Borrowed Youth: Demographic Change and Multinational Location Advantage

Why can manufacturers with the same plant and the same foreign finalists choose different hosts? The article distinguishes an opening workforce from control over continuing international recruitment. A shared affiliate can change one group's choice. A commercial agency offering the same terms can attract several firms to the host.

Jaguar Land Rover chose Nitra over Puebla for a twenty-year investment. Under the stated hiring and tenure assumptions, a UK-and-India hiring agreement makes that choice more likely than recruiting additional workers only through India. The Bayesian comparison gives the two offers equal prior odds and models Nitra's other funded bid terms with a normal distribution centred on EUR7 million. With a EUR30 million standard deviation the posterior probability of the two-origin offer ranges from 54.5% to 60.0% across four pre-choice migration reconstructions. These are conditional probabilities across two proposed offers. JLR's hiring agreement is unobserved.

Separate country and multinational analyses quantify the probability of a positive shift in the relative investment response to arrivals. Under the reference priors these probabilities are 41% in the primary country panel and 71% in the primary firm panel. The probability of a relative gain of at least ten per cent is 58% in the primary firm comparison. On the same 299 allocations the probability of any gain is 69% under industry machinery lives and 11% under parent accounting lives. The article explains the change in direction. The four-price analysis gives less than 0.1% probability to both workforce sources attracting capital at both horizons and arrivals gaining relative weight under the stated joint prior.

Online Appendix K and the bayesian_host_choice module provide the sources, likelihoods, priors, alternative scales, numerical checks and the code for Figure 8. The anonymous manuscript, online appendix, LaTeX, editable tables and replication materials are available here. README.txt gives restoration instructions. Licensed WRDS observations require subscriber access. Public aggregate estimates and reconstruction code are included. The release has 2,079 replication files, 2,078 checksums, 9,964 words across all main PDF pages and 20 full substantive pages before references.

ChatGPT and Codex (OpenAI) were used in October 2026 for coding help, literature search, and red teaming and review. All figures were drawn with AI-assisted code from the model, demographic data and statistical estimates.