Bank Mergers Keep Lending to the Poor, Even When No One Is Watching
By Dr. Boris Houenou ·
The suspicion every regulator carries
Ask anyone who has sat on a banking supervision committee what they think happens when a bank is trying to get a merger approved. They will tell you the same thing. The bank suddenly discovers the poor. Loan files to underserved neighbourhoods appear. Branch openings get announced in districts nobody thought about last year. Then the approval lands and, the suspicion goes, everything quietly reverts.
That suspicion has a formal name in the American literature. The Community Reinvestment Act, enacted in 1977, requires insured banks and thrifts to meet the credit needs of their entire communities, including low- and moderate-income neighbourhoods. Because regulators use CRA ratings as a criterion when approving mergers, acquisitions and branch expansions, some observers describe the CRA as a tax on M&A. Bostic, Mehran, Paulson and Saidenberg showed in 2002 that a higher share of mortgage originations going to low- and moderate-income borrowers raises the likelihood that a bank acquires another bank the following year. Anticipation, in other words. Banks lend to look good before the file goes in.
My new paper asks the follow-up question that matters more for policy. Does the behaviour survive after the approval?
Looking one year after the deal closes
The test is simple in concept. Instead of studying lending before a merger application, I study the CRA examination conducted one year after a completed merger. At that point the merger review is done. Absent another deal in the pipeline, a bank has little reason to inflate lending to impress an examiner. Whatever you observe then is closer to what the bank actually does.
The data come from three sources: merger records from Wharton Research Data Services covering 1987 to 2001, inter-agency CRA rating files from the FFIEC covering 1990 through 2017, and bank information from the FDIC institutions database. The estimation window is locked to 1991 to 2017. The primary sample is 76,622 bank-year observations across 17,089 bank clusters. I use the inter-agency rating deliberately, because the Federal Reserve, the OCC, the FDIC and the former OTS each emphasise different things, and a single-agency rating would carry that agency’s bias.
One detail matters for reading everything that follows. The CRA rating runs from 1 to 4, where 1 is outstanding and 4 is substantial non-compliance. A negative coefficient on the merger variable therefore means better ratings for acquirers.
What the numbers say
Acquirers have roughly a 20 percent predicted probability of an Outstanding rating, against about 15 percent for non-acquirers, in the random-effects and correlated-random-effects models. The pooled ordered logit puts it at 19 versus 15. Most banks of both kinds sit at satisfactory: around 0.82 probability for non-acquirers, about 0.79 for acquirers.
The bottom of the distribution moves the same way. Acquirers carry a lower predicted probability of a needs-improvement rating, roughly 0.024 to 0.027 across the three main estimators against 0.035 for non-acquirers, and a lower probability of substantial non-compliance. The race to the bottom is run by the banks that did not merge.
The pattern holds across random-effects, correlated-random-effects, pooled ordered-logit, control-function and instrumental-variable specifications, and it survives a fully out-of-sample replication on post-2015 mergers drawn from Federal Reserve Board records. In that later cohort acquirers show a 29 to 34 percent chance of Outstanding against 15 percent for non-acquirers.
I am not claiming a clean natural experiment, and the paper says so plainly. Mergers are chosen, not assigned. The instrument I use, the bank’s year of establishment, is strong in the first stage but its exclusion restriction cannot be proven, because older banks may carry reputational, scale or compliance advantages that bear on ratings directly. The model is just-identified, so exclusion is untestable. What I rest on instead is the breadth and stability of the result across six estimation strategies and an independent later sample. A single unobserved confounder would struggle to survive all of that with the same sign.
Why the mechanism is about cost, not virtue
The theory in the paper is not a story about corporate conscience. It is a cost story.
Serving low-income borrowers is unprofitable at the margin for a standalone bank. That is precisely why a mandate is needed. But a merger that delivers economies of scale and scope in loan production and in compliance pushes the marginal cost of serving that segment down. No new facilities. Labour specialisation. Scope across loan types. The merged bank can reach a broader assessment area and a higher volume of low-income lending at the same or higher total profit.
If that is what is happening, the lending is not a strategic gesture. It is a by-product of the consolidation itself, a positive or pecuniary externality that persists because the economics changed, not because someone is being watched.
One caveat belongs in the open. The data do not show low-income lending directly. They show the examiner’s rating. The bridge between them is that lending activity carries roughly 75 percent of the rating weight, a figure documented by Litan and co-authors in their 2001 Treasury report. Higher rating implies, in expectation, higher lending to the target segment. That is a maintained assumption, not a measurement.
What African regulators should take from this
This is a paper about American banks. The reason it should interest anyone in Lagos, Abidjan or Nairobi is that the structural question is identical and largely unanswered on our side.
Banking consolidation is a live policy instrument across the continent. Capital requirements get raised, weak institutions get absorbed, and supervisors approve or block the resulting deals. In almost every one of those reviews, financial inclusion appears somewhere in the file as a stated national objective. What almost never appears is a measured, published, comparable score of how well each institution actually serves underserved customers, and a rule that makes that score binding at the moment of approval.
That is my view, not a finding of this paper. But the American evidence makes the case concrete. If inclusion performance is scored consistently, published, and used as a decision rule in merger review, and if consolidation genuinely lowers the unit cost of serving thin-file customers, then the inclusion gains do not evaporate once the deal is signed. They persist because the merged institution can now afford them.
There is a second lesson, about how you score. The American system works partly because ratings are harmonised across agencies rather than left to each supervisor’s preference. Where several bodies touch the same institution, as I read the West African picture, where the central bank, national supervisors and regional authorities all touch the same institution, a single common scale is worth more than any individual agency’s rigour.
And the framing lesson. If regulators treat inclusion conditions purely as a cost imposed on dealmakers, they will price them defensively and grant exemptions under pressure. If the evidence says inclusion lending is a durable by-product of scale rather than a temporary tax, then the welfare arithmetic of conditioning approval on it is more favourable than the tax framing implies.
The practical move
For a supervisor designing a consolidation regime, the sequence is straightforward. Define a small number of observable inclusion metrics before the merger wave starts, not during it. Publish the scores. Apply them uniformly across supervisory bodies. Then, critically, examine the institution again a year after the deal closes, when nobody has an incentive to perform.
That post-deal examination is the entire test. It is what separates a genuine change in the cost of serving the underserved from a well-timed performance. The American record suggests the change is real. Ours is worth measuring.