Figure 112. Who held NYC taxi medallion loans, and where they moved, 2000 to 2025

Read with care

• No dollar amounts. Every number is a count of medallions (fractional where a loan is shared), not loan dollars. UCC filings record no loan principal.
• Holder of record: a sale is dated by its assignment filing.
• A sale can go unfiled, so some moves are missing or late.
• Exits are mostly lapses, dated up to 5 years after payoff.
• Public (no lender) = a medallion with no open UCC loan on record at that year end: not yet financed, loan terminated (released), loan lapsed (5-year UCC expiry, the largest share), or between loans. It is a state, not a group of people; lenders + Public = 13,122 medallions every year. Medallions never named by any UCC filing (~600) are outside the data. Not proof the owner is debt-free.
• Colour by days able to operate: share of held days on the TLC Authorized list (stage 213a, 2013 onward; holders by all open loans, so its holder-years differ from the drawn stocks), relative to Public that year; cells under 30 medallion-years shown as no data. Days on the list, not driving or hours; lenders' books were not randomly assigned (Figures 99, 107).
• Independent / Fleet split: each medallion's type is fixed at its spine value (1,330 drifted toward owner-must-drive count under their spine value); moves among small lenders inside Other lenders are not split by type.
• Concentration strip: shares of lender-held medallions by lender family (each medallion once, under its latest loan); not comparable with Figure 82's HHI, which counts loans.
• Moves among the small lenders inside "Other lenders" are drawn as a dotted self-loop on that node ("N moved among small lenders"); they are not arrows to named lenders.
• 2000 to 2001: the filing record is thin (Figure 88), so new-loan counts understate lending in those years.
• Follow a group (stage 255b): groups were not randomly assigned to buyers (Figures 99, 107), so a group's path describes, it does not prove a lender effect. Cohorts count every medallion a loan names, so they run above the registry (Melrose to DePalma 2,402 against 2,031). Holder of record; TLC owner changes can lag the transfer.
• Dashed exit arrows = lien lapsed (dated 5 years after the last filing, not when the loan ended); solid = a filed termination.
• Persistence = share of the start-of-year book still with the same holder at year end; moves within "Other lenders" count as staying.
• Self-loop: arrow from a holder back to itself; width = the share (Markov share) or number (Medallions moved) of its start-of-year book that stayed, on the same scale as the outgoing arrows.
• Self-loops stay year-specific in every window, including Memory and Since 2000.
• Memory: each year's moves count half as much after every half-life (default 2 years), so recent moves are drawn strongest.
• Memory colour strength = recency of the edge's last big year (a year carrying at least 20% of its largest single-year flow); it halves every 1.5 years after that year, at any half-life.
• Memory share = decayed moves ÷ the sender's decayed start-of-year books (same half-life).
• Timeline dates come from the project registries (_timeline.py, stages 229 and 232); stage 232 wins where they differ. MRP dates and the 2023 PenFed re-filing are from the project timeline notes.
• Markov share = medallions moved ÷ sender's book at the start of the year.
• Market share denominator = medallions held by any lender at the start of the year; Public excluded. Smooth: trailing 12 months ÷ lender-held medallions 12 months earlier. Self-loops in Market share = medallions that stayed ÷ the same lender-held total.
• Rank badges and Biggest moves rank year-moves (one sender, one receiver, one calendar year) by medallions; Smooth ranks the trailing 12 months against the same all-time list.
• "Since 2000" pools the share: total moved ÷ sum of the sender's start-of-year books.
• Arrow width grows with the square root of the flow, so small flows stay visible. Read exact counts in the inspector.
• All lenders: every family that moved 50 or more medallions over 2000 to 2025; the rest sit in "Other lenders".

Source: NY DOS UCC filings via Lexis. Holders and moves from stage 251 (holder of record, latest-loan rule); networks from stage 255 (holder_network.json, holder_network_full.json); page from stage 257.