Counting firms' internal data as capital raises measured IT-capital's contribution to U.S. growth by about one-third over 2002–2024, though the boost is uneven across industries.

Main Finding

Capitalizing own-account data and databases (OAD) — as recommended in SNA 2025 — materially changes measured capital composition and growth in the U.S. (2002–2024). Including OAD raises the share of intangible/ICT capital, increases measured capital-input growth (especially in data‑intensive industries), and produces a small net increase in measured GDP growth (≈8.2 basis points on average) while reducing measured TFP growth (≈2.1 basis points) over 2002–2024. The effect is strongly heterogeneous across industries.

Key Points

• What was added: Own-account data and databases (OAD) are treated as a new intangible asset class (digital recordings organized for production use >1 year), valued by a sum-of-costs approach (excluding analysis/use activities and excluding purchased data and R&D).
• Magnitude and composition effects:
  • OAD averaged a 7.85% share of capital services (2002–2024, nominal shares).
  • Intellectual property products (IPPs) share rises from 17% to 23% after OAD capitalization (∆ ≈ +6.4 percentage points).
  • If OAD is counted as part of ICT, ICT capital share rises from 11.58% to 18.33% (∆ ≈ +6.75 ppts).
  • Including OAD raises the contribution of IT‑related capital assets to GDP growth by about one‑third over 2002–2024.
• Growth and productivity summary:
  • Average GDP growth is ~8.2 basis points faster when OAD is capitalized.
  • This is driven by ~12 bps faster measured capital growth (notably intangibles and ICT) and partially offset by ~2.1 bps lower observed TFP growth.
• Industry heterogeneity:
  • Largest upward revisions to capital and value‑added growth: management of companies and enterprises; professional, scientific, and technical services (notably computer systems design); finance and insurance (e.g., securities and investments); and educational services.
  • Smaller or negative revisions occur in some administrative/waste sectors and some manufacturing subsectors — reflecting adoption timing and early‑adopter dynamics.
• Government treatment: for general government, output remains valued at cost; capitalization increases consumption of fixed capital (CFC) and affects income-side accounting, while general government TFP is constrained by valuation rules (zero TFP by definition).
• Limitations flagged by authors: PIM benchmark sensitivity (early-year stock profiles), simplified capital aggregation (weighted-average geometric depreciation vs. full vintage hyperbolic profiles), coverage gaps (no OAD investment imputed for farms due to OEWS limitations), and use of experimental estimates rather than fully implemented national-account methods.

Data & Methods

• Data coverage: Experimental industry-level OAD investment and price series (GFCF) run 1997–2024; analysis focuses on productivity impacts 2002–2024. Modified ILPA constructed for 63 industries (summary sectors) used in growth accounting.
• Asset breakdown: 10 aggregate asset classes — structures; equipment (5 types); and IPPs (R&D, software, ELAO, and OAD).
• Valuation of OAD: Sum-of-costs approach allocating planning/recording/digitizing/organizing/storing costs; excludes use/analysis and purchased data.
• Capital stock and services:
  • Perpetual inventory method (PIM) to convert flows (GFCF) into net productive stocks and consumption of fixed capital (CFC).
  • Simplified aggregation: weighted-average geometric depreciation profile approximates vintage accounting used in official ILPA.
  • Annual asset rental prices estimated from industry net stock distributions, depreciation, and residual capital costs (value added minus labor costs).
  • Capital and input quantity indices built using Törnqvist–Theil indexes; capital services aggregated across asset classes.
• Productivity accounting:
  • Growth accounting at the industry level: ΔlnTFP = ΔlnGrossOutput − sum(wj Δln quantity_j), where wj are compensation (income) shares (Törnqvist–Theil weights).
  • Domar aggregation used to compute contributions of industry-level sources to aggregate value‑added growth.
• Model is a simplified production-account implementation rather than full ILPA methodology; purpose is to approximate the impact of capitalizing OAD on measured inputs, output, and TFP.

Implications for AI Economics

• Data as capital changes measured capital intensity and thus the inferred returns to capital vs. labor. Studies that omit own‑account data will understate capital’s role in sectors adopting data‑driven AI and overstate TFP (or misattribute gains).
• Reassessment of AI contributions: Capitalizing OAD moves some growth previously attributed to TFP into measured capital accumulation. Empirical estimates of AI’s productivity impact, complementarities between AI/data assets and human capital, and the timing of AI gains should control for this measurement change.
• Sectoral analyses of AI impacts: Heterogenous revisions mean AI‑driven gains will be measured differently across industries — largest measured capital effects in management, professional services, finance, data processing and related services. Wage and labor‑demand studies should account for higher measured capital shares in these sectors.
• Policy and tax implications: Treating data as capital affects investment measures, depreciation allowances, and potentially tax bases; policy designed to incentivize data/AI investment or to tax capital returns must account for this reclassification.
• Measurement and research priorities for AI economics:
  • Improve price indices, depreciation profiles, and lifetime estimates for data assets used in AI (important for correct capital service estimation).
  • Distinguish own‑account vs purchased data in empirical work (different valuation and balance‑sheet implications).
  • Better microdata on firm‑level OAD creation, reuse across products/units, and the extent of data reuse in AI models to capture joint production and network/externality effects.
  • Reevaluate macro/micro models of technological change (e.g., growth, misallocation, skill bias) using the updated capital composition.
• Caution for trend interpretation: Because PIM benchmarks, early adoption timing, and simplified aggregation matter, comparisons across pre/post capitalization periods require adjustment; researchers should treat short-run TFP declines after capitalization as measurement changes rather than necessarily true slowdowns in technological progress.

If you want, I can: - Extract the industry‑level tables/figures (capital share changes, top industry revisions) into a concise table or CSV for further analysis; - Draft a short methods appendix linking these measurement changes to common econometric specifications used in AI productivity research (e.g., value‑added regressions, production‑function estimates).