R&D Tax Incentives: A Comparison of the Incentive Effects of Refundable and Non-refundable Tax Credits. Summer Research Paper

R&D Tax Incentives: A Comparison of the Incentive Effects of Refundable and Non-refundable Tax Credits Summer Research Paper Christy MacDonald University of Waterloo Abstract: Tax incentives to encourage corporate investment in R&D can be structured in a variety of ways. Few studies have examined the differences in incentive effects between the alternative design choices for these R&D incentives. This study investigates whether firms respond differently to a tax credit that is structured to be refundable in comparison to one that is non-refundable. I explore this question by comparing R&D expenditure changes in each of the ten Canadian provinces. This comparison allows me to exploit the differences in the provinces tax credit structures. Using data from 1984 to 2001 in a fixed-effects regression model, I find evidence that can only suggest that firms respond stronger to the presence of refundability. The estimated model also provides some evidence that the size of the firm does not alter the firm s response to refundability. The overall lack of statistical significance in the results may be overcome in the future through the increase in the number of observations or the use of firm-level data. 1

1. Introduction Both the Canadian and U.S. governments use tax incentives to encourage companies to invest in research and development (R&D) due to the significant economic externalities associated with private companies performing R&D (Klassen et al, 2003). These special R&D tax incentives have resulted in an extensive amount of research. This extant research has mainly focused on the policy debate of whether these R&D tax incentives are cost-effective in generating R&D while very few studies have examined how firms respond to alternative design structures of the incentives. This paper investigates whether the incentive effects differ between a tax credit that is structured as refundable and one that is non-refundable. The significance of being refundable is that the receipt of the tax benefit is not dependent on the existence of tax liabilities whereas non-refundable credit is dependent on tax liabilities. Through an analysis of a basic R&D investment model, I hypothesize that firms will respond stronger to a refundable tax credit than to a non-refundable. In addition, I also expect that small firms will respond differently to refundability than larger firms. These hypotheses are tested using provincial R&D data from 1984 to 2001 for each of the 10 provinces in Canada. Many of the Canadian provinces offer tax credit incentives for R&D at various credit rates either with or without refundability. This provides the cross-sectional variation required to estimate the model described below. A significant advantage of examining multiple jurisdictions within one country is the similarities in economies and firms. Using a fixed effects model, I regress changes in provincial R&D expenditures on proxies of the provincial tax credit, an indicator of refundability, their interaction and 2

control variables. My estimates from this model suggest that firms respond stronger to the presence of a refundable tax credit in comparison to a non-refundable credit. However, due to insufficient sample size, the results are not statistically significant. In addition, the estimated model implies that small and large firms do not respond differently to a refundable incentive. The paper proceeds as follows: Section 2 provides details of research that has been done previously in the area of R&D tax incentives, which is followed in section 3 by a description of the R&D tax policies of the Canadian federal and provincial governments. Section 4 outlines a basic R&D investment model in an effort to better understand the implications of the design choice of refundability in comparison to non-refundability of R&D tax credits. Sections 5 and 6 describe the research design and data sample while the seventh section reports the results of the tests of hypothesis. Section 8 outlines alternative research design specifications and the final section presents the limitations and conclusions. 2. Literature Review Extensive research has been performed on the effectiveness of R&D tax incentives in generating R&D expenditures, particularly in the U.S. Hall and Van Reenan (1999) provide a summary of many of these studies that have been conducted both in the U.S. and in other countries. This summary highlights that prior research has mainly focused on the tax policy debate of whether the R&D tax incentive are cost-effective. Tax incentives are generally considered to be cost-effective if the increase in R&D expenditures due to the incentives exceeds the tax revenue foregone. Based on the review of the literature, Hall and Van Reenan conclude that the R&D tax credit in the 3

U.S. produces roughly a dollar-for-dollar increase in reported R&D spending on the margin. Earlier studies found that the incentive effects were insignificant (Eisner et al., 1984). But later studies by Berger (1993), Hall (1993) and Hines (1993) reported that the estimated cost-effectiveness ratio ranges anywhere from 1.0 to 2.0. The most recent studies by Klassen et al. (2003) and Gupta et al (2004) estimated the cost-effectiveness ratio to be 2.96 and 2.40, respectively. Studies on the Canadian R&D tax incentive system by Klassen et al. (2003), the Canadian Department of Finance (1997) and Berstein (1986) have reported similar findings in that the benefits of the incentives generally exceed the costs but the costeffectiveness ratio is smaller than the U.S. ratios. The more recent studies of the effectiveness of the R&D tax credit have begun to investigate how firms respond to alternative design specifications of the R&D tax credit. Klassen et al. (2003) and Gupta et al. (2004) provide the only empirical research addressing alternative design choices. Klassen et al. (2003) investigate the differences in incentive effects between a permanent system that provides a tax credit for all R&D expenditures in Canada and a temporary system that provides for a tax credit on incremental R&D expenditures in the U.S. Using data from a matched sample of Canadian and U.S. firms, the study finds that on average the Canadian tax credit system generates $1.30 of additional R&D spending per dollar of tax revenue foregone while the U.S. system generates $2.96 of additional spending. This implies that the incremental U.S. system has a stronger incentive effect than the Canadian system. Gupta et al. (2004) examine the effect of the change in the definition of the base amount used to determine incremental R&D expenditures under the U.S. tax incentive 4

structure. The definition was changed from a moving average base to a fixed-base under the Omnibus Budget Reconciliation Act of 1989 (OBRA89). Using firm-specific data over a 14-year period, R&D intensity, defined as R&D expenses divided by sales, is regressed on measures of both tax and non-tax factors. The authors find that the structural change of the R&D tax incentive due to OBRA89 increased the overall R&D intensities but the effect was greater effect on high-tech firms than on firms in other industries. An R&D tax incentive design choice that has not been examined is the difference in incentive effects between a refundable and non-refundable R&D tax credits. A nonrefundable tax credit provides tax incentives to a firm only if it has current or future tax liabilities. On the other hand, a refundable tax credit allows a firm to obtain the benefit of the tax incentive regardless of whether or not the firm has tax liabilities. From a government s perspective, providing tax credits that are refundable is an expensive policy choice as all costs are incurred immediately. The U.S federal tax policies surrounding R&D tax credits have remained nonrefundable since the policies inception and so studies focusing on U.S. tax policies have not addressed the issue of refundability. However, in Canada, the federal and some provincial governments provide a limited degree of refundability on some R&D tax credits. Although studies have been performed on the Canadian R&D credit system as discussed above, these studies have not examined the incentive effect of the design choice of refundable of R&D tax credits. This study attempts to address this question. 5

3. Tax Incentives for R&D In Canada, the federal government s R&D tax incentive program allows for a 100 per cent deduction for qualifying current and capital R&D expenditures also referred to as Scientific Research and Development (SR&ED) expenditures. In addition, this program provides firms an investment tax credit (ITC) on these qualified expenditures. The rate of the credit is 20 per cent and is available to all corporations in Canada. An increased rate of 35 per cent is available on expenditures of up to $2 million to Canadian controlled-private corporations (CCPCs) that have taxable income less than $0.3 million 1 and taxable capital less than $25 million. ITCs that are not deducted against federal taxes in the year in which they are earned can be carried over. Additionally, small CCPCs are eligible to obtain a refund of unused ITCs earned in a fiscal period. 2 The amount of ITC used is included income in the following year. All provinces generally follow federal rules in respect of the deductibility of current and capital expenditures on R&D. However, the provinces differ in their policies for providing firms with additional R&D incentives. Currently, Newfoundland, New Brunswick, Nova Scotia, Quebec, Ontario, Manitoba, Saskatchewan, and British Columbia provide additional R&D tax incentives. Alberta and Prince Edward Island do not offer any R&D tax incentives at the provincial level but rely solely on the federal R&D tax incentives. A summary of the provincial R&D incentives are provided in Table 1. Most provinces provide additional R&D tax incentives in the form of ITCs. As with federal rules, these provincial ITCs increase taxable income in the following year in 1 Prior to 2003, the taxable income limit was $0.2 million. 2 For small CCPCs, unused ITCs earned at the 35 per cent rate are 100 percent refundable whereas unused ITCs earned at the 20 per cent rate are refundable at a general rate of 40 per cent. 6

all provinces except Quebec. For Quebec tax purposes, the Quebec R&D tax credit is not included in taxable income. In addition to a tax credit, Ontario has also provided a special deduction termed the superallowance. This superallowance provides an additional 25 to 52.5 percent writeoff on top of the 100 percent deduction of qualifying expenditures mentioned previously. The Ontario superallowance was not added to taxable income in the following year for federal or Ontario purposes until 2001. At that time the federal government decided to tax the superallowance and so in response, the superallowance was suspended by Ontario. As Table 1 shows, the provinces vary both in the rate of credit provided and the provision for refundability. To determine how firms respond to an incentive that includes refundable rather than non-refundable R&D tax credits, I analyze a basic R&D investment model in the following section. 4. R&D Investment Model The profitability index is one of the approaches that can be used by managers to select from available R&D investments. 3 Using the profitability index, Klassen et al. (2003) develop a basic R&D investment model that incorporates tax considerations. 4 This model shows that the pre-tax cash flows (both the investment outflows and expected future inflows) are adjusted by a tax adjustment factor of 1/(1-CR) where CR is the statutory rate at which the R&D investment receives a credit. This tax adjustment factor will be greater than one with the presence of an R&D tax credit. As a result, more R&D 3 The profitability index is calculated as the net present value of the project divided by the project investment (Klassen et al, 2003). 4 Refer to Klassen et al. (2003) for further details of the R&D investment model. 7

investments should be undertaken with R&D tax credits as the after-tax profitability index will be greater than the before-tax profitability index. An important element that needs to be incorporated into the tax adjustment factor is whether the R&D tax credit is refundable. If the credit is non-refundable and the firm is currently not in a taxable position, the CR must be discounted from the expected time when the firm has sufficient income tax liabilities against which to apply the tax credits. Letting the discount factor be denoted as θ, the tax adjustment factor becomes 1/(1-θCR). The discount factor reduces the value of the credit and hence, the index for all of the firm s projects, leading to less R&D investment then if the firm had current tax liabilities. The profitability index is further reduced if tax benefits are lost due to insufficient future tax liabilities to fully utilize all tax credits before the expiry of the carryover period. 5 If tax credits are lost, the discount factor for those credits would be zero leading to a tax adjustment factor of one. In other words, the tax incentives would have no impact on the R&D investment. However, if the firm is currently taxable, the tax adjustment will simply be 1/(1-CR) as θ = 1. On the other hand, if the tax credit is refundable, the CR does not need to be adjusted for the firm s current tax position. With refundability, a firm obtains the tax benefit regardless of whether there are sufficient tax liabilities to be offset by the credit so the discount factor, θ, will always be equal to one. With the certainty in the value of refundable R&D tax credits, firms should invest more in R&D than if the firm earned non-refundable tax credits. Thus based on the differences in the effect of the type of tax credits, I posit the following hypothesis: 5 The provinces that provide non-refundable R&D tax credits allow unused credits to be carried back 3 years and carried forward either 7 or 10 years. 8

H1: Firms respond more strongly to refundable than non-refundable tax credits. Large firms are more likely to be established and more diversified than a small firm. This provides larger firms with other sources of income from prior years or other divisions that can generate income tax liabilities against which the tax credits can be applied. As a result, small firms may be more likely to be in a non-taxable position and so may obtain more benefit from a refundable tax credit leading to the second hypothesis: H2: Small firms respond differently than larger firms to refundable than nonrefundable tax credits. 5. Research Design To test the prediction of these hypotheses, the changes in the level of provincial R&D expenditures are regressed on measures of the provincial tax credit, an indicator of refundability, and the interaction of these two variables. The cross-sectional variation required to estimate this model is found in the differences in the tax incentives offered by the provinces as discussed previously. Federal SR&ED tax credits are a cross-sectional constant and will not be included in the analysis. My empirical analysis includes two other important design specifications. First, the provincial fixed-effects are included in the regression as the panel of data consists of the full set of provinces in Canada. Second, I estimate the regression separately for expenditures from large and small firms as the applicable tax and R&D credit rate is dependent on the size of the firm for many of the provinces. A third regression is estimated using total provincial R&D expenditures as the dependent variable to avoid the arbitrary allocation of expenditures between large and small firms discussed in Section 6. 9

The independent tax variables will reflect measures related to large firms as the majority of R&D expenditures are made by larger firms. Based on these research design considerations, I estimate the following regression model: ln(r&d t ) ln(r&d t-1 ) = α1 + α2cred it + α3refund it + α4credxrefund it + α5tax it + α6 ln(prof it ) + α7ln(pop it ) + + α8 ln(gdp it ) + ΒYEAR + µt + ζit 6 (1) The regression variables are defined as follows: ln (R&D it ) CRED it REFUND it TAX it ln(prof it ) ln(pop it ) ln(gdp it ) YEAR t The natural logarithm of provincial R&D expenditures (in 000s) for year t. The provincial R&D tax credit rate. An indicator variable taking on the value of 1 if CRED is refundable; 0 otherwise. The current provincial corporate tax rate. The change in the natural logarithm of the number of professionals engaged in R&D. The natural logarithm of the provincial population (in 000 s) for year t. The natural logarithm of provincial GDP $ value (in 000s) measured at current prices. A vector of 17 indicator variables, each taking on the value of 1 when the observation is from a given year, 1985 2001, 0 otherwise. µt The provincial fixed-effect. The tax variables included in the model are CRED, REFUND, CRED*REFUND and TAX. The CRED variable and its interaction with REFUND are the main test 6 A variable to control for the difference in industry concentration between provinces was originally included in the model but was found to be a highly insignificant explanatory variable. 10

variables. A positive and significant estimate of α4 would be consistent with hypothesis H1 while a positive and significant difference between the estimates for a small and large firm of α4 would support hypothesis H2. The variable, CRED, will be calculated as the provincial R&D credit rate multiplied by (1-τ), where τ is the provincial corporate tax rate. 7 However, there are two exceptions to this calculation. First, all R&D expenditures qualify for provincial credit provisions except in Quebec where the credit only applies to salaries and wages related to R&D activities. As a result, the Quebec credit rate is multiplied by 50 per cent since according to Warda (1997), the labour component of R&D expenditures hovers around 50 per cent in Canada. 8 Second, the Quebec provincial credit and the Ontario superallowance are not taxable during the period studied and so are not be adjusted by (1-τ). The corporate decision to incur R&D expenditures depends on factors beyond just the R&D tax incentives that are available. Prior research has identified a number of nontax factors. One factor that has been controlled for in previous research is a firm s propensity to conduct research (Swenson, 1992; Berger, 1993; Hall, 1993; Klassen et al., 2003; and Gupta et al., 2004). This control variable is generally measured using prior year s R&D expenditures as the amount of current period R&D expenditures will depend to some degree on the amount of R&D previously expended. However, since the explanatory variables in this study explain the change in R&D expenditures rather than the level of R&D expenditures, the lagged value of R&D is not required. 7 To simplify the calculation, the $2 million cap on the expenditures that qualify for enhanced R&D tax incentives in Quebec, Ontario, and British Columbia are ignored. In addition, no distinction is made between expenditures incurred by small firms and small CCPCs. 8 For the period 1984-2001, the percentage of wages and salaries varied between 47 and 51 per cent according to Statistics Canada data. 11

Berger (1993) identifies the level of technological progress in the economy, measured as Gross National Product, as having an effect on the level of R&D investment for each firm. As an index for technological progress, this study will include a measure of the change in the provincial Gross Domestic Product. Previous research has included a control for the investment opportunity available to a firm measured as Tobin s q 9 (Berger, 1993; Gupta et al., 2004; Klassen et al., 2003). A firm with a higher Tobin s q has greater investment opportunities and therefore, should conduct more R&D. This calculation cannot be calculated at a provincial level. Instead the change in the number of professionals engaged in R&D in the province will be used to proxy for the investment opportunities within a province. Provinces with a larger increase in professionals should conduct more R&D. As a final explanatory variable, the natural logarithm of the population of each province is included to control for the magnitude of the change in R&D expenditures due to the size of the province. 6. Sample and Descriptive Statistics Sample As firm level data on R&D expenditures by province are generally not disclosed, the analysis in this paper includes provincial level data on R&D expenditures. Data on provincial level R&D expenditures were collected from Statistics Canada (StatsCan) for the period of 1984 to 2001. StatsCan collects R&D expenditures at a provincial level from the annual survey entitled Research and Development in Canadian Industry. The advantage of using this data from StatsCan is that the survey specifies that the R&D 9 Tobin s q is computed as the sum of market value of common shares plus book value of preferred stock plus long-term and current debt, divided by total assets (Klassen et al, 2003). 12

reported must meet the definition of Scientific research and Experimental Development as defined in Section 37, Regulation 2900 of the Income Tax Act. This definition is also used by the provinces to determine qualifying expenditures. However, the R&D expenditures collected from the survey includes spending on land and buildings which cannot be claimed for income tax purposes. Since the percentage of expenditures on land and buildings are insignificant, no adjustment to qualifying R&D expenditures was made. 10 Data on provincial level R&D expenditures were collected for each of the 10 provinces in Canada. Although Alberta and PEI do not have R&D tax incentive programs at any time during the period of study, they are included in the analysis to provide a complete panel of data and to serve as a base control. The final sample contains 180 provincial observations. A drawback of using provincial level data from StatsCan is the lack of disclosure of expenditures by small and large firms. For some of the provinces, a firm is eligible for enhanced R&D tax credit rates or refundability if the firm qualifies as a small firm 11 and has expenditures under $2 million. To address this problem, expenditures made by small firms are determined by defining firm size based on the number of employees. According to Industry Canada, goods-producing firms are considered small if they have fewer than 100 employees, while service-producing firms are small if employees are less 10 Based on Statistics Canada data, the percentage of R&D expenditures on land building ranged from approximately 0.1 to 3.6 per cent over the period of 1984-2001. 11 For the period of study, a small firm is classified as having taxable paid-up capital of less than $25 million and taxable income less than $0.2 million. For the Ontario OITC up to 1999 and the British Columbia R&D tax credit, the small firm must also be a Canadian Controlled Private Corporation (CCPC). 13

than 50. For purposes of this study, firms are assumed to meet the definition of a small firm for R&D purposes if they have fewer than 100 employees. 12 To determine the amount of R&D expenditures by size of firm, data were obtained on the level of R&D expenditures for each employment size category in Canada. 13 In addition, data were found on the number of firms in the same categories of employment size for each province. Based on this data, the following calculations were used to determine expenditures by firm size category for each province: Estimated Provincial R&D for the Size Category Total Canadian R&D Expenditures for = the Size Category x Total Number of Firms for the Size Category Provincial Number of Firms for the Size Category (2) Estimated Provincial R&D Expenditures for the Size Category = Estimated Provincial R&D for the Size Category x Total Actual Provincial R&D Total Estimated Provincial R&D (3) The ratio of actual provincial expenditures to total estimated expenditures captured in the third term of equation (3) is required to preserve the actual total R&D expended in each province. The R&D expenditures made by small firms is calculated as the sum of the estimated provincial R&D expenditures for the size categories 0-49 and 50-99. Table 4 provides an example of the calculation of R&D expenditures by size of firm using 2001 12 The model was estimated using a cutoff of 50 employees as well but no significant differences were noted. 13 Expenditures were allocated to the following categories based on the number of employees in firm: 0-49, 50-99, 100-499, and 500+. 14

data for Ontario. The estimated 2001 R&D expenditures for Ontario for small firms are $1,448.42, the sum of the first two rows of Table 4. This calculation was repeated for each province for every year included in the study. Descriptive Statistics Panels A, B, C and D of Table 2 provide descriptive statistics for all the variables used in the tests. The mean value of R&D expenditures, $391.51 million and $128.60 million for large and small firms respectively, show that large firms expend more on R&D as anticipated. The statistics also show that large firms are on average entitled to a lower R&D credit rate of 4.2 percent while small firms received on average a rate of 5.7 percent. Large firms pay tax at a higher rate as shown by the mean tax rate of 13.7 percent compared to a rate of 7.7 percent faced by small firms. The lower R&D tax credit and higher tax rate for larger firms are consistent with Canadian provincial governments generally providing more favourable tax rules for smaller firms. 7. Results Table 5 presents the results of the fixed-effects regression specified in equation (1). The first two columns report regression estimates for provincial R&D expenditures made by small and large firms while the final column presents estimates for total provincial R&D expenditures. Beginning with the results from the first two columns, both the credit rate and the indicator of refundability have a coefficient that is negative in each of the regressions, but not statistically different from zero. The interaction variable, CREDxREFUND, has a coefficient of 1.59 (t-statistic of 1.14) and 2.06 (t-statistic of 0.56) for small and large 15

firms, respectively. Although the sign of the coefficient is positive as predicted by the hypothesis, the coefficient is not statistically different from zero in any of the regressions. This can only suggest but not conclude that a firm responds more strongly to an R&D tax incentive structure with refundability. The magnitude of the coefficient appears reasonable in comparison to the findings of other Canadian studies (Klassen et al, 2003; Canadian Department of Finance, 1997; and Berstein, 1986) but my sample has an insufficient number of observations to reduce the standard errors to improve the t- statistic. The second hypothesis predicted that small firms would respond differently to a refundable incentive than a larger firm. The estimated coefficients from the fixed-effects regression appear to suggest that large firms respond stronger to refundability as the magnitude of the coefficient on the interaction variable was higher for larger firms than for smaller firms. But interpretation of the difference in magnitude is difficult as the t- statistics for either coefficient is not significantly different from zero. As a formal test of the second hypothesis, equation (1) is estimated by including expenditures by small and large firms in the same regression, increasing the number of observations to 360. An indicator variable for size 14 and its interaction with CRED, REFUND and CREDxREFUND are added to equation (1). A statistically significant estimate of the coefficient on the three-way interaction of SIZExCREDxREFUND would be consistent with hypothesis H2. From the untabulated results, the coefficient on the three-way interaction is not significantly different from zero suggesting that there is no significant difference between the response of small and large firms to refundability as expected from previous results. 14 The size indicator variable takes on the value of 1 if expenditures are by a large firm; 0 otherwise. 16

Looking at the remaining explanatory variables, the tax rate has a coefficient that is negative and statistically significant from zero at the 1 percent level for small firms and at the 10 percent level for large firms. As the R&D investment model would predict that the coefficient on the tax rate variable should be insignificant, it is unclear how to interpret a negative coefficient on the tax rate. However, a negative coefficient on the tax rate variable is consistent with the findings of Klassen et al. (2003). The change in professionals engaged in R&D was also positive as predicted and statistically significant at the 1 percent level for both small and large firms. The final two control variables for size and growth in technological progress are both positive as predicted but are not statistically different from zero. As shown in the third column of Table 5, the regression using the total provincial expenditures as the dependent variable produces similar coefficients and t-statistics for each variable in the model to those estimated by the regression using expenditures by large firms as the dependent variable. 8. Alternative Research Design Specifications As is common in other literature, the regression equation is estimated using levels of R&D expenditures rather than changes in the R&D expenditures as the dependent variable. With a levels regression, the independent variables explain the value of the R&D expenditures rather than just the change. As alternative design specifications, equation (1) is estimated using the level of provincial R&D expenditures with and without the lagged value of R&D expenditures as an explanatory variable. In addition, the provincial fixed-effects are removed from both the changes and levels regressions through the use of OLS with robust standard errors. The fixed effects 17

are removed to see if the fixed effects are picking up too much of the cross-jurisdiction variation that may be related to differences in tax incentives. Most of these alternative specifications result in positive but insignificant values for the interaction between the tax credit rate and the indicator of refundability. The only exception is the levels model estimated without the fixed effects and the lagged value of R&D. This model resulted in a negative coefficient for the interaction variable that was significantly different from zero. However, it is difficult to interpret this finding as coefficient may be reflecting variations from the omitted fixed effects or lagged value of R&D. 9. Limitations and Conclusion Limitations The provincial level data used in this study presents a number of limitations. First, since Canada only has 10 provincial jurisdictions, the number of observations was very restricted and could only be increased by expanding the sample to include additional years. Unfortunately, data on R&D expenditures was only available from 1984 to 2001. As a result, the estimated coefficients were mostly insignificant as the sample size was too small to reduce the standard errors. Second, the provincial level data did not distinguish between expenditures by small and large firms. Although a rough calculation using firm size defined by the number of employees is used to estimate expenditures by small firms, the estimation may not be reflective of actual expenditures by small firms. Third, the inclusion of a firm s tax status is not feasible with a dependent measure that used provincial level data. A firm s tax status is required to identify whether or not a 18

firm is able to use a non-refundable tax credit. This identification is important in the calculation of the cost-effectiveness of the R&D tax credit. A refundable tax credit is a more expensive tax policy choice than a non-refundable tax credit as a refundable credit is always realized by a firm. The inclusion of the tax payer status would have allowed an estimation of the difference in cost-effectiveness between the two types of credit. These limitations could be avoided by obtaining firm-level data. Ideally, the incentive effects of the design choice to refund tax credits should be investigated with firm-level data that includes the firm s tax status and R&D expenditure by province. Although this data is collected by the Canada Revenue Agency from tax returns, the data is currently unavailable to the public. Conclusion This study examines whether firms respond differently to alternative design structures of the R&D tax credit. Few studies have addressed this question as much of the previous research related to R&D tax incentives has concentrated on the costeffectiveness of these incentives. My research focuses on the design choice of whether to provide tax credits that are refundable or non-refundable. To examine firms response to this design choice, I compare the changes in provincial R&D expenditures in order to take advantage of the cross-jurisdictional differences in the tax policies towards refundable and non-refundable tax credits. Due to the limitations of the data, the results from the fixed-effects regression are not conclusive. However, the positive coefficient on the interaction variables suggest that firms do respond stronger to a refundable than a non-refundable R&D tax credit as predicted by the R&D investment model. The analysis also suggests 19

that the response of small and large firms to the existence of refundability is not statistically significant. In conclusion, this study did not find strong evidence of a difference in the incentive effects between a refundable and non-refundable tax credit. But future research may find stronger evidence of the incentive effects by increasing the number of observations or obtaining firm-level data. 20

References Berger, P.G. 1993. Explicit and implicit tax effects of the R&D tax credit. Journal of Accounting Research, 31(Autumn): 131-71. Berstein, J.I. 1986. The effect of direct and indirect tax incentives on Canadian industrial R&D expenditures. Canadian Public Policy, 12(3): 438 448. Canadian Department of Finance. 1997. The Federal System of Income Tax Incentives for Scientific Research and Experimental Development: Evaluation Report. Ottawa: Department of Finance. Eisner, R., S.H. Albert, and M.A. Sullivan. 1984. The New Incremental Tax Credit for R&D: Incentive or Disincentive? National Tax Journal. XXXVII(2): 171 183. Gupta, S., Y.Hwang, A.Schmidt. 2004. R&D Spending Fools? An analysis of the R&D Credit s Incentive Effects after the Omnibus Budget Reconciliation Act of 1989. Working Paper. Hall, B.H. 1993. R&D tax policy during the 1980s: Success or failure? In J.M. Poterba (Ed.), Tax Policy and the Economy. Cambridge, MA: The MIT Press, 7:1 35. Hall, B. and J. Van Reenan. 2000. How effective are fiscal incentives for R&D? A review of the evidence. Research Policy, 29: 449-469. Hines, J.R. 1993. On the sensitivity of R&D to delicate tax changes: The behavior of U.S. multinationals in the 1980s. In A. Giovannini, R.G. Hubbard, & J. Slemrod (Eds.), Studies in International Taxation (pp. 149 93). Chicago, IL: University of Chicago Press. Klassen, K.J., J.A. Pittman, M.P. Reed. 2003. A Cross-National Comparison of R&D Expenditure Decisions: Tax Incentives and Financial Constraints. Contemporary Accounting Research. (forthcoming) Swenson, C.W. 1992. Some tests of the incentive effects of the research and experimentation tax credit. Journal of Public Economics, 49(2): 203 18. Warda, J. 1997. R&D Tax Treatment in Canada: A Provincial Comparison. Conference Board of Canada. 21

Table 1: Summary of Provincial R&D Tax Incentives PROVINCE YEAR AVAILABLE TAX INCENTIVE RATE Newfoundland 1996 Fully refundable ITC 15% Nova Scotia 1984 1993 1994 New Brunswick 1994 2003 Quebec 1983 1988 1988 Ontario 1988 2000 Non-refundable ITC Fully refundable ITC Non-refundable ITC Fully refundable ITC Fully refundable ITC for R&D salaries Fully refundable ITC for R&D salaries Superallowance Deduction 10% 15% 10% 15% 10% 40% -- for small firms on R&D salaries up to $2 million 20% -- for large firms CCPCs: 35% up to base amount Non-CCPCs: 25% up to base amount 1995 Fully refundable ITC Manitoba 1992 Non-refundable ITC 15% Saskatchewan 1998 Non-refundable ITC 15% 10% -- for small firms (prior to 1999, only small CCPCs) on SR&ED expenditures up to $2 million British Columbia 1999 Fully refundable ITC Non-refundable ITC 10% -- for small CCPCs for SR&ED expenditures up to $2 million 10% -- for all other firms 22

Table 2: Calculation of Expenditures by Firm size using 2001 data for Ontario Firm Size Category (# of Employees) Average Canadian R&D Expenditure Number of Firms in Ontario Estimated R&D for Ontario (in 000s) Adjusted Estimation for Ontario 0-49 $ 0.0016 315,421 $ 504.67 $868.37 50-99 0.05124 6,579 337.11 580.05 100-499 0.11747 5,296 622.12 1,070.46 500+ 0.94326 1,586 1,496.01 2,574.18 Total 328,882 $2,959.91 $5,093.07 Total Ontario R&D Expenditures in 2001 (Actual in 000s) $5,093.00 Adjustment Ratio ($5,093/$2,959.91) 1.7207 23

Table 3: Descriptive Statistics Panel A: Continuous Variables Small Firms Variable Mean Std Dev Median Min Max R&D t 128.60 254.27 5.497 0.043 1446.80 ln(r&d t ) 2.259 2.740 1.704-3.148 7.277 ln(r&d t-1 ) 2.147 2.747 1.663-3.225 6.873 ln(r&d t )-ln(r&d t-1 ) 0.113 0.379 0.078-1.938 2.289 R&D credit rate 0.057 0.069 0.000 0.000 0.200 Tax rate 0.077 0.022 0.080 0.030 0.110 Panel B: Continuous Variables Large Firms Variable Mean Std Dev Median Min Max R&D t 391.51 690.11 49.350 0.939 3646.21 ln(r&d t ) 4.217 2.160 3.899-0.063 8.201 ln(r&d t-1 ) 4.135 2.160 3.857-0.063 7.964 ln(r&d t )-ln(r&d t-1 ) 0.082 0.322 0.076-1.781 1.762 R&D credit rate 0.042 0.052 0.000 0.000 0.1425 Tax rate 0.137 0.039 0.150 0.050 0.170 Panel C: Other Continuous Explanatory Variables All Firms Variable Mean Std Dev Median Min Max ln(prof t )-ln(prof t-1 ) 0.084 0.214 0.057-0.511 1.253 ln(pop) 7.281 1.234 6.968 4.855 9.384 ln(gdp t )-ln(gdp t-1 ) 0.049 0.035 0.050-0.142 0.217 Panel D: Discrete Variable All Firms Small Firms Large Firms Variable Number Percent Number Percent Refund 40 23.5 32 18.8 Notes: The variables are defined as follows: R&D is the total R&D expenditures for a province. The tax rate is the provincial corporate income tax rate. R&D credit rate is the tax credit rate multiplied by (1-τ) where τ is the tax rate. PROF is the number of professionals engaged in R&D in the province and POP is the provincial population. GDP is the provincial level GDP at current prices. 24

Table 4: Correlations of Explanatory Variables Tax rate R&D Credit rate Refund ln(prof) ln(pop) ln(gdp) Large Firm Observations Tax rate -0.155-0.577-0.119-0.047-0.015 R&D credit rate -0.395 0.616-0.002 0.064-0.164 Refund -0.571 0.718 0.027 0.203-0.055 ln(prof) 0.002-0.003 0.038-0.065 0.065 Small Firm Observations ln(pop) -0.143 0.270 0.351-0.065 0.038 ln(gdp) 0.064-0.168-0.056 0.065 0.038 Notes: Correlations of the variables are presented. Correlations of large firm observations are presented above and right of the diagonal and correlations of small firm observations are presented below and left of the diagonal. 25

Table 5: Regression of R&D on tax incentives ln(r&d t ) ln(r&d t-1 ) = α1 + α2cred it + α3refund it + α4credxrefund it + α5tax it + α6 ln(prof it ) + α7ln(pop it ) + α8 ln(gdp it ) + ΒYEAR + µt + ζit Variable Small Firms Large Firms Total Intercept? -4.729 (-0.98) -5.580 (-1.01) -5.202 (-0.94) CRED it + -0.099 (-0.11) -0.091 (-0.10) -0.102 (-0.11) REFUND it + -0.272 (-1.31) -0.243 (-0.49) -0.207 (-0.42) CRED it xrefund it + 1.593 (1.14) 2.061 (0.56) 1.793 (0.49) TAX it? -5.377 (-2.66)** ln(prof it ) + 0.624 (4.58)** ln(pop it ) + 0.719 (1.06) ln(gdp it ) + 1.161 (1.38) -2.312 (-1.91)* 0.536 (3.85)** 0.815 (1.06) 1.115 (1.32) -2.280 (-1.89)* 0.526 (3.79)** 0.763 (1.00) 1.073 (1.28) Year indicators Yes Yes Yes R 2 0.4739 0.2587 0.2727 N 170 170 170 Notes: This table presents results of a fixed-effects regression. The variables are defined in Table 2. Signs for the predicted variables coefficients are presented to the right of the variables. The coefficients of the interaction variable provide evidence on whether firms respond differently to a refundable R&D tax credit than to a non-refundable R&D tax credit. ** significant at the 1% level using a one-sided test * significant at the 10% level using a one-sided test 26