Microsimulation anywhere

In the second week of July 2026, five countries acquired public, inspectable models of their own tax and benefit systems. None of them had one before. The work took about a week.

For most of this field's history, a national tax-benefit model has been the work of years — a team, a budget, a data-sharing agreement, a maintenance commitment measured in decades. That is why the models chapter 2 described lived inside finance ministries and a handful of well-funded institutes, and why most of the world has none. A researcher in Lusaka or Kampala who wanted to know whether a proposed change fell hardest on the poorest fifth of households had, until recently, to wait for a ministry to answer or to take the answer on faith. The week in July is the story of that wait collapsing — and, just as much, of what has to be true for the collapse to be worth anything.

Five countries is not a large number, and this is not a finished project. But the count was never the point. Two things had to be true at once for the week to mean anything: that a model could be built this fast, and that each one arrived carrying the evidence for whether it was right. Most of this chapter is about the second condition, because without it the first is a liability.

It started on July 8, with two countries finished on the same day. Ghana came first: its 2025 tax-and-benefit system, encoded provision by provision and tested across its full surface against GHAMOD, the Ghanaian member of SOUTHMOD — UNU-WIDER's family of tax-benefit models covering fourteen developing economies .^[1] The comparison produced thirteen findings, thirteen catalogued places where the encoded statute and the reference model disagreed, each one run down, explained, and posted to a public ledger anyone can open and read .^[2]

Uganda was finished the same day, in a single day of work, reusing the recipe Ghana had just proven. Its ledger recorded no findings at all. The encoding was, in the phrase recorded in that ledger, statutorily exact everywhere tested; the widest disagreement anywhere in the suite was four-tenths of one Ugandan shilling per year — rounding, and nothing else.

Thirteen and zero, produced the same day by the same method. Neither is the number you would reach for as a headline, and both are good news — but only once you know what a finding is, which is most of this chapter.

By the end of the week there were five. Zambia landed on the 9th. Ethiopia landed overnight. Rwanda closed it out. Two more sat at the edge of the week, and they matter more than the tidy five.

---

The rest of the week

Zambia's system was encoded in about a day and checked against MicroZAMOD, and it returned eight findings. It was the first of the five to turn up divergences on both the tax and the benefit sides of the system rather than the tax side alone — among them a definition of the pay-as-you-earn base, and a set of excise rates that had been carried a year stale. Eight findings on a first pass across an entire tax-benefit system is a low number, and the fact that they straddled both sides is itself informative: it is where the two halves of a system meet that encodings and reference models most often drift apart, and Zambia was the first case with enough surface tested to see the drift on both.

Ethiopia was done in a single overnight run and compared against ETMOD. Two findings, and both sat in the same place: the new minimum alternative tax introduced only months earlier, by Proclamation 1395/2025. A statute that new has barely been implemented by anyone, and two first-pass disagreements about how it works are close to what you would predict — both of the understandable kind, reasonable readings of a fresh law that differ, rather than arithmetic that is simply wrong.

Rwanda closed the week with five findings against RWAMOD, and it was the first lane where the comparison was executed live rather than replayed from a fixture: the encoded model and the reference model were handed the same eighty-six cases and asked to agree. The run surfaced the five findings, and once those were dispositioned the two agreed across all eighty-six.

Then the two that matter most. Tanzania is under way. Nigeria has begun — and Nigeria has no oracle. It is not one of the SOUTHMOD countries, and no other reference model or administrative ground truth is available to check the encoding against. This book has held to one rule about when a simulation may be believed: only where its chain of verification ends in something real — a reference model it matches, a statute it reproduces exactly, a number the world will later publish. Nigeria's chain does not yet terminate in any of those. The encoding can still be built, and it has been; but the claim that it is correct stays weaker than the same claim for Ghana or Uganda, and the honest thing is to say so in the same breath that reports the encoding exists. The boundary of the method is part of the method.

---

What a finding is

A finding is not a bug report and not a failure notice. It is a dispositioned divergence: a specific case where the encoded statute and the reference model produce different numbers, chased until someone can say why, and recorded together with the evidence for the explanation. "Explained" here is a checked artifact, not a shrug — an arithmetic reconciliation, a citation to the governing text, or a documented convention of the reference model. A divergence nobody can account for is not allowed to sit quietly; it is the one kind of result the process treats as unfinished.

The explanation can land on either side. Sometimes the encoding is wrong — a misread threshold, a missing phase-out, a rate applied to the wrong base — and the fix goes into the encoding. Sometimes the reference model is the one that is wrong, or out of date, or has made a defensible modeling choice the statute does not actually require; and then the finding travels the other way, written up with its arithmetic and its citation and reported upstream to UNU-WIDER, who maintain the SOUTHMOD models. The reference models are not adversaries in this. They are the instruments that make the check possible at all, and like any instrument they are improved by being read carefully against their source. The countries that never had a public model get one; the reference models get free, evidence-backed quality assurance in return. Openness runs in both directions, or it is not openness.

There is a quiet reversal of trust in that arrangement, and it is worth dwelling on. The usual assumption is that an established, decade-old reference model is the authority and a freshly generated encoding is the thing on trial. Most of the time that is right, and most findings are the encoding's to fix. But not all of them. When a divergence is chased to ground and the statute plainly says one thing while the reference model does another, the encoding becomes the instrument that caught it, and the finding is a contribution back to a model that researchers across the field rely on. Neither model is presumed correct; the statute is, and both are measured against it. That is what keeps the exercise collaborative rather than competitive — nobody is grading anybody, and everybody is grading the law.

Each country's ledger is a public GitHub discussion. Ghana's thirteen are there to read in full ;^[2] so are Uganda's zero ,^[3] Zambia's eight ,^[4] and Ethiopia's two ,^[5] with Rwanda's five alongside them citation pending. Which is what makes Uganda's empty ledger worth as much as Ghana's full one.

Zero findings is not the absence of a result. It is one.

UGAMOD turned out to be statutorily exact everywhere the suite reached, and the encoding matched it to the shilling — a fact established by evidence, not assumed by silence. A model that has been checked against an independent reference and agrees is in a different epistemic state from a model no one has checked, even when the two would print identical numbers. Uganda's zero is a compliment paid to a reference model, in evidence, and it is in its way the most reassuring of the five results: it says the method can tell the difference between right and merely unexamined.

---

The recipe

The recipe was the same in every country. Capture the governing acts from the official gazette, each provision anchored to a page in a signed source manifest, so that every rule traces back to the exact text it came from. Encode the system one instrument at a time. Put every change through the same merge-blocking gates the previous chapters described — if it does not compile, does not pass its proofs, does not clear the oracle comparison, it does not merge. Run the oracle suites first case by case, then across a whole simulated population, so that agreement holds not only on the examples someone thought to write but on the shape of the distribution. Publish the findings ledger. Report what it turned up back upstream. And once a country reaches parity, the gates ratchet: coverage can only rise and unexplained gaps can only fall, so the model cannot quietly rot back to wrong. The cost of all this, per country, was measured in days of agent time — not the years of institutional effort that a national tax-benefit model has meant for the entire prior history of the field. Across five countries with five different statutes, the recipe barely changed.

---

The moat was never only American

Chapter 2 called it the analytical moat: government tax models run on the universe of actual returns, while outside analysts get public-use files that are sampled, top-coded, and blurred, and no amount of open-source software fully closes the gap. That was told as an American story, with an American statute — Section 6103 — as its cause.

But the moat was never only American, and in most of the world it is not really a moat at all. It is a vacuum. The situation is not a gap between the government's model and a challenger's model; it is the absence of any public, inspectable model on either side. A citizen of most countries on earth cannot check what a tax change would do to a household like theirs, because no one outside a ministry has built the tool, and often no one inside one has published it. The asymmetry chapter 2 diagnosed in Washington — where at least half a dozen institutions could argue over a tax bill within days of its introduction — is, across most of the map, closer to silence.

Consider what a public model changes for the people around a single tax decision. When a finance ministry proposes to broaden a value-added tax, or to replace a fuel subsidy with a cash grant, the question everyone actually cares about — who gains, who loses, and by how much across the income distribution — has an answer that lives inside a model. Where that model is the ministry's alone, the answer arrives as an assertion, and the opposition, the press, the civil-society analyst, and the affected household can only accept it or reject it whole. Where the model is public and checkable, the same people can run the reform themselves, disagree about assumptions in the open, and be wrong in ways that others can catch. The point is not to take the tool away from the ministry. It is to end the ministry's monopoly on being able to answer at all.

What the week in July changed is the marginal cost of ending that silence. Building a national tax-benefit model went from a multi-year institutional undertaking to a job of days, and there are fourteen countries in the SOUTHMOD family alone with reference models standing ready to check the work. That much, on its own, would be worth reporting. But a model that costs days and that no one can check is not an advance over a model that costs years — it is only a faster way to be confidently wrong. What actually traveled from Ghana to Uganda to Zambia was not the encoding speed by itself. It was the discipline bolted to it: the gazette manifests, the merge gates, the oracle suites, the public ledger of everything that did not match. The verification shipped in the same box as the model. That is the entire point of the exercise, and it is the only part worth being emphatic about.

---

When the data can't travel

Rules are the easy half. A statute is public by definition; encoding it needs the text and a way to check the result, and this chapter has been about how cheap both of those have become. The harder half is the people. A tax-benefit model is only as good as its picture of who actually lives under the rules — at what incomes, in what households, drawing which benefits — and that picture comes from microdata, and microdata is where the openness usually runs out.

The asymmetry is not incidental; it is structural, and it runs the opposite way from what you might guess. A statute is written to be public — promulgated, gazetted, binding precisely because anyone can read it — so encoding it in the open takes nothing that was ever meant to be hidden. Microdata is the reverse: it is a record of actual people, protected by law and by promise, and its value to a model comes from exactly the granular, individual detail that makes it impossible to release. Law is portable because it is already public; microdata is stuck because it is already private. Any honest attempt to build models everywhere has to be designed around that fact rather than in spite of it.

So the summer's second experiment asked a harder question than the SOUTHMOD week did. Take the calibrated synthetic population that the data layer builds for the United States — populace, the commons introduced in Part II — and adapt it to a country it was never built for. Then grade it, honestly, against that country's real data. The United Kingdom was the natural test case, because it has both an open reference model and a high-quality household survey to grade against. The adapted population was run through a pre-registered evaluation harness — the scoring rules fixed and published before any of the answers were known — and scored against held-out ground truth from Britain's Family Resources Survey citation pending.

The verdict was mixed, and it was reported as mixed: a credible first pass for aggregate, earnings-centric analysis — not a substitute for native microdata on benefits, distribution tails, or subnational detail. Two specifics carried it. One was a defect found and fixed: a stale data feed had zeroed out state pensions, and the error cascaded downstream into pension-credit calculations; it was caught, traced to its source, and corrected. The other was a defect found and deliberately left open, because it cannot be papered over. Capital gains had been carried across one-to-one into a survey surface that captures almost none of them, and the honest fix is not to copy realizations from one country into another but to model asset positions and the rules under which gains are realized — a rebuild, designed but not yet done. The transferred population is good at what surveys are good at and blind where surveys are blind, and pretending otherwise would have been easy and wrong.

One smaller episode from the same work is worth naming, because it is part of why the verdict can be trusted at all. A summary statistic in an early draft of the write-up turned out to be fabricated — fluent, plausible, and unsupported by the run it claimed to describe — and it was caught at a verification gate before publication and struck. The gates in this book are aimed first at our own output, not at anyone else's. A discipline that only checks other people's work is not a discipline; it is an accusation.

Where both halves can be built, they close tightly. In Belgium, the same machinery was calibrated to twenty-one administrative targets and hit all twenty-one within 1.8 percent; and on the same population, per-record social-security contributions agreed with the EUROMOD reference model to within three eurocents per person per year citation pending. That is what a trustworthy result looks like when the reference model exists and the population can be built and run: not a promise that the model is right, but a small, checkable number that states exactly how close it came.

---

An open referee

Belgium points straight at the harder case. For many countries the microdata simply cannot leave the building. National statistical offices hold survey and administrative records under legal obligations that do not bend for a research project; you cannot download the file, and you should not want to be able to. That looks, at first, like the end of the road for open modeling in those places: you can encode their laws in the open, but you can never open their data, so how would anyone outside ever know whether a transferred population resembles the real one?

The answer is to stop trying to move the data and to move the referee instead. Give the institution that holds the microdata two things: the transferred population, and a pre-registered evaluation pack — the scoring code pinned to an exact version, the configuration frozen, and a scorecard schema that fixes in advance which numbers will be reported. The institution runs the pack inside its own walls, against its own protected records, and publishes only the scorecard: a short, low-dimensional set of results saying how well the transferred population matched the ground truth it was graded against. Because the scorecard was specified before the run, no one can fish the data for a flattering statistic after the fact; the referee's questions are fixed before it ever sees the answers. No record leaves the building. The score does. An outside analyst never sees a single citizen's data and still learns, in public and on terms agreed in advance, whether the open model can be trusted for that country.

A scorecard is a deliberately small thing — a handful of numbers, fixed in advance, that say how closely the transferred population reproduced the real one on the dimensions that matter: how the income distribution lines up, how many households the model places on each benefit, how far the tails diverge. It carries none of the underlying records and none of their risk, and it is exactly the kind of low-dimensional summary a verification chain is allowed to end in. You do not need to see a country's microdata to trust a model of it. You need a referee who has seen it, a test written down before the answers were known, and a score published where anyone can read it. That is less than full openness, and it is enough.

Open microsimulation anywhere doesn't require open microdata anywhere — it requires an open referee.

That sentence is the thesis of the chapter, and it dissolves the tension the chapter has been carrying. The encoded law can be fully open, because law is public. The microdata can stay fully closed, because it must. What sits between them — the referee — is a small, honest, pre-committed test that either party can run and everyone can read. The verification chain still terminates in something real: not a dataset anyone downloaded, but a scorecard a custodian ran and stands behind.

Notice what made every step of this possible. The rules, the data, and the prediction never had to travel together. Ghana's statutes could be encoded in the open while Ghana's household records stayed home. A population built for the United States could be shipped to the United Kingdom and graded there. A statistical office could run a referee no outsider was ever allowed to watch. Each of those moves works only because the law, the population, and the forecast are genuinely separable things — built by different methods, checked against different ground truth, and, it turns out, best kept in different hands. So far that separation has been a technical convenience. The next chapter is about what happens when an organization decides to take it seriously enough to build itself around it.

---