewer than 10 contracts |
| Lease | Freehold or 999-year | Never: 99-year is neutral | — |
| How nearby homes' prices moved: each nearby project's median price in its first 12 months of sales (data from Oct 2021) vs its resales and sub-sales in the last 12 months, against the URA private non-landed price index over the same quarters | 60% or more of nearby projects beat the market | 40% or fewer | Fewer than 3 projects with 5 early and 5 recent sales, 2 or more years apart |
| MRT: nearest station in service (LTA exits) and nearest planned station (Master Plan, no exit yet) | Within 500 m | Over 1 km from both | No location |
| Competing homes finishing nearby: units in other unfinished projects within 1.5 km due within 2 years of this project's TOP (URA pipeline) | None | 1,000 homes or more | No TOP year |
| Land cost vs launch price: this project's psf ÷ the likely breakeven, taken as the land price per sq ft of plot ratio × 1.4 (land plus construction, financing and fees; a rule of thumb, consultants' estimates differ) | Under 1.15× breakeven | 1.3× or more | Land price not found by research |
| Developer's track record: housing projects completed in Singapore, and delays, disputes or licence actions reported by established news in the last 10 years | 5 or more completed, none reported | Any reported | Not researched yet |
Land price, developer record, nearby developments and growth areas come from PropGPT research of official and established news sources, each shown with its source and the date it was checked. The land price is a past sale and does not change; the developer record, nearby developments and growth areas are checked again every 6 months. Not judged, because they predict profit poorly: sales so far, project size, government land sales nearby and growth-area plans. Cost to buy, hold and sell shows stamp duty, seller's stamp duty if sold within 4 years, and for ECs the 5-year minimum occupation period.
PropGPT financial score
One number from 0 to 100 that adds up the lines of “Is it worth buying?” that can be measured. Each part scores from 0 (its worst end) to 100 (its best end) in a straight line between them. Parts with no data are left out and the others scaled up; the score shows how much of the weight was known. Below 40% known, there is no score. Confidence is high from 75% known, medium from 50%, otherwise low.
65 or more: strong case on the numbers. 45 to 64: fair case on the numbers. Under 45: weak case on the numbers.
| Part | Weight | Scores | For |
|---|---|---|---|
| Price vs similar homes nearby | 20 | 10 percentage points below the usual gap scores full marks, 10 above scores none | All |
| Price vs other new launches nearby | 10 | 10% cheaper scores full marks, 10% dearer none | New launches |
| Price vs the region (launches) or town (existing condos) | 10 | 15% below the typical price scores full marks, 15% above none | All |
| Price vs the developer's likely breakeven | 10 | 1.05 × breakeven or less scores full marks, 1.45 × or more none | New launches |
| Rental yield vs typical | 15 | 0.6 percentage points above the typical yield scores full marks, 0.6 below none | All |
| Value lost to the lease in the next 10 years | 10 | freehold or 999-year scores full marks; a lease losing 10% or more of its value in 10 years (SLA table) none | All |
| Nearby homes that beat the market | 10 | 80% of nearby projects beating the market scores full marks, 20% none | All |
| This condo's own prices vs the market | 10 | 15 percentage points ahead of the market scores full marks, 15 behind none | Existing condos |
| Distance to the MRT | 10 | 300 m or less scores full marks, 1.2 km or more none (nearest of in service or planned) | All |
| Competing homes finishing nearby | 5 | none scores full marks, 2,000 homes or more none | All |
| Developer's track record | 5 | ▲ full marks, ● half, ▼ none | New launches |
The score is a PropGPT calculation from public data, not a valuation, a forecast or advice. It leaves out what only you can judge: the layout, the view, schools that matter to you, how long you will stay. The parts and cut-offs were chosen by PropGPT and checked against all launches and existing condos so that each verdict holds a meaningful share; they are a reasoned judgement, not a statistical model of returns.
Existing condos
Every private condo, apartment and EC with a resale or sub-sale in URA's records since October 2021 has a page, built from those records. A launch PropGPT lists keeps its own page instead. URA records each home's floor area, not its bedrooms, so PropGPT guesses the bedroom count from size: 1 bedroom under 650 sq ft; 2 bedrooms 650–999 sq ft; 3 bedrooms 1,000–1,399 sq ft; 4 bedrooms 1,400–1,899 sq ft; 5+ bedrooms 1,900 sq ft or more. Older condos are larger for the same bedroom count, so the guess can be one off.
The price PropGPT shows for a size is the median of that size's sales in the last 12 months (or 24 months when there were fewer than two), or the lowest current asking price when that is lower. Asking prices come from EdgeProp's condo pages, a sample of about ten listings, checked about every two months with slow, polite requests; agents' details are not kept. Asking prices are what sellers hope for, not what homes sold for. The financial score for an existing condo compares it with nearby resales and with other existing condos in the same town (URA planning area), each scaled to its lease (nearby resales also to its building age when both completion years are known), and adds how its own prices moved against the market.
Towns are URA Master Plan 2019 planning areas (data.gov.sg). A home is in the town its map point falls in.
What PropGPT does not do
- It does not predict future property prices.
- It does not give financial, legal or investment advice. Speak to a licensed professional before you commit.
- It does not guess missing data. Missing values show as “Not available”.
- It does not rank projects by advertising. PropGPT has no paid placements today; any in future will be labelled.