After floods, carmakers add engine production sites rather than retreat

A factory can recover from a flood without its owner returning to business as usual. Research on the global car industry suggests that manufacturers respond by spreading some engine production across additional sites—usually within their existing international networks—rather than abandoning the disrupted plant.

Flood resilience need not mean moving a factory to higher ground. In a preliminary preprint, Michele Fioretti, Sophia Praetorius and Thierry Mayer find that flood-disrupted engine designs subsequently become more likely to be manufactured at multiple plants. The original sites show no persistent output decline, suggesting an adjustment through additional production locations rather than wholesale relocation.

The paper, marked “Preliminary and Incomplete,” also identifies a constraint: manufacturers appear to spread production mainly when they already have a relatively inexpensive place to put it. The findings do not establish that these additional sites actually rescue customers during subsequent emergencies.

Following products through production networks

The researchers combine S&P Global Mobility production data for 2000–2024 with a hand-collected record of disruptions assembled from news reports, company statements and filings. The production data cover 1,196 vehicle assembly plants, 443 engine plants and 350 transmission plants, and identify which component plants supply which assembly plants.

Rather than following broad vehicle nameplates alone, the analysis tracks specific engine designs, transmission designs and generations of car models. This distinction helps avoid mistaking the replacement of an older car generation by a newer one for duplication of the same product.

The disruption record includes both closures associated with flooding at or around a plant and stoppages at dry plants whose supplies were interrupted by floods elsewhere. The main sample contains 216 affected engine programs, 121 transmission programs and 421 vehicle model versions.

Using a staggered difference-in-differences design, the authors compare changes after each product’s first documented disruption with changes among comparable products not yet affected. Comparisons match launch cohorts and the number of production plants before the event, with control products’ largest plants in other countries. Matching the initial plant count matters because products already made at several sites were expanding differently before floods.

More sites, not a permanent retreat

Across years five through ten after a disruption, affected engine programs were 19.4 percentage points more likely than controls to be made in at least two plants. The reported standard error is 6.1 percentage points. Plants that made these programs before the event generally retained them, with no estimated lasting reduction in their output.

At the whole-plant level, the estimated flood-year loss for engine factories was about 11% of the previous year’s output relative to controls, although that estimate is imprecise. Output recovered the following year.

The added locations mostly served different customers. Of 78 observed additions after disruptions, 63 were at existing engine plants in another country. Only 5% of the added plants’ volume went to assembly plants previously supplied by the disrupted sites. These were therefore not simply duplicate delivery routes for the same cars.

The authors interpret the pattern as a potential form of insurance: a running production operation elsewhere could become useful during a prolonged interruption. But annual data cannot demonstrate that such emergency switching occurs, and the observed disruptions were generally too short to test that capability directly.

Existing technology makes adaptation easier

Among affected engine programs initially made at one plant, the long-run increase in duplication was about 36 percentage points when the owner already produced the same engine family abroad. Programs without that option showed no corresponding increase. Related engines share tooling and processes, making an additional program less costly to introduce.

Using public investment announcements to estimate capital costs, the authors price an additional production location after a flood at $26.8 million on average. This is an inferred cost, not directly observed spending on each addition. It excludes possible extra operating costs and treats some additions to existing lines as requiring no capital expenditure, so the authors describe it as a lower bound.

The pattern differs for other products. Car model versions show temporary duplication three to four years after a flood, but no lasting increase. The authors suggest that compatible plants may lack nearby buyers for those particular cars. Transmissions show no clear overall duplication response; large production scales offer a possible explanation, although transmission estimates are noisy and affected products already differed from controls before disruption.

Important uncertainties remain

The main long-run engine estimate rests on only 11 owners at year five. Disruptions in 2011 feature prominently, and some events in the original record were disputed or misclassified. A stricter event definition nevertheless produces a similar long-run estimate of roughly 17 percentage points.

Separate tests suggest more duplication where insurance may be expensive and less where flood defences are mapped nearby. Those tests use broader flood signals, insurance-market concentration rather than actual premiums, and undated defence maps. Their sensitivity to alternative specifications makes them suggestive, not firm policy evidence.

The central implication is narrower but substantial: recovered output can conceal a lasting change in production geography, and the feasibility of that adaptation depends on the network a manufacturer already has.