## wp1784

## Source details

**Canonical URL:** [wp1784](https://www.imf.org/-/media/files/publications/wp/2017/wp1784.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp1784.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp1784.pdf.json)

---

### Research question and identification
- Estimate the causal effect of R&D tax incentives on qualifying R&D expenditure in the United Kingdom.
- Quasi-experimental identification:
  - Exploits a 2008 expansion in the SME definition that doubled thresholds in employment, turnover, and total assets for R&D tax credit eligibility.
  - Treatment: companies reclassified from large-company status to SME status for R&D tax relief in 2008.
  - Empirical approach: difference-in-differences (diff-in-diff) leveraging cross-sectional and time-series variation; placebo tests in pre-reform years to assess parallel trends.

### Data and empirical advantages
- Data sources and features:
  - Linked HM Revenue and Customs corporation tax assessments (CT600), HMRC Specialist R&D Units qualifying R&D spending data, and Bureau van Dijk’s FAME annual company accounts.
  - Fixed sample period: 2002–11.
  - Fixed dataset includes 30,056 firm-years for companies that undertook positive qualifying R&D both before and after the 2008 reform.
  - Intent-to-treat unbalanced sample: 247 treated firms and 335 control firms; each firm observed at least once before and once after the reform.
- Advantages over prior work:
  - Observes tax-return-based SME/large status, avoiding proxy misclassification.
  - Analyzes a volume-based enhanced deduction scheme, reducing complications from incremental schemes.

### Policy context and mechanics
- U.K. R&D tax incentive characteristics (sample period 2002–11):
  - Form: enhanced deductions from taxable income.
  - SME enhanced deduction rate changes: 150 percent (pre-2008) → 175 percent (April 2008) → 200 percent (2011).
  - Large company enhanced deduction rate changes: 125 percent (pre-2008) → 130 percent (April 2008).
- User-cost effects:
  - For a profitable firm, incorporating all 2008 tax changes, the overall reduction in the tax component of the user cost of R&D capital amounts to 10–31 percent for a newly classified SME, depending on firm size and tax position.
  - Companies that shifted from large-company relief before the reform to SME relief after the reform experienced an average reduction in their user cost of R&D by about 21 percent between 2007 and 2009.
  - Firms eligible to the large company relief both before and after the reform experienced a reduction in their user cost of capital of 0.3 percent.
- Comparative context:
  - U.K. scheme described as permanent, relatively simple, low administrative costs compared with other jurisdictions.
  - U.S. R&D tax credit was only made permanent in December 2015 and has historically been incremental and administratively complex.
  - During 2012, about half of U.K. funding for business R&D was channeled through tax incentives (OECD R&D tax incentive statistics).

### Conceptual framework
- Treats R&D as knowledge capital analogous to physical capital in a neoclassical optimal capital accumulation framework (Hall and Jorgenson 1967; Jorgenson 1963; Griliches 1979).
- First-order conditions equate marginal product of R&D capital to its user cost; reductions in user cost via tax incentives raise optimal R&D capital and observed R&D flows.
- Theoretical expression yields a positive response of R&D capital to increases in tax generosity (Equation 8 in the source).

### Main empirical findings
- Average treatment effect (preferred intent-to-treat):
  - Diff-in-diff implied increase in qualifying R&D spending: 26.4 percent (preferred specification).
  - Diff-in-diff coefficients in baseline Table 3:
    - Column (1): 0.287, SE = (0.112)
    - Column (2): 0.286, SE = (0.109)
    - Column (3): 0.264, SE = (0.109)
    - Column (4): 0.264, SE = (0.109)
  - Interpretation: differential increase in R&D spending for treated firms of at least 26.4 percent; significant at the 5 percent level.
- Implied elasticities and user-cost estimates:
  - Diff-in-diff implied elasticity of R&D spending with respect to its user cost: -1.55.
  - Earlier version estimated user-cost elasticity around -2.3 (Guceri and Liu 2015); revision leaves predictions and conclusions unaffected.
  - Rate-change experiment (SMEs that remained SMEs) yields: 23.4 percent increase in R&D in response to a 9 percent drop in the tax component → implied elasticity around -2.9.
  - Direct user-cost regressions (Table 11, N = 29,552):
    - Column (1): coefficient on user cost = -2.361 (standard errors large).
    - Column (2): user cost based on previous-year marginal tax rate = -2.395 (standard errors large).
    - Column (3): user cost based on "before-R&D spending" marginal tax rate = -1.746**, standard error (0.796).

### Magnitudes and "bang-for-the-buck"
- For each pound of foregone tax revenue:
  - Qualifying R&D spending increases by around £1.30 for a company paying at the main rate of 28 percent.
  - Qualifying R&D spending increases by around £0.8 for a company paying at the “small profits” rate of 21 percent.
- Example calculations:
  - With treated firms increasing R&D by 0.264R and enhanced deduction moving from 1.25 to 1.75, foregone tax revenue per firm is 0.202R at the small profits rate and 0.335R at the main rate, producing a bang-for-the-buck of £0.80–£1.30.
- Representative firm experience:
  - Decrease in R&D user cost around 21 percent between 2007 and 2009 if paying at the small profits rate, and around 15 percent if paying at the main rate.
- Average observed tax-component of user cost:
  - Medium-sized (treated) firms: decline from 0.92 (pre-treatment) to 0.76 (post-treatment).
  - Large (control) firms: remained at 0.94 throughout the sample period.
- Enhanced deduction outcomes post-reform:
  - Treated firms: could deduct up to £175 for every £100 spent in R&D (post-reform SME treatment).
  - Firms remaining large: deductible £130 for every £100 of R&D spending (large company scheme).

### Estimation strategy and samples
- Model: Difference-in-differences specification (Equation 9) on qualifying R&D spending in 2009 prices with interaction D_i T_t capturing treatment effect δ_I.
- Controls and estimation:
  - Firm fixed effects; time fixed effects; first lags of real revenue and real revenue growth; standard errors clustered by firm.
  - Estimator: Poisson Pseudo-Maximum-Likelihood (PPML) used for skewed dependent variable with mass at zero.
- Key sample counts across specifications:
  - Main intent-to-treat sample: N = 4,799; No of Treated Firms = 247; No of Control Firms = 335.
  - Placebo specifications: N = 5,353.
  - Robustness and alternative samples: N = 4,756; N = 4,253; N = 10,106 (alternative “actual treatment” with HMRC-declared status; treated = 185, control = 1,102).

### Robustness checks and sensitivity
- Placebo reforms (Table 2) placebo diff-in-diff sequence: -0.171, 0.079, 0.129, 0.194, 0.19; N = 5,353 in each placebo year specification.
- Removing 2007 and 2008 fiscal years (Table 4):
  - Diff-in-diff: 0.305**, 0.306**, 0.281**, 0.281** (SEs: (0.124), (0.121), (0.124), (0.124)); preferred estimate ~28.1 percent; N = 4,253.
- Dropping largest 1 percent of control group firms (Table 5):
  - Diff-in-diff: 0.275**, 0.275**, 0.254**, 0.254** (SEs: (0.115), (0.112), (0.111), (0.111)); N = 4,756.
- Winsorising top and bottom percentiles of R&D spending (Table 6):
  - Diff-in-diff: 0.210*, 0.210**, 0.203*, 0.203* (SEs: (0.109), (0.106), (0.106), (0.106)); reported magnitude ~20 percent (significant at 10 percent); N = 4,799.
- Alternative “actual treatment” (HMRC-declared status) (Table 10):
  - Diff-in-diff: 0.400**, 0.420**, 0.397**, 0.397** (SEs: (0.192), (0.186), (0.185), (0.185)); preferred specification estimate 39.7 percent; N = 10,106.
- Robustness to bunching and outliers:
  - Excluding employment between 240–260 and 230–270 yields similar results.
  - Dropping largest percentile of firms and winsorising outcomes does not remove positive, significant effects.

### Heterogeneity and other outcomes
- Consistent R&D performers (Table 7):
  - Consistent * diff-in-diff: 0.406, 0.424, 0.480*, 0.480* (SEs: (0.289), (0.284), (0.266), (0.266)).
  - Sample counts: Treated Consistent = 116; Treated Not Consistent = 131; Control Consistent = 171; Control Not Consistent = 164.
  - Interpretation: consistent performers respond differentially more than intermittent performers.
- Young firms (Table 8):
  - Young * diff-in-diff: 0.608**, 0.608**, 0.586**, 0.586** (SEs: (0.241), (0.238), (0.238), (0.238)).
  - Young * Post 2008: -0.430**, -0.377**, -0.352*, -0.352* (SEs: (0.179), (0.184), (0.187), (0.187)).
  - Sample counts: No of Treated, Young = 41; No of Treated, Not Young = 206; No of Control, Young = 53; No of Control, Not Young = 282.
  - Interpretation: younger firms respond differentially more to the policy change.
- Tests for relabeling and other outcomes (Table 9):
  - Investment dependent variable (Column (1)): Diff-in-diff = -0.126, SE = (0.103).
  - R&D in same sample (Column (2)): Diff-in-diff = 0.255**, SE = (0.111).
  - Non-R&D Cost Ratio (Column (3)): Diff-in-diff = 0.074, SE = (0.076).
  - R&D in same sample (Column (4)): Diff-in-diff = 0.262**, SE = (0.109).
  - Interpretation: no robust evidence of systematic relabeling of ordinary investment expenditure or non-R&D input costs into qualifying R&D categories; R&D increases remain positive and significant.

### Appendices and supplementary experiments
- Appendix 3 (Rate Increase Experiment, SMEs that remained SMEs after 2008):
  - Appendix Table 3.1 (removing 2007 and 2008): Diff-in-diff = 0.317***, 0.318***, 0.285**, 0.286**; Post 2008 = 0.238***; N = 10,572.
  - Appendix Table 3.2 (excluding 2008): Diff-in-diff = 0.260***, 0.264***, 0.234**, 0.234**; Post 2008 = 0.190**; N = 11,968.
- Appendix 4 (Cash Credits for SMEs):
  - HMRC provides a cash refund up to 24 percent of total R&D spending, capped by PAYE or NIC liabilities.
  - The total cash available to a company was kept at around 24–25 percent of total R&D spending across periods of different enhanced deduction rates.
  - Cash refund rate: up to 24 percent of total R&D spending.

### Interpretation, policy implications, and limitations
- Main substantive finding: preferred intent-to-treat estimate ≈ 26 percent increase in qualifying R&D spending for medium-sized firms reclassified as SMEs after the 2008 reform (reported as 26.4 percent).
- Alternative actual-treatment estimate: preferred specification at 39.7 percent.
- Empirical elasticities range from around -1.55 to -2.9 depending on experiment and specification.
- Fiscal implications:
  - Estimated bang-for-the-buck between £0.80 and £1.30 of additional R&D per £1 of foregone corporation tax, varying by taxpayer marginal rate.
- Robustness: results robust to anticipation-effect exclusions, lagged turnover controls, firm and year fixed effects, outlier exclusions, and placebo checks.
- Limitations and future work:
  - Short sample period prevents analysis of links between observed increases in R&D spending and long-run productivity growth or R&D spillovers; left for future research.
  - Endogeneity concerns in direct user-cost regressions addressed via previous-year and “before-R&D spending” marginal tax rate constructions; coefficients and standard errors vary accordingly.

*Source: wp1784 - References / wp1784 - 2008. The positive and significant effect of the change in R&D tax incentive is robust to / Appendix 2 / Section II / Appendix 4 (wp1784.pdf)*

### References .............................................................................................................

### wp1784 - References

### Content inventory (Tables, Figures, Appendices)
- Tables
  - 1. Marginal Corporation Tax Rate in the United Kingdom, 2002–14
  - 2. Placebo Reforms. Treatment: Large Companies Reclassified as SMEs after 2008
  - 3. Baseline Results. Treatment: Large Companies Reclassified as SMEs after 2008
  - 4. Baseline Results. Treatment: Large Companies Reclassified as SMEs after 2008, Removing Both 2007 and 2008 Fiscal Years
  - 5. Robustness—Dropping The Largest Percentile of Firms from the Control Group
  - 6. Robustness—Winsorising the Outcome for Highest and Lowest Spenders (Top and Bottom Percentiles) in R&D
  - 7. Heterogeneous Responses to the Policy, Consistent Performers of R&D
  - 8. Heterogeneous Responses to the Policy, Young Firms. Treatment: Large Companies Reclassified as SMEs after 2008
  - 9. Effect of Policy on Other Outcomes than R&D. Treatment: Large Companies Reclassified as SMEs after 2008
  - 10. Baseline Results. Treatment: Large Companies Reclassified as SMEs after 2008
  - 11. Direct Estimates of User Cost Elasticity

- Figures
  - 1. Evolution of R&D Tax Relief Deduction Rates
  - 2. Size Thresholds for the SME Tax Relief
  - 3. Tax Component of the User Cost of R&D on Current Spending
  - 4. Average User Cost of Capital across Treatment and Control Groups, Tax Component
  - 5. Average R&D Spending across Groups, Relative to 2007 R&D Spending

- Appendices
  - 1. Summary of Recent Related Literature
  - 2. Sample Characteristics
  - 3. Treatment II: SMEs that Remained as SMEs after 2008
  - 4. Cash Credits for SMEs

- Appendix Tables and Figures
  - Appendix Tables 1.1, 3.1, 3.2
  - Appendix Figure 4.1 Cash Credit Rates for Loss-Making R&D Performers

### Research question and identification strategy
- Research objective
  - Estimate the causal effect of R&D tax incentives on qualifying R&D expenditure in the United Kingdom.
- Quasi-experimental identification
  - Exploits a 2008 expansion in the SME definition that doubled thresholds in employment, turnover, and total assets for R&D tax credit eligibility.
  - Treatment: companies reclassified from large-company status to SME status for R&D tax relief in 2008.
  - Empirical approach: difference-in-differences (diff-in-diff) leveraging cross-sectional and time-series variation; placebo tests in pre-reform years to assess parallel trends.

### Data and empirical advantages
- Data sources and features
  - Large-scale administrative dataset linking corporation tax records, qualifying R&D expenditures, and financial statements.
  - Precise indicators: firm eligibility (SME vs large) based on tax returns and employment-size from accounting data.
  - Ability to measure the marginal tax rates and the rate at which firms obtain R&D tax incentives.
- Advantages over prior work
  - Avoids proxy misclassification by observing tax-return-based SME/large status.
  - Simpler U.K. scheme: enhanced deductions applied to total R&D each year (volume-based), reducing complications from incremental schemes.

### Policy context and mechanics
- U.K. R&D tax incentive characteristics (sample period 2002–11)
  - Form: enhanced deductions from taxable income.
  - SMEs enjoyed substantially more generous deduction rates than large companies.
  - 2008 reform: expansion of SME definition effective 2008; no concurrent national policy changes targeted at the same group.
- Tax changes and user cost effects
  - Incorporating all 2008 tax changes, for a profitable firm the overall reduction in the tax component of the user cost of R&D capital amounts to 10–31 percent for a newly classified SME, depending on firm size and tax position.
  - In the data, companies that shifted from large-company relief before the reform to SME relief after the reform experienced an average reduction in their user cost of R&D by about 21 percent between 2007 and 2009.
  - Firms eligible to the large company relief both before and after the reform experienced a reduction in their user cost of capital of 0.3 percent (considered negligible).

### Main empirical findings
- Average treatment effect
  - Treatment group increased R&D spending by about 26.4 percent in response to the increased generosity of tax incentives.
- Robustness and heterogeneity checks (items listed in tables)
  - Placebo reforms and pre-trend checks (Table 2).
  - Multiple baseline specifications and exclusion of 2007 and 2008 fiscal years (Table 4).
  - Robustness to dropping largest percentile of firms in control group (Table 5).
  - Robustness to winsorising highest and lowest R&D spenders (Table 6).
  - Heterogeneous responses for consistent R&D performers and young firms (Tables 7 and 8).
  - Effects on outcomes other than R&D (Table 9).
  - Direct estimates of user cost elasticity (Table 11).
  - Additional experiment: rate increases for SMEs that remained SMEs after 2008 (Appendix 3; Appendix Tables 3.1 and 3.2).
  - Cash credits for SMEs discussed in Appendix 4; Appendix Figure 4.1 documents Cash Credit Rates for Loss-Making R&D Performers.

### Literature and contributions
- Three main contributions
  - Identification: uses SME definition reform that is limited to the R&D tax credit to generate exogenous variation in R&D user cost.
  - Data: exploits administrative linkage of tax records and qualifying R&D expenditure for precise measurement of incentives and eligibility.
  - Institutional clarity: analyzes a volume-based enhanced deduction scheme in the U.K., alleviating issues of salience and scheme complexity.
- Related literature coverage
  - References to methodological challenges (Hall and Van Reenen (2000)) and recent quasi-experimental studies (Bronzini and Iachini 2014; Agrawal and others 2014; Dechezlepretre and others 2016; Guceri 2017).
  - Appendix 1: Summary of Recent Related Literature; Appendix Table 1.1 lists Related Literature (Since 2000).

*Source: wp1784 - References*

### 2008. The positive and significant effect of the change in R&D tax incentive is robust to

### wp1784 - 2008. The positive and significant effect of the change in R&D tax incentive is robust to

### Main empirical findings
- Tax incentives for R&D have a statistically significant, positive effect on the level of R&D spending for companies that were actively performing qualifying R&D both before and after the reform.
- Diff-in-diff implied elasticity of R&D spending with respect to its user cost: -1.55.
- An earlier version estimated the user-cost elasticity to be around -2.3 (Guceri and Liu 2015); the revision leaves paper predictions and conclusions unaffected.
- For each pound of foregone tax revenue:
  - Qualifying R&D spending increases by around £1.30 for a company paying at the main rate of 28 percent.
  - Qualifying R&D spending increases by around £0.8 for a company paying at the “small profits” rate of 21 percent.
- Companies do not systematically relabel ordinary investment expenditure as R&D to benefit from the larger tax deduction.

### Magnitude of tax-price and user-cost changes
- Key deduction-rate changes:
  - SME enhanced deduction: 150 percent (pre-2008) → 175 percent (April 2008) → 200 percent (2011).
  - Large company enhanced deduction: 125 percent (pre-2008) → 130 percent (April 2008).
- Representative firm that benefited from the SME definition change experienced a decrease in the R&D user cost by around 21 percent between 2007 and 2009 if paying at the small profits rate, and around 15 percent if paying at the main rate.
- Average tax-component of the user cost (tax component) observed:
  - Medium-sized (treated) firms: decline from 0.92 (pre-treatment) to 0.76 (post-treatment).
  - Large (control) firms: remained at 0.94 throughout the sample period.
- Enhanced deduction outcomes (post-reform comparison):
  - Treated firms: could deduct up to £175 for every £100 spent in R&D (post-reform SME treatment).
  - Firms remaining large: deductible £130 for every £100 of R&D spending (large company scheme).

### Policy background and comparative context
- U.K. R&D tax relief features during 2002–11: permanent, relatively simple, low administrative costs compared with other jurisdictions in the discussion.
- U.K. scheme contrasts with the U.S. R&D tax credit which was only made permanent in December 2015 and has historically been incremental and administratively complex.
- French R&D tax credit evolved from an incremental design to a volume-based scheme in 2008.
- Many jurisdictions (for example, Canada, France, and Norway) apply maximum spending limits that may limit effectiveness for spending above thresholds.
- During 2012, about half of U.K. funding for business R&D was channeled through tax incentives (OECD R&D tax incentive statistics).

### Conceptual framework (summary)
- Framework: analogy between investment in physical capital and knowledge capital within a neoclassical optimal capital accumulation framework (Hall and Jorgenson 1967; Jorgenson 1963; Griliches 1979).
- Treating R&D like capital, firm first-order conditions equate marginal product of R&D capital to its user cost; reductions in user cost (via tax incentives) raise optimal R&D capital and hence observed R&D flows.
- Theoretical expression yields a positive response of R&D capital to increases in tax generosity (Equation 8 in the source).

### Data, sample, and empirical approach
- Datasets linked:
  - HM Revenue and Customs corporation tax assessments (CT600).
  - HMRC Specialist R&D Units comprehensive R&D spending data.
  - Bureau van Dijk’s FAME annual company accounts.
- Fixed sample period: 2002–11.
- Sample coverage and size:
  - Fixed dataset includes 30,056 firm-years for companies that undertook positive qualifying R&D both before and after the 2008 reform.
  - Dataset for analyzing the 2008 SME definition change (intent-to-treat using employment size) is an unbalanced sample with 247 treated firms and 335 control firms; each firm observed at least once before and once after the reform.
- Aggregate benchmarks and trends:
  - Total current R&D spending by all U.K. businesses: around £13.7 billion in 2005 and around £17.4 billion in 2011 (nominal terms).
  - Ratio of total qualifying R&D observed in HMRC data to aggregate current spending in BERD rose from just over 50 percent to 70 percent between 2005 and 2011.
  - Number of beneficiary companies increased from 6,120 to 12,050 between 2005 and 2011.
- Empirical identification:
  - Main treatment exploited: August 2008 SME definition change (doubling thresholds) that reclassified some large firms as SMEs, giving access to more generous SME enhanced deduction rates.
  - Control group: firms that remained labeled as large in both pre- and post-reform periods and that carried out qualifying R&D in at least one year before 2008 and at least one year after 2008.
  - Intent-to-treat refined using employment size (250–499 employees for “medium sized companies” treated group) to avoid post-reform status endogeneity.
  - Placebo tests show no significant differential increases for treated group in any period before the reform.

### Additional evidence and caveats
- The paper documents that the observed increase in qualifying R&D spending is not driven by relabelling of ordinary investment as R&D.
- The study focuses on increases in private R&D spending due to reduction in tax price; the extent of productivity effects from additional R&D remains an important area for future research.
- The diff-in-diff approach addresses the intensive margin (firms already performing R&D before reform) and does not identify responses at the extensive margin (firms that begin R&D only after reform).

*Source: wp1784 - 2008. The positive and significant effect of the change in R&D tax incentive is robust to*

### Appendix 2 provides further descriptive statistics on the treated and the control groups as

### wp1784 - Appendix 2 provides further descriptive statistics on the treated and the control groups as

### Estimation strategy
- Model: Difference-in-differences specification (Equation 9) on qualifying R&D spending ܴ௜௧ in 2009 prices with interaction D_i T_t capturing the treatment effect δ_I.
- Interpretation: δ_I can be directly interpreted as the percentage change in qualifying R&D spending.
- Identification features:
  - Firm fixed effects to absorb time-invariant heterogeneity.
  - Time fixed effects ߶௧ to control for aggregate shocks (including the great recession).
  - Controls include first lags of real revenue and real revenue growth; standard errors clustered by firm.
- Estimator: Poisson Pseudo-Maximum-Likelihood (PPML) recommended for variables with long right-tail and mass at zero; used clustered standard errors to correct for over-dispersion.
- Sample framing:
  - Main “intent-to-treat” uses pre-reform employment size groups (treated: large firms reclassified as SMEs after 2008; control: firms remaining Large).
  - Middle of the 2008 tax year removed from regressions.
  - Alternative “actual treatment” uses HMRC-declared relief status (robustness check).

### Baseline results (main experiment)
- Main baseline Table 3 diff-in-diff estimates (level of qualifying R&D spending):
  - Column (1): Diff-in-diff = 0.287, SE = (0.112)
  - Column (2): Diff-in-diff = 0.286, SE = (0.109)
  - Column (3): Diff-in-diff = 0.264, SE = (0.109)
  - Column (4): Diff-in-diff = 0.264, SE = (0.109)
- Interpretation reported: diff-in-diff coefficient is significant at the 5 percent level, indicating a differential increase in R&D spending for treated firms of at least 26.4 percent.
- Sample counts reported in baseline:
  - N = 4,799 (columns with firm fixed effects and controls)
  - No of Treated Firms = 247
  - No of Control Firms = 335
- Post-2008 (uninteracted) reported example: Post 2008 = 0.182, SE = (0.075) in one specification.

### Robustness and sensitivity checks
- Placebo reforms (Table 2): placebo diff-in-diff estimates by imposing placebo treatment years 2003–2007:
  - Placebo diff-in-diff sequence: -0.171, 0.079, 0.129, 0.194, 0.19
  - N = 5,353 in each placebo year specification
  - Additional controls and fixed effects included; lag revenue and lag revenue growth statistically insignificant.
- Removing 2007 and 2008 fiscal years (Table 4):
  - Diff-in-diff: 0.305**; 0.306**; 0.281**; 0.281** (SEs: (0.124), (0.121), (0.124), (0.124))
  - Preferred specification coefficient size reported as 28.1 percent and significant at the 5 percent level.
  - N = 4,253; No of Treated Firms = 247; No of Control Firms = 335
- Dropping largest 1 percent of control group firms (Table 5):
  - Diff-in-diff: 0.275**, 0.275**, 0.254**, 0.254** (SEs: (0.115), (0.112), (0.111), (0.111))
  - Post 2008 example: 0.193**, SE = (0.080)
  - N = 4,756
- Winsorising top and bottom percentiles of R&D spending (Table 6):
  - Diff-in-diff: 0.210*, 0.210**, 0.203*, 0.203* (SEs: (0.109), (0.106), (0.106), (0.106))
  - Post 2008 example: 0.258***, SE = (0.070)
  - N = 4,799
  - Reported magnitude with winsorized sample: around 20 percent (significant at 10 percent).
- Alternative “actual treatment” (HMRC-declared status) (Table 10):
  - Diff-in-diff: 0.400**, 0.420**, 0.397**, 0.397** (SEs: (0.192), (0.186), (0.185), (0.185))
  - Preferred specification estimate reported as 39.7 percent and highly significant.
  - This revised unbalanced sample: 185 firms in the treated group and 1,102 firms in the control group; regressions reported with N = 10,106.

### Heterogeneous responses across firms
- Consistent performers of R&D (Table 7):
  - Consistent * diff-in-diff: 0.406, 0.424, 0.480*, 0.480* (SEs: (0.289), (0.284), (0.266), (0.266))
  - Findings: consistent performers respond differentially more to the policy than intermittent performers; sample splits reported:
    - Number of Treated, Consistent = 116
    - Number of Treated, Not Consistent = 131
    - Number of Control, Consistent = 171
    - Number of Control, Not Consistent = 164
- Young firms (bottom quartile of age distribution) (Table 8):
  - Young * diff-in-diff: 0.608**, 0.608**, 0.586**, 0.586** (SEs: (0.241), (0.238), (0.238), (0.238))
  - Young * Post 2008: -0.430**, -0.377**, -0.352*, -0.352* (SEs: (0.179), (0.184), (0.187), (0.187))
  - Sample counts:
    - No of Treated, Young = 41
    - No of Treated, Not Young = 206
    - No of Control, Young = 53
    - No of Control, Not Young = 282
  - Interpretation: younger firms respond differentially more to the policy change.

### Testing for relabeling and other outcomes (Table 9)
- Diff-in-diff estimates on other outcomes (investment and non-R&D cost ratio) and corresponding R&D checks:
  - Dependent variable: Investment (Column (1)) — Diff-in-diff = -0.126, SE = (0.103)
    - In this sample, revenue control is significant at 10 percent.
  - Dependent variable: R&D in same sample as Column (1) (Column (2)) — Diff-in-diff = 0.255**, SE = (0.111)
  - Dependent variable: Non-R&D Cost Ratio (Column (3)) — Diff-in-diff = 0.074, SE = (0.076)
    - In this sample, revenue growth control is significant at 5 percent.
  - Dependent variable: R&D in same sample as Column (3) (Column (4)) — Diff-in-diff = 0.262**, SE = (0.109)
- Interpretation: negative and insignificant effect on investment; no robust evidence of systematic relabeling of ordinary investment expenditure or non-R&D input costs into qualifying R&D categories. R&D diff-in-diff estimates remain positive and significant at 5 percent in these subsamples.

### Key descriptive/comparative statistics and data behavior
- Only 40 percent of companies claiming R&D spending during the observed period claim R&D tax relief continuously in all years; remaining firms stop claiming at least once (interpreted as zero R&D expenditure in some years).
- Robustness to bunching around employment threshold:
  - Excluding firms with employment between 240 and 260 yields similar results.
  - Wider exclusion band (230–270) also yields similar results.
- Robustness to outliers in firm size and R&D spending: main findings remain positive and statistically significant across checks (drop largest percentile, winsorize top/bottom percentiles).
- Sample sizes reported across specifications: N = 4,799; 4,756; 4,253; 5,353; 10,106 depending on specification and robustness check.
- Firm counts repeatedly reported: Treated = 247 and Control = 335 in main intent-to-treat sample.

### Main substantive findings and interpretation
- Preferred intent-to-treat estimate: approximately a 26 percent increase in qualifying R&D spending for medium-sized firms reclassified as SMEs after the 2008 reform (reported as 26.4 percent in preferred specification).
- Alternative actual-treatment estimate: larger effect, preferred specification at 39.7 percent.
- Effects are robust to:
  - Removing 2007–08 to limit anticipation effects (preferred estimate ~28.1 percent).
  - Dropping largest control-group percentile (most conservative specification ~25.4 percent).
  - Winsorizing outcome for extreme R&D spenders (effect ~20 percent, significant at 10 percent).
- Heterogeneity:
  - Stronger responses for “consistent” R&D performers (Consistent * diff-in-diff up to 0.480*).
  - Stronger responses for young firms (Young * diff-in-diff up to 0.608**).
- No clear evidence of systematic relabeling of ordinary investment or non-R&D costs into qualifying R&D in the data examined.

*Source: wp1784 - Appendix 2 provides further descriptive statistics on the treated and the control groups as (wp1784.pdf)*

### Section II. First, the increase in the generosity of the SME eligibility threshold allows medium-

### wp1784 - Section II. First, the increase in the generosity of the SME eligibility threshold allows medium-

### Policy experiments and identification
- Two exogenous 2008 reforms increased the generosity of the R&D enhanced deduction: (i) an SME eligibility threshold change that reclassified some medium-sized firms as SMEs; and (ii) an increase in the SME enhanced deduction rate from 150 to 175 percent in 2008.
- The main experiment exploits the SME definition change: treated firms newly qualify as SMEs and therefore move from claiming R&D at the 130 percent rate to claiming at the 150 percent SME enhanced deduction rate (and after 2008 to 175 percent).
- A second (rate-change) experiment focuses on firms that remained SMEs throughout the sample to isolate the effect of the increase in the enhanced deduction rate.

### Main empirical findings
- The SME definition change (main experiment) is associated with:
  - "26.4 percent" increase in qualifying R&D spending in response to "around an average 17 percent" drop in the tax component of the user cost.
  - Implied elasticity of R&D with respect to its user cost of "around -1.55".
- The rate-change experiment (SMEs that remained SMEs) yields:
  - On average, a "23.4 percent" increase in qualifying R&D spending in response to a "9 percent" drop in the tax component of the user cost.
  - Implied R&D user cost elasticity estimate of "around -2.9".
- Bang-for-the-buck (foregone corporation tax → additional R&D):
  - Each "£1" foregone in corporation tax generates around "£1.30" in R&D for taxpayers at the small profits rate ("21 percent" in the post-reform period).
  - Each "£1" foregone in corporation tax generates around "£0.80" in R&D for taxpayers at the main rate ("28 percent" in 2008–10).
  - Example calculation: with treated firms increasing R&D by 0.264R and enhanced deduction moving from 1.25 to 1.75, foregone tax revenue per firm is "0.202R" at the small profits rate and "0.335R" at the main rate, producing a bang-for-the-buck of "£0.80–£1.30".

### Direct estimates of user cost elasticity and robustness
- Replacing the discrete interaction with a continuous R&D user cost (firm-specific statutory marginal tax rate, dependent on current-year taxable profit) produces direct elasticity estimates:
  - Column (1): coefficient on user cost = "-2.361" (standard errors large).
  - Column (2): user cost based on previous-year marginal tax rate = "-2.395" (standard errors large).
  - Column (3): user cost based on "before-R&D spending" marginal tax rate (tax component) = "-1.746**" (standard error "(0.796)").
  - Sample size ("N") in Table 11: "29,552".
- Endogeneity concerns and addressing strategies:
  - R&D spending reduces current-year taxable profit and marginal tax rate, biasing simultaneous regressions toward zero.
  - Assuming marginal tax rate is zero for loss-making companies that do not receive cash may understate incentives for firms that carry forward tax savings, biasing coefficients away from zero.
  - Two alternative user cost constructions used to mitigate endogeneity: (i) previous-year taxable profit marginal tax rate; (ii) “before-R&D spending” marginal tax rate. Both yield negative user cost elasticities and findings consistent with diff-in-diff results.

### Additional robustness and subgroup results
- The positive effect of more generous R&D tax incentives is robust to:
  - Accounting for anticipation effects (analyses excluding 2007 and/or 2008).
  - Controls for lagged real turnover and lagged real turnover growth (reported as statistically insignificant at conventional levels).
  - Firm and year fixed effects.
- Subgroup findings:
  - Strong increases in R&D spending observed in both consistent and intermittent spenders, with larger effects for consistent R&D spenders.
  - Young firms responded very strongly by increasing R&D spending after the reform.
  - Tests rule out relabeling of ordinary physical investment or non-R&D expenses as the driver of the observed increase in qualifying R&D spending.

### Appendix regression point estimates (selected)
- Appendix Table 3.1 (Rate Increase Experiment, removing 2007 and 2008 fiscal years):
  - Diff-in-diff coefficient estimates reported as "0.317***", "0.318***", "0.285**", "0.286**" across specifications; Post 2008 coefficient "0.238***".
  - Observations ("N") reported as "10,572" for these specifications.
- Appendix Table 3.2 (Rate Increase Experiment, excluding 2008):
  - Diff-in-diff coefficient estimates reported as "0.260***", "0.264***", "0.234**", "0.234**" across specifications; Post 2008 coefficient "0.190**".
  - Observations ("N") reported as "11,968".
- Sensitivity: removing observations in 2007 and 2008 yields a point estimate of the diff-in-diff coefficient of "0.197" (significant at the "5 percent" level).

### Interpretation and policy implications
- R&D tax incentives can substantially lower the R&D user cost and generate sizable increases in qualifying R&D spending: empirical elasticities range from around "-1.55" to "-2.9" depending on experiment and specification.
- Fiscal cost per additional pound of R&D varies by tax rate: estimated bang-for-the-buck between "£0.80" and "£1.30" of additional R&D per "£1" of foregone corporation tax, implying that the policy generated more R&D than the immediate revenue cost for some firms (especially at the small profits rate).
- Robustness checks and direct user-cost regressions support the negative relationship between R&D user cost and R&D spending, although endogeneity concerns remain and alternative user-cost constructions change magnitudes and standard errors.

### Limitations and avenues for future work
- Short sample period prevents analysis of the link between the observed increases in R&D spending and long-run productivity growth or R&D spillovers; these are left for future research.

*IMF Working Paper (excerpt): wp1784 - Section II.*

### Appendix 4. Cash Credits for SMEs

### Appendix 4. Cash Credits for SMEs

### Design and purpose of the cash component
- From its inception, the SME scheme has featured a cash component for companies which do not have taxable profits and hence cannot benefit from the enhanced deduction in the year in which the R&D expenditure has been made.
- HMRC provides a cash refund up to 24 percent of the amount of the total R&D spending of the firm in cash, which is an amount capped by the PAYE or NIC liabilities of the company.
- If the company is not cash constrained, it has an incentive to carry forward its losses and use the full deduction amount in a future period when it becomes profitable; however, a company with liquidity constraints would choose the cash option which can be claimed immediately.

### Evolution of the cash calculation and overall generosity
- The calculation of the cash amount changed over time, which is depicted in Appendix Figure 4.1.
- The total amount of cash available to a company was kept at around 24–25 percent of total R&D spending across periods of different enhanced deduction rates.

### Key statistics
- Cash refund rate: up to 24 percent of total R&D spending.
- Overall cash generosity maintained: around 24–25 percent of total R&D spending.

*Source: Appendix 4. Cash Credits for SMEs (wp1784 - Appendix 4. Cash Credits for SMEs).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp1784.pdf_