Capital Deepening and Employment Dynamics in UK Information-Intensive Services: Evidence from SVAR Analysis

Abstract

This paper documents a fundamental sectoral divergence in capital–employment relationships using UK quarterly data (2014Q1–2024Q4, N = 44). While manufacturing automation studies consistently find negative employment effects, we show that information-intensive service sectors (SIC J: Information and Communication; K: Financial and Insurance; M: Professional/Scientific/Technical) exhibit robust positive co-movement between capital formation and employment. Structural vector autoregression analysis reveals persistent positive employment responses following capital shocks, with effects peaking at 5–6 quarters and remaining significant through 10 quarters. This pattern holds across eight alternative specifications with varying lag structure, variable ordering, and subsample periods. Granger causality tests reveal bidirectional temporal relationships (capital → employment: F = 3.932, p = 0.028; employment → capital: F = 5.659, p = 0.007), indicating joint determination from anticipated demand growth rather than unidirectional technology-driven dynamics. This finding—while complicating causal interpretation—strengthens the contribution by providing honest empirical characterization of coordination mechanisms in information-intensive sectors. Our capital formation proxy measures all investment in AI-intensive sectors (buildings, equipment, conventional IT, emerging AI systems) rather than AI expenditure specifically, creating measurement ambiguity we acknowledge transparently. The sectoral focus (J+K+M sectors with 22–34% AI adoption rates exceeding the 15% economy-wide average) provides indicative evidence that patterns relate to advanced technology deployment, but measurement breadth prevents definitive AI-specific conclusions. The contribution lies not in establishing AI-specific causality—which aggregate time-series methods cannot achieve—but in documenting robust sectoral heterogeneity using methodology comparable to manufacturing displacement studies. The positive association in information-intensive services contrasts sharply with manufacturing’s negative relationship, suggesting technology–employment dynamics vary fundamentally across sectors with different task structures. Three limitations constrain interpretation: (i) recursive identification cannot definitively rule out common demand shocks, (ii) the 44-quarter sample provides limited statistical power for precise magnitude estimation, and (iii) external validity to other countries, time periods, or service sectors remains uncertain. The findings motivate sector-specific rather than economy-wide technology policy approaches, recognizing that extrapolating manufacturing evidence to service-dominated economies may systematically mischaracterize employment dynamics.