How Many Actually Go Home — and Why

Start with the honest answer, because the honesty is the point of the piece. Nobody knows. There is no register of Western retirees who moved to Thailand or the Philippines and then left, because no institution is built to record it. The number that would most clarify the entire decision is the one number the entire apparatus is structured not to produce. Any source that gives you a clean percentage is either guessing or quoting a study of a different population in a different decade.

That is not a reason to say nothing. It is a reason to triangulate carefully, label every inference as an inference, and refuse to launder a soft number into a hard one. What follows is the silent cohort bounded from three independent directions. None of the three is “the return rate.” All three point the same way, and the way they point is not the zero the brochure implies. (A companion piece takes these bounded inputs and builds them into an explicit cohort-decay model — the stay-curve, with its scenario bands stated.)

Why the number does not exist

The data on living abroad is collected in two shapes, and neither is a return.

The first is stock. The US Social Security Administration publishes how many beneficiaries receive payments in each country. But a beneficiary who gives up and goes home does not generate a “return” record. They change a deposit address, and the count quietly decrements with no cause attached. Stock is also not what it looks like: the Philippines figure is heavily inflated by Filipino-American returnees and their dependents, not Western retirement migrants, so even the stock cannot be read as “Westerners who retired here.” The number that looks relevant is measuring someone else.

The second shape is deaths. Consular services count the people who die abroad. They do not count the people who pack up and leave. So the institutional record has exactly two states for a retiree abroad: present, or dead. The third state, went home, the one that matters most to someone deciding, falls through the gap between the two datasets and is recorded nowhere. The absence is not a footnote to apologise for before getting to the real data. It is the finding. The decision is sold against a number its own evidence system is designed not to collect.

The one hard cohort figure, and what it is not

The cleanest measurement available is not from Southeast Asia. A 2025 Population Studies follow-up of Dutch pensioners living abroad tracked intention against behaviour over three years. Fewer than 5% said they intended to return. About 9% did. Hold both halves of that, because the gap between them is the more useful number than either figure alone. Roughly twice as many people went back as ever planned to. Return is not, mostly, a plan. It is a thing that happens to people who did not plan it, which is exactly the category the relocation pitch assures you that you are not in.

Nine per cent over three years is not a Thailand number and must not be passed off as one. The Dutch retire into a different welfare and healthcare backstop, frequently inside the EU uprating and reciprocal-care zone, with a softer floor under failure than a Briton in Chiang Mai has. If anything that makes 9% a low analogue for the SE-Asia case, not a high one. And it is three years. Retirement is not three years. Whatever the annual hazard of return is, it does not switch off at year three; it compounds across a twenty-five-year horizon, and a single-digit triennial rate becomes a large fraction of any starting cohort over the length of an actual retirement.

A separate, narrower measurement points the same way. A qualitative study of older Japanese migrants in greater New York put eventual return of that cohort near 30%. Again: not transferable as a point rate, different people, different place. Cited for one reason only. Every honest measurement of older-migrant return that exists lands somewhere between “meaningful minority” and “roughly a third,” and not one lands near zero, which is the figure the arrival videos implicitly assert by never mentioning it.

Why the peak is the retiree

There is a structural result in the migration literature worth stating plainly because it inverts the pitch. Return-migration probability is not flat across a life. It is maximised at or near retirement, and the link is strongest for people who moved close to retirement age. Most international retirement migrants move between about 65 and 75. So the moment the move is sold as final is the moment the data says reversal is most likely. The brochure markets permanence to precisely the cohort whose measured behaviour is least permanent.

The shape of that reversal is not a single hump. It is two, and they are different events with different victims.

The first peak is early. Years one to three. The novelty has gone, the structural problems it was supposed to fix did not move, and the person still has the capital, the health, and the home-country footing to reverse the decision relatively cleanly. This is the survivable exit. It is also the one the persona channels quietly absorb as “it wasn’t for them,” as if fit were the only variable.

The second peak is late, and it is not a decision. It is an evacuation. It arrives when the money runs out, or the cover lapses, or the spouse who was the entire support plan dies first, and going home stops being a choice and becomes the only remaining option exercised under duress, often with the assets that would have funded it already spent. The drawdown model names the year that arrives. The frozen-pension arithmetic names one of the mechanisms that brings it forward. This piece names what it looks like in the aggregate: a person in their late seventies or eighties leaving a country that was chosen on the assumption this would not happen.

The cohort that never went home, because they died there

There is one part of this that is not soft, and it is the darkest part, so it is stated without ornament. A large share of the people who do not appear in any return statistic are absent for a reason no brochure will show: they did not leave.

The FCDO records several hundred British deaths in Thailand each year, with 597 new death cases in 2022 and 430 in the first nine and a half months of 2023, and the Philippines reported as a country where British nationals are especially likely to die, attributed to a large elderly British expat population. DFAT handled 2,054 Australian overseas-death cases in 2024-25, up 7% year on year, with Thailand the single top destination for Australian consular cases and the Philippines third, and its own reporting attributes the rise to more older Australians retiring in both countries. These are two nationalities. The pattern repeats across the others.

Read what that is, in the frame of this piece. It is the exit no dataset files as an exit. “Did they go home?” has a third answer the question is not built to return, and the deaths-abroad totals are its lower bound. They are not a return rate and must not be reported as one. They are the floor under it: the people removed from the visible cohort not by repatriation but by mortality, in numbers that are not marginal and that rise mechanically as the retired population ages in place. Every one of them was, at some earlier point, a data point in someone’s “most people who move here stay” comfort statistic.

The cause breakdown

The literature that exists separates returns into two kinds, and the split matters more than the headline rate. There is the chosen return and the forced one. The Dutch follow-up gives the cleanest evidence that the second dominates: when twice as many people leave as ever intended to, most return is not the exercise of a preference. It is a response to an event. Group the events and you get four, in rough order of how late and how irreversible they arrive.

Financial failure is the largest and the most predictable, because it is arithmetic rather than chance. It is the drawdown model reaching its zero year and the frozen-pension arithmetic bringing that year forward. This return is rarely filed by the person as “I went broke.” It is filed as “we decided it was time to be near family,” which is true and is also the financial cause wearing a softer label. The euphemism is part of why the cohort stays invisible: the people in it do not describe themselves as a failure statistic, and the channels have no incentive to.

Health is the second, and it bifurcates. One branch is the person who goes home while they still can, because a diagnosis made the calculation about local cost and uninsurability concrete. The other branch does not go home at all, and is counted in the deaths-abroad figures above rather than in any return. Which branch a person lands in is largely decided by timing and by whether cover survived the age it was most needed, not by character.

Bereavement is the third and the most structurally brutal, because for a large share of Western men in the region the support plan was a younger local spouse, and the plan has no redundancy. When that person dies or leaves, the practical infrastructure of daily life collapses at the same moment as the emotional one, and the surviving partner returns not to a home they kept but to whatever is left of one, often a decade older than when they last navigated it.

Family pull-back is the fourth and the quietest: an adult child’s decision as much as the retiree’s, a slow gravitational return as grandchildren arrive or a sibling fails, frequently indistinguishable in the data from the financial return it often accompanies. It is the one return that is sometimes genuinely chosen and genuinely good, and it is also the one most often cited to make all the others sound like it.

None of these is the return rate. They are its anatomy. And every one of them clusters at the same end of the timeline: late, at the age the move was sold as final, exercised under an event rather than a preference.

The triangulation

No single number. Three vectors, each labelled, converging.

Vector	What it establishes	What it is not	Bend
Intention vs behaviour	About 9% of Dutch pensioners abroad returned within three years, against fewer than 5% who intended to.	Not a Thailand rate; a softer welfare backstop; triennial, so it compounds over a real retirement.	Understates SE Asia
The retirement peak	Return probability is maximised at the very age the move is sold as permanent.	A direction, not a magnitude.	Peak = the sold age
The deaths-abroad floor	Hundreds of Western deaths a year in Thailand alone, rising as the cohort ages.	Not a return rate; a lower bound on the exit recorded only as mortality.	A rising floor

Three different populations or proxies. None transferable as a clean percentage. All three bend the same way, and none lands near the zero the arrival content implies.

Source: Population Studies 2025 (Dutch); IZA J Dev Migration 2:20; FCDO (Hansard 2024) + DFAT Consular State of Play 2024-25 · checked 2026-05

Each is a different population or a different proxy. Not one is transferable to “Western retirees in Thailand and the Philippines” as a clean percentage, and this piece will not pretend otherwise. What they jointly establish is not a figure. It is a refutation. The implied claim of the entire arrival-content economy is that the silent cohort is negligible, that essentially everyone who moves stays and thrives, because that is what you see. Three independent, conservatively-read data directions say the silent cohort is large, that it leaves or dies disproportionately at the age the move was sold as final, and that the reason you cannot see it is not that it is small. It is that it is, by construction, invisible to the only sources you were shown.

Why you were never shown it

The mechanism is not a conspiracy. It is a filter, and it is measurable. Every visible source on this decision is generated by people who are currently there and currently coping. The cost-of-living aggregator entry is updated by the retiree it is still working for. The channel is monetised on the arrival and goes quiet on the fifth year. The Facebook group’s active members are, definitionally, the ones still in the group. The person who ran out and went home stops contributing. The person who died there never edits the spreadsheet again. So the median you are shown is computed only over survivors, sampled at the front of the curve, with every adverse outcome silently removed from the denominator before you ever ran the query. Paradise is a place you visit; the data about it is written almost entirely by people for whom the bill has not yet arrived.

That is the whole of it. The question “how many actually go home” has no clean answer, and the absence of one is not neutral. It is the single most load-bearing piece of survivorship bias in the entire decision, and it runs in exactly one direction: toward making the move look safer than the measurable behaviour of people who made it.

Contest the model

The credibility here is the uncertainty, not in spite of it. Every input and the direction it bends.

• Dutch 9%/3yr. Source: Population Studies (2025). Different welfare backstop, almost certainly understates the SE-Asia hazard; triennial, compounds over a full retirement. Pulls the true figure up, not down.
• ~30% older-Japanese-NY. Source: PMC qualitative (2023). Narrow, qualitative, not SE Asia. Cited for convergence, not magnitude.
• Retirement-peak return. Source: IZA J Dev Migration 2:20. A direction, not a rate. Robust across the literature.
• Deaths-abroad floor. Source: FCDO (Hansard, 2024); DFAT Consular State of Play (2024-25). A lower bound on the unrecorded exit, not a return rate. Rises mechanically with cohort age.
• The data gap. Stock and deaths are recorded; return is not, anywhere. This is the load-bearing assumption and it is not an assumption; it is the structure of the available data. If a clean return register existed this piece would cite it instead. None does.

Rebuild it with better data the day better data exists. Until then the honest statement is the bounded one, and the bounded one already contradicts the brochure. The figure you were implicitly sold was zero. Every conservative reading of every real proxy says it is not.

---

This piece cites deaths-abroad data and is analytical, not advice or counsel. If you or someone you know is struggling, contact a local crisis line or, internationally, findahelpline.com. Figures are sourced and dated to 2026 and label their own populations and uncertainty; they are bounds and proxies, not a measured return rate, and should be read as such.