Abstract

Understanding how innovation-intensive activities evolve across space is a central challenge for regional statistics and place-based economic policy. While national and sectoral employment forecasts are widely available, they are typically produced at coarse spatial scales and rely on static industrial classifications, limiting their usefulness for analysing how emerging activities translate into local labour market outcomes. For instance, using static industrial classifications, it is not possible to find companies which engage in such new activities like artificial intelligence, quantum computing or biopharma. This paper presents a new statistical framework that links real-time firm-level sectoral classifications of firms with macro-consistent forecasting models to generate granular, forward-looking employment, turnover and GVA estimates for cities.
The framework combines complementary strengths from Data City and Oxford Economics. Data City provides real-time sectoral classifications derived from large-scale web data, administrative sources and machine-learning techniques, enabling firms to be mapped to multiple activities rather than a single industrial code. Employment is disaggregated within firms to full address site level using Lightcast’s information on the location of employees present on LinkedIN.
To capture short-term labour market dynamics, the framework incorporates web-scraped firm announcements on planned layoffs and major hiring rounds. These signals are identified using AI-assisted text classification applied to company websites, press releases and news sources.
Firm-level employment forecasting is conducted using an econometric time-series approach. For each firm, employment is modelled within an ARIMAX framework that includes lagged turnover, past employment, online job postings, and sector-specific controls reflecting broader industry conditions.
To ensure aggregate consistency, firm-level forecasts are aligned with Oxford Economics’ Global Industry Model. This model is a globally integrated system of macroeconomic and industry-specific econometric models used to forecast economic trends, analyse scenarios, and assess impacts across countries, regions and sectors. National sub-sectoral industry employment forecasts from the Industry Model define aggregate growth paths, which are matched to firm-level projections using econometric allocation rules. This guarantees that granular spatial outputs remain fully consistent with the macroeconomic outlook while preserving heterogeneity across firms and locations. Spatial allocation is based on observed firm footprints, implicitly assuming incremental growth at existing sites rather than large-scale relocation.
The framework is applied to the United Kingdom’s Modern Industrial Strategy “frontier sectors”, generating city-level employment, turnover and GVA forecasts to 2030. Results show that frontier employment is highly concentrated in a small number of large urban areas, yet forecast growth is more widely distributed and rarely driven by a single sector. Cities with diverse frontier ecosystems—spanning digital technologies, life sciences, creative industries, advanced manufacturing and clean energy—tend to exhibit stronger and more resilient growth trajectories. Case studies of Manchester and Cambridge illustrate contrasting development paths shaped by differences in scale, diversity and research intensity.
From a regional statistics perspective, the paper demonstrates how real-time sectoral classification, AI-enabled webscraping and firm-level econometric forecasting can be combined with macro-consistent industry models to bridge the gap between national projections and local outcomes.