Corporate profitability and effective tax rate: the enforcement effect of large taxpayer units

Abstract

This paper examines how Large Taxpayer Units (LTUs), a commonly-used tool for enforcing tax compliance, affect large firms’ reported profitability and effective tax rate. Increased scrutiny may either improve reporting and compliance efforts, or lead to adverse reactions from large taxpayers such as profit shifting to reduce tax liabilities. As a source of exogenous enforcement shock, we exploit the actions of Jamaica's LTU around its large-taxpayer eligibility cutoff using a before-during regression discontinuity approach. We find the LTU increases pre-tax profit margin by 2–3 percentage points. Increased effective tax rates are also evidenced, albeit less robustly.

Keywords:

• large taxpayers
• profitability
• tax compliance
• regression discontinuity

JEL Classification:

• H29
• H32
• M210

Acknowledgment

The authors are grateful to two anonymous referees and the associate editor of this journal, Martin Jacob, for constructive comments which substantially improved our paper. We also thank participants at the 2018 Midwest Economics Association Annual Meeting for their useful feedback.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Hereafter, monetary values are reported in millions of Jamaican dollars. The 2009 exchange rate of J$88.28 to US$1 can be used for conversion.

2 Vehorn (2011) explains that ‘the tax officials surveyed may have responded to a questionnaire from the IMF by telling the IMF what they thought the IMF wanted to hear … With respect to sample bias, the surveys were sent to 38 of the 57 countries that have LTUs and 33 countries (58 percent of the population) responded by answering at least some of the questions.’

3 Statistics sourced from Jacobs et al. (2012) and OECD (2018).

4 For the US tax reporting compliance rate, see https://www.irs.gov/newsroom/irs-releases-2006-tax-gap-estimates.

5 The tax reform project also led to the establishment of the TAJ as a semi-autonomous revenue authority. TAJ therefore enjoys internal autonomy and control over daily operations (e.g. budgetary control and human resource management), but still reports directly to the Financial Secretary of the Ministry of Finance and Public Service.

6 In the Jamaican context, corporate income tax is assessed as per company, rather than as a consolidated group.

7 Taxpayer-to-staff ratios for 2009 and 2011 are based on the number of tax staff and taxpayers, available from the Jamaica Information Service (2009), Jacobs et al. (2012), and personal correspondence with TAJ officials.

8 The Memorandum of Understanding was signed in 2015 with the TAJ and the Ministry of Finance and Public Service. The agreement sets out the conditions for the use and disclosure of the data provided. It states that the data can be used only for educational and research purposes, within three years of signing the agreement. It also indicates that the information be used in a manner that minimises privacy risks. As such, data confidentiality is preserved to safeguard the protection and privacy of the corporate taxpayers. The data used in the analysis are aggregated and therefore do not allow for the identification of individual taxpayers.

9 See Ministry of Finance and Planning (2012) for the White Paper on tax reform in Jamaica.

10 We rely on firms with 2010 sales to compute the running (forcing) variable used in our ‘sharp’ BD-RD approach.

11 This allows us to employ the simpler, more efficient ‘sharp’ BD-RD design described in section 4 rather than the ‘fuzzy’ BD-RD design. In any case, estimates would be similar because of the small proportion of firms excluded.

12 It is not clear why, but the data provided by TAJ did not include firms with sales between J$250M and J$350M. Nevertheless, this is not critical to our methodology since observations farther away from the sales threshold naturally have less influence on RD estimates. However, since more of such observations increase the statistical power with which we estimate the polynomial functions used in our BD-RD design, we include firms with sales between J$100M and J$250M. For robustness, we consider narrower sales bands that gradually exclude these firms (with sales between J$100M and J$250M), and found similar results (table 3).

13 We acknowledge that doing this ignores the largest tax contributors. Still, it is worth getting more precise and unbiased estimates by focusing on a more comparable subset of firms. Moreover, even large firms close to the cutoff are targeted by the LTU, and therefore are still relevant for identifying LTU effects.

14 Although 2009 is the LTU’s inception year, we retain observations in that year in our analysis because initial contact with LTUs began in the fourth quarter and significant efforts to address tax avoidance/evasion started post-2009. Moreover, we prefer to rely on two years of available data rather than one (2008). In any case, excluding the 2009 transition year results in a similar effect (0.025 with a linear fit at the 5% significance level) compared to the baseline results in table 2.

15 Some net profit (or loss) margins are near or above 100%, and are likely reported by holding or trust companies that tend to include investment income in their net profit calculations. Both trusts and holding companies are typically associated with profit shifting and manipulative practices. Also, since two-thirds of firms with extreme net profit margins are not large (i.e. below the sales cutoff), some reported margin values could be noise. Dupas & Robinson (2013) note that noisy data are a common issue when measuring business outcomes in studies of small firms. Data entry errors are also a possibility. As of 2015, the TAJ still has no desk-auditing (cross-checking) system and a small fraction of taxpayers get audited, thus much information in tax returns is accepted as is (TAJ 2015).

16 We also used 1% trimming and achieved a similar estimate (about 0.020 with a linear fit) compared to the baseline results in table 2; however, the standard error was inflated. The problem is 1% trimming still allows for firms with unusually high profits/losses (with no extreme sales levels). Such outliers (with low leverage) tend to inflate standard errors rather than bias coefficient estimates.

17 We are cautious about using the ETR measure because it is subjected to carry forward losses, credits, special deductions and waivers. However, these deductions do not similarly influence net profit before tax.

18 A firm with no tax liabilities in a given year would still be targeted by the LTU if its sales exceeded J$500M during 2010–2011. However, the firm drops from our analysis if it has no tax liabilities pre- or during-LTU.

19 Accounting for period effects is important because Jamaica experienced a sharp increase in GDP in 2011, a reflection of firms recovering from the worst years of the financial crisis. So, based on equation (1), we use the growth experienced by comparison firms close to the cutoff to separate out the targeting effect of LTU on treated firms close to the cutoff.

20 This is a ‘sharp’ design because whether a firm is targeted by the LTU is a deterministic function of gross sales.

21 Jacob et al. (2012) suggest using polynomial functions to generate the main results when there is not a large mass of firms close to the cutoff (as in our study); but also recommend complementing the results with nonparametric RD estimation. This method involves estimating: $\mathrm{\Delta }{y}_i={\alpha }_0+{\alpha }_{1}1\left\{S_i\ge c\right\}+{\alpha }_2\left(S_i-c\right)+{\alpha }_{3}1\left\{S_i\ge c\right\}\times \left(S_i-c\right)+{\epsilon }_i$ using weighted least-squares with weights given by the triangular kernel function $K_h\left(s\right)=\left(1-\left|s/h\right|\right)\times \left(\left|s\right|\le h\right)$, where $s=S_i-c$ and $h$ is a bandwidth parameter that controls the window width around the cutoff $c$. In the empirical analysis, we set $h=1$ (and then $h=0.5$) to ensure the sample around the cutoff is about one-half (and then one-fourth) of all firms.

22 Firm-level information on firms with foreign operations is not available. So, we identify sectors with the largest number of foreign affiliates namely, financial, manufacturing, mining, telecommunication, and tourism sector.

23 The authors find that bunching is higher in heavy manufacturing (equipment and metals), transportation, wholesale trade, and primary sectors, where there tend to be more paper trail.

24 Formally, we estimate this fuzzy BD-RD model: $\mathrm{\Delta }{y}_i={\alpha }_0+{\alpha }_1\widehat{p\left(D_i\right)}+\delta \left({\tilde{S}}_i-c\right)+e_i$, where $D_i=1\left\{S_i\ge c\right\}$ and $\widehat{p\left(D_i\right)}$ is the predicted value of $D_i$ from regressing it on $\delta \left({\tilde{S}}_i-c\right)$ and 1 $\left\{{\tilde{S}}_i\ge c\right\}$.

25 We recognise there is zero density between −0.7 and −0.4 range of sales (centred and scaled). As explained before, the data provided by TAJ did not include firms with sales between J$250M and J$350M. Again, this is not critical because observations farther away from the cutoff have less influence on RD estimates, albeit increase the statistical power of polynomial regressions. In fact, using a $\pm 0.5$ window width as in table 3 (or narrower) gives qualitatively similar results (but with lower precision).

26 We find similar and significant effects for even narrower window widths when using a uniform kernel (rather than a triangular one), which assigns an equal weight to each data point around the cutoff.

27 The MSE-optimal bandwidth of Calonico et al. (2017) is 0.544 and 0.844 for pre-tax margin and ETR, respectively. Based on these bandwidths, the estimated LTU effect is 0.025 (s.e. = 0.013) for pre-tax margin and 0.060 (s.e. = 0.021) for the ETR. These estimates are qualitatively unchanged when corrected for smoothing bias.

28 These industries attract high foreign direct investment (Bureau of Economic, Energy and Business Affairs 2011).