Population Projections under Rapid Technological Change — A Proposed Biological-Social-Cognitive (BSC) Upstream Information Framework and Evaluation Design with Reference to UN World Population Prospects 2024 / 快速技术变化条件下的人口预测
Rapid technological change may alter the timing and age pattern of childbearing through marriage formation, employment and income expectations, housing burden, assisted reproductive technology, and AI-driven changes in tasks and occupational structure. Yet the United Nations World Population Prospects 2024 (WPP 2024) already rests on a full cohort-component accounting system (CCMPP), annual single-age demographic processes, Bayesian fertility estimation, and probabilistic population projection. This paper therefore does not seek to replace WPP with a coarse “structural-variable model.” It asks a narrower, falsifiable question: while retaining a WPP-like demographic accounting backbone, can explicitly qualified biological-social-cognitive (BSC) upstream information improve out-of-sample forecasts of age-specific fertility rates (ASFR) and births? 快速技术变化可能通过婚姻形成、就业与收入预期、住房负担、辅助生殖技术以及 AI 对劳动任务与职业结构的影响,改变生育发生的时间和年龄分布。然而,联合国《世界人口展望2024》(WPP 2024)已经建立在完整的队列要素法(CCMPP)、年度单岁年龄人口过程、贝叶斯生育率估计以及概率人口预测体系之上。本文因此不试图以一个粗糙的“结构变量模型”取代 WPP,而提出一个更窄、可证伪的问题:在保留 WPP 类人口核算母体的前提下,显式加入经定义、来源、时间和模型角色审查的生物—社会—认知(BSC)候选上游信息,是否能够改善年龄别生育率(ASFR)及出生数的样本外预测。 Version 0.2 proposes two competing architectures. Model A forecasts the dynamic path of ASFR directly. Model B introduces marriage/partnership states and parity-specific events as intermediate states and constructs births through event hazards. The planned core sample comprises China, Japan, and the Republic of Korea, with Germany as a candidate external-validation setting. Evaluation uses pseudo-real-time rolling-origin backtests that strictly separate information available at each forecast origin, the ex-post evaluation reference series, and the historical WPP vintage available at that origin. B, S, and C variables enter formal comparisons only when the variable-data-artifact-experiment combination passes qualification. AI exposure, actual AI adoption, and labor-market outcomes are kept distinct; post-2023 generative-AI effects are primarily treated as conditional scenarios and stress tests rather than retroactively injected into earlier historical backtests. v0.2 提出两类竞争架构:A 型模型直接预测 ASFR 的动态路径;B 型模型以婚姻/伴侣状态及孩次事件为中间状态,并通过事件风险形成出生过程。研究拟以中国、日本和韩国为核心样本,以德国作为候选外部验证样本,采用伪实时 rolling-origin 回测,严格区分预测起点可获得信息、事后采用的评价参考序列以及对应 WPP 历史版本。B/S/C 各层只在变量—数据制品—实验组合通过资格审查后进入正式比较。AI 暴露、AI 实际采用和劳动市场结果被分别建模;2023 年以后生成式 AI 的影响主要进入条件情景和压力测试,而不被倒灌进更早历史回测。 The paper also proposes an auditable data-governance structure consisting of a Variable Concept ledger (VC), Release/Version ledger (REL), Data Series/Artifact ledger (DS), and Experiment Use ledger (EXP). Definitions, source lineage, release dates, vintages, transformation status, forecast-origin eligibility, and model permissions are recorded explicitly, with discovery evidence separated from verification evidence. Version 0.2 contains no principal empirical results; the Results section only freezes what must later be reported and under what conditions the proposed extension should be regarded as failing. The goal is not to pre-prove that all three BSC layers are “important,” but to give each layer a fair opportunity to earn empirical qualification. 本文同时提出一个可审计的数据治理结构:概念账本(VC)、发布/版本账本(REL)、数据制品账本(DS)和实验使用账本(EXP)。研究将记录定义、来源血缘、发布日期、版本、转换状态、预测起点资格和模型权限,并区分“发现证据”与“确认证据”。v0.2 不包含主要实证结果;结果章节仅冻结未来必须报告的指标和失败条件。其目标不是预先证明 BSC 三层“重要”,而是给每一层取得被公平检验的资格。
Authors
- Ai Chen (ORCID: https://orcid.org/0009-0001-8078-5762)
- Gemini
- ChatGPT
Institutions
- Asian Growth Research Institute (JP)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24267461
- Primary Topic
- Insurance, Mortality, Demography, Risk Management
- Type
- preprint