Macro-economic modelling is becoming cheap

We do not mean that serious economic research is becoming trivial, or that model output is suddenly reliable simply because it can be produced quickly. We mean something more specific: the coding effort required to build, test and run econometric models is collapsing.

With a few well-structured prompts, a macro-economist can now assemble in days what used to require weeks or months of programming work. A purchasing power parity model, a sovereign risk classifier, a nowcasting pipeline, an early-warning indicator, a fair-value model for government bond yields: the basic modelling infrastructure of an economic research unit can increasingly be built at remarkable speed. This changes the economics of macro research.

When coding capacity was scarce, a large part of the institutional advantage came from having the people and systems required to turn ideas into working models. If coding becomes abundant, that advantage shifts. The moat moves away from simply being able to build the model and toward the parts of the value chain that make the model useful, reliable and differentiated.

We think three things become more important: infrastructure, domain expertise and curiosity.

1. Infrastructure Becomes the Foundation

Putting together the code for a PPP-based exchange rate valuation model is now the easy part. But it is only easy if two foundations already exist. The first is data infrastructure. Generative AI can help write the code, but it still needs to work with data that can be discovered, understood and retrieved reliably. For macro-economic modelling, this is not a small requirement. Macro statistics are full of quirks: different publication dates and observation dates, vintages and revisions, mixed frequencies, multi-dimensional datasets, metadata issues, seasonal adjustment choices, country-code inconsistencies and changing definitions over time.

This is why we believe there is a strong case for purpose-built macro data infrastructure such as CountryData.io. Generic data warehouses are powerful, but macro data has a structure and history that often require dedicated handling. A database that is built for macro-economic statistics, and that can be queried naturally by both humans and AI agents, becomes a critical asset.

The second foundation is production infrastructure. A quickly built model is useful only if it can survive outside the notebook. It needs automatic updates, data audit trails, version control, transparent model outputs and robust delivery channels. The output has to be available through APIs, MCP-compatible tools and user-friendly dashboards. It has to be monitored. It has to be reproducible. It has to be explainable when the model changes its mind.

In other words, cheap modelling code does not reduce the importance of infrastructure. It increases it.

2. Domain Expertise Becomes More Valuable

AI can generate a model. It cannot reliably know whether that model is the right one.

The human role shifts from writing every line of code to defining the task, supervising the modelling choices and knowing where things can go wrong. This requires domain expertise and technical knowledge at the same time.

A macro-economist using AI needs to be able to ask precise questions. What is the economic relationship we are trying to capture? Which countries should be in the sample? Which variables are actually comparable? Which transformations make sense? What should happen when data is missing? Which frequency should be used? Are we measuring a level, a flow, a ratio, a change or a deviation from trend?

It is easy to get a result from an AI-constructed model. It is much harder to know whether that result is meaningful. The risks are not abstract. Problems can enter through data ingestion, country mapping, unit standardization, frequency conversion, sample selection, outlier handling, transformations, lag structures, parameter choices or evaluation windows. A model may run successfully and still be economically wrong.

This is where experienced economists with technical fluency matter. They can spot suspicious results. They can challenge the model specification. They can ask why a coefficient has the wrong sign, why a country behaves differently from peers, why a signal appears only after a data revision, or why a model performs well in backtests but poorly in real time.

As modelling becomes cheaper, judgment becomes more valuable.

3. Creativity and Curiosity Become the Real Differentiators

If more people can build models, the harder question becomes: which models should we build?

This is where creativity and curiosity matter. The most valuable research questions are often not the consensus questions. They emerge from noticing a tension, a pattern, a contradiction or a market narrative that does not quite fit the data.

Cheap modelling makes it easier to explore these questions. We can test alternative hypotheses faster. We can build scenario tools more quickly. We can compare countries, time periods and indicators with less friction. We can ask what would need to be true for a market narrative to be wrong.

But the starting point still has to come from a curious analyst. The advantage will belong to teams that can formulate original questions, explore unconventional explanations and move faster than the consensus narrative. In macro, this matters because the market often converges around a story before the story is fully tested. Being able to build and interrogate alternative scenarios quickly is a major advantage.

AI lowers the cost of exploration. It does not replace the imagination required to decide where to explore.

Why This Is Exciting for CountryRisk.io

For us at CountryRisk.io, these are exciting times to be macro-economists.

We benefit directly from the new abundance of modelling capacity. We have a modern technology stack, no legacy infrastructure to defend and a strong interest in making macro-economic analysis more systematic, transparent and scalable.

But the most important point is simpler: we are genuinely curious. We want to understand how the macro world works. We want to know why countries diverge, why risks build, why markets reprice and why some signals matter only in certain regimes. We want to use the new tools not just to produce more models, but to ask better questions.

Macro-economic modelling is becoming cheap. That does not make macro research less valuable. It changes where the value is created. The future moat in macro research will not be the ability to produce code. It will be the infrastructure that makes models reliable, the expertise that makes them credible and the curiosity that makes them worth building in the first place.

CountryRisk.io does not provide research views. We help organisations ask better macro-economic questions and explore them faster than other market participants. Reach out to us to learn more.