Auditor choice and information asymmetry: evidence from international syndicated loans

Abstract

Analyzing a large sample of non-US public firms from 31 countries that obtain private loans, we find that loan syndicates that lend to borrowers that employ Big N auditors are larger and less concentrated and that the lead arrangers and largest investors of these syndicates are able to hold a lower proportion of the loan after issuance. Further analysis demonstrates that this effect exists only in countries with strong creditor rights and in those countries with high levels of societal trust, suggesting that both sound formal and informal institutional factors are prerequisites for lenders and borrowers to benefit from differential audit quality on loan syndicate structure efficiency. Furthermore, we find that the loan syndicate structure benefits for borrowers that employ Big N auditors are higher for borrowers with greater information asymmetry problems, but we do not find that Big N audits are able to address the information asymmetry and moral hazard issues between the lenders themselves.

Keywords:

• big N auditors
• international debt markets
• loan syndicate structure
• creditor rights
• trust

JEL Classification:

• G01
• M4
• M49

Acknowledgments

We thank an anonymous reviewer, Gary Biddle, Peter Chen, Mark Clatworthy (editor), Jeffrey Cohen, Mark DeFond, Peter Clarkson, Sadok El Ghoul, Mingyi Hung, John Lyon, and Miguel Minutti-Meza for helpful comments and suggestions.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Supplemental data

Supplemental data for this article can be accessed at https://doi.org/10.1080/00014788.2018.1507810

Notes

1. We follow prior studies and refer to increased loan “efficiency” throughout the paper to indicate (1) the ability for lead arrangers to hold a smaller portion of the loan, and (2) larger loan syndicate sizes, with both conditions arising from a decrease in information asymmetry between loan lead arrangers and other syndicate participants (see, e.g., Leland and Pyle 1977).

2. Loan issuances in the syndicated loan market continue to grow and were approximately $2 trillion in 2013, with $700 billion made to US borrowers and $1.3 trillion made to non-US borrowers (Lee et al. 2017).

3. Our results are robust to (1) excluding influential countries and the global financial crisis (GFC, hereafter) period; (2) using various different panel specifications that control for time invariant and slow-moving characteristics; (3) alternative measures of high quality auditors and (4) addressing omitted variable issues and the selection issue related to choosing Big N auditors. Although none of our robustness tests is individually able to rule out all concerns about endogeneity, given that our results hold in variety of analyses that use disparate techniques, it is unlikely that our results are purely driven by unobservable factors. We discuss these results in more detail in Sections 4 and 5.

4. One exception is Chin et al. (2014) based on the unique auditor data from Taiwan only.

5. We acknowledge that lenders may use a set of modified accounting numbers (Leftwich 1983, ElGazzar and Pastena 1990, Li 2010, Baylis et al. 2017, Dyreng et al. 2017), but this does not eliminate the demand for high quality financial reporting because these numbers are likely to be the starting point in contracting (e.g., Li 2016) and will be associated with the overall information environment of the firm (e.g., Kim et al. 2011).

6. Early studies (e.g., DeAngelo 1981, Datar et al. 1991) argue that large, prestigious public accounting firms (Big N auditing firms) have incentives to protect their investment in reputation capital and are more likely than other auditors to supply a high-quality audit.

7. Consistent with this view, previous studies provide evidence that Big N auditors improve financial reporting quality and perform an external monitoring role using samples of non-US firms (e.g., Choi and Wong 2007, Francis and Wang 2008).

8. We follow prior studies, such as Kim and Song (2011), and use the number of lenders as one of our measures of syndicate structure efficiency. An alternative view on the association between audit quality and the number of lenders suggests that poor auditing could lead banks to attempt to share high risk loans with more parties. It is also possible that lenders could use a diffuse syndicate to mitigate a possible strategic default from a client, occurring, for example, when a larger fraction of a loan is held by the lead arranger. Therefore, we also examine whether high quality auditors lead to lower loan ownership concentration and lead arranger ownership, which mitigates concerns related to this alternative view.

9. We note that there also exists a contra argument, namely that when creditor rights are weak lenders need to form concentrated syndicates to monitor borrowers. When creditor rights are strong, however, such monitoring is not needed because lenders have greater control over bankrupt borrowers’ assets and cash flow. However, the findings in prior studies do not support this argument.

10. Other studies show that social trust promotes investment, trade, and economic growth, encourages financial development and investors’ participation in the stock market, and facilitates venture capital investment, corporate financing, and cross-border mergers and acquisitions (Knack and Keefer 1997, Zak and Knack 2001, Guiso et al. 2004, 2008, Bottazzi et al. 2011, Ahern et al. 2015).

11. To make our study more comparable with prior studies investigating loan syndicate structure (e.g., Qian and Strahan 2007, Sufi 2007), we conduct our analysis using OLS. To mitigate the concern that using a discrete and non-negative variable (e.g., the number of lenders) as the dependent variable leads to incorrect inferences, we also check the robustness of our results using a negative Binominal regression (see, e.g., Rock et al. 2000, Kim and Song 2011), and our inferences do not change. In addition, we also consider whether our results are robust in a GMM model that could more efficiently corrects heteroskedastic (and/or auto-correlated) errors. Furthermore, we repeat our analysis and run a weighted-least squares (WLS) regression that helps to address the disproportionate representation of countries as in Choi and Wong (2007). Our results hold with no change to our inferences (untabulated).

12. In an untabulated robustness check, we redefine Big N to include Big Four auditors only (i.e., Ernst & Young, Deloitte & Touche, KPMG, and PricewaterhouseCoopers) to mitigate concerns that our findings are driven by the relatively small, merged, or fraudulent auditors that were the predecessors of the Big Four auditing firms in the later sample period. We find similar results (with no change in inferences) using this alternative definition. In addition, to mitigate the concern that our measure of high quality auditors using a Big N indicator variable contains measurement error, particularly in a cross-country setting, we also define high quality auditors as industry specialists (instead of Big Four auditors) for auditors that have the largest market share (measured as the number of clients) in a certain industry. Our inference that high quality auditors improve loan structure efficiency does not change (untabulated).

13. We follow Sufi (2007) and use two variables in order to categorize lenders as either lead arrangers (“Lenders-Lead Arranger”) or non-lead arranger participants (“Lenders-All Lenders”) from the Dealscan database. If the variable “Lenders-Lead Arranger” is populated, we categorize the lender listed as the lead arranger, and all other lenders are considered participants. If this variable is not available, then any lender designated as having a “Lead Role” in “Lenders-All Lenders” is designated as a lead arranger.

14. Syndicated loans often bundle multiple facilities into one transaction. These different facilities have different contract terms but are syndicated as a single deal. We average variables that are measured at the facility level.

15. Because we cluster standard errors at the country level, we automatically control for clustering at the lower level that is contained in a country (e.g., borrower-level) (Bertrand et al. 2004, Dinc 2005, Cameron and Miller 2011). Our main results, however, hold if we (i) cluster at the borrower or lead arranger-level (e.g., Ball et al. 2008), (ii) adopt two-way clustering (borrower and year, lead arranger and year, industry and year, or country and year) (Petersen 2009, Gow et al. 2010), or (iii) use Huber-White standard errors without clustering.

16. We measure societal trust based on the following question from the WVS: “Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people?” We recode the response to this question to 1 if a survey participant reports that most people can be trusted (>50%), and 0 otherwise, and then calculate the mean of the response in each country year.

17. Our proposed mechanism through which the positive Big N effect on loan structure works depends on the extent to which market participants trust the monitoring of Big N auditors; therefore, it is more appropriate to measure this variable based on an auditor’s country. However, we are not able to identify whether the country of syndication is the country of the auditor using our data. Instead, we assume that the country of syndication is likely the same as the country of the auditor, which has been the empirical approach adopted in other cross-country studies of Big N auditors (e.g., DeFond et al. 2000, Choi and Wong 2007, Francis and Wang 2008) and measure social trust based on the borrower’s country.

18. Our results are not affected by including or omitting Korean, Japanese and Indian borrowers given the findings of Francis and Wang (2008) who point out that there may be potential miscoding of the auditor identification variable in these countries. Our results also continue to hold when (1) we exclude post-2006 years from the analysis, or (2) we use historical auditor data from Worldscope until 2009 in order to avoid measurement error stemming from the auditor identity miscoding in Compustat Global starting from 2005 (El Ghoul et al. 2016). In addition, because our sample period overlaps with the GFC, it is possible that our results may reflect crisis-specific behavior on the part of Big N auditors, borrowers, or lenders. Thus, we repeat our analyses after deleting observations from 2007 to 2008. We classify 2007 and 2008 as the crisis period following prior studies (e.g., Ryan 2008), and all results hold after omitting these observations.

19. This requirement results in a large sample of borrowers from Taiwan. In our additional tests section, we provide evidence that our results are unaffected by excluding Taiwanese borrowers. In addition, requiring non-missing cost of borrowing data substantially reduces our sample size. Our results are not affected if we do not require non-missing cost of borrowing.

20. 12.75% = exp(0.12) - 1, where 0.12 is the coefficient on Big N in Column (1). We use the natural log of the number of participants in the loan syndicate, Log (Number of Lenders), as the dependent variable.

21. −12.47% = −2.17/17.40, where −2.17 is the coefficient on Big N in column (3), 17.40 is mean value of Lead Arranger Shares for non-Big N sample.

22. Although we include a series of control variables that are found to affect the structure of loans, first, it is possible that some omitted but unobservable borrower characteristics drive the differences we find (e.g., borrower’s country or other slow-moving firm characteristics); second, we also note that some lenders are more likely to ask for disclosure covenants or other requirements that could potentially be correlated with using a Big N auditor. To address those concerns, we have done the following to mitigate the concern that our results are driven by an omitted variable issue with no change to our inferences: (1) added borrower firm fixed effects; (2) included lead arranger fixed effects; or (3) controlled for the likelihood of a loan having disclosure covenants. For this last test, we conduct a principal component analysis to measure the likelihood of disclosure covenant inclusion using the set of determinant variables that significantly predict the usage of disclosure covenants following Table 3 of Carrizosa and Ryan (2017).

23. Our results are not affected if we focus on a subsample in which lead arrangers and borrowers are in the same country, although our testing power is significantly affected because the sample size is reduced by 25%.

24. Our results are not affected if we use firm size (Size), or firm age as alternative measures (e.g., Krishnaswami et al. 1999).

25. Kim and Song (2011) use prior borrower-lender relationships and credit ratings. However, these two measures are highly associated with information asymmetry between borrowers and lenders (Sufi 2007; Chava et al. 2009), though they could also indirectly impact the information asymmetry among lenders.

26. A factor limiting lead arranger moral hazard and information asymmetry between lead arrangers and other participants is the lead arranger’s reputation. Because lead arrangers are responsible for ex ante due diligence, allocation of the loan to other syndicate members, and ex post monitoring; banks in the syndicate will often rely on the lead bank's reputation in making lending decisions (Ross 2010). Because the lead arrangers and syndicate participants are repeat players in the loan syndication market, if the lead arranger shirks in their due diligence and monitoring activities, it faces a credible threat of loss of reputation and future income (Pichler and Wilhelm 2001). Banks engaged as lead arrangers need to build trust with potential syndicate participants in order to retain substantial fee income from subsequent syndicated loan arranging activities.

27. We would expect to find the positive effect of Big N to be more pronounced for loans in countries with lower levels of societal trust (i.e., information asymmetry among lenders is greater if lenders are less likely trust with each other) if the second channel is true. We, however, fail to find supporting evidence for this explanation as documented in Table 5.

28. There are several recent studies which report that lenders express preferences for larger auditors. For example, the UK Competition Commission Investigation of the Market for Statutory Audit Services, available at https://www.gov.uk/cma-cases/statutory-audit-services-market-investigation.

29. As we discussed before, our results are also similar when we add lead arranger fixed effects or include an indicator variable for loans that have disclosure covenants.

30. Alternatively, matching non-Big N auditors with Big N auditors at the firm-year level also yields similar results.

31. For example, suppose that two firms sign debt contracts in 2004 and are audited by the same non-Big Four auditor. Further suppose that Firm A initially hired the auditor in 1994, whereas Firm B did not do so until 2003. It follows that any bias in the coefficient estimates arising from endogeneity is likely to be worse for Firm B because this auditor choice occurred shortly before its debt issuance. For example, Firm B may have been audited prior to 2003 by a Big N auditor, which may have resigned after concluding that Firm B had become a high-risk client because of its low accounting quality. Alternatively, Firm B may have dismissed its incumbent auditor in 2003 in favor of appointing a lower quality non-Big N auditor during a period with higher business risk. In either case, endogeneity is likely to be more serious for Firm B because there is a shorter lag between the choice of auditor and its decision to issue debt.

32. To explore the sensitivity of our results to the two-year cutoff, we alternatively definite Short using cutoffs of two, four, or five years. Unreported results show that the effects of Big N on syndicate structure do not change.

33. We convert all non-ratio variables reported by other currencies into US dollars based on the exchange rate from Compustat Global at the end of the corresponding fiscal year.