---
title: "BIS MaRCo@BIS: What Its AI Agent Skills Teach Central Banks"
slug: "bis-marco-ai-agent-skill-framework-central-banks"
author: "Miguel Amigot"
date: "2026-11-19 12:00:00"
category: "Premium"
sector: "financial-services"
topics: "MaRCo@BIS, BIS Working Paper 1395, Bank for International Settlements AI, agentic AI central banks, AI agent skills, macroeconometric AI agent, AI for economic research, human in the loop AI, self-hosted AI for banks, model-agnostic AI financial services"
summary: "BIS economists released MaRCo@BIS on November 12, 2026, an Apache-2.0 framework of 12 plain-text skills, 30 verified model specifications and a tested library that turns an AI coding agent into a supervised macroeconometric research assistant. Here is how it works, what it found, and where the data goes."
banner: ""
thumbnail: ""
linkedin: |
  Three economists at the Bank for International Settlements just published MaRCo@BIS, an open-source framework that turns an off-the-shelf AI coding agent into a research assistant for central bank macroeconometrics.

  It is not a new model. It is 12 plain-text workflow skills, a dictionary of 30 model specifications verified against the original papers, and a tested library of 24 estimation modules. The agent proposes, the economist confirms or overrides every consequential decision, and every choice is logged.

  The design lesson travels well beyond economics: move everything that can be fixed out of the model and into files and tested code, and gate everything that needs judgement. The authors note a model upgrade then calls for re-validation, probably without a complete rebuild.

  One line in the README matters most to any regulated institution: agent sessions send prompts and the files the agent reads to the model provider's API, so confidential data needs institutional clearance first.

  With ibl.ai you own all the code and the data, run any model, and keep agents, instructions and logs on infrastructure you control, with no per-seat pricing.

  #iblai #AgenticAI #EnterpriseAI #CentralBanking #AIAgents #FinancialServices
faqs:
  - question: "What is MaRCo@BIS?"
    answer: "MaRCo@BIS, the MAcroeconometric Research Companion, is a framework published with BIS Working Paper 1395 on November 12, 2026 by Marco Lombardi, Taejin Park and Fernando Perez-Cruz. It combines plain-text workflow skills, a verified dictionary of canonical model specifications and a tested estimation library, read by an off-the-shelf agentic coding environment."
  - question: "What licence is MaRCo@BIS released under?"
    answer: "Apache-2.0, without warranty. The paper calls it a proof of concept, and the GitHub repository states that it is not maintained and that issues or pull requests are likely to remain unresolved."
  - question: "Which AI model did the BIS use to build MaRCo@BIS?"
    answer: "The paper notes the work was based on Claude Code from Anthropic, running Claude Opus 4.8 and Fable 5, accessed May to July 2026. The authors say the framework's knowledge lives in plain-text files and code, so a model upgrade or replacement calls for re-validation, probably without a complete rebuild."
  - question: "Does MaRCo@BIS send data to an AI provider?"
    answer: "Its README says agent sessions send prompts and any file contents the agent reads, including data files, to the model provider's API, which is Anthropic's for Claude Code. It tells users to clear confidential or unpublished data with their institution's policy, including the approved deployment route and data-retention terms."
---

## The Short Answer

**BIS economists released MaRCo@BIS on November 12, 2026: an Apache-2.0 framework of 12 plain-text skills, 30 verified model specifications and a tested library that turns an AI coding agent into a supervised research assistant for central bank economists. Its lesson for regulated institutions: fix what can be fixed and gate the rest. On ibl.ai you own all the code and the data, so that pattern runs inside your perimeter.**

The release accompanies [BIS Working Paper 1395, "MaRCo@BIS: AI-agentic econometric analysis for central bankers"](https://www.bis.org/publications/working-paper-1395-marcobis-ai-agentic-econometric-analysis-central-bankers), by Marco Lombardi, Taejin Park and Fernando Perez-Cruz of the BIS Monetary and Economic Department. The code is on GitHub at [bis-med-it/marco-BIS](https://github.com/bis-med-it/marco-BIS).

Some context before the details. The BIS is [owned by 63 central banks](https://www.bis.org/about/index.htm) that together account for about 95% of world GDP, and working papers carry the authors' views, not necessarily those of the BIS.

## What did the Bank for International Settlements release with MaRCo@BIS?

MaRCo@BIS stands for the MAcroeconometric Research Companion.

According to the [working paper](https://www.bis.org/publications/working-paper-1395-marcobis-ai-agentic-econometric-analysis-central-bankers.pdf), the release holds a tested reference library of 24 estimation modules, 12 workflow skills, a dictionary of 30 canonical model specifications verified against the original papers, and a data layer with runtime connectors.

The library spans VARs, Bayesian VARs and local projections, with recursive, long-run, sign, narrative, proxy and heteroskedasticity-based identification.

Pre-collected panels include the FRED-MD and FRED-QD datasets. The repository README says the reference library is covered by more than 1,000 tests.

The 12 skills listed in the [repository](https://github.com/bis-med-it/marco-BIS) are knowledge-base, orchestrator, model-selection, model-dictionary, specification, data-retrieval, diagnostics, results-presentation, benchmarking, consistency-audit, advisor-protocol and research-log.

The package is distributed under the Apache-2.0 licence, without warranty.

## How does the BIS framework stop an AI agent from making silent decisions?

The paper's first design principle is a guided, research-led workflow.

At every consequential decision, such as the model class, the variable set, the identification scheme, the sample or a data substitution, the agent proposes a choice with its rationale and waits for the economist to confirm or override it.

A session starts in one of three modes: replicate a paper already in the dictionary (mode A), ingest a new paper and verify its specification (mode B), or design a specification for an open question (mode C).

All three run the same staged loop of specification, data, estimation, an economic sensibility check, reporting and a final disposition.

Two rules govern specification. Variables and identification are never chosen by statistical fit, which settles only tuning choices like lag order.

And specification search is disclosed: the final report counts sensibility-driven revisions and shows first-pass results next to final ones.

Every decision enters a research log with its rationale, the alternatives considered and the human confirmation. Four days later, Federal Reserve Governor Michael Barr asked the same of AI decisions in banks: that they be [explainable and replicable](/blog/fed-barr-ai-agents-financial-stability-bank-controls).

## What changed when the BIS ran the same request with and without MaRCo@BIS?

The authors ran paired sessions of two replication exercises on the same day and the same underlying model: once with the framework, once in a plain coding-agent session in an empty folder. Their Table 4 summarises the difference, condensed here.

<table style="width:100%; border-collapse:collapse; margin:1.5rem 0; font-size:0.95rem;">
  <thead>
    <tr style="background:#f5f5f0; border-bottom:2px solid #2175C5;">
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">Dimension</th>
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">With MaRCo@BIS</th>
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">Plain coding agent</th>
    </tr>
  </thead>
  <tbody>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Decisions put to the researcher</strong></td>
      <td style="padding:0.75rem; background:#f0f9ff;">Gates at the binding moments: data source, substitutions, disposition of new code</td>
      <td style="padding:0.75rem;">Deep design calls made autonomously</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Data sourcing</strong></td>
      <td style="padding:0.75rem; background:#f0f9ff;">Known sources, confirmed by the researcher</td>
      <td style="padding:0.75rem;">Web search at run time</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Estimation code</strong></td>
      <td style="padding:0.75rem; background:#f0f9ff;">Shared library backed by the test suite</td>
      <td style="padding:0.75rem;">Several hundred lines of fresh, untested code per run</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Validation</strong></td>
      <td style="padding:0.75rem; background:#f0f9ff;">Machine-checked targets from the dictionary entry</td>
      <td style="padding:0.75rem;">Self-asserted match to the paper's figures</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Diagnostics</strong></td>
      <td style="padding:0.75rem; background:#f0f9ff;">Sensibility gate plus residual diagnostics</td>
      <td style="padding:0.75rem;">None run in either session</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Record left behind</strong></td>
      <td style="padding:0.75rem; background:#f0f9ff;">Research log, run state, data registry, disposition</td>
      <td style="padding:0.75rem;">Clean code and a readme; the decision trail lives only in the chat</td>
    </tr>
  </tbody>
</table>

Both setups reproduced the headline finding. The difference was traceability. In one plain session the agent estimated the VAR with a constant term the original paper explicitly excluded, an error that leaves the headline intact and is easy to miss.

The framework did not come free. The paper reports that on the first exercise both sessions used about 1.6 million weighted tokens; on the second, the framework session used about 3.7 million against about 2.2 million without it.

The authors say two short sessions cannot settle which effect dominates.

## How well did the BIS AI agent replicate published macroeconomic models?

The authors benchmarked the agent's replications of three widely used structural VARs against the originals and ran five exercises, from a verified replication to an open design question.

The [BIS summary](https://www.bis.org/publications/working-paper-1395-marcobis-ai-agentic-econometric-analysis-central-bankers) states it reproduced published findings where the specification and data were well defined.

It also highlighted the greater uncertainty that comes with set identification and incomplete data.

In the open fiscal policy exercise, the impact multiplier stayed roughly between 0.6 and 1.1 across most lag and shrinkage checks (a one-lag version gave 0.43), while the eight-quarter cumulative multiplier swung from -0.56 to +1.04 depending on the specification.

The agent recommended headlining the impact result and reporting the rest as unidentified.

When the authors chose a Bayesian specification it had argued against, it built the run as instructed and wrote its disagreement into the output. The BIS summary of the paper concludes that AI should support, rather than replace, expert judgment.

## Where does central bank data go when an AI research agent reads it?

This is the question a risk officer should ask first, and the MaRCo@BIS [README](https://github.com/bis-med-it/marco-BIS) answers it plainly. Agent sessions send prompts and any file contents the agent reads, including data files, to the model provider's API, which for Claude Code is Anthropic's.

The README tells users to clear confidential or unpublished data with their institution's policy first, including which deployment route and data-retention terms it has approved.

An optional advisor protocol can send math questions to a second model; the README says it is meant to carry equations, never data, and calls that an instruction to the agent rather than a technical filter.

That is honest documentation, and it puts the decision where it belongs. For public FRED data the question is easy. For supervisory data, unpublished forecasts or a bank's own positions, the route the data takes is the whole question.

## Why does MaRCo@BIS store its knowledge in plain-text files instead of a model?

The paper's fifth design principle divides labour between the stochastic and the deterministic. The AI handles judgement.

Canonical specifications live in the dictionary, estimation code in the tested library, and the conduct of each stage in written instructions the agent follows.

The authors report fewer errors once tasks moved out of the model into fixed components, because errors there could then be corrected centrally.

The same structure makes the framework more general across AI models: there is no server and no bespoke model, so an upgrade or replacement calls for re-validation, probably without a complete rebuild.

When their own model was updated during development, the instructions, specifications and code stayed usable.

The paper notes the work was built on Claude Code running Claude Opus 4.8 and Fable 5 between May and July 2026. Skills that outlive the model are the same idea we covered in [agent skill catalogs as a supply chain](/blog/agent-skill-catalogs-are-a-supply-chain).

## How can a bank or central bank run a MaRCo@BIS-style agent on infrastructure it owns?

The pattern has five parts: plain-text skills, verified reference knowledge, confirmation gates, a decision log and model portability. None of the first four depends on where the model runs. What depends on hosting is the data path the README warns about, and who holds the logs.

On ibl.ai you own all the code and the data.

Institutions deploy the platform in their own cloud, on-premise or fully air-gapped, run it model-agnostic across commercial or open-weight LLMs, and keep AI agents, their instructions and their logs on infrastructure they run. It all runs with no per-seat pricing.

<table style="width:100%; border-collapse:collapse; margin:1.5rem 0; font-size:0.95rem;">
  <thead>
    <tr style="background:#f5f5f0; border-bottom:2px solid #2175C5;">
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">MaRCo@BIS principle</th>
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">The question for a regulated deployment</th>
    </tr>
  </thead>
  <tbody>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>12 plain-text skills</strong></td>
      <td style="padding:0.75rem;">Who can change an agent's instructions, and is the change versioned?</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>30 verified specifications</strong></td>
      <td style="padding:0.75rem;">Where does the reference knowledge live, and who curates it?</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Confirmation gates</strong></td>
      <td style="padding:0.75rem;">Which actions can an agent take without a named human approving them?</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Research log</strong></td>
      <td style="padding:0.75rem;">Does the decision trail stay on systems the institution controls?</td>
    </tr>
    <tr style="background:#f0f9ff; border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Model portability</strong></td>
      <td style="padding:0.75rem;">Can the institution switch models and re-validate without rebuilding?</td>
    </tr>
  </tbody>
</table>

ibl.ai is family-owned and operated from New York, NY. For banks and supervisors weighing these questions, our [financial services page](/solutions/financial-services) covers deployment options, and [Agentic OS](/product/agentic-os) is where the agents, their permissions and their logs run.

## What does MaRCo@BIS not prove about AI agents in central banking?

It is a proof of concept, in the paper's own words, and the repository states it is not maintained, with issues and pull requests likely to remain unresolved. The value is the published design and benchmark, not a supported product.

The authors list the risks themselves. A user can bypass the safeguards against specification search by starting a fresh session and reporting only the chosen result.

Automation bias remains, so the framework is meant for trained economists, and outputs required expert review that, in their exercises, materially changed the specifications.

The README adds that verification is best-effort and that many methods rest on the library's own simulated-data tests rather than independent implementations.

Read MaRCo@BIS as evidence that disciplined agent design works in a conservative institution, not as evidence that the judgement can be delegated.

## Why does owning the AI stack matter?

**ibl.ai is the agentic AI platform where you own all the code and the data. You self-host the entire stack inside your own perimeter, run it model-agnostic across any LLM and switch anytime, and pay by usage with no per-seat pricing — so you can deploy anywhere: your cloud, on-premise, GovCloud, or fully air-gapped.**

- **You own all the code and the data.** Full source code under a perpetual license, running on your infrastructure. Not API access to someone else's platform — the stack itself is yours.
- **Model-agnostic.** Run any LLM — Claude, GPT, Gemini, Llama, Command, or your own fine-tune — and switch providers without rewriting the platform.
- **No per-seat pricing.** Usage-based billing against a budget cap you set. Cost tracks what your organization actually uses, not how many people you employ.
- **Deploy anywhere.** Your cloud, your VPC, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity.

1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University.

ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.

[Explore sovereignfinancialservices.ai](https://sovereignfinancialservices.ai/)
