Estimating the Incremental Impacts of a Provincial R&D Tax Credit on Business R&D Expenditures Using a Natural Experiment in British Columbia

This publication is also available online in HTML at
http://www.ic.gc.ca/eic/site/eas-aes.nsf/eng/h_ra02224.html. To obtain a copy of this publication or an alternate format (Braille, large print, etc.), please contact:

Economic Research and Policy Analysis Branch
Innovation, Science and Economic Development Canada
C.D. Howe Building
235 Queen Street
Ottawa, ON Canada
K1A 0H5

Telephone: 613-952-6411
Facsimile: 613-991-1261 and
E-mail: ic.erpa_reap.ic@canada.ca

Permission to Reproduce

Except as otherwise specifically noted, the information in this publication may be reproduced, in part or in whole and by any means, without charge or further permission from Industry Canada, provided that due diligence is exercised in ensuring the accuracy of the information reproduced; that Innovation, Science and Economic Development Canada is identified as the source institution; and that the reproduction is not represented as an official version of the information reproduced, nor as having been made in affiliation with, or with the endorsement of, Innovation, Science and Economic Development Canada.

For permission to reproduce the information in this publication for commercial purposes, please contact the:

Web Services Centre
Innovation, Science and Economic Development Canada
C.D. Howe Building
235 Queen Street
Ottawa, ON Canada
K1A 0H5

Telephone (toll-free in Canada): 1-800-328-6189
Telephone (Ottawa): 613-954-5031
Fax: 613-954-2340
TTY (for hearing-impaired): 1-866-694-8389
Business hours: 8:30 a.m. to 5:00 p.m. (Eastern Time)

Email: ic.info-info.ic@canada.ca

© Her Majesty the Queen in Right of Canada,
represented by the Minister of Industry, (2014)
Cat. No. Iu172-1/2013-2E-PDF
ISBN 978-1-100-22790-0

Aussi offert en français sous le titre Utilisation d'une expérience naturelle pour estimer l'impact différentiel du crédit d'impôt provincial pour la R-D sur les dépenses en R-D des entreprises de la Colombie-Britannique.

The views and opinions expressed in the research paper are those of the author alone and do not represent, in any way, the views or opinions of the Department of Innovation, Science and Economic Development or of the Government of Canada.

Footnote *
Innovation, Science and Economic Development Canada

Abstract

The introduction of the 1999 R&D tax credit in British Columbia (BC) is used to estimate the average treatment effect (ATE) of a provincial R&D tax credit on BC enterprises' R&D expenditure. The ATE estimated in this study is the average percentage change in firm-level R&D expenditures before and after the introduction of the BC tax credit compared to the average percentage change from a control group (Alberta R&D performing firms) that had not witnessed any fiscal change in the period studied (1998–2002). Results show that the new tax credit increased average enterprise-level R&D expenditures by 9%–19% between 2000 and 2002, but these ATEs are not statistically different from zero. This suggests weak evidence that the new BC tax credit had some impact on R&D expenditures in BC three years after its introduction. The results are nevertheless consistent with the hypothesis of an inelastic business demand for innovation in Canada.

, revised in 2013

Table of contents

  1. Introduction
  2. Review of Literature
  3. Natural Experiment in Canada
  4. Data
  5. Results
  6. Discussion
  7. Conclusion

1. Introduction

Some recent reports (Council of Canadian Academies, 2009; McFetridge, 2008; Expert Review Panel on Research and Development, 2011) have identified the low level of business innovation in Canada as a major factor explaining the low labour productivity growth relative to the United States and other OECD countries. Identifying the main drivers of innovation is thus an important first step to address the labour productivity growth problem in Canada.

R&D expenditures is an important input in the complex process of innovation. In Canada, indirect support for business R&D is much more intensively used compared to other OECD countries (see Figure 1). The figure also shows that overall support for R&D is relatively high in Canada. However, despite this generous support, business expenditures on R&D (BERD) in Canada trailed behind BERD of other OECD countries (OECD, 2010a). Given the amount of public funds involved, it is important to assess the effectiveness of R&D tax credits to induce new investments in R&D.

Figure 1: Direct and Indirect Government Funding of BERD
OECD Countries, 2008 of latest – As a % of GDP

Graph of Direct and Indirect Government Funding of BERD, OECD Countries, 2008 of latest – As a % of GDP (the long description is located below the image)
Description of Figure 1
Figure 1: Direct and Indirect Government Funding of BERD - OECD Countries, 2008 of latest – As a % of GDP
country Indirect government support through R&D tax incentives Direct government funding of BERD
United States (2008) 0.05 0.18
France (2008) 0.08 0.15
Korea (2008) 0.19 0.15
Czech Republic 0.03 0.13
Spain 0.03 0.12
Sweden 0.00 0.11
Austria 0.09 0.10
Finland 0.00 0.09
Norway (2008) 0.04 0.09
Germany 0.00 0.08
United Kingdom (2008) 0.06 0.08
Iceland (2008) 0.00 0.08
Belgium 0.14 0.07
Luxembourg 0.00 0.05
Denmark (2008) 0.06 0.05
Hungary 0.08 0.05
Australia (2006) 0.05 0.05
Ireland 0.09 0.05
New Zealand 0.00 0.05
Italy 0.00 0.04
Switzerland (2008) 0.00 0.04
Japan 0.12 0.03
Turkey 0.00 0.03
Slovak Republic (2008) 0.00 0.03
Netherlands 0.08 0.02
Canada (2008) 0.22 0.02
Portugal 0.06 0.02
Poland 0.00 0.02
Mexico 0.00 0.01
Greece (2005) 0.00 0.01
  1. The R&D tax expenditures estimates do not cover sub-national R&D tax incentives.
  2. The United States estimate covers the research tax credit but excludes the expensing of R&D
  3. The Austria estimate covers the refundable research premium but exclude other R&D allowances.
  4. Italy, Turkey and Greece provided R&D tax incentives in 2007 but the cost of those incentives was not available.

Source: OECD2010b. Based on national estimates from the Working Party of National Experts in Science and Technology (NESTI) R&D tax incentives questionnaire, January 2010; and OECD, Main Science and Technology Indicators Database, December 2009.

Consequently, this paper focuses on estimating the incremental impact of additional tax credits on enterprise-level R&D expenditures. For this purpose, the introduction in 1999 of the provincial R&D tax credit in British Columbia is used as a natural experiment. The control group consists of enterprises located in Alberta, the most comparable province mainly because no change in R&D fiscal incentives occurred during the period under study. The results of this study are mixed. On average, enterprise-level R&D expenditures in British Columbia increased by 9%–19% between 2000 and 2002 relative to 1998–1999, but these estimates are not statistically different from zero. Results from this paper apply to the British Columbia tax credit only. The results can not be directly applied to other provincial tax credits nor to the federal Scientific Research and Experimental Development (SR&ED) tax credit.

Although they depart from most other results for Canada found in the empirical literature, the results are consistent with the hypothesis of an inelastic demand for innovation in Canada. Since R&D tax credits lower the cost to perform R&D activities, a low response from enterprises could be interpreted as a demand-side instead of a supply-side problem.

The remainder of the paper is organized as follows. The second section provides an overview of the literature on the impacts of R&D tax credits on BERD while the natural experiment used in this study is described in the third section. The fourth section presents the dataset and the fifth the results. The sixth section focuses on the policy implications of the results and the last one concludes.

2. Review of Literature

This section reviews several studies on the impacts of R&D tax credits on BERD and innovation. The first part focuses on Canada while the second emphasizes other countries. To the knowledge of the author, there is no study analyzing R&D tax credit in Canada using a natural experiment. For more comprehensive surveys of fiscal incentives studies, see Hall and van Reenen (2000), Parsons and Phillips (2007) and Brouillette (2010b). Results for subsidies are reported in David et al. (2000), García-Quevedo (2004) and Brouillette (2010b).

2.1 Results for Canada

There is little empirical evidence on the impact of provincial R&D tax credits on R&D expenditures. Mohnen and Baghana (2008) estimate that the short-term price-tax elasticity for manufacturing enterprises in Québec was −0.10 (1997–2003). The long-term elasticity was −0.14 which is much lower than the international average of long-run elasticities (−1) reported by Hall and van Reenen (2000).

Mohnen and Baghana also compute an incremental ratio (IR) associated with a 10% increase in tax credit rates. The IR measures the change in R&D per dollar of tax revenue foregone. In the literature, it is generally considered that a successful policy induces an incremental ratio above one. Their simulations show this is the case for the level-based R&D tax credit for small enterprises (less than 50 employees) and for the short-lived Québec incremental R&D tax credit. However, the incremental ratio for large enterprises would fall below one only a few years after the change. This inefficiency is partly due to the presence of a dead weight loss resulting from subsidizing activities that would have been conducted anyway. Nevertheless, Lokshin and Mohnen (2009b) argue that a policy with an incremental ratio below one is not necessarily cost-ineffective if spillovers are large enough.

The efficiency of level-based provincial R&D tax credit is also challenged by Dahlby (2005). He develops a simple framework to determine the necessary conditions to justify the implementation of a provincial R&D tax credit. Under the assumption that the marginal cost of public funds is $1.40 for each dollar of tax collected, he estimates that the incremental ratio would have to be about $2 and the external rate of return on R&D projects around 30% to justify the existence of a provincial R&D tax credit.

At the federal level, estimated price-tax elasticities and incremental ratio suggest a positive relation between SR&ED and R&D expenditures. Short-term (long-term) price-tax elasticities estimated by Bernstein (1986b) and Dagenais et al. (2004) are, respectively −0.07 (−0.32) and −0.13 (−1.09). All incremental ratio studies report a ratio close to or higher than one: $0.98 (Dagenais et al., 2004), $1.38 (Department of Finance and Revenue Canada, 1997) and $0.83–$1.73 (Bernstein, 1986a). Similarly, Klassen et al. (2004) estimate that a one percentage point increase in R&D tax credits (federal and provincial combined) is associated with a 1.30% increase in enterprise-level R&D expenditures.

Other evidence suggests that SR&ED positively affects other dimensions of business innovation. Czarnitzki et al. (2004) report that SR&ED has increased by 29 percentage points the incidence of R&D activities in Canadian enterprises between 1997 and 1999. They also find that recipients of SR&ED are more successful in commercializing new product innovations. Bérubé and Mohnen (2009) also find that R&D subsidies increase the effectiveness of SR&ED, i.e. that enterprises who received both R&D subsidies and SR&ED introduce more product innovations and have more commercial success relative to enterprises only claiming SR&ED.

Most of the aforementioned analyses ignore factors such as the financing of distortionary taxes, spillovers and administrative and compliance costs which also affect the cost-benefit welfare analysis of tax credits. Taking these factors into account, Parsons and Phillips (2007) estimate that for each dollar of forgone tax revenue, the SR&ED tax credit yields a small positive net welfare effect of about 11 cents. However, Russo (2004) conducts a similar welfare analysis and concludes that increasing SR&ED rates is the most effective way to increase research effort and welfare in Canada compared to other fiscal policy such as lowering corporate income taxes or increasing investment tax credits for high-tech inputs. He also finds that an incremental SR&ED tax credit would dominate the actual level-based SR&ED in terms of welfare gains in the long run, a result consistent with the conclusion of Mohnen and Baghana (2008) for the Québec incremental R&D tax credit.

2.2 Results for Other Countries

This section emphasizes the results of three most relevant foreign natural experiments that use enterprise-level data, two for the United States and one in Norway. More details on quantitative impacts for other countries are reported in Table A1 of Appendix A.

Natural Experiments
Paff (2005) uses the 1997 increase in R&D tax credits in California as a natural experiment. She finds that the higher R&D tax credits generated, on average, a 40% increase of R&D expenditures per year for enterprises in California between 1994–1996 and 1997–1999. Her analysis is limited to enterprises in pharmaceutical and prepackaged software industries. The control group consists of enterprises in Massachusetts, a state comparable to California in terms of R&D activities in the selected industries.

Another natural experiment is the change of the U.S. federal tax credit incremental basis in 1989. This change was made to increase the eligibility to the tax credit of enterprises with rapid R&D expenditures growth and low sales. Gupta et al. (2006) show that after 1989, the eligibility of high-tech enterprises increases relative to other enterprises. In their analysis, they define the treatment group as eligible U.S. high-tech enterprises. The implicit assumption is that their R&D intensity would evolve at the same pace as for eligible U.S. low-tech enterprises in the absence of the reform. They find that the base change is associated with a $2.10 incremental ratio for high-tech enterprises for the 1990–1994 period.

The last natural experiment reviewed in this section is the introduction of a R&D tax credit in Norway in 2002 (Hægeland and Møen, 2007). The tax credit is characterized by an annual cap of 4 million kroner on eligible R&D. The identification strategy consists in comparing enterprises with R&D expenditures over and below the cap. Such a comparison is valid if R&D expenditures of enterprises above and below the cap would grow at the same rate in the absence of the introduction of the R&D tax credit. Hægeland and Møen estimate that the incremental ratio is around 2.

Other Results
Most other enterprise-level evidence suggests that R&D tax credits have a positive impact on business innovation in France (Duguet, 2008; Mairesse and Mulkay, 2004), Japan (Koga, 2003), Norway (Cappelen et al., 2007, 2008) and the Netherlands (Lokshin and Mohnen, 2007, 2009a). See Table A1 for details.

Results from state-level studies in the United States are mixed. Wu (2005, 2008) finds that the introduction of state R&D tax credits had a positive impact on business innovation, but Wilson (2007) reports that state fiscal incentives encourage relocation of R&D activities between states. For OECD countries, the range of long-run price-tax elasticities reported by Bloom et al. (2002), Falk (2006) and McKenzie and Sershun (2005) is −0.5 to −1. Finally, Guellec and van Pottelsberghe de la Potterie (2003) report a slightly lower elasticity (−0.31) and also that tax incentives and direct government funding are substitutes, not complements.

3. Natural Experiment in Canada

A natural experiment is usually a policy change that affects some enterprises (treatment group) while being irrelevant for other enterprises (control group). It is also required that no other, or at least as few as possible, policy changes affect enterprises during the experimental period because this could make it difficult to disentangle the effects of several policy changes.

Canadian R&D fiscal incentives are complex because provincial and federal governments have several non-exclusive provisions. Combined with frequent policy changes, a natural experiment in Canada must be carefully defined. In this paper, the introduction of a provincial R&D tax credit in British Columbia in 1999 is used as a natural experiment. The counterfactual group consists of enterprises located in Alberta. It is the most comparable available province over the experimental period (1997–2002) as explained below.

3.1 The Treatment Group

The first thing to consider is where BERD is performed in Canada. Table 1 shows that, between 1997 and 2004, four provinces accounted for 97% of BERD in Canada: Ontario, Québec, Alberta (AB) and British Columbia (BC). As the total share of these four provinces is high, only these four provinces will be considered for this natural experiment.

Second, changes in provincial R&D tax credit schemes have to be considered. Québec and Ontario are interesting candidates due to their high share of BERD in Canada. Unfortunately, many major changes to R&D tax credits occurred in these provinces since their introduction in the late '80s (see Table 2), meaning that Ontario and Québec R&D fiscal regimes hardly meet the criterions for a natural experiment.

R&D tax credits were introduced in British Columbia in 1999 and in Alberta 2009 (the latter is not shown in Table 2). Unfortunately, it is too early to assess the impact of the Alberta tax credit. The case of British Columbia is more promising. This 10% tax credit is refundable and applies to all SR&ED eligible expenditures performed in BC. No R&D tax credit was in place before 1999 and no change occurred after 1999, although the tax credit is supposed to be repealed in 2014. This means that the experiment could span over several years before and after 1999. Having identified a potential treatment group, enterprises located in BC, the next critical step is to find a valid counterfactual. This issue is discussed in the next section.

Table 1: Provincial Share of Total BERD in Canada, Selected Provinces
Year Ontario Quebec Alberta British Columbia Subtotal of
these four
Provinces
1997 55% 28% 6% 7% 97%
1998 57% 28% 6% 6% 97%
1999 55% 30% 4% 7% 97%
2000 58% 28% 5% 7% 97%
2001 55% 30% 5% 8% 97%
2002 53% 31% 6% 8% 97%
2003 53% 31% 6% 8% 97%
2004 52% 29% 6% 9% 96%
Average 55% 29% 5% 7% 97%
Source: Statistics Canada (2000, 2001a, 2001 b, 2002, 2003, 2005, 2006, 2007).
Table 2: Changes in R&D Fiscal Incentives in Canada, 1987–2007
Year Canada Quebec Ontario British Columbia Alberta
1987 Symbol representing a star Symbol representing a star
1988 Symbol representing a star Symbol representing a star
1989 Symbol representing a star
1990
1991 Symbol representing a star Symbol representing a star
1992 Symbol representing a star Symbol representing a star Symbol representing a star
1993 Symbol representing a star Symbol representing a star
1994 Symbol representing a star
1995 Symbol representing a star Symbol representing a star Symbol representing a star
1996 Symbol representing a star
1997 Symbol representing a star
1998 Symbol representing a star Symbol representing a star
1999 Symbol representing a star Symbol representing a star Symbol representing a star
2000 Symbol representing a star
2001 Symbol representing a star Symbol representing a star
2002
2003 Symbol representing a star Symbol representing a star Symbol representing a star
2004 Symbol representing a star Symbol representing a star
2005 Symbol representing a star Symbol representing a star
2006 Symbol representing a star Symbol representing a star
2007 Symbol representing a star Symbol representing a star Symbol representing a star
The star indicates the year in which a change in federal or provincial R&D fiscal incentives occurred.
Source: Madore (2006), Québec's Additional Information on the Budgetary Measures (1987–2002), Ontario Budget Papers (1987–2007) and British Columbia Department of Finance website. For more information, see Brouillette (2010a).

3.2 The Control Group

Enterprises in Québec and Ontario can not be included in the control group for the same reason as these provinces can not be included in the treatment: there were too many changes in these provinces R&D tax credits. Consequently, the only province that can effectively be used as a counterfactual is Alberta.

There are three reasons why Alberta is a valid counterfactual to British Columbia. First, as shown in Table 1, the share of both provinces in total BERD in Canada are similar (5.4% vs 7.4%). Second, the economies of both provinces rely significantly on natural resources. Third, the Government of Alberta never granted fiscal incentives to support BERD before 2009, a situation identical to the pre-reform period in BC. Consequently, as far as Alberta alone is concerned, the natural experiment could span over a long period before 1999 and up to 2009.

3.3 Scientific Research and Experimental Development (SR&ED) Tax Credits

The SR&ED tax credit rate is 35% for small Canadian-controlled private corporations (CCPCs) applicable on all eligible current and capital expenditures. Depending on the type of business and expenditures, SR&ED tax credits are refundable (see Table A2 in Appendix A). For other corporations, the rate is 20% and the credits are not refundable.

It is a priori irrelevant to care about SR&ED when comparing provinces since SR&ED applies equally to them. However, a natural experiment is best defined in a stable environment, so it is desirable to chose an experimental period with few changes to the federal fiscal incentives. No changes in SR&ED rates were implemented since 1987, but some other changes occurred in 1993–1995, 1998, 2000 and 2003 as shown in Table 2. Table A3 of Appendix A shows the details. The thresholds for CCPCs were changed in 1993 and 2003, so 1994–2002 is potentially a valid experimental period.

Changes in 1994 and 2000 concern the special Atlantic tax credit and the provincial super-deductions which are irrelevant for Alberta and British Columbia. However, the changes in 1995 and 1998 affected both provinces, but it is difficult to find evidence on whether the impacts of these changes, if any, were different across provinces. In this analysis, it is thus assumed that the 1995 and 1998 changes (see Table A3) had the same impact in both provinces, meaning that the difference-indifference (DD) estimator will eliminate these effects.

In conclusion, based on the provincial and federal fiscal incentives for R&D, the experiment could span from 1994 to 2002. However, the last factor to consider is data availability, which is the subject of the following section.


top of page

4. Data

4.1 The Database

The data used in this analysis comes from the Statistics Canada Research and Development in Canadian Industry (RDCI) database. It is a rich source of information because it consists of a quasi-census of enterprises performing R&D in Canada.

From 1997 to 2008, large performers, i.e. enterprises with annual R&D expenditures over $1 per year ($1.5M after 2005), received the detailed RDCI questionnaire. Information for small performers come from administrative tax data collected by the Canadian Revenue Agency (CRA). Consequently, only enterprises conducting R&D activities and claiming SR&ED tax credits are included in the database.

Essential to the present project is the availability of total intramural R&D expenditures (current and capital) by province. This information is readily available (self-reported) for large performers through their responses to the questionnaire. For small performers, the SR&ED form does not require the claimant to allocate qualified expenditures between provinces, so it is assumed that all their R&D expenditures are performed in only one province.

Important changes in the RDCI methodology occurred in 1997 and 2008. Prior to 1997, all R&D enterprises performing or funding R&D in Canada were surveyed. From 1997 to 2007, only large performers were surveyed and R&D for small performers were imputed from administrative data. In 2008, the $1.5M threshold was removed and a stratified sample strategy was implemented to achieve a better industry coverage, in particular for the service sector. This suggests that enterprise-level R&D expenditures are not comparable before and after 1997. Given the discussion on Canadian R&D fiscal incentives presented in the previous section, the experimental period begins in 1997 and ends in 2002. The first post-reform year is 2000 because the BC tax credit became effective in September 1999, meaning that all R&D activities performed before September 1999 were not eligible to the tax credit.

4.2 The Sample

Several points were considered to construct the sample. For each enterprise, at least one year of data in the pre-reform period (1997–1999) and one year in the post-reform period (2000–2002) should be available for the DD estimator to be valid. Moreover, in one empirical specification, the lagged dependent variable is instrumented by the two-periods lagged dependent variable. This implies that each year of data included in the regression requires three consecutive years of data. Combining these requirements, observed data must match one of the three following patterns for an enterprise to be included in the sample: 1997–2000, 1997–2001 or 1997–2002.

R&D funders, i.e. organizations financing R&D activities without performing R&D activities themselves, have been excluded because it is not known where the R&D was performed. Non-profit organizations were also excluded as well as enterprises with positive R&D expenditures in both BC and AB. The last restriction is necessary because the province of location is the key variable in identifying the average treatment effect. Enterprises located outside BC or AB but performing some R&D activities in BC or AB were also removed to eliminate the relocation effect, but a robustness analysis was conducted to assess the impact of including them (see Section 5.4). Finally, a search for extreme or inconsistent values has been performed (see Table A4 in Appendix A). The final sample contains 230 enterprises. It consists of a six years (1997–2002) unbalanced panel with 1212 observations (enterprise/year), 432 located in AB and 780 located in BC.

4.3 Descriptive Statistics

The dependent variable is enterprise-level intramural R&D expenditures, deflated using the GDP deflator (CANSIM Table 380-0056). R&D expenditures in level are used instead of R&D intensity because the latter might change even if R&D expenditures are constant.

Figure 2 shows the average R&D expenditures in BC and in AB during the experiment. Note that the averages presented in Figure 2 are different from the official numbers published by Statistics Canada (see e.g. Statistics Canada, 2007) because of the restrictions imposed on the sample (see Section 4.2). The Y-axis was suppressed to preserve confidentiality.

Figure 2: Average R&D Expenditures 1997–2002

Graph of Average R&D Expenditures 1997–2002, in B.C. and Alberta (the long description is located below the image)

Source: Research and Development in Canadian Industry database.

Description of Figure 2

The figure shows the average enterprise-level research and development (R&D) expenditures for enterprises located in Alberta and British Columbia over the 1997–2002 period. Average R&D expenditures for British Columbia increased from 1997 to 1998, decreased slightly between 1998 and 1999, and increased from 1999 to 2002. Average R&D expenditures increased sharply from 2000 to 2002. Average R&D expenditures for Alberta decreased from 1997 to 1999 and then increased from 1999 to 2002. As for British Columbia, the increase was sharper for the 2000–2002 period. Compared to average R&D expenditures in British Columbia, average R&D expenditures in Alberta were higher in 1997 and 1999 and lower afterward. However, average R&D expenditures were almost the same in both provinces in 2002.

Looking at Figure 2, average enterprise-level R&D expenditures are higher in AB than in BC in 1997 and 1998, but the picture reverses after 1998 and the two lines converge in 2002. It is possible that a combination of macro-economic conditions and industrial mix can explaining this trend as discussed in Section 5.4.

A drawback of the RDCI database is the limited number of control variables available. In addition to time, province and industry dummy variables, two enterprise-level variables are included in the empirical specifications: R&D personnel (PERS) in the province (BC or AB) and total employment in Canada (EMPL). PERS includes all R&D-related workers, both scientific and administrative. These variables are introduced to control for the R&D capacity and the size of the enterprise.

Descriptive statistics for the control variables are shown in Table 3.

On average over 1997–2002, enterprises in AB tend to be larger than enterprises in BC as measured by the total number of employees in Canada. R&D personnel is comparable between provinces. In terms of the provincial industrial mix, enterprises in BC seemed to be slightly more oriented toward high-tech activities.

Table 3: Sample Descriptive Statistics 1997–2002
Variable Alberta
N = 432
British Columbia
N = 780
Mean S.D. Mean S.D.
Total Employment (EMPL) 267.03 1008.44 119.51 501.25
R&D Personnel (PERS) 9.23 18.89 10.66 23.02
Industrial CompositionFootnote 2
High-Tech (Group1) 5.32% 8.72%
Med/High-Tech (Group2) 44.21% 47.05%
Med/Low-Tech (Group3) 40.05% 26.03%
Low-Tech (Group4) 10.41% 18.21%
N : number of enterprise-year.
Research and Development in Canadian Industry database.
S.D.: Standard Deviation

4.4 Requirements for DD Estimator

As mentioned in the introduction, a DD estimator with unobservable effects is used to evaluate the impacts of the new BC tax credit. This means that the following two assumptions must hold (see e.g. Blundell and Costa Dias, 2002): 1) there is a common time trend in both provinces; and 2) there is no change in groups composition over time.

The implication of the first assumption is that, except for the new tax credit, all other macroeconomic shocks have the same impact in both provinces. For example, in this model, it is impossible to include a province-specific trend for industries because this would not be separately identified from the treatment effect.

The implication of the second assumption is that, a change in firms' distribution before and after the BC R&D tax credit introduction would create a bias because the treatment and control groups would not be comparable anymore.

Since the seminal works of Aghion and Howitt (1992), Grossman and Helpman (1993) and Romer (1990), firms heterogeneity plays an important role in explaining the differences in firms' ability to generate business innovation.

However, one possible interpretation for the second assumption to hold is if it is assumed that the distribution of firms heterogeneity does not vary during the experiment.

The distribution of heterogeneity is unfortunately unknown. Indirect evidence on whether this assumption holds can be gathered by testing the stability over time of observable variables distributions (Table 4).

Table 4: P-Values for Tests on Distribution Over Time

Table 4: P-Values for Tests on Distribution Over Time - Tests by ProvinceFootnote 3 (Reference = 1997)
Year Industrial CompositionFootnote 4 EMPLFootnote 5 PERSFootnote 5
AB BB AB BC AB BC
1998 0.850 0.966 1.000 0.663 0.903 0.376
1999 0.940 0.899 0.290 0.139 0.392 0.078Footnote c
2000 0.965 0.936 0.653 0.139 0.515 0.010Footnote a
2001 0.510 0.637 0.122 0.156 0.143 0.008Footnote a
2002 0.264 0.957 0.021Footnote b 0.134 0.116 0.017Footnote b
Table 4: P-Values for Tests on Distribution Over Time Part 2 -Tests Between Provinces by YearFootnote 6
Year Industrial CompositionFootnote 4 EMPLFootnote 5 PERSFootnote 5
1997 0.833 0.094Footnote c 0.953
1998 0.954 0.137 0.602
1999 0.652 0.241 0.332
2000 0.776 0.735 0.247
2001 0.687 0.683 0.883
2002 0.237 0.173 1.000
It was not possible to disclose the share by province and year due to confidentiality issues.

The P-values reported in the upper part of Table 4 show that for each year considered, the provincial distribution of industrial groups is not systematically different from the distribution in 1997. The same conclusions apply for the other variables, except for PERS in BC. However, the P-values in the lower part of the table show that there is no difference for a given year between province for all distributions including PERS.

Overall, there are few significant changes in time and between provinces in the distribution of key explanatory variables. This is good news, although it is not a formal test on the distribution of the unobserved effects. One way to temper this issue is to include time constant unobserved effects in the regressions. At least, these time-constant effects will be washed out by the DD estimator.

5. Results

5.1 Empirical Specifications

The impacts of the introduction of the BC provincial tax credit are estimated using panel regressions with time-constant enterprise-specific unobserved effects. The general empirical specification is given by Equation (1).

Equation 1

Graph of Formula showing Empirical Specifications of impacts of the introduction of the BC provincial tax credit (the long description is located below the image)
Description of Equation 1

natural logarithm RD subscript it equals delta subscript 0 plus alpha open big square bracket for vector three by one first line natural logarithm RD second line PERS third line EMPL close big square bracket subscript it-1 plus gamma open big square bracket for vector three by one first line Group subscript 1 second line Group subscript 2 third line Group subscript 3 close big square bracket subscript it plus beta open big square bracket for vector seven by one first line BC second line Y subscript 2000 third line Y subscript 2001 fourth line Y subscript 2002 fifth line Y subscript 2000 multiplied by BC sixth line Y subscript 2001 multiplied by BC seventh line Y subscript 2002 multiplied by BC close big square bracket subscript i plus open big square bracket vector two by one first line eta subscript i second line epsilon subscript it close big square bracket

where i denotes the enterprise, t the time, ηi the unobserved effect and εit the usual error term. The dependent variable is the natural logarithm of enterprise-level R&D expenditures (ln RDit). The past lagged dependent variable (ln RDit−1) is included to capture the persistence of enterprise-level R&D. PERS, EMPL and the industrial groups are defined as in the last section and low-tech industries (Group4) are the reference group. All these variables come from RDCI.

In a panel regression, it is widely acknowledged that including lagged endogenous variables as regressors is likely to bias the estimated parameters (Baltagi, 2001). Therefore, in one specification, ln RDit−1 and PERSit−1 are instrumented by ln RDit−2 and PERSit−2 using a simple instrumental variables technique (XTIVREG in Stata).

BC is a province indicator equals to 1 for enterprises located in BC. Yj are time dummy variables for 2000, 2001 and 2002, 1998–1999 being the reference years. The parameters of interest are βY jBC , i.e. the interaction parameters between time and province. If the necessary assumptions for the DD estimators are met, these parameters give the impacts of the introduction of the BC R&D tax credit on ln RDit. Following Kennedy (1981), the average treatment effect (ATE) on the level of R&D expenditures (RDit) is

Equation 1

Graph of Formula showing average treatment effect (ATE) on the level of R&D expenditures (the long description is located below the image)
Description of Equation 1

ATE subscript j equals open big square bracket exponential open big parenthesis beta subscript YjBC minus one over two V open parenthesis beta subscript YjBC close parenthesis close big parenthesis close big square bracket multiplied by one hundred

with j = (2000, 2001, 2002) and V (βY jBC ) is the variance of βY jBC . AT Ej gives the average percentage change on the level of R&D expenditures by BC R&D performing firms in year j relative to R&D spending in 1998–1999, compared to the average percentage change by Alberta R&D performing firms before and after the introduction of the R&D tax credit in BC.

5.2 Estimated Parameters

In a first step, Equation (1) is estimated without instrumenting ln RDit−1 and PERSit−1. Table A6 in the Appendix presents the results for fixed effects (FE) regressions. FE were chosen over random effects (RE) based on the results of a Hausman test (P-value = 0.0000). Moreover, the estimated variance of the unobserved effects ηi in all RE specification is negative. According to Wooldridge (2002, p. 261), this is a sign that the error terms are negatively serially correlated, which violates the RE model assumptions.

Sample I corresponds to the unbalanced panel sample described in section 4.2. Sample II consists in a balanced sub-sample of Sample I. Only enterprises performing R&D activities in every year between 1997 and 2002 are included. In Samples III and IV, the Sample I is divided between large performers (surveyed enterprises) and small performers (data coming from the CRA administrative tax files).

Overall, there are very few significant parameters in the FE models reported in Table A6 except for ln RDit−1. Given the log-log empirical specification, this parameter is the short-term elasticity of past R&D on current R&D. The estimated elasticities, ranging from 0.1 to 0.2, are consistent to what is reported by Hægeland and Møen (2007) for Norwegian enterprises (0.10).

The within regression R2 is quite low and a χ2 test does not reject the null hypothesis that all parameters except ln RDit−1 and PERSit−1 are equal to zero for all models. This has to be kept in mind when interpreting the ATEs. However, even if they are not significant, most estimated parameters have the expected sign for ln RDit−1 , PERSit−1 and EMPLit−1. Provincial GDP has been initially included in the specifications to capture the effect of overall economic activity in each province, but has to be dropped due to collinearity with year and time interaction. Industrial groups and province dummy variables were also dropped as time-constant variables can not be included in FE regressions.

5.3 Estimated ATEs

Basic Model
Because the RDCI database does not include information on whether enterprises in BC have claimed the provincial R&D tax credit, the results must be interpreted as the average effect of being offered the opportunity to claim the provincial tax credit.

ATEs were computed using the interaction parameters from Table A6. ATEs are represented in Figure 3 by the navy dot. The two green squares (red hollow diamonds) denote the upper and lower bound of the 10% (5%) confidence interval computed using a block bootstrap procedure with 500 replications as described in Cameron and Trivedi (2005). The reference period is 1998–1999.

For the unbalanced sample (Sample I), ATEs indicate that the reform has increased enterprise-level R&D expenditures by 10%–20% over the 2000–2002 period as shown in Figure 3a. However, all the confidence intervals overlap the zero line, thus none of these ATEs are significantly different from zero.

Results for the balanced panel, i.e. the sample of persistent performers, are presented in Figure 3b. The estimated ATEs suggest that the BC tax credit did not have much positive impacts on R&D activities of these enterprises. Indeed, the only significant ATE is negative and the other are close to zero.

The latter results call for three comments. First, since the main effect of R&D tax credits is to reduce the cost of performing R&D, this suggests that the demand for R&D of persistent performers is inelastic. Second, it is also possible that persistent performers R&D investments are mainly driven by long-term business strategies. In other words, decisions to invest in R&D are based on expected future benefits associated with these strategies rather than on short-term variation of R&D tax credit rates. Unfortunately, the reduced-form model used in this analysis is not informative to differentiate between those two hypotheses. Third, the lack of additional investments in R&D highlights a major drawback of level-based tax credits which is to subsidize activities that would have been conducted anyway.

Figures 3c and Figure 3d contrast the ATEs for large and small performers. The story for large performers is similar to the one for persistent performers: the impacts are close to zero (to a lesser extent in 2001) and non-significant. However, small performers seem more sensitive to R&D tax credit. Indeed, ATEs are around 20% in 2000 and 2001, although they are not significantly different from zero.

Figure 3(a): Basic Model ATEs – Reference Years = 1998–1999 - Unbalanced Panel
(Sample I)

Graph of Basic Model ATEs – Reference Years = 1998–1999 - Unbalanced Panel - (Sample I) (the long description is located below the image)
Description of Figure 3(a)
Unbalanced Panel (Sample I)
Year ATE (Blue Dot) Lower Bound of the 5% Confidence Interval (Lower Red Diamond) Upper Bound of the 5% Confidence Interval (Upper Red Diamond) Lower Bound of the 10% Confidence Interval (Lower Green Square) Upper Bound of the 10% Confidence Interval (Upper Green Square)
2000 9.36 −7.908251 22.71107 −4.063813 20.13136
2001 19.22 −5.571737 39.85341 −0.910302 36.50937
2002 11.7 −14.92195 33.14993 −8.386572 29.70828

Figure 3(b): Basic Model ATEs – Reference Years = 1998–1999 - Balanced Panel
(Sample II)

Graph of Basic Model ATEs – Reference Years = 1998–1999 - Balanced Panel - (Sample II) (the long description is located below the image)
Description of Figure 3(b)
Balanced Panel (Sample II)
Year ATE (Blue Dot) Lower Bound of the 5% Confidence Interval (Lower Red Hollow Diamond) Upper Bound of the 5% Confidence Interval (Upper Red Hollow Diamond) Lower Bound of the 10% Confidence Interval (Lower Green Square) Upper Bound of the 10% Confidence Interval (Upper Green Square)
2000 -12.31 −27.30802 −0.3249623 −25.52378 -2.626566
2001 6.33 −21.76422 33.23296 −16.35881 26.09813
2002 2.12 −23.706 23.39614 −20.50728 18.47101

Figure 3(c): Basic Model ATEs – Reference Years = 1998–1999 - Survey
(Sample III)

Graph of Basic Model ATEs – Reference Years = 1998–1999 - Survey -(Sample III) (the long description is located below the image)
Description of Figure 3(c)
Survey (Sample III)
Year ATE (Blue Dot) Lower Bound of the 5% Confidence Interval (Lower Red Hollow Diamond) Upper Bound of the 5% Confidence Interval (Upper Red Hollow Diamond) Lower Bound of the 10% Confidence Interval (Lower Green Square) Upper Bound of the 10% Confidence Interval (Upper Green Square)
2000 5.773009 −20.6981 31.08926 −14.99045 25.37798
2001 10.28864 −18.43271 38.46165 −14.39147 32.95726
2002 −5.66201 −29.80531 18.92475 −25.6592 13.99949

Figure 3(d): Basic Model ATEs – Reference Years = 1998–1999 - CRA Enterprises – (Sample IV)

Graph of Basic Model ATEs – Reference Years = 1998–1999 - CRA Enterprises -(Sample IV) (the long description is located below the image)
Description of Figure 3(d)
Figure 3(d): CRA Enterprises (Sample IV)
Year ATE (Blue Dot) Lower Bound of the 5% Confidence Interval (Lower Red Hollow Diamond) Upper Bound of the 5% Confidence Interval (Upper Red Hollow Diamond) Lower Bound of the 10% Confidence Interval (Lower Green Square) Upper Bound of the 10% Confidence Interval (Upper Green Square)
2000 2.98782 −17.81985 17.92142 −12.80089 14.30753
2001 20.12102 −8.782679 46.45125 −4.5243 41.58513
2002 21.09487 −11.89383 50.83244 −6.794808 44.06605

While the comparison of large and small performers is interesting, it is not clear why small performers appear more reactive to R&D tax credits. A small performer is not necessarily a small enterprise, performer status is based upon R&D expenditures not size of enterprise. Therefore, explanations like financial constraints on credit market or the greater generosity of SR&ED for small enterprises (see Table A2) are not likely to apply here. More research is definitely needed to answer this question.

Overall, there is weak evidence that the introduction of the BC R&D tax credit in 1999 had little positive impact on R&D expenditures of BC enterprises between 2000 and 2002. Some positive impacts are found, e.g. for small performers, but they are not statistically different from zero. Other estimates are either close to zero or negative.

Receipt of Direct Government Support
The RDCI database also contains information on whether enterprises received direct support from the government. This includes grants from provincial and federal governments as well as federal government contracts. ATEs for enterprises who did not receive direct support and for those who did for at least one year between 1997 and 2002 are presented, respectively, in Figure 4a and Figure 4b. Note that these results are conditional upon receiving (or not) direct government support. In other words, like in the case of large and small performers, the results come from two different regressions. Estimated parameters are presented in Table A7.

Figure 4(a): Basic Model ATEs – Reference Years = 1998–1999 - Without Subsidies – (Unbalanced)

Graph of Basic Model ATEs – Reference Years = 1998–1999 - Without Subsidies - (Unbalanced) (the long description is located below the image)
Description of Figure 4(a)
Basic Model ATEs Without Subsidies (Unbalanced)
Year ATE (Blue Dot) Lower Bound of the 5% Confidence Interval (Lower Red Diamond) Upper Bound of the 5% Confidence Interval (Upper Red Diamond) Lower Bound of the 10% Confidence Interval (Lower Green Square) Upper Bound of the 10% Confidence Interval (Upper Green Square)
2000 −0.9909723 −24.06787 17.38173 −18.85968 13.11743
2001 3.624745 −25.50663 28.97592 −18.83067 24.15689
2002 34.24704 −7.063188 74.13982 0.8069816 64.09592

Figure 4(b): Basic Model ATEs – Reference Years = 1998–1999 - With Subsidies – (Unbalanced)

Graph of Basic Model ATEs – Reference Years = 1998–1999 - With Subsidies -(Unbalanced) (the long description is located below the image)
Description of Figure 4(b)
Basic Model ATEs With Subsidies (Unbalanced)
Year ATE (Blue Dot) Lower Bound of the 5% Confidence Interval (Lower Red Diamond) Upper Bound of the 5% Confidence Interval (Upper Red Diamond) Lower Bound of the 10% Confidence Interval (Lower Green Square) Upper Bound of the 10% Confidence Interval (Upper Green Square)
2000 32.79523 −0.1339879 60.75795 3.728635 55.3472
2001 32.0663 −3.135853 66.62777 2.879223 58.7019
2002 −21.17645 −47.34653 6.071978 −43.03597 -1.686669

Estimated ATEs for enterprises without direct support are nil for 2000 and 2001. In contrast, enterprises with subsidies seem more affected by the introduction of the BC tax credit. ATEs for this group are around 30% and significant, suggesting that direct and indirect support for R&D are complements. This evidence is consistent with the findings of Bérubé and Mohnen (2009) who report a similar correlation using data from Statistics Canada's Survey of Innovation 2005.

However, for both groups, the relation reverses in 2002: the ATE is positive for enterprises without subsidies and negative for enterprises with subsidies. Both effects are statistically significant at the 0.10 level. It is difficult to explain these effects within this framework, but it could be related to some macro-economic factors challenging the validity of the experiment as discussed in the next section. More research is definitely needed to investigate the link between direct and indirect support for R&D at the enterprise-level in Canada.

5.4 Caveats

This section emphasizes on some important issues affecting the validity and robustness of the estimates such as assumptions made, data limitations and the choice of the empirical specification.

Timing of the experiment
A major concern for this study is the timing of the experiment. The burst of the ICT bubble in 2001 could have impacted more severely BC's economy than AB's. In parallel, the oil boom in AB could have had the reverse impact. This hypothesis is supported by Figure 1 where AB provincial BERD caught up with BC in 2002. Therefore, the combination of these two macroeconomic shocks could challenge the common trend assumption. However, the BERD share of BC and AB in Canada remains constant during the experimental period (Table 1) and to the extent that province and time dummy variables (in place of GDP) capture a good part of these macroeconomic factors, the experiment is valid. Unfortunately, no other province can be used as a control group. As explained in Section 3, the ATEs in BC would not be identified if Québec or Ontario were included.

The focus of this paper is on R&D tax credits, but other public policies can also affect R&D. A good example is corporate taxation. As reported by McKenzie and Sershun (2005), high production taxes have a negative impact on business innovation that partly nullify the positive effect of R&D tax credits. Another example is government spending on innovation. If the government of Alberta spent more on higher education in response to the increase in revenue due to the oil boom, this could have increased BERD if spillovers effects were strong enough. Consequently, a thorough analysis of these factors should be undertaken to strengthen the conclusion of this study.

Instrumental Variables
The results from the instrumental variables (IV) specifications (with the unbalanced panel) are problematic. First, a Hausman test suggests that the IV RE model is a better choice than the IV FE model (P-value = 0.997). However, all specifications estimated with IV RE yield a negative estimated variance for the firm-specific unobserved effects ηi even when using the Swamy and Arora (1972) correction for small unbalanced panel.

Second, the coefficients estimated by IV FE for some variables are dubious. As shown in the last column of Table A6, the estimated coefficients for ln RDt−1 is equal to 3.21 although it is not statistically significant. This short-term BERD elasticity seems high and it is in stark contrast with the results of non-instrumented FE regressions. Beside, both estimated coefficients for PERSt−1 and EMPLt−1 are negative, but not statistically different from zero. Finally, the ATEs for this model (28%, 62% and 0% for 2000, 2001 and 2002) are not comparable with ATEs in Figure 3a.

A possible explanation for the poor performance of IV is the small number of variables included. The RDCI database includes few non-R&D enterprises characteristics. The instrumental procedure may simply impose to much structure on the model for the information available in the dataset. It could also be the case that other IV methods would be more appropriate.

Enterprises Located Outside BC or AB
Enterprises not located in BC or AB, but performing some R&D activities in these provinces, were excluded from the previous analysis in an attempt to remove any relocation effect. This effect is present if large performers conducting R&D activities in multiple establishments scattered across Canada relocate their R&D activities between provinces in response to changes in provincial fiscal incentives.

No study investigates this issue in Canada, but Wilson (2009) reports the presence of a strong sub-national R&D relocation effect in the United States due to higher U.S. state R&D tax credit rates. His results show that higher state rates have a zero impact on overall U.S. BERD as increases observed within states are compensated by decreases in neighbour states. Moreover, Bloom and Griffith (2001) report that BERD in one country is sensitive to the price of R&D in its major commercial partners.

The presence of a relocation effect in Canada between 2000 and 2002 can not a priori be ruled out. In 2001, Québec and Ontario abolished their super-deduction on R&D expenditures. Given that the other fiscal incentives remained the same in these provinces, the introduction of the BC tax credit could have increased the attractiveness of BC as a place to perform R&D.

Panel B of Table 5 shows the effect on ATEs of including enterprises located outside BC or AB who perform R&D activities in BC or AB. Only five enterprises were added and compared to the results in Panel A (Figure 3a), the most marked differences are for the sample of surveyed enterprises (Sample III). This was to be expected because these five extra enterprises are large performers and FE regressions are sensitive to outliers. Note that the potential relocation effect from BC/AB to other provinces is ignored because large BC/AB performers with R&D activities in other provinces could not be identified. This suggests that one should be cautious when analyzing the investment behaviour of large R&D performers, especially because R&D expenditures are highly concentrated in Canada (see e.g. Statistics Canada, 2010).

Table 5: Comparison of ATEs
Model Year Sample
Unbalanced Balanced Survey CRA
A Basic Model (Figure 3a)
(Reference Year = 1998–1999)
2000 9.36 −12.31Footnote b 5.77 2.99
2001 19.22 6.33 10.29 20.12
2002 11.70 2.12 −566 21.09
B Basic Model Including Enterprises Outside BC/AB 2000 12.84 −2.93 28.26 2.39
2001 14.42 8.11 11.43 17.30
2002 16.95 10.82 18.63 20.35
C Basic Model (Reference Year = 1998) 2000 4.07 −27.97Footnote b 8.30 −5.21
2001 13.47 −12.43 12.63 10.55
2002 6.38 −16.04 −3.48 11.56
D Basic Model
(Reference Year = 1999)
2000 14.44Footnote c 5.85 −0.70 11.05
2001 24.71Footnote c 27.85 3.49 29.39Footnote c
2002 16.83 22.78 −11.91 30.51

Different Reference Year
To obtain ATEs with 1998 (1999) as the reference year, it is sufficient to add a dummy variable for 1999 (1998) in the empirical specification. Panel C and D of Table 5 show that the results are different depending on which year is selected as the reference year. Since the same sample and empirical specification are used, the differences are only due to the choice of the reference year.

ATEs with year 1998 as the reference year (Panel C) are lower than in the case where 1999 is the reference (Panel D), except for the large performers samples. These ATEs are also different from the ones in Panel A, although the statistical significance is comparable. Nevertheless, the average of ATEs in 1998 and 1999 is close to the ATEs reported in Panel A. For example, for the unbalanced sample in 2000, the 1998–1999 average from Panels C and D is 9.29 ((4.07 + 14.44)/2) while the ATE in Panel A is 9.36. This relation holds true for all the other samples.

This analysis highlights that with DD estimator, it is important to perform a robustness check of the results because the estimates could be sensitive to the choice of the reference year as illustrated by the ATEs reported in Panels C and D of Table 5.

5.5 Comparability with other Natural Experiments

Finally, it is interesting to contrast the results of this study with the results found by Paff (2005). Among the three natural experiment papers cited in Section 2.2, Paff 's study is the most comparable. Two main reasons explain this likeness.

First, both studies use a DD estimator with a natural experiment and, second, the reforms analyzed are changes in provincial (state) R&D tax credits. However, a major difference is that the Californian R&D tax credit is incremental, not level-based like in BC.

Paff reports that the increase of in-house and basic research R&D tax credits in California augmented enterprise-level R&D expenditures by about 40% annually (statistically significant) between 1997 and 1999. Despite the results being not statistically significant, the introduction of the BC R&D tax credit increases enterprise-level R&D expenditures by 9%–19% per year between 2000 and 2002. Sample size is comparable.

Three factors could explain this difference. First, California is a major player in terms of BERD. This state accounts for 20% of total BERD in the United States while British Columbia accounts for 7% of BERD in Canada (see Table 1). Second, Paff 's study covers only two high-tech industries, pharmaceutical and prepackaged softwares. In contrast, the present study includes all industries available no matter whether they are oriented toward high-technology. Combining these two factors, it is reasonable to assume that enterprises in Paff 's sample are either more responsive to fiscal incentives or have more business opportunity than enterprises in the Canadian sample. Third, the reform in California is an increase of existing credit rates, not the introduction of a new credit. Although the BC tax credit used the same rules as the federal SR&ED tax credit, there may have been some delay before enterprises in BC could fully take advantage of the new credit.

6. Discussion

Policy Implications
The results have at least two important implications for the development of government innovation support policy. First, they suggest that an inelastic business demand for innovation could be a source of the lagging BERD performance in Canada (McFetridge, 2008; Council of Canadian Academies, 2009). Figure 5 depicts such a situation. As the main impact of a R&D tax credit is to lower the price of R&D (the supply side), it shifts the equilibrium from Point A to Point B. Consequently, a relatively important decrease in price results in a small increase of BERD. This would not only be consistent with the main results found in this paper, but also explain why BERD in Canada is low while fiscal incentives are high.

Figure 5: Business Demand and Supply for R&D

Graph of Business Demand and Supply for R&D (the long description is located below the image)
Description of Figure 5

The figure shows a linear demand curve for research and development (R&D) and two linear supply curves for R&D. The Y-axis shows the price of R&D and the X-axis the quantity of R&D performed. The demand curve has a steep and negative slope. The supply curves have a positive slope, which is the same for both curves. In absolute value, the slope of the supply curves is smaller compared to the slope of the demand curve. The initial supply curve denoted by Cost 1 intersects the demand curve at a point denoted by A. There is a shift in the supply curve to the right and below the Cost 1 curve, and the new supply curve is denoted by Cost 2. The intersection point is now denoted by B. Given the relative slopes of the demand and supply curves, the displacement of the equilibrium from Point A to Point B results in the large decrease in the price of R&D but in a much smaller increase in the quantity of R&D.

Unfortunately little is known about the drivers of business demand for innovation in Canada. Some recent reports (Competition Policy Review Panel, 2009; Council of Canadian Academies, 2009) have identified potential culprits such as structural factors (e.g. small market size in Canada), business attitude of Canadian entrepreneurs and a lack of complementarity assets to innovation (e.g. skills, intangibles). It is unclear what is the importance of these factors in explaining the low level of business innovation in Canada and more empirical evidence is required.

The second policy implication revolves around the optimal policy mix for Canada. The results of the analysis conditional on the receipt of subsidies (Section 5.3) along with the findings of Bérubé and Mohnen (2009) suggest that enterprises who claimed R&D tax credits and received subsidies performed better than firms only claiming R&D tax credits. Given the high share of fiscal incentives in Canada compared to other OECD countries (see Figure 1); the overall poor performance of Canada in terms of BERD and finally the apparent complementarity between both types of support, it could be the case that a shift toward direct intervention as called for in Expert Review Panel on Research and Development, 2011 or toward broader support to invest in complementary assets such as technology adoption, skills and formation could increase BERD and innovation in Canada. Again, this conjecture needs to be supported by more robust empirical evidence.

7. Conclusion

This paper presents an estimation of the impacts of the introduction of the provincial R&D tax credit in British Columbia in 1999 on R&D expenditures using a natural experiment. The counterfactual consists of enterprises located in Alberta and the experiment spans from 1997 to 2002.

Subject to the caveats listed in Section 5, the results from a panel DD estimator provide weak evidence that enterprises in BC did not increase their R&D expenditures following the introduction of the provincial tax credit. For the unbalanced sample, the estimated average treatment effects (ATE) vary between 9% and 19% (2000–2002), but are not statistically different from zero. ATE is defined as the % change in the level of enterprises R&D expenditures following the introduction of the new provincial R&D tax credit. Other results suggest that small and large R&D performers are affected differently by the new tax credit and that receipt of both tax credits and subsidies are complements. However, most of those additional results are not statistically significant.

In terms of policy implications, the absence of statistically significant results are consistent with an inelastic demand for innovation. This would explain why BERD is low in Canada despite high incentives (see Figure 4). Since the actual policy to support innovation is mainly oriented toward supply side interventions, e.g. decreasing the cost of R&D by using tax credits, this suggests a revision of government array of business R&D support. This conclusion is reinforced by another results of this paper, namely that R&D tax credits and subsidies seem to be complements for business R&D activities. However, more empirical evidence is needed to help develop better business innovation support policy for Canada.

Finally, the results of this paper, along with the findings of Mohnen and Baghana (2008), suggest that R&D tax credits have smaller impacts on R&D expenditures compared to the results of previous enterprise-level Canadian studies. Part of the explanation could lie in the use of Statistics Canada Research and Development in Canadian Industry (RDCI) database. RDCI is the most comprehensive and reliable database on enterprise-level R&D expenditures in Canada. Replicating the previous study, e.g. Dagenais et al. (2004), with RDCI data could confirm the validity of this assumption. This emphasizes that access to better data is critical to provide more empirical evidence to support innovation policy in Canada.

References

  • Aghion, Philippe, and Peter Howitt (1992) 'A Model of Growth through Creative Destruction.' Econometrica 60, 323–351
  • Baltagi, Badi H. (2001) Econometric Analysis of Panel Data (John Wiley & Sons, Ltd)
  • Bernstein, Jeffrey I. (1986a) 'The Effect of Direct and Indirect Tax Incentives on Canadian Industrial R&D Expenditures.' Canadian Public Policy 12(3), 438–448

    — (1986b) Research and Development, Production, Financing and Taxation (Toronto University Press)

  • Bérubé, Charles, and Pierre Mohnen (2009) 'Are Firms That Received R&D Subsidies More Innovative?' Canadian Journal of Economics 42(1), 206–225
  • Bloom, Nick, and Rachel Griffith (2001) 'The Internationalisation of UK R&D.' Fiscal Studies 22(3), 337–355
  • Bloom, Nick, Rachel Griffith, and John Van Reenen (2002) 'Do R&D tax credits work? Evidence from a panel of countries 1979–1997.' Journal of Public Economics 85, 1–31
  • Blundell, Richard, and Monica Costa Dias (2002) 'Alternative Approaches to Evaluation in Empirical Microeconomics,.' The Institute for Fiscal Studies, Working paper 10/02
  • Brouillette, Dany (2010a) 'A Descriptive Note on R&D Tax Credits in Canada and the United States.' Mimeo, Industry Canada (2010b) 'Impact of Business Innovation Support Program on Business R&D Expenditures: Review of Literature.' Mimeo, Industry Canada
  • Cameron, A. Colin, and Pravin K. Trivedi (2005) Microeconometrics: Methods and Applications (Cambridge University Press)
  • Cappelen, Ådne, Arvid Raknerud, and Marina Rybalka (2007) 'The Effects of R&D Tax Credits on Firm Performance.' Statistics Norway Report No. 2007/22
  • — (2008) 'The Effects of R&D Tax Credits on Patenting and Innovation.' Statistics Norway Discussion paper No.565
  • Competition Policy Review Panel (2009) 'Compete to Win.' Government of Canada
  • Council of Canadian Academies (2009) 'Innovation and Business Strategies: Why Canada Falls Short.' Report of the Expert Panel on Business Innovation
  • Czarnitzki, Dirk, Petr Hanel, and Julio Miguel Rosa (2004) 'Evaluating the Impact of R&D Tax Credits on Innovation: A Microeconometric Study on Canadian Firms.' Groupe de Recherche en Économie et Développement International, Working paper 05-01
  • Dagenais, Marcel, Pierre Mohnen, and Pierre Therrien (2004) 'Les Firmes Canadiennes Répondent-elles aux Incitations Fiscales à la Recherche-Développement?' L'Actualité Économique 80(2-3), 175–205
  • Dahlby, Bev (2005) 'A Framework for Evaluating Provincial R&D Tax Subsidies.' Canadian Public Policy 31(1), 45–58
  • David, Paul A., Bronwyn Hall, and Andrew A. Toole (2000) 'Is Public R&D a complement or substitute for private R&D? A review of the econometric evidence.' Research Policy 29, 497–529
  • Department of Finance, and Revenue Canada (1997) 'Tax Incentives for Scientific Reseach and Experimental Development.' Ottawa
  • Department of Finance, Government of Canada (2007) 'Tax Incentives for Scientific Reseach and Experimental Development.' Consultation paper, Ottawa
  • Duguet, Emmanuel (2008) 'The Effect of the R&D Tax Credit on the Private Funding of Research: An Econometric Evaluation.' Université D'Evry
  • Expert Review Panel on Research and Development (2011) 'Innovation Canada: A Call to Action' Review of Federal Support to Research and Development – Expert Panel Report, Ottawa: Industry Canada
  • Falk, Martin (2006) 'What Drives Research and Development (R&D) Intensity Across Organisation for Economic Co-operation and Development (OECD).' Applied Economics 38, 533–547
  • García-Quevedo, José (2004) 'Do Public Subsidies Complement Business R&D? A Meta-Analysis of the Econometric Evidence.' Kyklos 57(1), 87–102
  • Grossman, Gene M., and Elhanan Helpman (1993) Innovation and Growth in the Global Economy (MIT Press)
  • Guellec, Dominique, and Bruno van Pottelsberghe de la Potterie (2003) 'The Impact of Public R&D Expenditure on Business R&D.' Economics of Innovation and New Technology 12(3), 225–243
  • Gupta, Sanjay, Yuhchang Hwang, and Andrew Schmidt (2006) 'An Analysis of the Availability and Incentive Effects of the R&D Tax Credit after the Omnibus Budget Reconciliation Act of 1989.' Working paper
  • Hægeland, Torbjørn, and Jarle Møen (2007) 'Input Additionality in the Norwegian R&D Tax Credit Scheme.' Statistics Norway Report No.2007/47
  • Hall, Bronwyn, and John van Reenen (2000) 'How effective are fiscal incentives for R&D? A review of the evidence.' Research Policy 29, 449–469
  • Hatzichronoglou, Thomas (1996) 'Revision of the High-Technology Sector and Product Classification.' OECD STI Working paper No.1997/2
  • Kennedy, Peter E. (1981) 'Estimation with Correctly Interpreted Dummy Variables in Semilogarithmic Equations.' American Economic Review 71(4), 801
  • Klassen, Kenneth J., Jeffrey A. Pittman, and Margaret P. Reed (2004) 'A Crossnational Comparison of R&D Expenditure Decisions: Tax Incentives and Financial Constraints.' Contenporary Accounting Research 21(3), 639–680
  • Koga, Tadahisa (2003) 'Firm size and R&D tax incentives.' Technovation 23, 643–648
  • Lokshin, Boris, and Pierre Mohnen (2007) 'Measuring the Effectiveness of R&D Credits in the Netherlands.' United Nations University MERIT Working paper No.2007-025

    — (2009a) 'How Effective Are Level-Based R&D Tax Credits? Evidence from the Netherlands.' University of Maastricht

    — (2009b) 'What Does It Take for R&D Tax Incentives Policy to Be Effective?' United Nations University MERIT Working paper No.2009-014

  • Madore, Odette (2006) 'Scientific Research and Experimental Development: Tax Policy.' Library of Parliament of Canada
  • Mairesse, Jacques, and Benoît Mulkay (2004) 'Une évaluation du crédit d'impôt recherche en France, 1980–1997.' Cahier de la recherche de l'INSEE no. 2004-43
  • McFetridge, Donald G. (2008) 'Innovation and the Productivity Problem. Any Solutions?'
  • McKenzie, Kenneth J., and Natalia Sershun (2005) 'Taxation and R&D: An Investigation of Push and Pull Effects.' Institute for Advanced Policy Research, Technical paper TP-05015
  • Mohnen, Pierre, and Ruffin Baghana (2008) 'Effectiveness of R&D Tax Incentives in Small and Large Entreprise in Québec.' UNU-Merit University Working paper 2009-001
  • OECD (2010a) 'Main Science and Techology Indicators.' Volume 2010/2

    — (2010b) 'STI Outlook: Policy Trends'

  • Paff, Lolita (2005) 'State-Level R&D Tax Credits: A Firm-Level Analysis.' Topics in Economic Analysis & Policy 5(1), 1–25
  • Parsons, Mark, and Nicholas Phillips (2007) 'An Evaluation of the Federal Tax Credit for Scientific Research and Experimental Development.' Department of Finance Working Paper 2007-08
  • Romer, Paul (1990) 'Endogenous Technological Change.' Journal of Political Economy 98(5), S71–S102
  • Russo, Benjamin (2004) 'A Cost-Benefit Analysis of R&D Tax Incentives.' Canadian Journal of Economics 37(2), 313–335
  • Statistics Canada (2000) 'Industrial Research and Development – 1999 Intentions.' Catalogue no.88-202-XIE

    — (2001a) 'Industrial Research and Development – 2000 Intentions.' Catalogue no.88-202-XIE

    — (2001b) 'Industrial Research and Development – 2001 Intentions.' Catalogue no.88-202-XIE

    — (2002) 'Industrial Research and Development – 2002 Intentions.' Catalogue no.88-202-XIE

    — (2003) 'Industrial Research and Development – 2003 Intentions.' Catalogue no.88-202-XIE

    — (2005) 'Industrial Research and Development – 2004 Intentions.' Catalogue no.88-202-XIE

    — (2006) 'Industrial Research and Development – 2005 Intentions.' Catalogue no.88-202-XIE

    — (2007) 'Industrial Research and Development – 2006 Intentions.' Catalogue no.88-202-XIE

    — (2010) 'Industrial Research and Development – 2009 Intentions.' Catalogue no.88-202-X

  • Swamy, P. A. V. B., and S. S. Arora (1972) 'The Exact Finite Sample Properties of the Estimates of Coefficients in the Error Components Regression Models.' Economic of Innovation and New Technology 40, 643–657
  • Wilson, Daniel J. (2007) 'Beggar thy Neighbor? The In-State vs. Out-of-State Impact of State R&D Tax Credits.' Federal Reserve Bank of San Francisco, Working paper 2005-08

    — (2009) 'Beggar thy Neighbor? The In-State, Out-of-State, and Aggregate Effects of R&D Tax Credits.' The Review of Economics and Statistics 91(2), 431–436

  • Wooldridge, Jeffrey M. (2002) Econometric Analysis of Cross Section and Panel Data (MIT Press)
  • Wu, Yonghong (2005) 'The Effects of State R&D Tax Credits in Stimulating Private R&D Expenditure: A Cross-state Empirical analysis.' Journal of Policy Analysis and Management 24(4), 785–802

    — (2008) 'State R&D Tax Credits and High-Technology Establishment.' Economic Development Quarterly 22(2), 136–148

Appendix A: Tables

Table A1: Review of the Literature – Other Countries
Studies Country Impacts
Dugu (2008) France
(1993–2003)
Incrementality ratio: 1 €.
Mairesse and Mulkay (2004) France
(1980–1997)
A permanent increase of 10% in the R&D tax credit would increase R&D investment and capital by 3% to 5.5%.
Koga (2003) Japan
(1989–1998)
ηLT: −0.68 (all); −1.03 (large enterprises).
Cappellen et al. (2007) Norway
(2001–2004)
The new tax credit increases productivity. The rate of return is about 4% per year.
Cappellen et al. (2008) Norway
(2001–2004)
The new tax credit increase process innovations and to a lesser extent product innovations.
Lokshin and Mohnen (2007) Netherlands
(1996–2004)
ηST: −0.121 to −0.573; ηLT: −0.284 to −1.323. Incrementality ratio for large enterprises is below 1.
Lokshin and Mohnen (2009) Netherlands
(1996–2004)
ηST: −0.247; ηLT: −0.464.
Wu (2005) United States
(1979–1995)
Introduction of state tax credit increases annual per capita R&D spending by about $75–$120.
Wu (2008) United States
(1994–2002)
Introduction of state tax increases high-tech establishment in a state by 1.35% to 1.47%.
Wilson (2007) United States
(1981–2002)
Complete relocation effect between states following an increase in state R&D tax credit. ηin–stateLT : −2.5; ηbetween–stateLT: +2.5.
Bloom et al. (2002) 9 OECD countries
(1979–1997)
ηST : −0.1; ηLT: −1.
Falk (2006) 21 OECD countries
(1980–2002)
ηST: −0.22; ηLT: −0.84. HERD to BERD elasticity: +0.24 (short-term); +1.00 (long-term).
McKenzie and Sherchun (2005) 9 OECD countries
(1979–1997)
Push effect: ηST = −0.15 to −0.22; ηLT = −0.46 to −0.77. Pull effect: μST = −0.19 to −0.31; μLT = −0.58 to −0.93.
Guellec and van Pottelsberghe de la Potterie (2003) 17 OECD countries
(1983–1996)
+1$ in R&D government contracts increase BERD by +0.70$. Tax incentives and government fundings are substitutes.
ηST (ηCT) : short-term (long-term) price-tax elasticity.
ηST (ηCT ) : short-term (long-term) production tax elasticity on BERD.
Table A2 : SR&ED Tax Credit Rate 1987–2002
Business Type Credit Rates
(%)
Refundability (%)
Current Capital
Expenditures
Unincorporated business 20 40 40
CCPCs with prior-year taxable income ≤ $400,000 and
prior-year taxable capital employed in Canada ≤ $10 M
Expenditures up to expenditure limitFootnote 9 35 100 40
Expenditures over expenditure limit 20 40 40
CCPCs with prior-year taxable income between
$400,000 and $600,000 or prior year taxable capital
employed in Canada between $10 M and $15 M
Expenditures up to expenditure limitFootnote 10Footnote 11 35 100 40
Expenditures over expenditure limit 20 0 0
All other corporations 20 0 0

Source: Department of Finance (2007).

Table A3: Changes in R&D Federal Fiscal Incentives 1992–2003
Year Description
1993
  • The tax credit available to small corporations is extended to CCPCs with taxable income between $200,000 and $400,000.
1994
  • The special 30% tax credit rate applicable in the Atlantic and Gaspé regions is eliminated. The exemptions applying to sole-purpose SR&ED performers are eliminated. A time limit is set for identifying SR&ED expenditures incurred in previous years.
1995
  • Changes are made with respect to information technology expenditures, contract research and non-arm’s length transaction, third-party payments, and unpaid amount.
1998
  • Eligible SR&ED expenditures must be reduced by the revenue gained from the sale of a product of an SR&ED project. A review of the administration of tax incentives for SR&ED is undertaken and a new, simplified form for the tax credit is developed.
2000
  • Provincial super-deductions must be excluded from the calculation of eligible expenditures for federal SR&ED tax purpose.
2003
  • The small business limit for CCPCs is raised from $200,000 to $300,000 and the tax credit is extended to businesses with taxable incomes ranging from$300,000 to $500,000.
Source: Madore (2006).
Table A4: RDCI Sample – Extreme Value Analysis
Firms exhibiting sharp changes in R&D expenditures or R&D employment not matched with consistent changes in other variables were identify for deletion. Industrial classification was also considered in this exercise. The following two fictitious cases, one acceptable and one unacceptable, illustrate the principle applied.
Case for Deletion (NAICS 3313)
Year 1997 1998 1999 2000 2001 2002
R&D Expenditures ($ thousand) 2100 2200 2000 2000 2200 2300
Total Employment 3000 3000 3000 3100 1 2900
Total Sales ($ thousand) 14000 15000 15200 15000 15500 15900
Comment : Employment in 2001 inconsistent with other variables and industry.
Table A4: RDCI Sample – Extreme Value Analysis - Acceptable Case (NAICS 5417)
Acceptable Case (SCIAN 5417)
Year 1997 1998 1999 2000 2001 2002
R&D Expenditures ($ thousand) 10 20 50 100 120 150
Total Employment 5 5 7 10 12 15
Total Sales ($ thousand) 100 100 100 500 550 600
Comment : Variables pattern consistent a with start-up enterprise in this industry.

Table A5: Definition of Industrial Groups

High Tech (Group1)
Pharmaceutical and medicine Computer and peripheral equipment
Communication equipment Aerospace products and parts
Med–High Tech (Group2)
Management, scientific and technical consulting services
Navigational, measuring, medical and control instrument
Electrical equipment, appliance and components
Semiconductors and other electronic components
Architectural, engineering and related services
Computer system design and related services
Other computers and electronic products
Other chemicals Motor vehicle and parts
All other transportation equipment Transportation and warehousing
Finance, insurance and real estate Scientific research and development
Med–Low Tech (Group3)
Oil and gas extraction Mining
Electrical power Petroleum and coal products
Plastic products Rubber products
Non-metallic mineral products Primary metal (ferrous)
Primary metal (non-ferrous) Fabricated metal products
Machinery Furniture and related products
Information and cultural industries Other manufacturing industries
Health care and social assistance All other services
Low Tech (Group4)
Agriculture Forestry and logging
Fishing, hunting and trapping Construction
Food, beverage and tobacco Textile
Paper and wood product  
Classification based on d'Hatzichronoglou (1996).
Source: Statistics Canada internal industry classification.
Table A6: Estimated Parameters: Basic and Instrumented Models
Variable Fixed Effects Models
Basic Model InstrumentedFootnote 13
(Unbalanced)
Sample IFootnote 12
(Unbalanced)
Sample IIFootnote 12
(Balanced)
Sample IIIFootnote 12
(Survey)
Sample IVFootnote 12
(CRA)
lnRDt−1 0.1319Footnote b
(0.0526)
0.1869Footnote a
(0.0648)
0.2088Footnote c
(0.1157)
0.0997
(0.0626)
3.2063
(6.0370)
PERSt−1 0.0056
(0.0037)
0.0044
(0.0035)
0.0031
(0.0032)
0.0078
(0.0121)
−0.1333
(0.1955)
EMPLt−1 0.0001
(0.0001)
0.0000
(0.0001)
−0.0001
(0.0001)
0.0002Footnote a
(0.0000)
−0.0007
(0.0016)
Y2000 −0.026
(0.0864)
0.1639Footnote c
(0.0852)
−0.0200
(0.1578)
0.0065
(0.0965)
−0.3673
(0.6551)
Y2001 −0.0624
(0.1106)
0.0326
(0.1568)
0.0068
(0.1547)
−0.0979
(0.1394)
−0.6048
(0.9572)
Y2002 0.0540
(0.1359)
0.1245
(0.1474)
0.1666
(0.1387)
−0.0441
(0.1943)
−0.0822
(0.4839)
Y2000 × BC 0.0948
(0.1034)
−0.1254
(0.1097)
0.0712
(0.1738)
0.0366
(0.1199)
0.3535
(0.5419)
Y2001 × BC 0.1851
(0.1361)
0.0787
(0.1863)
0.1128
(0.1725)
0.1986
(0.1748)
0.6007
(0.6371)
Y2002 × BC 0.1225
(0.1543)
0.0345
(0.1647)
−0.0434
(0.1724)
0.2146
(0.2153)
0.1647
(0.5956)
Constant 4.5644Footnote a
(0.2643)
4.5597Footnote a
(0.3463)
6.2439Footnote a
(0.9080)
4.1206Footnote a
(0.2631)
−10.2102
(30.0007)
ση 1.4869 1.4302 0.7111 1.1441 3.0017
σε 0.6109 0.5866 0.3747 0.6468 1.8913
ρ 0.8556 0.8560 0.7827 0.7578 0.7158
# Obs. 982 580 215 767 752
# FirmsFootnote 14 230 116 58 192 230
R2 (within) 0.0452 0.0779 0.1690 0.0216
Table A7: Estimated Parameters: Subsidies and Robustness Analysis
Variable Fixed Effects Models (Unbalanced Panel)
Subsidies Basic Model
WithoutFootnote 15 WithFootnote 15 Multi-province Enterprises Reference Year = 1998 Reference Year = 1999
lnRDt−1, 0.0656
(0.0821)
0.1906
(0.0615)
0.1262
(0.0515)
0.1301
(0.0531)
0.1301
(0.0531)
PERSt−1. 0.0025
(0.0033)
0.0097
(0.0049)
0.0043
(0.0032)
0.0057
(0.0038)
0.0057
(0.0038)
EMPLt−1, 0.0000
(0.0001)
0.0003
(0.0003)
0.0000
(0.0001)
0.0001
(0.0001)
0.0001
(0.0001)
Y1998 −0.0967
(0.1065)
Y1999 0.0967
(0.1065)
Y2000 0.0520
(0.0984)
−0.2208
(0.1595)
−0.0549
(0.0921)
0.0227
(0.1154)
−0.0740
(0.0858)
Y2001 −0.0325
(0.1479)
−0.1091
(0.1392)
−0.0164
(0.1077)
−0.0137
(0.1331)
−0.1103
(0.1119)
Y2002 −0.0490
(0.1799)
0.2478
(0.2009)
0.0133
(0.1403)
0.1025
(0.1542)
0.0058
(0.1379)
Y1998 × BC 0.0922
(0.1207)
Y1999 × BC −0.0922
(0.1207)
Y2000 × BC −0.0007
(0.1358)
0.2986
(0.1728)
0.1267
(0.1090)
0.0484
(0.1311)
0.1407
(0.1075)
Y2001 × BC 0.0542
(0.1931)
0.2940
(0.1781)
0.1438
(0.1346)
0.1386
(0.1564)
0.2309
(0.1414)
Y2002 × BC 0.3154
(0.2045)
−0.2117
(0.2298)
0.1692
(0.1594)
0.0764
(0.1702)
0.1686
(0.1616)
Constant 4.7494
(0.3998)
4.4224
(0.3279)
4.6426
(0.2609)
4.5536
(0.2614)
4.5908
(0.2734)
ση 1.6976 1.2359 1.5325 1.4904 1.4859
σε 0.6142 0.6004 0.6214 0.6113 0.6113
ρ 0.8843 0.8091 0.8588 0.8560 0.8553
# Obs. 544 438 1004 982 982
# FirmsFootnote 14 128 102 235 230 230
R2 (within) 0.0211 0.1178 0.0393 0.0465 0.0465
Date modified: