Cash or Credit? Customer and Supplier Tax Rates and Trade Credit Use. Margot Howard and Kenneth Njoroge College of William & Mary.

Transcription

1 Cash or Credit? Customer and Supplier Tax Rates and Trade Credit Use Margot Howard and Kenneth Njoroge College of William & Mary May 2017 PRELIMINARY DRAFT DO NOT CITE OR CIRCULATE WITHOUT AUTHORS PERMISSION Abstract We investigate whether a supplier s use of trade credit varies in relation to the tax rates of its major customer(s). We find evidence that the lower the customer s tax rate, the longer the period of time the supplier extends trade credit. These results suggest that customers with already-low tax rates look to trade credit as a way to use the high-tax rate interest shields of their suppliers. This relation holds even after controlling for both supplier and customer financial constraints. Overall our study contributes to the literature on the determinants of trade credit and the literature on tax-coordinated supply chains. 1

2 I. INTRODUCTION Trade credit is an increasingly important source of financing and a relevant portion of the balance sheet for many firms. According to the Federal Reserve, the total trade receivables of nonfinancial corporate businesses exceeded $2.6 trillion in Trade credit theories seek to explain why industrial firms extend so much trade credit in economies with well-developed capital markets, where specialized financial institutions could provide financing. The vast majority of early studies in the literature use information imperfections to explain the flow of trade credit from large suppliers to smaller buyers. The underlying premise is that small buyers turn to suppliers for trade credit when financial institutions are unavailable or costly to access. This theory is supported by strong empirical evidence in the trade credit literature (Petersen and Rajan 1997). Yet, as more recent trade credit literature documents, large highly-rated buyers receive trade credit from their much smaller and often lower-rated suppliers (Fabri and Klapper 2008; Klapper, Laeven, and Rajan 2012; Murfin and Njoroge 2015). For example, Wal-Mart, an investment grade-rated firm that appears as an important customer in our sample, reported accounts payable of over $41 billion on the balance sheet ending January 31, 2017, accounting for over one-third of the its total liabilities, over half of its book-value of equity, and more than its total debt ($39 billion). It is a lot more difficult to reconcile the financing constraint explanation with the flow of trade credit from small low-rated suppliers to large well-rated buyers like Wal-Mart. In this paper, we focus on a relatively straightforward question: why do large creditworthy firms borrow trade credit from their much smaller, and often lower-rated, suppliers? 2

3 Brick and Fung (1984) provide a tax-based theoretical model that explains the flow of trade credit, without invoking credit market imperfections. The model shows that the relative magnitudes of the buyer and supplier s effective tax rates (ETR) determine the flow of trade credit for a given transaction. The model is predicated on the idea that trade credit provides a channel through which suppliers with high ETRs can transfer their more valuable interest tax shields to buyers with low ETRs. The model shows that when a buyer s ETR is higher than the supplier s ETR, the buyer will find its own interest tax shield more attractive than the implicit interest tax shield offered by the supplier through trade credit. In this case, the buyer will prefer cash terms to trade credit terms. In contrast, when the buyer s ETR is lower than the supplier s ETR, the buyer will find it more attractive to exploit the supplier s larger interest tax shield through trade credit. Brick and Fung s model specifically predicts that, all things equal, buyers with relatively lower ETRs will prefer trade credit compared to buyers with relatively higher ETRs. Likewise, suppliers with high ETRs will supply more trade credit relative to similar suppliers with lower ETRs. Brick and Fung s model provides a compelling tax-based explanation for the flow of trade credit from small suppliers to their much larger and creditworthy buyers as long as the small supplier s ETR is sufficiently higher than the large buyer s ETR. More importantly, neither party need be financially constrained. This tax-based explanation, which does not invoke information imperfections in financial or product markets, is our primary focus in this paper. It is important to emphasize that there are a wide variety of activities that can fall under the broad umbrella of tax planning, ranging from relatively uncontroversial methods such as depreciation to illegal tax shelters, which a firm can engage in to arrive at its ETR. In this paper, we do not examine what tax planning methods a firm may use to arrive at its ETR or the determinants of 3

4 that tax planning. Instead we focus on the possible consequences of the differential tax planning for trade credit flow in the cross section of buyer-supplier relationships. Desai, Foley, and Hines (2016) argue that the Brick and Fung model derives from a cash accounting regime, in which transactions are taxed when payments occur. According to Brick and Fung, however, cash accounting is merely a simplifying assumption intended to standardize tax payment dates in their model. Nevertheless, Desai et al., (2016) point out that under a strict accrual accounting method, which is required for all publicly traded firms, different predictions obtain. They develop an alternative model, which makes the opposite prediction that high buyer tax rates encourage trade credit borrowing. Their model predicts that firms with low tax rates are more likely to find it profitable to use trade credit to reallocate capital to firms with higher tax rates and higher pre-tax marginal products of capital. The authors document supporting empirical evidence, showing that U.S. multinational firms use trade credit to internally reallocate investment from subsidiaries in low-tax countries to subsidiaries in higher tax locations. Whether this finding is generalizable to explaining the flow of trade credit between unrelated parties is an answered empirical question. Ultimately, the Desai et al. model offers a competing tax-based explanation for the flow trade credit one in which small suppliers with low ETRs have incentives to offer trade credit to large buyers with higher ETRs. Recent tax planning literature hypothesizes and finds an association between a firm s tax planning and financial constraints. Law and Mills (2015) develop a new measure of financial constraints based on a firm s qualitative disclosures and find a positive relation between that measure and measures of aggressive tax planning. They also use the exogenous shock of local bank closures to examine how external financial constraints are related to a firm s tax avoidance strategies. Edwards, Schwab and Shevlin (2016) also investigate the relation between financial 4

5 constraints and tax planning. Their results suggest that financially constrained firms look to internally generate funds through increased tax planning, reducing their ETRs. In addition, based on the very idea of financial constraint it is not unreasonable to expect that constrained buyers are more likely to prefer trade credit to a cash transaction, while unconstrained suppliers are more likely to offer trade credit. Along with the results of these two studies this provides an alternative explanation for the predictions in Brick and Fung (1984). It may appear that trade credit is based on supplier-customer ETR differences, when really trade credit is just naturally flowing from unconstrained (high-etr) suppliers to constrained (low-etr) customers. A closer look, however, reveals that the financing constraint explanation is difficult to reconcile with the flow of trade credit for the overwhelming majority of buyer-supplier relationships in our sample. Each customer in our sample accounts for at least 10% of the supplier s revenues meaning the customers are typically much larger and more highly-rated than their suppliers. 1 Of the 3,243 different supplier firms in our sample only 10% are S&P 500 firms, compared to 40% of our 975 customer firms. The median customer-year in our sample has over $7.6 billion in total assets, over $7.3 billion in sales, and a market value of equity (MVE) of over $7.1 billion. In comparison, the median supplier-year in our sample has only $217 million in total assets, $207 million in sales, and a MVE of $211 million. Of our 16,787 supplier-years, only 24% are rated, compared to 74% of our 6,351 customer-years. Of the rated supplier-years, only 48% are investment grade, with a median rating of BB+. Of the rated customer-years, over 75% are investment grade, with a median rating of BBB+. Since access to external financing is easier for customers relative to suppliers for the vast majority of the buyer-supplier relationships in our sample, it would not appear that financial constraints alone could explain the flow of trade credit 1 Throughout the paper the terms rated and rating are used to refer to the S&P Domestic Long Term Issuer Credit Rating per the Compustat Ratings file. 5

6 in our setting. Nevertheless, we are careful to control for both supplier and customer financial constraints in all our tests and we conduct robustness tests using a variety of proxies for financial constraints. Our paper makes several contributions to the literature. First, our empirical results shed more light on why differences in ETRs could play an important role in the flow of trade credit from small suppliers to large customers. Murfin and Njoroge (2015) find that when large customers borrow trade credit from small suppliers, slower payment terms are associated with lower investment at the supplier level, particularly in financially constrained times/for financially constrained firms. Our results suggest there is a tax-based explanation for why major customers may borrow from small suppliers and find it attractive to pay more slowly. Our results also complement Petersen and Rajan (1997), who focus on small customers who borrow trade credit from large suppliers when access to credit from financial institutions is constrained. Their results suggest that suppliers use different information, compared to financial institutions, when evaluating whether to extend trade credit to low rated customers. Second, we contribute to the existing literature on tax avoidance. Specifically, we explore the use of trade credit as a form of tax planning. We find that although using trade credit may not directly lower the customer s ETR, large customers not only borrow trade credit from small suppliers, but more importantly, customers with low ETRs repay trade credit more slowly. This result appears consistent with the hypothesis that large low-etr customers find it more attractive to take advantage of the supplier s greater benefit from the tax deduction for interest. While there is a growing literature on methods of tax planning, such as the recent string of papers focused on transfer pricing (e.g., Dyreng and Markle 2016 and De Simone 2016), most of the literature focuses on tax planning completely internal to the firm. A notable exception is Cen et 6

7 al. (2016) who explore corporate tax avoidance in separately-owned suppliers and customers. Their main results suggest that both dependent suppliers and important customers have significantly lower ETRs than other firms. Supplemental tests in the paper support the authors assertion that tax havens may be one mechanism used in this supplier-customer tax avoidance. Similar to Cen et al. (2016) we investigate tax-motivated coordination in the supply chain. However, we explore a completely different mechanism that operates under a completely different set of circumstances. The coordination explored in Cen et al. relates to lower ETRs for both the supplier and customer while the trade credit mechanism we focus on in this paper does not lower a customer s ETR. Rather it allows a customer to take advantage of its supplier s more beneficial interest tax shield. As such we shed light on a new form of supplier-customer coordinated tax avoidance. Petersen and Rajan (1997) point out that the problem in testing theories of trade credit is attributable to the paucity of data. This may explain why tax-based theories of trade credit largely remain untested. A notable exception is Desai et al., (2016) who use low frequency data from the U.S. Bureau of Economic Analysis (BEA) survey of U.S. direct investment abroad in which detailed data on trade credit are available only every five years. The level of detail in this data does not identify non-affiliated buyers or suppliers. In addition, since Desai et al. focus on how U.S. multinationals use trade credit to internally reallocate capital between locations with different tax rates, it is difficult to generalize their findings to explaining the flow of trade credit between unrelated parties. In this study, we use annual data from the Compustat Segments Customer file to match firms to their major customers. Within the trade credit literature, our study stands alone in empirically testing tax-based theories that explain the flow of trade credit from small suppliers to their unrelated, and typically larger and higher-rated, customers. 7

8 II. HYPOTHESIS DEVELOPMENT Our research question is most closely associated with Brick and Fung (1984) (hereinafter BF) and Desai et al. (2016), which have opposite predictions. In the BF model the supplier offers the buyer a discount for paying cash today instead of using trade credit. The supplier structures the discount such that it is indifferent between receiving a smaller amount of cash that it can use/invest immediately (cash transaction) and a larger amount of cash at some point in the future, thereby forgoing the use/investment of that cash in the meantime (trade credit transaction). Ultimately, the prediction of the BF model hinges on the supplier s tax rate, ts, relative to the buyer s tax rate, tb. If ts exceeds tb the buyer will prefer trade credit because it shifts the tax sheltering to the supplier, who will derive a larger tax benefit under these conditions. Take a hypothetical situation involving Cisco (low-etr customer) and eplus (high-etr supplier). Cisco is about to make a purchase from eplus, but both firms need the cash related to that sale immediately. There are two options: (1) Cisco can borrow externally to pay eplus in cash today for the sale. (2) eplus can extend trade credit to Cisco and borrow externally to fund operations while waiting to receive cash from Cisco. A critical difference between these scenarios is where the tax-deductible interest expense is located. In first (second) scenario Cisco (eplus) will be able to deduct the interest expense for tax purposes. Since Cisco has a much lower ETR the after-tax cost of the interest payments (assuming both firms borrow at the same rate) will be much higher for Cisco. 2 It would be more beneficial overall to choose the second scenario since 2 In our empirical tests, rather than assuming the customer and seller could borrow at the same rate, we control for customer and seller financial constraints. 8

9 eplus will have a much lower after-tax interest expense. This conclusion would be reversed in a scenario with a low-etr supplier and a high-etr customer. Desai et al. create their own model, centered on the marginal products of capital for a U.S. parent company buyer and its international subsidiary supplier. There is an incentive to get cash to the party with the higher pre-tax marginal product of capital. If the buyer has the higher pretax marginal product of capital it is more productive for the buyer to keep its cash and delay payment (trade credit transaction). If the supplier has the higher pre-tax marginal product of capital it is more productive for the supplier to get the cash immediately (cash transaction). In the trade credit scenario the buyer will pay later, but will pay additional interest as a condition of the trade credit agreement. Assuming the buyer and supplier have equal after-tax marginal products of capital, the model has the opposite prediction from BF that the benefit of using trade credit over cash increases as the buyer s ETR exceeds the supplier s ETR since the buyer will be getting a benefit from the deduction of that interest at a higher rate than the supplier will be paying tax on that interest income. We will now return to Cisco (now a high-etr customer), but instead of buying from an unrelated supplier it will make a purchase from its Cayman Islands subsidiary, Cisco Holdings Cayman Ltd (Holdings a low-etr supplier). Under the Desai et al. model if Cisco uses trade credit rather than cash for its purchase it will pay additional interest to Holdings. Since Cisco will be subject to a higher rate of tax in the U.S. than Holdings will be in the Cayman Islands, the tax deduction Cisco receives will exceed the tax payment Holdings makes on the same interest. If Cisco purchased from a subsidiary in a high-tax country the situation would reverse and Cisco would prefer to pay cash. As mentioned in Section I, Desai et al. note that BF use a cash accounting regime in their model, while publicly-traded firms must use accrual accounting. BF use cash accounting to 9

10 simplify the exposition of their model, but acknowledge (for example, in footnote four of their paper) that as a result their model may not hold in all real-life scenarios. Although Desai et al. find results that support their model, their data is from U.S. parent firms and their international subsidiaries. Their model assumes an equal cost of capital for both buyer and supplier, which is much less likely to hold in scenarios where the buyer and supplier do not share common ownership. Therefore, it remains an open question as to whether either model will hold for unrelated suppliers and customers. Given the opposite predictions from the BF model and the Desai et al. model, our null hypothesis is: There is no relation between a supplier s receivable days and the excess of that supplier s ETR over the ETR(s) of its principal customer(s), ceteris paribus. III. RESEARCH DESIGN AND DATA We identify supplier-customer pairs (referred to as dependent suppliers and principal customers in Cen et al. (2016)) using the Compustat Segment Customer file. The data is based on supplier firms annual disclosures of significant customers, indicating that the customer accounts for a meaningful portion of the supplier s sales in that year. We then match both the customer and supplier firms to Compustat data to calculate all of our variables. 3 Following Murfin and Njoroge (2015), we calculate receivable days (Rec_Days), our dependent variable, as average trade accounts receivable divided by sales and multiplied by We also use the variable SPLITCRM from the Compustat Ratings file to construct the ratings-related variables used in the financial constraint robustness tests, as described in Section IV. 10

11 To control for how receivable days relates to other parts of the supplier s financing cycle we also calculate payable days (Pay_Days), and inventory days (Inv_Days). 4 Hanlon and Heitzman (2010) discuss the various tax measures used in the literature and note that researchers should take care in using the measure(s) appropriate for a given research question. We focus on GAAP ETR in our main tests because survey results Graham et al. (2014) show that a large majority of publicly-traded firms in their sample value GAAP ETR as a metric at least as much as cash ETR, with almost 50 percent of surveyed publicly-traded firms considering GAAP ETR to be the more important metric. We form our initial regressions with these trade credit and tax variables 5 : Sup_Rec_Daysi,t = β0 + β1sup_etri,t-1 + β2cust_etrj,t-1 + Σβ3Sup_Trade_Credit_Controlsi,t + Σβ4Cust_Trade_Credit_Controls j,t + β5sup_sizei,t-1 + β6cust_size j,t-1 + β7sup_nolcfi,t-1 + β8cust_nolcf j,t-1 + β9sup_hp_indexi,t + β10cust_hp_index j,t + β11sup_roe i,t-1 + β12cust_roe j,t-1 + β13sup_lossi,t + εi,t (1) Sup_ETR and Cust_ETR represent GAAP ETR for the supplier and customer, respectively. 6 We expect the coefficient on Sup_ETR to be positive and the coefficient on Cust_ETR, our main variable of interest, to be negative. BF predict that if the supplier s ETR exceeds that of the customer both parties will prefer a trade credit transaction. If the customer s ETR exceeds that of the supplier both parties will prefer a cash transaction. If we think of the difference in ETRs as the supplier s ETR less the customer s ETR we can see that the higher the supplier s ETR, the 4 See Appendix A for a full description of the variables. 5 We also include year indicator variables in our regressions, omitted here for brevity. 6 Following the prior literature we censure ETR at 0 and 1. 11

12 higher the difference in favor of trade credit. Conversely, the higher the customer s ETR, the lower that difference is, lowering the advantage of trade credit over cash. Sup_Trade_Credit_Controls include whatever other pieces of the cash to cash cycle are not included in the dependent variable. For example, when our dependent variable is Rec_Days we include Pay_Days and Inv_Days as control variables. We include customer receivable days and inventory days to control for the customer s own trade credit practices. We also control for both supplier and customer size, measured as the natural log of total assets indexed for inflation. We expect the coefficient on supplier (customer) size to be negative (positive) since as the supplier (customer) gets larger we expect the firm to use its bargaining power to shorten (lengthen) the trade credit period. Sup_NOLCFi, t-1 and Cust_NOLCFi, t-1 control for tax loss carryforwards at the supplier and customer, respectively. We expect the coefficients on these variables to be the opposite of the ETR variables since they control for firm NOLs which can be used to lower income tax liabilities. We control for both supplier and customer financial constraints (HP_Index), as previously discussed, by using the index developed by Hadlock and Pierce (2010). In their model Desai et al. (2016) assume the buyer and supplier will have equal aftertax marginal products of capital. Since we are examining unrelated suppliers and customers we make no such assumption. Therefore, we include supplier and customer return on equity (ROE) as control variables. Finally, Sup_Loss is an indicator variable equal to one if the supplier has a pre-tax loss that year, zero otherwise. IV. RESULTS We begin with an analysis of the entire Compustat database for years 1990 through In looking at Compustat as a whole we cannot identify whether a firm is a dependent supplier, an 12

13 important customer, or both. Therefore, we slightly modify equation (1) by only including the supplier variables so that we may examine whether there is a relation between a firm s receivable/payable days and its own ETR: 7 Rec_Daysi,t = β0 + β1etri,t-1 + Σβ2Trade_Credit_Controlsi,t + β3sizei,t-1 + β4nolcfi,t-1 + β5hp_indexi,t + β6roe i,t-1 + β7lossi,t + εi,t (2) Even without being able to pair a firm to its suppliers and/or customers, based on the predictions in BF we would generally expect relatively higher-etr firms to have higher receivable days and lower payable days. Column 1 (Column 2) of Table 1 shows the results of the regression with Rec_Days (Pay_Days) as the dependent variable. The coefficient of interest is both columns is ETR. In column 1 (column 2) we see that the coefficient on ETR is positive (negative) and statistically significant. These results confirm our predictions: firms with higher ETRs tend to have receivables (payables) outstanding for a longer (shorter) period of time, even after controlling for a variety of firm characteristics. Next we focus on our identified supplier-customer relationships. Table 2 provides a comparison between supplier and customer firm-years for a few key variables. Whether based on total assets, sales, or market value of equity (MVE) the customer firms are much larger than the supplier firms. This is to be expected considering the data is based on supplier firms reporting customers that account for a significant portion of their sales. We also see that on average our supplier firms have lower ETRs than our customer firms. Finally, on average our supplier firms extend trade credit longer as suppliers, but pay slightly faster as purchasers 7 We recognize that there is potential for endogeneity issues since a firm may simultaneously determine its own receivable/payable days and its ETR. This is only to give us a sense of how our predictions hold when looking at the general population of firms. Our later tests should address any potential endogeneity issues. 13

14 compared to our customer firms. We have 3,243 supplier firms in our sample. Over half of those suppliers (1,636 supplier firms) mention more than one customer during our sample period. We have 975 customer firms, 514 of which show up as a customer for multiple suppliers. Table 3 shows the results for equation (1) using our supplier-customer sample. The coefficients on our main variable of interest is Cust_ETR negative and statistically significant at the one percent level. These results suggest that as the customer s ETR rises that supplier extends trade credit for a shorter period of time, as per our prediction. The coefficients on our other control variables are also generally as expected, although several of them are not statistically significant. The prediction in BF is about the supplier s ETR relative to the customer s ETR. Therefore, we next divide our sample into two groups: observations where Sup_ETR is greater than Cust_ETR and observations where Sup_ETR is lower than Cust_ETR. We expect the relation between Sup_Rec_Days and Cust_ETR as shown in Table 3 to be concentrated in the subsample where the supplier has the higher ETR. Table 4 has the results of this subsample analysis. Columns 1 and 2 show the results of equation (4) for observations where Sup_ETR exceeds Cust_ETR and Cust_ETR exceeds Sup_ETR, respectively. As expected, the relation between Cust_ETR and Sup_Rec_Days is stronger for those observations where the supplier has the higher ETR. Although the coefficient on Cust_ETR remains significant and negative in column 2 the absolute value of the coefficient is smaller and the significance is only at the ten percent level. 8 8 Supplemental analysis suggests that the difference between the coefficients on Cust_ETR in columns 1 and 2 is statistically significant. 14

15 We are not aware of any omitted variable that would account for the relation Sup_Rec_Days and Cust_ETR, particularly given that those two variables are coming from two separate firms with no common ownership. However, to provide further support for our results we examine an exogenous shock to ETRs. The domestic production activities deduction (DPAD) was enacted in Section 199 of the Internal Revenue Code as part of the American Jobs Creation Act (AJCA) of The DPAD essentially allows taxpayers an additional deduction in calculating taxable income, based on qualifying production activities in the United States. 10 We have four possible scenarios with our supplier-customer pairs in regards to the DPAD: (1) both customer and supplier claim the DPAD, (2) neither customer nor supplier claim the DPAD, (3) only the customer claims the DPAD, and (4) only the supplier claims the DPAD. These scenarios will have varying effects on the relative differences in supplier-customer ETRs. For simplicity, we start with a pre-dpad scenario where the supplier s ETR is higher than its customer s ETR. Under both scenarios (1) and (2) we generally would not expect a change in the relative difference between customer and supplier ETR because either both ETRs would decrease or neither ETR would decrease. Under scenario (3) only the customer s ETR would be reduced by the DPAD, increasing the difference between supplier and customer ETRs. Based on our previous results this would only make trade credit more attractive. Only the supplier s ETR would be reduced under scenario (4), shrinking the difference between supplier and customer ETRs, which would make trade credit less attractive. Thus we have predictions on how we expect supplier receivable days to change in the post-dpad period depending on whether the supplier and/or the customer claim the DPAD. 9 The AJCA contained other tax provisions, but we do not expect any of those provisions would map onto our sample in the same manner as the DPAD. 10 For a more detailed discussion of the DPAD see Deloitte Tax LLP (2005). 15

16 Using IRS corporate statistics of income we identify the 18 two-digit SIC industry groups that account for over 80% of the DPAD deductions from active corporations from 2005 through 2013 (the post-ajca period for which DPAD data is available from the IRS). The DPAD indicator equals one for firms (customer or supplier) that are in those 18 industries. We then create indicator variables based on which of the four scenarios applies: (1) both the customer and supplier are in these DPAD industries (BOTH equals one, zero otherwise), (2) neither the customer nor the supplier is in a DPAD industry (NONE equals one, zero otherwise), (3) only the customer is in a DPAD industry (C_ONLY equals one, zero otherwise), and (4) only the supplier is in a DPAD industry (S_ONLY equals one, zero otherwise). We also add POST, an indicator for fiscal years after the enactment of the AJCA. We interact POST with each of the scenario indicators and then run equation (1) for each scenario, replacing our ETR variables with POST, the scenario variable, and the interaction variable. The first four columns of Table 5 show the results for each scenario. 11 In column 1 the coefficients on both POST and BOTH are statistically insignificant while the coefficient on the interaction term is positive and significant at the five percent level. This suggests that when both the customer and supplier qualify for the DPAD the supplier extends credit to the customer for a longer period of time after the DPAD is enacted. Column 2 shows that only the coefficient on NONE is significant, suggesting that the enactment of the DPAD had no effect on receivable days for scenarios where neither party qualifies for the DPAD. In column 3 we see that the coefficient on POST is insignificant while the coefficients on C_ONLY and POST*C_ONLY are positive and significant at the one and five percent level, respectively. This suggests that in general when the only the customer is in a DPAD industry the supplier tends to extend credit to that customer for a longer period of time, 11 We only include the variables of interest in Table 5 due to space constraints. The sign and significance of other coefficients are the similar to Table 3. 16

17 particularly after the DPAD is enacted. This result matches our prediction: that if only the customer qualifies for the DPAD the enactment of the AJCA lowered the customer s ETR relative to the supplier s ETR, making trade credit more attractive. In column 4 we see that while the coefficients on POST and S_Only are not statistically significant, the coefficient on POST*S_ONLY is negative and significant at the one percent level. Again, this matches our earlier prediction: that if only the supplier qualifies for the DPAD the enactment of the AJCA lowers the supplier s ETR relative to the customer s ETR, making trade credit less attractive. In column 5 we simultaneously test all the scenarios by including indicators for BOTH, C_ONLY, and S_ONLY; POST; and the related interaction terms. The only statistically significant coefficients in this comprehensive model are C_ONLY, which is positive, and POST*S_ONLY, which is negative. This suggests that the AJCA only significantly affected Sup_Rec_Days in cases where the exogenous shock of the DPAD applied to the supplier, but not the customer. To assess the validity of this DPAD test rather than dividing firms based on IRS DPAD data we randomly pick 18 SIC codes from our sample as pseudo-dpad industries. When we run our regressions using these pseudo-dpad industries (results untabulated) the results from Table 5 do not hold. These results support our conclusion that the ETR effects from the exogenous shock of the DPAD are related to changes in supplier receivable days. Specifically, in scenarios where only the supplier s ETR was lowered by the DPAD these suppliers respond by collecting receivables sooner now that the tax advantage of trade credit has been diminished. Finally, we conduct (untabulated) robustness tests where we substitute in other measures of financial constraint for HP_Index. First we look to the two firm-specific financial constraint methods used in Edwards et al. (2016), the Altman (1968) Z-score and the KZ Index as developed by Kaplan and Zingales (1997). We then create two variables based on a firm s credit 17

18 rating. Invt_grade is equal to one if the firm s credit rating in that year is BBB or higher, zero otherwise. Rated is equal to one if the firm has a credit rating in that year, zero otherwise. Substituting any of these four variables for HP_Index in equation (1) does not materially alter our results. Also, to ensure that our results are not simply picking up on a general trend in financial constraints we rerun equation (1) after restricting our sample to only include observations where neither the supplier nor the customer experiences a change in Rated over our entire sample period. Our results remain qualitatively similar to those in Table 3. In total these results give us confidence there is a negative relation between a supplier s receivable days and its customer(s) ETR, in line with the BF theoretical model. V. CONCLUSION In conclusion, we find evidence that supplier firms extend trade credit for longer periods of time as their own ETRs exceed the ETRs of their customers. This is an important contribution to the literature because it identifies another important determinant of trade credit, which is often a major source of assets (accounts receivable) and/or liabilities (accounts payable) on a firm s balance sheet. In particular we provide evidence on why large non-financially constrained firms may want to obtain trade credit from their smaller suppliers. This paper also contributes to the tax avoidance literature by identifying a form of tax planning that does not lower the customer s ETR, but rather allows that customer to benefit from their supplier s relatively higher ETR. We also contribute to the literature on supplier-customer relationships, showing how taxes can be an important part of that relationship. 18

19 REFERENCES Altman, E Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of Finance 23 (4): Brick, I., and W. Fung Taxes and the theory of trade debt. The Journal of Finance 39 (4): Cen, L., E. Maydew, L. Zhang, and L. Zuo Supplier-customer relationships and corporate tax avoidance. Journal of Financial Economics 123 (1): De Simone, L Does a Common Set of Accounting Standards Affect Tax-Motivated Income Shifting for Multinational Firms? Journal of Accounting and Economics 61 (1): Deloitte Tax LLP. Producing Results: An Analysis of the New Production Activities Deduction. Tax Analysts Special Report February 21, 2005: Desai, M., C. Foley, and J. Hines Jr Trade credit and taxes. The Review of Economics and Statistics 98 (1): Dyreng, S. and K. Markle The Effect of Financial Constraints on Income Shifting by U.S. Multinationals. The Accounting Review 91 (6): Edwards, A., C. Schwab, and T. Shevlin Financial Constraints and Cash Tax Savings. The Accounting Review 91 (3): Fabri, D., L. Klapper Trade Credit and the Supply Chain. Working paper, University of Amsterdam. Graham, J., M. Hanlon, T. Shevlin, and N. Shroff Incentives for Tax Planning and Avoidance: Evidence from the Field. The Accounting Review 89 (3): Hadlock, C. and J. Pierce New Evidence on Measuring Financial Constraints: Moving Beyond the KZ Index. The Review of Financial Studies 23(5): Hanlon, M. and S. Heitzman A Review of Tax Research. Journal of Accounting and Economics 50 (2/3): Kaplan, S., and L. Zingales Do investment-cash flow sensitivities provide useful measures of financing constraints? Quarterly Journal of Economics 112 (1): Klapper, L., L. Laeven, and R. Rajan Trade credit contracts. The Review of Financial Studies 25 (3): Law, K. and L. Mills Taxes and Financial Constraints: Evidence from Linguistic Cues. Journal of Accounting Research 53 (4):

20 Murfin, J. and K. Njoroge The implicit costs of trade borrowing by large firms. The Review of Financial Studies 28 (1): Petersen, M. and R. Rajan Trade Credit: Theories and Evidence. The Review of Financial Studies 10 (3):

21 Appendix A Variable Definitions Rec_Days Average accounts receivable divided by sales, multiplied by 365 Pay_Days Average accounts payable divided by purchases (cost of goods sold plus the change in inventory), multiplied by 365 Inv_Days Average inventory divided by cost of goods sold, multiplied by 365 ETR Total taxes paid, divided by pre-tax income less special items Size Natural logarithm of total inflation-deflated assets NOLCF Year t tax loss carryforward, divided by year t-1 total assets HP_Index -(0.737*Size) + (0.043*Size 2 ) - (0.04*Age), where Age is determined based on the first year the firm appears in Compustat ROE Net income divided by average stockholders equity Loss Indicator variable equal to one if the firm has a pre-tax loss for the year, zero otherwise Post Indicator variable equal to one for years 2005 and later, zero otherwise Both Indicator variable equal to one when both the supplier and customer are in DPAD industries (based on two-digit SIC codes), zero otherwise None Indicator variable equal to one when neither the supplier nor the customer are in DPAD industries (based on two-digit SIC codes), zero otherwise C_Only Indicator variable equal to one when only the customer is in a DPAD industry (based on two-digit SIC codes), zero otherwise S_Only Indicator variable equal to one when only the supplier is in a DPAD industry (based on two-digit SIC codes), zero otherwise 21

22 Table 1 Full Compustat Sample (1) (2) Rec_Days Pay_Days GAAP_ETR * *** (1.2882) (1.4197) Invt_Days *** *** (0.0050) (0.0058) Pay_Days *** (0.0097) Rec_Days *** (0.0134) Size *** *** (0.2443) (0.2844) NOLCF *** *** (0.2137) (0.3964) HP_Index *** *** (0.5970) (0.6634) ROE (0.0035) (0.0066) Loss *** *** (0.4358) (0.4856) Intercept *** *** (1.7497) (1.7714) N 90,548 90,548 Variables as defined in Appendix A. Standard errors, clustered by firm, in parentheses. Coefficients for year indicator variables are omitted for brevity. ***, **, and * denote statistical significance at the one percent, five percent, and ten percent level, respectively. 22

23 Table 2 Descriptive Statistics Total Assets Sales MVE Supplier Customer Supplier Customer Supplier Customer Mean 2,573 23,096 *** 2,039 20,499 *** 3,070 23,657 *** Median 218 7, , ,159 Minimum Maximum 552, , , , , ,550 ETR Rec_Days Pay_Days Supplier Customer Supplier Customer Supplier Customer Mean *** *** ** Median Minimum Maximum N = 16,787 for suppliers and 6,351 for customers for all variables (except for MVE: supplier N = 16,768 and customer N = 6,342). Total Assets, Sales, and MVE are in millions of dollars. *** and ** indicate the difference between the supplier and customer mean for the variable is statistically significant at the one and five percent level, respectively. 23

24 Table 3 Main Regression Supplier-Customer Pair Sample Sup_Rec_Days Sup_ETR *** (1.0623) Cust_ETR *** (1.4859) Sup_Pay_Days *** (0.0094) Sup_Inv_Days *** (0.0061) Cust_Rec_Days *** (0.0068) Cust_Inv_Days (0.0074) Sup_Size *** (0.3038) Cust_Size *** (0.3104) Sup_NOLCF *** (0.2523) Cust_NOLCF (1.9951) Sup_HPIndex (0.7716) Cust_HPIndex *** (1.0702) Sup_ROE (0.2361) Cust_ROE (0.5433) Sup_Loss (0.6486) Intercept *** (4.0812) N 24,226 Variables as defined in Appendix A. Standard errors, clustered by supplier-customer pair, in parentheses. Coefficients for year indicator variables are omitted for brevity. ***, **, and * denote statistical significance at the one percent, five percent, and ten percent level, respectively. 24

25 Table 4 Supplier/Customer ETR Comparison (1) (2) Sup_ETR > Cust_ETR Cust_ETR > Sup_ETR Sup_ETR *** (1.5206) (2.6521) Cust_ETR *** * (2.8957) (1.9022) Sup_Pay_Days *** *** (0.0161) (0.0100) Sup_Inv_Days *** *** (0.0089) (0.0067) Cust_Rec_Days ** *** (0.0076) (0.0080) Cust_Inv_Days (0.0107) (0.0080) Sup_Size *** *** (0.3859) (0.3385) Cust_Size *** *** (0.3659) (0.3556) Sup_NOLCF *** *** (0.4923) (0.2611) Cust_NOLCF (2.4199) (2.3022) Sup_HPIndex (0.9652) (0.8623) Cust_HPIndex *** *** (1.1772) (1.2543) Sup_ROE (0.4246) (0.2689) Cust_ROE (0.6531) (0.7522) Sup_Loss * (0.8329) (0.8567) Intercept *** *** (4.9432) (4.8345) N 9,477 15,336 25

26 Dependent variable is Sup_Rec_Days. Variables as defined in Appendix A. Standard errors, clustered by supplier-customer pair, in parentheses. Coefficients for year indicator variables are omitted for brevity. ***, **, and * denote statistical significance at the one percent, five percent, and ten percent level, respectively. 26

27 Table 5 Domestic Production Activities Deduction (1) (2) (3) (4) (5) Both None C_Only S_Only Both, C_Only, S_Only Post (1.9984) (1.9097) (1.9212) (1.9124) (2.4609) Both (0.9142) (1.1927) Post*Both ** (1.2438) (1.8786) None ** (1.0712) None*Both (1.7681) C_Only *** *** (1.1626) (1.4151) Post*C_Only ** (1.7934) (2.3245) S_Only (1.0692) (1.3330) Post*S_Only *** ** (1.3968) (2.0300) Dependent variable is Sup_Rec_Days. Variables as defined in Appendix A. N = 24,226 for all specifications. Standard errors, clustered by supplier-customer pair, in parentheses. Coefficients for year indicator variables are omitted for brevity. ***, **, and * denote statistical significance at the one percent, five percent, and ten percent level, respectively. 27