Korean Public Auction Intelligence
Market research and product specification for a Korean public auction scoring tool targeting KAMCO's 온비드 platform. 86,875 unsold listings in 2024 — 2× the 2023 count. Primary signal: 유찰횟수 (failed auction rounds), which depresses reserve prices in a structured, predictable pattern. All four scoring fields confirmed in the KAMCO API. One API test away from build decision.
Overview
KAMCO's 온비드 platform lists tens of thousands of public auction properties — 압류재산, 국유재산, 체납처분재산 — but provides no analytical scoring to help investors identify undervalued listings. The primary investment signal — 유찰횟수, the number of times an auction has failed — is present in the API response but never surfaced in a structured comparative format. Each failed round reduces the reserve price by a statutory percentage, creating a predictable price-discovery mechanism that systematic scoring can exploit. This project maps the opportunity: market size, buyer evidence, API structure, competitive landscape, and scoring model design.
Problem
온비드 listed 86,875 unsold properties in 2024 — double the 2023 count — representing a structurally growing inventory of assets clearing at below-market prices. Investors who identify high-유찰횟수 properties in target categories (압류재산 주거용) before they are widely noticed can acquire at prices reflecting multiple reserve-price reductions. No private platform provides systematic 유찰횟수-based comparative scoring. The closest competitor (땅집고옥션) claims AI 공매 analysis but depth review reveals it is a chatbot layer over a filtered listing view with no independent 유찰횟수 scoring methodology.
Constraints
- The KAMCO API field name for failed auction rounds was documented incorrectly in secondary sources and community forums — PBCT_FAIL_CNT does not exist; the correct field is USCBD_CNT. Required verification from PublicDataReader open-source client source code.
- Python User-Agent spoofing may be required to access the KAMCO API programmatically — the government endpoint's behavior under standard Python request headers is the remaining unknown
- KAMCO signed a data-sharing MOU with 9 unnamed private platforms in November 2024 — one or more may build a 유찰횟수 scoring feature before this project reaches market
- Retail investor WTP (₩29–59K/month by analogy to comparable tools) must be confirmed by direct interviews before committing to a consumer-facing pricing model
Approach
Three-round adversarial research, including a follow-through verification round. Competitive landscape first: identified 땅집고옥션 as the most credible competitor and reclassified it from High to Low-Medium threat after confirming its 공매 feature is a chatbot layer without independent scoring methodology. Field names verified from PublicDataReader open-source client source — the only reliable source after API documentation proved incorrect. All four scoring fields confirmed present in the list endpoint (USCBD_CNT, 최저매각가, 물건종류, 소재지) — no per-listing detail endpoint fan-out required for v0. API test script written and ready to run.
Key Decisions
USCBD_CNT (유찰횟수) as the primary scoring signal over price deviation from appraisal value
Price deviation from appraisal value is the standard signal for undervaluation — but the appraisal value itself may be stale or inaccurate. 유찰횟수 is a cleaner signal: each failed round mechanically reduces the reserve price by a statutory percentage, creating a price-discovery process that is directly observable and requires no appraisal assumption. A property at 3 유찰 has had three rounds of market clearing fail — the reserve price has been cut three times and buyers have still not appeared. That is structurally informative.
- Appraisal-to-reserve deviation — depends on appraisal quality, which varies; less clean than counting failed rounds
- Days on market — not a KAMCO API field; unavailable without scraping
List-endpoint-only architecture for v0
All four fields needed for the v0 scoring model (유찰횟수, 최저매각가, 물건종류, 소재지) are present in the KAMCO list endpoint response. No per-listing detail endpoint fan-out is required. Building a fan-out architecture would multiply API calls, hit rate limits faster, and add complexity with no v0 benefit.
- Detail-endpoint fan-out for all listings — 86,875 detail calls for a full inventory scan; rate limit risk and unnecessary for v0
- Scraping as fallback — ToS risk; avoided by using the official API
Tech Stack
- Python 3.11
- KAMCO 온비드 OpenAPI (USCBD_CNT field)
- PublicDataReader (field name verification source)
- pandas
Result & Impact
- 86,875Market size (2024 unsold listings)
- 2× 2023Year-on-year growth
- 4 / 4Scoring fields confirmed in API
- Low-Medium (땅집고옥션 reclassified)Competitor threat level
The market size, buyer evidence, API structure, field name correction, and competitor assessment are complete. One remaining human action — registering a data.go.kr API key and running the test script to confirm Python User-Agent bypass and USCBD_CNT field presence in live responses — separates this from a build decision. The test takes 1–2 hours. The API test script is written and in the repository.
Learnings
- Government API documentation describing field names is unreliable. PBCT_FAIL_CNT appearing in documentation and community forums does not mean it is the correct field — the actual field name (USCBD_CNT) was only confirmed by reading the PublicDataReader open-source client source code. Primary source verification for API field names means running the API, not reading the docs.
- Reclassifying a competitor from High to Low-Medium threat based on depth research (not surface-level claim reading) is the right move, but the reclassification should be documented with the specific evidence so it can be revisited if the competitor ships a genuine feature.
- Market size (86,875 unsold listings, 2× growth) is a compelling entry signal, but the KAMCO MOU with 9 unnamed platforms in Nov 2024 is the competitive risk that warrants a watch cadence — not an immediate action, but a quarterly check.