How the Merger Finder works

A plain, complete description of the model behind the Merger Finder. It names no credit union. As of 2025Q4.

What this is — and what it is not

The Merger Finder scores how closely a credit union’s finances resemble the historical profile of a credit union that acquires another, and (separately) the profile of one that becomes a merger / succession partner. It is a propensity — a ranking against a historical pattern — not a prediction that any particular credit union will merge, and not a statement about its safety, soundness, or regulatory standing. About 3% of credit unions merge in a given year, so even a top-ranked credit union most likely will not.

Data

Two public sources, both used exactly as reported: the NCUA 5300 Call Report (each credit union’s quarterly financials) and the NCUA Merger Activity & Insurance Report (every completed merger, with the continuing and merging credit unions). Nothing is estimated or blended; every input is a regulator-reported number.

The model (transparent by design — no black box)

A logistic-regression scorecard on the features below, each read from a credit union’s most recent complete fiscal year. Features are trimmed at the 1st/99th percentile (so one outlier can’t distort the scale) and standardized. Each feature’s effect is a fixed, disclosed weight; a credit union’s score is just the sum of its features’ contributions, which is why every result on the Finder shows its own factor breakdown.

In plain terms: acquirers tend to be larger, have prior acquisitions, and run efficiently; succession candidates tend to be smaller, slower-growing, and thinner on earnings. Direction and relative weight are shown; the exact coefficients live in the code (cu411/mergers.py).

How it was validated (before any score was shown)

The model was trained only on 2019–2023 mergers and tested out-of-time on mergers it had never seen (2024–2025). It is a ranking tool, so we measure ranking quality, not false precision:

“Top-decile lift 3.6×” means the 10% of credit unions the model rated highest went on to do roughly 3.6 times as many deals as an average group — a useful ranking, never a certainty about any one credit union. We show a fit band (weak / moderate / strong), never a per-credit-union percentage.

Bank targets — a separate, rules-based screen

The Bank targets tab is different, and we flag it plainly. There is no public, labeled set of credit-union-buys-bank deals to train and validate on (those ~20-a-year transactions are compiled by hand from press coverage, not a downloadable file), so that ranking is not a backtested model like the two above. It is a disclosed, rules-based attractiveness score: a fixed weighting of deposit franchise (30%), size fit (20%), capital (20%), credit quality (15%), and profitability (15%), each scored from public FDIC Call Report data for community banks (the relative factors as a percentile among community banks; size fit as an absolute target range). It is a screening heuristic to focus a buyer’s homework — never a prediction, and never a claim that a bank is for sale.

Fairness & privacy guardrails

• Mission-designated credit unions are exempt from the succession screen: an NCUA low-income or CDFI-certified credit union is small-by-design in service of its mission, so flagging it as a “candidate” would be wrong. They are removed from that ranking entirely.
• Nothing about a named credit union’s merger likelihood is public. The public pages show only completed mergers (the merger tracker) and a factual “merged into X” note on a merged credit union’s profile. The forward-looking screens are members-only and gated.
• The succession ranking is held behind a legal review before it names any credit union.
• The model is refreshed each year as new Call Report and merger data publish.