Research & Methodology
Valuation Methodology
LuxMetrix produces fair market value estimates for luxury assets. This document explains exactly how — no black boxes.
01Data Sources
We collect pricing data from two types of sources, each offering a fundamentally different signal about what an asset is worth.
Marketplace Asks
Supply SignalActive listings on Chrono24 — what sellers are currently asking. These reflect supply-side expectations but may be aspirational.
Auction Sold Prices
Demand SignalCompleted sales at Phillips and Sotheby's — what buyers actually paid at arm's length.
02Ask vs. Sold Pricing
Asking prices tell you what sellers hope to get. Sold prices tell you what the market actually paid. Both are useful, but they mean different things.
A watch listed at $35,000 that sells at auction for $28,000 tells a more complete story than either number alone. We use both, but weight sold prices more heavily because they represent real transactions.
03Normalization
Raw data from each source is cleaned and standardized before entering the valuation model.
Reference Matching
Each record is scored for how confidently it matches the target asset. Only exact and high-confidence matches are used.
Currency Conversion
All prices are converted to USD at scrape-time exchange rates. We store both original and converted amounts.
Deduplication
Duplicate records from the same source are identified by listing ID, URL, or title hash and merged.
Condition Mapping
Free-text descriptions are mapped to a controlled vocabulary: new, unworn, excellent, very good, good, fair.
04Outlier Filtering
Before computing medians, we remove outlier prices using the IQR (Interquartile Range) method — a standard statistical technique that's robust against extreme values.
Q1 = 25th percentile
Q3 = 75th percentile
IQR = Q3 − Q1
Keep prices within [ Q1 − 1.5 × IQR , Q3 + 1.5 × IQR ]If fewer than 4 data points are available, we skip filtering entirely to avoid discarding valid observations from thin markets.
05Valuation Formula
After filtering, we compute the median price for each group and blend them with fixed weights:
fair_value = (0.60 × sold_median) + (0.40 × ask_median)Auction results receive 60% weight because they represent completed transactions between willing participants. If no auction data is available, we use the ask median alone.
Value Range
The low/high range corresponds to the 25th and 75th percentiles of all retained comps — the middle 50% of the market. Intuitive, robust, and easy to explain.
06Confidence Scoring
Each valuation receives a 0–100 confidence score based on five factors. The score is rule-based and deterministic — no machine learning.
| Factor | Max Points | What It Measures |
|---|---|---|
| Comp Volume | 30 | Total number of comparable prices |
| Sold Comp Bonus | 20 | Having actual transaction data |
| Source Diversity | 15 | Data from multiple independent sources |
| Recency | 20 | Freshness of the most recent data |
| Price Spread | 15 | Market consensus (tighter = better) |
High
70–100
Medium
40–69
Low
0–39
07Update Cadence
Scrapers run daily. Each run collects the latest marketplace listings and any newly published auction results. A new valuation snapshot is computed after each cycle. Trends are calculated by comparing the current snapshot to snapshots from 30 and 90 days prior.
08Transparency Disclaimer
LuxMetrix valuations are estimates, not appraisals. They are based on publicly available market data and a deterministic formula. We do not account for individual item condition, provenance, or variant-specific premiums in v1. Actual transaction prices may differ. This is not financial advice.