This paper evaluates the causal impact of Generative Artificial Intelligence (GenAI) adoption on productivity and systemic risk in the U.S. banking sector. Using a novel dataset linking SEC 10-Q filings to Federal Reserve regulatory data for 809 financial institutions over 2018--2025, we employ two complementary identification strategies: Dynamic Spatial Durbin Models (DSDM) to capture network spillovers and Synthetic Difference-in-Differences (SDID) for causal inference using the November 2022 ChatGPT release as an exogenous shock. Our findings reveal a striking ``Productivity Paradox'': while DSDM estimates show that AI-adopting banks are high performers ($β> 0$), the causal SDID analysis documents a significant ``Implementation Tax'' -- adopting banks experience a 428-basis-point decline in ROE as they absorb GenAI integration costs. This tax falls disproportionately on smaller institutions, with bottom-quartile banks suffering a 517-basis-point ROE decline compared to 129 basis points for larger banks, suggesting that economies of scale provide significant advantages in AI implementation. Most critically, our DSDM analysis reveals significant positive spillovers ($θ= 0.161$ for ROA, $p < 0.01$; $θ= 0.679$ for ROE, $p < 0.05$), with spillovers among large banks reaching $θ= 3.13$ for ROE, indicating that the U.S. banking system is becoming ``algorithmically coupled.'' This synchronization of AI-driven decision-making creates a new channel for systemic contagion: a technical failure in widely-adopted AI models could trigger correlated shocks across the entire financial network.
Alex: Welcome to another episode of ResearchPod. Today we're diving into some intriguing research on banks and artificial intelligence. Sam, what paper are we looking at?
Sam: This is a study by Tatsuru Kikuchi called "The Innovation Tax: Generative AI Adoption, Productivity Paradox, and Systemic Risk in the U.S. Banking Sector." It examines how U.S. banks are starting to use powerful AI tools like ChatGPT, and whether that boosts their profits or creates new problems.
Alex: Okay, so the big puzzle here is why banks jumping on AI might actually see their profits drop at first, even though they look like top performers?
Sam: That's right. The paper finds a "productivity paradox"—banks that adopt this AI show higher overall performance in one view of the data. But when you look closely at cause-and-effect, adoption leads to a sharp short-term drop in profits, around 4 percentage points lower return on equity for adopters. Return on equity is just a measure of how much profit a bank makes compared to the money invested in it—like checking if your lemonade stand earns more than the cost of lemons and cups.
Alex: And this hits smaller banks harder?
Sam: Yes, smaller banks see even bigger drops—over 5 percentage points—because they lack the scale to spread out the setup costs, like buying powerful computers or hiring experts. Larger banks handle it better. But there's a twist: when one bank adopts, it actually helps connected rivals through shared knowledge, pulling the whole network closer together.
Alex: So it's like an "implementation tax" upfront, tanking profits, but then everyone starts moving in sync?
Sam: Exactly. The study calls these benefits "spillovers." For big banks, they're substantial—meaning if AI goes wrong for one, it could ripple widely. This "algorithmic coupling" synchronizes decisions across the system, raising risks we haven't seen before.
Alex: What makes one bank's AI choice help its rivals, rather than hurt them?
Sam: Banks with similar sizes and assets—like those chasing the same customers—influence each other strongly. One way is through sharing know-how: employees switch jobs carrying AI tips, or banks swap notes at meetings and through shared suppliers. The other way is competition—if a rival gets faster with AI, they might steal business—but the study finds sharing wins out. Connected banks see a clear ROE lift from peers' adoption.
Alex: And it's stronger for the big players?
Sam: Yes. For large banks, the effect is notably bigger—a spillover three times the average—because they operate in overlapping high-stakes areas like big loans. This creates what's called strategic complementarity: one adopts, boosting peers enough that they follow suit, amplifying the network effect. It fits a pattern called the J-curve—profits dip first from setup costs, then climb as benefits and spillovers kick in.
Alex: Like new running shoes slowing marathoners at first while they adjust, but then the group paces up together?
Sam: That's a solid way to see it. Micro-studies show AI cuts writing time by about a third for simple tasks. But in banking, those savings don't yet outweigh worker shifts or system overhauls—what economists term "so-so automation," where tech nudges productivity modestly but displaces jobs without huge output jumps.
Alex: So the small bank CEO watching big rivals rave about ChatGPT in reports jumps in... pays the upfront hit... but gets pulled along by the crowd?
Sam: Precisely. The evidence points to positive selection—top banks lead—yet short-term costs dominate direct effects, with spillovers turning the tide. This synchronization heightens shared failure risks across the system.
Alex: Positive selection means top banks pick AI first... but adoption still tanks their profits short-term? How do they pull those apart?
Sam: One part of the analysis looks at the full picture across all banks—showing adopters tend to outperform because they're already strong players who choose AI wisely, plus neighbors' choices lift everyone. But to isolate just the direct hit from adopting, they treat the sudden public launch of ChatGPT in late 2022 as a surprise everyone could grab, like a free tool dropping into stores overnight. Banks that jumped on it right after get compared to those that didn't. The math builds a "what if they skipped it" path by mixing similar non-adopters' paths to match pre-shock trends perfectly. This reveals the pure cost: adopters saw return on equity drop about 4 percentage points right away, from setup expenses outweighing early gains.
Alex: Like timing runners' slowdown from new gear, ignoring who picked the best shoes to start.
Sam: Precisely. And those neighbor lifts hold up strong—for every bit a similar bank adopts, your own profits get a meaningful nudge, about two-thirds as big on return on equity. For the largest banks, it's over three times average, since they overlap most in big deals and talent pools. This pulls the sector into lockstep: one glitch hits all.
Alex: If costs dominate at first, when do gains show?
Sam: The pattern suggests a J-curve—initial dip from the "tax," then recovery as savings build and spillovers compound. Micro evidence hints AI trims routine writing by a third, but banking's rules and shifts slow the full payoff. The network sync raises a caution: shared tech flaws could cascade widely.
Alex: And those spillovers create strategic complementarity? Like one bank's move makes it smarter for others to follow?
Sam: Yes. When a connected bank adopts AI, it lifts the adopter's peers' profits through shared tips on setup or tools—knowledge diffusion beats any rivalry. The math shows this clearly: for every unit a similar bank gains from their own adoption, neighbors get about two-thirds as much on return on equity.
Alex: How do they spot these connections in the data?
Sam: They pull bank details from Federal Reserve reports on assets and loans, plus SEC filings where executives discuss plans. AI adoption gets flagged by scanning those public reports for words like "ChatGPT" or "large language model"—simple keyword counts in quarterly updates. Weights link banks by matching sizes or locations, revealing how one's choices echo.
Alex: So public filings act like a trail of who's jumping in. That frontier selection plus costs and these links explain why top banks lead but still pay upfront.
Sam: Precisely. It extends past studies on banking tech waves—like early computers for loans or fintech apps—showing AI follows the same laggy path, but networks amplify both gains and risks. The study cautions this sync could spread errors system-wide if models falter together.
Alex: How exactly does one bank's AI setup help its neighbors—through shared tips or something more concrete?
Sam: Three main channels stand out. First, early adopters build standard ways to connect AI tools to bank systems—like shared recipes for plugging in software—that make it cheaper and faster for others to follow. Second, as more banks train staff, a bigger pool of experts forms, so hiring gets easier without starting from scratch. Third, suppliers improve their AI products for banks as demand grows, dropping prices and fixing bugs quicker. These turn individual efforts into group progress.
Alex: Like trailblazers clearing a path that everyone else uses. And these spillovers outweigh the direct gains by a lot?
Sam: Yes, the analysis finds that indirect effects from neighbors account for over 80 percent of the overall impact on profits. Direct benefits from your own adoption make up the rest—showing this is mostly a team effort, not solo wins.
Alex: Earlier you noted small banks pay a steeper upfront cost—does the data show why?
Sam: The causal estimates confirm it: smaller banks see profit drops over five times deeper relative to their size, around 5 percentage points on return on equity, versus just over one for larger ones. Big banks spread fixed costs like expert hires across huge operations and have deeper pockets to weather the switch. Smaller ones feel every expense more acutely, with less room to maneuver.
Alex: And that J-curve—does the timing show the dip before the turn?
Sam: Event study plots confirm the shape: profits slide in quarters before official adoption as banks prep infrastructure, hitting a low just prior—like gearing up wears you out first. Then at adoption, there's a sharp snap back, marking recovery start. Pre-trends match a classic lag pattern in tech rollouts.
Alex: So selection picks strong starters, treatment hits costs, spillovers dominate—and it all reconciles to that network-driven J-curve.
Sam: Exactly. The full picture: frontier banks self-select into AI, showing positive links in broad data; causal math isolates the setup tax; spillovers explain most gains and sync risks. It's a measured view of tech's uneven path in a connected sector.
Alex: Pulling it all together, we've got this clear picture: strong banks lead the charge into AI, pay a real upfront cost that dips profits, but network effects from peers drive most of the gains—and pull the system into tighter sync, with bigger risks for everyone.
Sam: Yes. The study cautions regulators might need new checks: stress tests for shared AI failures, rules on vendor reliance to avoid monocultures, or help for small banks like shared tech pools to ease entry. The short sample here catches only the early dip—no full recovery yet—and filings might miss quiet adoptions.
Alex: That's a clear, grounded look at generative AI in U.S. banking. Thanks, Sam. And thanks for listening.