Tax incentives steer corporate investment: South Korea's cut to automation tax credits curtailed robot purchases and lifted employment, showing subsidies speed adoption but risk overinvestment; Germany's R&D allowance raises research, patents and productivity, whereas complex tax systems erode firm value and blunt policy impact.

Main Finding

Taxation matters for firms’ real decisions: targeted tax incentives can meaningfully shift investment toward automation and R&D, but their effectiveness depends on firm heterogeneity and tax-system design; conversely, greater tax complexity reduces firm market value and dampens investment (including into AI-related R&D). Aggregate welfare implications hinge on policy design — simple, well‑targeted incentives encourage productive AI/automation adoption while complex rules and poorly calibrated credits can cause inefficiencies (over/underinvestment) and higher compliance costs.

Key Points

• Chapter 1 (Quasi-robot tax — South Korea)

  • A 2018 reduction of an automation investment tax credit for medium/large firms in South Korea led to a significant decline in industry and firm-level automation investments and a relative increase in employment in affected firms.
  • Heterogeneous effects: liquidity-constrained firms cut total investment, while unconstrained firms reallocated toward higher‑productivity uses; evidence that earlier generous credits induced some inefficient overinvestment in robots.
  • Identification leverages a quasi-experiment (differential exposure across firm sizes/industries) and event‑study / diff‑in‑diff comparisons (Korea vs. comparator Japan) at industry and firm levels.

• Chapter 2 (German Research Allowance / Forschungszulage)

  • The indirect R&D tax incentive (introduced Dec 2020) increased firms’ R&D expenditures, R&D employment and wages, new product development, patenting activity and firm productivity—effects concentrated among SMEs (majority of applicants).
  • Uptake: ~33,840 applications for ~42,231 projects; regional and sectoral distributions documented.
  • Descriptive and literature-synthesis approach: short-run increases in R&D spending observed, but Germany still short of the 3.5% GDP R&D target; provides critical assessment of program design and administrative features.

• Chapter 3 (Tax Complexity and Firm Value)

  • Country-level increases in tax complexity are associated with lower market valuations of publicly listed firms: counterfactuals imply aggregate market capitalization would be ~2% (≈ US$900bn) higher absent observed increases in complexity.
  • Effects are strongest for complexity related to anti‑avoidance rules and assessment procedures, and for firms with limited profit‑shifting scope, weak governance, or low internal information quality.
  • Mechanisms: higher compliance costs, increased uncertainty, and reduced responsiveness of investment (including R&D) to tax incentives.
  • Estimation uses two‑way fixed effects on MSCI World–listed firms and a tax‑complexity index (Hoppe et al., 2023).

Data & Methods

• Chapter 1

  • Data: IFR robot installation data (industry-level), firm-level financials from Orbis (BvD).
  • Methods: Difference‑in‑differences / event-study at industry and firm levels exploiting Korea’s 2018 tax-credit reduction; firm heterogeneity analyses by liquidity constraints; robustness checks (IV, sample restrictions, alternative financial constraint measures).

• Chapter 2

  • Data: Administrative/aggregated application data for the German Forschungszulage, national R&D statistics, patent and innovation indicators from German federal sources; literature review and synthesis of empirical studies on R&D tax incentives.
  • Methods: Descriptive analyses of uptake and short-/medium-term outcomes; synthesis of micro-econometric findings from previous studies to assess causal impacts where available.

• Chapter 3

  • Data: Publicly listed firms in MSCI World, country-level tax complexity index (Hoppe et al.), firm accounting and governance variables.
  • Methods: Panel two‑way fixed effects regressions (firm and year) with heterogeneity tests (profit shifting potential, governance, information quality), dynamic analyses, and robustness checks (alternative clustering, subcategory decompositions of complexity).

Implications for AI Economics

• Tax incentives shape the diffusion and direction of automation and AI adoption

  • Generous, targeted credits for automation/robotics can accelerate adoption but risk overinvestment if poorly designed; calibration should aim at productive AI adoption (complementarity with human capital, productivity gains) rather than mechanical capital accumulation.
  • R&D tax incentives increase AI-related R&D and complementary human capital in R&D roles—SME‑targeting can foster broader innovation diffusion but administrative simplicity is crucial for uptake.

• Design and sequencing matter

  • Combine incentives with policies easing financing constraints (e.g., grants, credit lines) to prevent liquidity-constrained firms from cutting productive investment when credits change.
  • Accompany automation/AI incentives with labor-market policies (retraining, transition support) to manage employment reallocation.

• Complexity undermines AI policy effectiveness

  • Complex tax systems reduce firms’ market value and blunt responsiveness to tax incentives, lowering investment in AI and digitalization. Simplifying rules, clearer assessment procedures, and predictable anti‑avoidance provisions improve policy leverage and reduce deadweight compliance costs.

• Distributional and firm‑level heterogeneity

  • Effects differ across firm size, governance quality, and international mobility of profits. Policies aiming to promote AI should account for heterogeneity: SMEs may need simplified access and administrative support; multinationals’ responses will depend on profit‑shifting opportunities.

• Measurement and evaluation

  • Evaluation should use high-frequency, micro-level adoption and R&D measures (robot installations, AI project counts, patent/R&D spending by category) and exploit quasi‑experiments or staged policy rollouts for causal inference.
  • Monitor both quantity (investment/R&D) and quality (productivity gains, employment effects, innovation outputs) to detect overinvestment or misallocation.

Limitations and caveats - Chapter 2’s empirical assessment is largely descriptive and synthesizes prior causal evidence; causal identification of the German Forschungszulage’s long-term effects requires more quasi‑experimental variation as more post‑implementation data accrue. - External validity: South Korea’s institutional and industrial context may differ from other countries (e.g., labor market institutions, capital markets), so magnitudes may not generalize directly.

Key policy takeaways for AI economics - Use targeted, simple, and evaluated tax incentives to promote productive AI and automation adoption. - Reduce unnecessary tax complexity and streamline procedures to preserve firms’ responsiveness to incentives and avoid value destruction. - Pair incentives with liquidity support and active labor policies to ensure equitable and efficient transitions.