Do Syndicated Loan Borrowers Trade-Off Real Activities Manipulation with Accrual-Based Earnings Management?

Abstract

This study investigates how managers choose between alternative earnings management mechanisms among syndicated loan borrowers. Specifically, it examines the trade-off between accrual-based earnings management (AEM) and real activities manipulation (RAM) during the period leading up to syndicated loan origination. The study also explores whether lender monitoring mechanisms influence subsequent earnings management behavior. The syndicated loan market, positioned between the private and public fixed income markets, offers a distinctive context for analyzing these strategic decisions. Using a propensity score-matched sample of syndicated and bilateral loans issued between 1989 and 2005, the study finds that firms obtaining syndicated loans are more likely to engage in earnings manipulation beforehand, relying more heavily on AEM than on RAM. Further analysis reveals that monitoring mechanisms—such as lender reputation, the number of syndicate members, loan size, and loan maturity—are significantly associated with future changes in AEM but show a weaker relationship with changes in RAM.

1. Introduction

This study explores how managers choose between different earnings management mechanisms—specifically, accrual-based earnings management (AEM) versus real activities manipulation (RAM)—in the period leading up to syndicated loan originations.1 Cohen and Zarowin (2010), in the context of seasoned equity offerings, argue that firms’ selection between these mechanisms depends on their capacity to employ accrual-based strategies and the relative costs associated with each option. Building on that insight, this study extends the investigation to a different but equally significant corporate finance event: syndicated loan origination. This setting is particularly relevant because it blends characteristics of both private and public debt markets (Allen & Gottesman, 2006; Wittenberg-Moerman, 2008), offering a distinct environment for examining how firms manage earnings under varying degrees of lender scrutiny.

The syndicated loan market emerged in the 1970s as a semi-structured alternative to bilateral lending, allowing borrowers to access larger pools of capital through a consortium of lenders (Gadanecz, 2004). In the syndicate lending process, after the closing of initial loans between the borrower and the lender (the primary market), loans are sold to multiple secondary lenders, known as the syndicate lenders. All lenders maintain a copy of the documentation of the borrower because each syndicate lender has a separate claim on the loan although there is only one loan contract.2 In the primary phase of this process, the borrower negotiates with a lead or primary lender—typically an institution with prior knowledge of or existing relationships with the firm (Allen & Gottesman, 2006; Sufi, 2007; Allen et al., 2008). Once the initial loan is structured, it is distributed among other syndicate participants, each holding individual claims under a unified loan agreement. While this syndication process allows for risk-sharing and broader participation, it also introduces variability in the effectiveness of monitoring, as oversight transitions from a concentrated private setting to a more dispersed structure—occasionally even involving secondary trading and public credit evaluations (Wittenberg-Moerman, 2008). This hybrid nature of syndicated loans makes them an ideal context to study firms’ strategic financial reporting behavior, particularly in balancing the use of AEM and RAM under differentiated monitoring conditions.

Prior research has established that firms facing higher levels of information asymmetry are more likely to obtain financing from banks or other private lenders, while those with lower information asymmetry tend to access public debt markets. Moreover, private lenders are generally considered more effective monitors than public investors, primarily due to their closer access to firm-specific information and more concentrated ownership structures (Denis & Mihov, 2003; Bharath et al., 2008b). In public debt markets, the monitoring capacity is weakened because of widely dispersed debt ownership, which limits the individual investor’s ability to effectively oversee the borrower. In the context of syndicated loans, once the loan is distributed among multiple lenders, the ability to monitor becomes more complex and is highly dependent on the lead lender’s informational advantage. As Bharath et al. (2008b) note, this dilution in monitoring makes it more difficult for borrowers to renegotiate contract terms and often results in higher borrowing costs.

While the literature has addressed earnings management by borrowers, prior studies have primarily examined this behavior in the context of private debt (e.g., Ahn & Choi, 2009) or public debt (e.g., Teoh et al., 1998a; Cohen & Zarowin, 2010; Ge & Kim, 2014). In the case of syndicated loans, research has largely focused on the secondary loan market, examining pricing, trading, and post-origination borrower behavior (Sufi, 2007; Ball et al., 2008; Gaul & Uysal, 2009; Wittenberg-Moerman, 2008; El Mahdy & Park, 2014). However, there has been limited attention to how borrowers manage earnings prior to syndicated loan originations in the primary market. This study addresses that gap by investigating whether firms strategically select between RAM and AEM mechanisms based on their ability to use accrual-based tools. By applying this theoretical lens to syndicated loan origination, a major and complex corporate financing event, this paper advances our understanding of how firms adapt their financial reporting behavior in response to varying monitoring environments and contractual dynamics.

To examine earnings management behavior in the context of syndicated loan originations, this study focuses on the primary loan market and identifies 2192 syndicated loan facilities issued between 1989 and 2005.3 Using a propensity score matching approach, this sample is matched to a comparable set of firms that issued non-syndicated (bilateral) loans during the same period. The analysis evaluates borrowers’ AEM and RAM in the year preceding loan origination.

At the point of origination, syndicated and bilateral loans share many structural similarities, suggesting that the underlying incentives for earnings management may be comparable across the two. However, the capacity to engage in such manipulation may differ due to variations in lender monitoring effectiveness. Private lenders in bilateral arrangements tend to be more effective monitors due to their concentrated relationship with the borrower (Denis & Mihov, 2003; Bharath et al., 2008b). In contrast, syndicated loans involve more dispersed ownership, which diminishes individual lenders’ ability to exert oversight (Denis & Mihov, 2003). Given this difference, borrowers in syndicated loan arrangements are expected to have greater leeway to manage earnings prior to loan issuance.

Building on the framework proposed by Cohen and Zarowin (2010), which suggests that the choice between AEM and RAM depends on a firm’s ability to implement accrual-based techniques and the associated costs, this study hypothesizes a greater reliance on AEM among syndicated loan borrowers. The rationale is that when monitoring is less stringent, firms are more likely to exploit discretionary accruals to influence financial reporting. The implications for RAM are less definitive, as its usage is often constrained by operational and cash flow considerations and may itself be influenced by the firm’s engagement in AEM. To further examine the role of lender monitoring, this study constructs proxy variables to capture the effectiveness of monitoring mechanisms within syndicated loan arrangements. These proxies include the number of syndicate participants, the presence of reputable lead lenders as measured by the lender’s Herfindahl–Hirschman Index, loan size, and loan maturity. These variables are used to assess their relationship with changes in earnings management behavior—specifically AEM and RAM—during the year following syndicated loan origination. The second hypothesis of this study posits that stronger lender monitoring, as indicated by these proxies, is negatively associated with the extent of AEM.

The empirical results support our expectations. First, we find that borrowers undertaking syndicated loans engage in significantly more upward AEM in the year leading up to loan origination, while the use of RAM appears less prominent or non-existent. This aligns with the theoretical framework presented by Cohen and Zarowin (2010), which suggests that firms favor AEM when external monitoring is weak, such as in environments characterized by dispersed debt ownership. The muted role of RAM may be attributed to its more direct and potentially adverse impact on operating performance and future cash flows—factors that are critical to maintaining favorable credit ratings.

Moreover, given that lead lenders in syndicated loans often possess prior knowledge of the borrower’s operations (Wittenberg-Moerman, 2008), borrowers may be less inclined to engage in RAM. This is because RAM is harder to conceal, more costly to implement, and, as our results show, is associated with a higher cost of debt. Further analyses reveal that the identified monitoring proxies are indeed associated with changes in AEM behavior after loan origination. Specifically, an increase in the strength of the monitoring mechanism—through factors such as syndicate size, lender reputation, and longer loan maturity—is linked to a subsequent reduction in AEM. These findings are consistent with the results of Chung et al. (2005), who find that enhanced monitoring constrains managerial opportunism in financial reporting. Collectively, our results suggest that while borrowers exploit weaker monitoring to engage in AEM ahead of loan issuance, stronger oversight mechanisms effectively reduce such behavior in the post-origination period.

This study contributes to the existing literature on earnings management around specific corporate finance events (Teoh et al., 1998a, 1998b; Cohen & Zarowin, 2010). More importantly, it is the first to provide empirical evidence on the existence and substitution effects of earnings management mechanisms—specifically the trade-off between AEM and RAM—prior to syndicated loan originations in the primary private market. It demonstrates that AEM remains a critical tool for borrowers in the private debt market, despite evidence that its prevalence has declined following the passage of the Sarbanes–Oxley Act (Cohen et al., 2008).

This study’s contribution is further supported by recent research. For example, Adam et al. (2020) show that managerial overconfidence influences the structure of syndicated loan contracts, particularly through performance-sensitive debt, suggesting that cognitive biases can drive financial decisions that align with the use of AEM to project optimistic performance expectations. Related, Mafrolla and D’Amico (2017) demonstrate that small and medium-size private borrowers in private loan markets from Italy, Portugal and Spain engage in earnings management—both before and after loan origination—to increase their borrowing capacity, especially in the post-Basel II regulatory environment where accounting-based assessments have become more central. Similarly, Gao et al. (2024) show that management earnings forecasts (MEFs) reduce loan spreads by corroborating private communications between borrowers and lenders. This suggests that AEM can also be used to manage lenders’ perceptions, particularly in opaque or information-asymmetric lending environments such as syndicated loans. Moreover, Ertan (2022) reveals that banks themselves engage in real earnings management through the timing and pricing of syndicated loans to meet earnings benchmarks, underscoring that opportunistic behavior permeates both sides of the lending relationship. Collectively, these studies reinforce the significance of this paper’s findings and situate them within a broader context of strategic financial reporting in private debt markets.

The rest of this paper is organized as follows. Section 2 reviews background studies and develops research hypotheses. Section 3 presents the research methodology. Section 4 reports empirical results and additional analyses. Conclusion and recommendations for future research are given in Section 5.

2. Background Study and Hypotheses Development

2.1. Syndicated Loans and the Alternative Earnings Management Mechanisms

Prior research (Allen & Gottesman, 2006; Wittenberg-Moerman, 2008; Ball et al., 2008; Gaul & Uysal, 2009; Dhaliwal et al., 2011) document a positive association between syndicated loan prices in the secondary market and the borrowers’ financial reporting quality; their results suggest that the loan prices in the secondary loan market reflect the impacts of borrowers’ financial reporting choices subsequent to loan syndication and resale in the secondary market. However, prior work is limited with respect to the association between syndicated loan prices and the financial reporting quality prior to the loan origination. Also, prior research is restricted with respect to the mechanism of how management manipulates earnings prior to syndicated loan originations. Ahn and Choi (2009) examine the association between banks’ monitoring mechanisms and earnings management among syndicated loan borrowers and find that earnings management decreases as banks’ monitoring mechanisms increase. Their study, however, focuses mainly on accrual-based earnings management without contrasting the alternative earnings management mechanisms.

In accounting literature, it is well-documented that earnings management can be carried out through alternative mechanisms. Accrual-based management (AEM) involves manipulation of accounting assumptions and estimates, and the timing of recognition of certain accruals (for example, by understating allowances for doubtful accounts or bad debt reserves); real activities management (RAM) alters business operating decisions for financial reporting purposes (Graham et al., 2005; Roychowdhury, 2006; Cohen & Zarowin, 2010; Zang, 2012).4 The alternative mechanisms lead to different subsequent accounting effects. AEM, with the inevitable accrual reversals, results in fluctuations in accounting earnings from period to period, but its impact on future cash flows is limited. RAM, however, results in immediate alteration in corporate operations, including cash flows (Graham et al., 2005; Roychowdhury, 2006; Gunny, 2010; Taylor & Xu, 2010) and cannot be easily challenged by auditors and regulators5.

More recent research shows that firms increasingly engage in RAM after the Sarbanes–Oxley Act (SOX) of 2002, because AEM exposes managers to higher litigation risk (Cohen et al., 2008; Cohen & Zarowin, 2010; Gupta et al., 2010; Zang, 2012; Chi et al., 2011). Related, Ge and Kim (2014) address RAM and the cost of corporate bonds and find that RAM impairs credit ratings and increases bond yield spreads for public debts. Part of their argument is that lenders have fixed claims such as periodic interest payments. Therefore, lenders tend to focus on future cash flows rather than on accounting earnings. In other words, the substitute effects and the possible driver of the substitute effects are documented. However, the choice between the alternative mechanisms given the specific setting of syndicated loan originations has not been addressed. This current study fills in this gap.

This study follows Cohen and Zarowin (2010) and Ahn and Choi (2009) in analyzing the management’s choice of earnings management mechanisms. On top of the unique setting of syndicated loan originations, it is different to previous studies in several aspects. First, we examine the existence of alternative earnings management activities (AEM versus RAM) among syndicated loans borrowers versus non-syndicated loans (bilateral loans) borrowers prior to syndication. Second, Ahn and Choi’s study investigates the association between bank lenders’ monitoring mechanisms on syndicate lenders’ AEM while this study examines the effects of lenders’ monitoring mechanism on future changes in AEM versus RAM. Third, Ahn and Choi’s study excludes the effect of SOX by limiting their sample to the year 2001, while our sample period extends post-SOX, since the documented trade-off between AEM and RAM is more pronounced post-SOX.

An important concern for earnings management among syndicated loan borrowers is the cash flow consequences arising from manipulating RAM. Bharath et al. (2008a) contrast the designs of debt contracts of private versus public debts and show that private debt lenders use both the price and non-price terms to customize the loan contracts, whereas public debts are only negotiated through interest costs. Following Bharath et al. (2008a) and Ge and Kim (2014), if bondholders, with only fixed claims, are more concerned about the uncertainty of future cash flows than about accounting earnings, it is reasonable to assume that this concern applies to private lenders, too. Since firms that manage earnings through RAM are more likely to suffer from higher cash flow uncertainties in the future (Gunny, 2010; Taylor & Xu, 2010; Zang, 2012; Ge & Kim, 2014), it is expected that syndicated loan borrowers have incentives to avoid manipulating earnings using RAM and to resort more heavily to AEM.

Also, since syndicated loans are negotiated privately upon origination and not subject to public and regulatory scrutiny like those for public debts, the constraints on AEM are not as strenuous as those for public debts. Since RAM results in direct negative consequences on future cash flows, when compared with AEM (Graham et al., 2005; Roychowdhury, 2006; Cohen & Zarowin, 2010; Zang, 2012; Kothari et al., 2016), it is hypothesized that syndicated loan borrowers have incentives to engage in AEM rather than RAM prior to loan syndication. When contrasting syndicated loan borrowers with non-syndicated loan borrowers, we hypothesize based on Denis and Mihov’s (2003) finding that the lenders’ monitoring ability diminishes when ownership is more widely dispersed. It is, therefore, expected that any effects of AEM are more pronounced for the syndicated loan borrowers.

H1. 

Syndicated loan borrowers manipulate earnings prior to the origination of syndicated loans, with more dependence on accrual-based earnings management.

2.2. Lenders’ Monitoring Mechanisms

This study also addresses the impacts of lenders’ monitoring on managerial opportunism prior to loan syndication. The incentives and abilities for syndicated loan borrowers to manipulate accounting earnings may not be the same as that for public debt borrowers (Wittenberg-Moerman, 2008). On the primary market, syndicated borrowing is a private event without public scrutiny. Without public scrutiny, it is up to the primary lender’s monitoring mechanisms to curb the borrower’s opportunistic behavior. Thus, it is interesting to understand the extent to which these monitoring mechanisms are effective in curbing managerial opportunism after the loan origination. For example, it is interesting to understand whether more extensive lender monitoring can lead borrowers to rely more on RAM vs. AEM, because AEM entails higher litigation risk (Chi et al., 2011). The answer to this question is unclear because RAM could result in real cash flow reductions, which would impair borrowers’ ability to repay the loans. Thus, it is possible that borrowers will still rely on AEM to a greater extent, despite its perceived litigation costs.

Prior research documents the effectiveness of external monitoring mechanisms in curbing income-increasing AEM (Chung et al., 2005). The structure of syndicate lending is unique and offers wide selections of provisions not traditionally offered by the primary debt market such as the primary lenders’ reputation, loan types, and number of syndicates (Sufi, 2007; Wittenberg-Moerman, 2008). Ahn and Choi (2009) argue that bank monitoring mechanisms (e.g., lender’s reputation, the magnitude of bank loan, loan maturity, number of syndication) are significantly associated with AEM. We therefore examine the association between these lender’s monitoring mechanisms and future change in earnings management among syndicated loan borrowers. We predict that the lender’s monitoring mechanisms will curb earnings management among syndicated loan borrowers as suggested by Ahn and Choi (2009) and prior literature. The monitoring mechanisms are discussed hereafter.

2.3. Number of Syndications

The primary lender has superior privilege over other syndicate arrangers, such as a priority to other claims on the firms’ assets, and the ability to impose restrictive covenants on the firm (Wittenberg-Moerman, 2008; Allen & Gottesman, 2006). Ball et al. (2008) argue that the proportion of the loan that is held by the primary lender is dependent on the increasing adverse selection and moral hazard problems created by information asymmetry. They also argue that accounting information mitigates information asymmetry, allowing the primary lender to hold a larger proportion of the loan. To the extent that the loan is risky, the primary lender will retain a smaller portion of the loan and syndicate the remainder in the secondary loan market to multiple arrangers, thereby reducing their exposure to credit risk and meanwhile increasing the number of syndicate lenders. It is expected that the higher the number of syndicate lenders, the more scrutiny by multiple arrangers. Therefore, we expect a negative relationship between the number of syndications and the future changes in earnings management.

2.4. Loan Size

A loan large in size necessitates rigorous monitoring mechanisms by lenders but a loan small in size does not require such rigorous monitoring. Larger loan sizes are also significantly associated with lower information asymmetry (Wittenberg-Moerman, 2008). Syndicate lenders are mostly sophisticated market participants (e.g., bank lenders or institutional investors) who possess incremental informational advantages over other financial intermediaries. Therefore, it is expected that syndicate lenders will be more diligent in providing corporate governance over the financial and non-financial measures of syndicate borrowers in order to decrease the borrower’s informational opacity. Ahn and Choi (2009) provide empirical evidence that the magnitude of bank loans is negatively associated with earnings management among syndicate borrowers; this is largely because lenders have a greater stake in the loan and greater monitoring capabilities over the borrowers’ financials. Larger loans also incentivize lenders to exert more efforts in governance and reduce opportunistic behaviors by the borrowers. This perspective has been consistently avowed in prior literature (Khalil & Parigi, 1998; Kang et al., 2000; Lee & Mullineaux, 2004; Sufi, 2007). We therefore expect the size of syndicate loans to be negatively associated with the future changes in earnings management.

2.5. Loan Maturity

The lenders’ incentives to effectively monitor syndicated loan borrowers are stronger for long-term loans than for short-term loans. Long-term loans allow lenders to regularly monitor borrowers to reduce agency costs related to debt6 (Ahn & Choi, 2009). Long-term loans also stimulate lenders to get access to private information of borrowers before syndication so that lenders become cognizant of borrowers’ opportunistic behaviors. The relationship between loan maturity and lender monitoring has been suggested in the prior literature (Rajan & Winton, 1995; Ongena & Smith, 1998; Dennis & Mullineaux, 2000; Ahn & Choi, 2009). Following the prior literature, we expect that loan maturity will be negatively associated with the future changes in earnings management.

2.6. Lenders’ Reputation

The reputation of the primary lender plays a major role in loan pricing in the secondary market. Bushman and Wittenberg-Moerman (2012) investigate the role of the primary lender’s reputation and the “certification” effect that a reputable primary lender can provide. Investors in the secondary market access information regarding the syndicated loans largely through the primary lenders. Reputable lenders are more likely to have informational advantages over non-reputable lenders, thereby allowing the former to more effectively assess credit risk of borrowers and to alleviate agency costs. Reputation is an important concern that lenders have to carefully manage in order to avoid the negative reputational damage that may lead to serious financial damage (Ahn & Choi, 2009). Reputable lenders are also linked to a high probability of syndication (Dennis & Mullineaux, 2000). Related, Billett et al. (1995) suggest that reputable lenders with higher credit ratings are significantly associated with abnormal borrowers’ returns, evidence supporting the proposition that the market reacts to the identity of lenders. In line with this predication, Wittenberg-Moerman (2008) finds that loans syndicated by more reputable lenders are associated significantly with lower information asymmetry, consistent with the prediction that reputable lenders see through borrowers’ private information. It is therefore expected that reputable lenders will be associated with significantly lower future changes in earnings management.

The market for syndicate loans involves sophisticated market players (e.g., bank lenders or institutional lenders) who presumably can see through the mere figures in the borrower’s financial statements. Lenders can make an informed decision on whether earnings are manipulated using AEM as a well-established earnings management technique. In this regard, prior research suggests a significant association between financial reporting quality and the borrowing costs (Francis et al., 2005) and information asymmetry (Wittenberg-Moerman, 2008). It is hypothesized that the lender’s monitoring mechanisms, such as lender’s reputation, loan size, number of syndications and loan maturity, have a differential effect on the future changes in earnings management activities. That is, we expect to see different levels of earnings management given different levels of lender monitoring.

H2. 

For syndicated loan borrowers, the lender’s monitoring mechanisms—such as lender reputation, loan size, number of syndications, and loan maturity—have differential effects on earnings management.

3. Research Methodology

3.1. Data and Sample Selection

We obtained data on syndicated loans from the Loanware database from 1989–2005.7 We further obtained data related to earnings management as well as the control variables from COMPUSTAT. We started with 2517 U.S. firms (16,676 facilities) of syndicated loans after removing non-U.S. firms and firms in the financial services and utilities industries. We then merged syndicated loan data with COMPUSTAT and obtained 8602 syndicate facilities. We initially obtained 4680 facilities that have private bilateral debt over the same sample period from Bloomberg and required these firms to be U.S.-based, public, and non-financial services or utilities industries and to have complete data about the issue date, maturity, and loan amount. We also required this private bilateral debt data to have complete information about earnings management and control variables, and this resulted in 3141 bilateral loans. The final full sample was composed of 11,743 facilities (3141 bilateral loans and 8602 syndicated loans). We then matched private bilateral (non-syndicated) to syndicated loans based on propensity score matching.8 The final matched sample was composed of 4384 facilities (2192 bilateral loans and 2192 syndicated loans).

3.2. Empirical Models

We first examined the association between syndicated loan originations and earnings management. We used a propensity score matched sample of firms that engaged in syndicated borrowing versus firms that engaged in bilateral borrowing between 1989 and 2005. In order to test our first hypothesis, we ran a logistic regression to calculate the propensity score of syndication, as shown in model (1), and then prepared a matched sample of firms with and without syndication based on 0.10 ranges of propensity scores.9 The prior literature suggests that debt contracting is associated with prior lending relationship between the lender and borrower (Sufi, 2007), and financial reporting quality of the borrower (Costello & Wittenberg-Moerman, 2011). When the information asymmetry between the lender and borrower is severe, the lead arranger will retain a large portion of the loan, and it becomes difficult to diversify the risk of such loans through syndication. Meanwhile, lower financial reporting quality signals uncertainties about the borrower’s future ability to pay off the cost of debt (e.g., interest expense) as well as the debt itself and therefore the lender would be closer to the borrower geographically and would prefer to lend to a borrower with a previous lending relationship.

Therefore, we expect the likelihood of syndication to be associated with the prior year’s syndication, audit quality, business complexity, liquidity, leverage, financial reporting quality, firm performance, firm size, and fiscal year end.10 As model (1) shows, we regressed current year syndication (SYND_it), which is an indicator variable equal to one if the loan is syndicated in year t and zero otherwise, on prior year’s syndication (SYND_it−1), audit quality as measured by Big 4 (BIG4), financial reporting quality as measured by restatement (REST), business complexity as measured by the existence of foreign transactions (FOR), firm performance as measured by returns on assets (ROA) and losses (LOSS), financial leverage (FINL) and the firm’s ability to pay off debt in the short term (LIQ), firm size as measured by assets (AT), whether the firm’s fiscal-year end is December (BUSY), industry categories (IND_CAT), and fixed-year effect (YEAR). Definitions of variables are in Appendix A.

SYND_it = β₀ + β₁ SYND_it−1 + β₂ BIG4_it + β₃ REST_it + β₄ FOR_it + β₅ ROA_it + β₆ FINL_it + β₇ LIQ_it
+ β₈ LOSS_it + β₉ AT_it + β₁₀ BUSY_it + Σβ_i IND_CAT_it + Σβ_t YEAR_it

(1)

After acquiring the propensity scores from Model (1), we created the matched sample based on 0.10 ranges of propensity scores. We then examined our first research hypothesis by running a Heteroscedasticity-Consistent Standard Error regression model with lagged earnings management variables11 (EM, we use AEM and RAM variables) as the dependent variable and current year’s syndication (SYND) as the independent variable of interest, with other control variables. Following Zang (2012) who argues that managers use AEM and RAM as substitutes, we control for the substitute effect (SUB) between alternative earnings management tools (e.g., RAM substitutes in AEM models are ACFO, ADISC, and APROD, and the AEM substitute in RAM models is DA_K, which is discretionary accruals according to Kothari et al., 2005). We also control for post-SOX period (SOX) because Cohen et al. (2008) find that firms resort to RAM after SOX. Prior research also suggests that earnings management is associated with the firm’s ability to use earnings management in prior years (Barton & Simko, 2002), firm performance (Gunny, 2010), whether the firm is in a litigious industry (Cohen & Zarowin, 2010), audit quality (Chi et al., 2011; Becker et al., 1998), firm leverage (Zamri et al., 2013), management’s incentives to avoid reporting losses (Matsumoto, 2002), and management forecasts (García Osma et al., 2023). We therefore control for the extent of the prior year’s earnings manipulations as measured by net operating assets (NOA), the firm’s complexity as measured by the presence of foreign transactions (FOR) or extraordinary activities (EXTRA), firm performance as measured by losses (LOSS) and returns on assets (ROA), financial leverage (FINL), probability of default as measured by Altman Z score (Z), firm growth as measured by market-to-book ratio (MTB), audit quality as measured by big 4 (BIG4), firm size as measured by assets (AT), economic fluctuations as measured by the change in Gross Domestic Product (ΔGDP), management forecasts (MF), litigious industries (LIT), industry categories (IND_CAT), and fixed-year effect (YEAR). Definitions of variables are in Appendix A.

EM_it = β₀ + β₁ SYND_it+ β₂ SOX_it + β₃ SUB_it + β₄ LOSS_it + β₅ NOA_it + β₆ EXTRA_it + Β₇ FOR_it +
β₈ ROA_it + β₉ FINL_it + β₁₀ Z_it + β₁₁ MTB_it + β₁₂ BIG4_it + β₁₃ AT_it + β₁₄ ΔGDP_it + β₁₅ MF_it +
β₁₆ LIT_it + β₁₇ MATUR_it + β₁₈ SIZE_it + Σβ_i IND_CAT_it + Σβ_t YEAR_it

(2)

To test our second research hypothesis, we ran an OLS regression model (3) with the future changes in earnings management variables (ΔEM_it+1, we use ΔAEM_it+1 and ΔRAM_it+1 variables) as the dependent variable and the lender’s monitoring mechanisms (LENDER_MONITOR) as the independent variable of interest and other control variables. Following Ahn and Choi (2009), we measured lender monitoring mechanisms with four proxies: lender’s Herfindahl–Hirschman Index (HHI) per two-digits industry, loan size (SIZE), number of syndications (SYNDICATES) and loan maturity (MATUR). We also added the same set of control variables used in model (2) to model (3). Definitions of variables are in Appendix A.

ΔEM_it+1 = β₀ + β₁ LENDER_MONITOR_it + β₂ SUB_it + β₃ SOX_it + β₄ LOSS_it + β₅ NOA_it +
β₆ EXTRA_it + β₇ FOR_it + Β₈ ROA_it + β₉ FINL_it + β₁₀ Z_it + β₁₁ MTB_it + β₁₂ BIG4_it + β₁₃ AT_it +
β₁₄ ΔGDP_it + β₁₅ MF_it + β₁₆ LIT_it + Σβ_i IND_cat_it + Σβ_t YEAR_it

(3)

We also ran several additional analyses. For example, we examined whether earnings management is associated with the cost of debt (COD). Therefore, we ran the OLS regression model (4) with the interaction term of lagged earnings management (EM) variables (AEM and RAM variables) and HHI as the dependent variables and current year’s cost of debt (COD) as the independent variable of interest. We also added the same set of control variables to model (4) in addition to a set of variables to control for loan characteristics (LOAN_CH)12 such as the borrower’s information asymmetry (SPREAD), the presence of financial debt covenants (COVEN), loan credit rating (DRATINGS), and the four measures of lender’s monitoring mechanisms (LENDER_MONITOR). Definitions of variables are in Appendix A.

COD_it = β₀ + β₁ EM_it−1*HHI_it + β₂ EM_it−1 + β₃ HHI_it+ Σβ₄ LENDER_MONITOR_it +
Σβ₅ LOAN_CH_it + β₆ SOX_it + β₇ LOSS_it + β₈ NOA_it + β₉ EXTRA_it + β₁₀ FOR_it +
β₁₁ ROA_it + β₁₂ FINL_it + β₁₃ Z_it + β₁₄ MTB_it + β₁₅ BIG4_it + β₁₆ AT_it + β₁₇ ΔGDP_it +
β₁₈ LIT_it + Σβ_i IND_CAT_it + Σβ_t YEAR_it

(4)

3.3. Descriptive Statistics

The means of lead, lag and current AEM and RAM are graphed in Figure 1, which shows that, on average, signed discretionary accruals (DA_K_t−1) according to Kothari et al. (2005) among syndicate borrowers are higher than that among non-syndicate borrowers. Figure 1 also suggests that syndicate borrowers are, on average, experiencing lower RAM compared to bilateral (non-syndicate) borrowers as measured by higher ACFO, and lower APROD and lower RM1 and RM2.

Table 1. Descriptive statistics.

Variable	Mean	STD	25th	50th	75th
ACFOt−1	0.1099	0.4102	−0.0165	0.0547	0.1695
ADISCt−1	0.0161	0.3150	−0.1966	−0.0246	0.1463
APRODt−1	−0.0346	0.3170	−0.1722	−0.0489	0.0865
DA_Kt−1	−0.1157	0.4095	−0.1361	−0.0430	0.0184
DA_Dt−1	0.0031	1.2122	−0.0598	0.0192	0.1395
RM1t−1	−0.0507	0.4721	−0.2984	−0.0293	0.2104
RM2t−1	−0.1260	0.5084	−0.2352	−0.0290	0.1054
SOX	0.4701	0.4992	0.0000	0.0000	1.0000
LOSS	0.1914	0.3934	0.0000	0.0000	0.0000
EXTRA	0.3141	0.4642	0.0000	0.0000	1.0000
FOR	0.6202	0.4854	0.0000	1.0000	1.0000
ROA	0.0519	0.1011	0.0127	0.0511	0.0964
NOA	0.7013	0.6363	0.3325	0.5750	0.8605
FINL	0.6016	0.2480	0.4514	0.5816	0.7247
Z	−1.1449	1.5785	−2.0797	−1.2849	−0.3544
MTB	2.7522	4.0155	1.1869	1.9763	3.4130
BIG4	0.7005	0.4581	0.0000	1.0000	1.0000
AT	6.7155	1.8321	5.4407	6.6690	7.9939
ΔGDP	5.3812	1.2647	4.9000	5.7000	6.5000
MF	0.0417	0.2000	0.0000	0.0000	0.0000
LIT	0.1620	0.3684	0.0000	0.0000	0.0000
MATUR	1.4855	0.5008	1.0977	1.6094	1.8986
SIZE	18.5059	1.6257	17.5044	18.6438	19.5993
HHI	0.3539	0.3479	0.1111	0.2500	0.5000

This table shows descriptive statistics of earnings management and loan—firm-specific variables. All continuous variables are winsorized at 1 and 99. Variables’ definitions and measurements are in Appendix A and Appendix B.

summarizes the descriptive statistics of earnings management measures, syndicated borrowers’ characteristics and the set of control variables. The average signed discretionary accruals DA_D_t−1 (DA_K_t−1) is 0.0031 (−0.1157) according to abnormal accruals obtained from the models of Dechow et al. (1995) and Kothari et al. (2005). Average abnormal cash flow from operations (ACFO_t−1), abnormal discretionary expenses (ADISC_t−1), and abnormal production costs (APROD_t−1) are 0.1099, 0.0161, and −0.0346, respectively. The average change in the GDP for the sample is 5%. A total of 47% of the sample is post-SOX 2002 period and almost 16% is drawn from litigious industries. Roughly 19% of the final sample incurred losses, 70% was audited by Big 4, 31% reported extraordinary activities, and 62% engaged in foreign transactions. The average ROA is 0.05 and 60% is the percentage of total debt to total assets. Roughly 4% of the total sample of firms issued management forecasts. The average Altman Z score is −1, indicating a likelihood of bankruptcy. The averages of MTB and loan size are 2.8 and USD 1805M respectively. The average (median) HHI is 0.3539 (0.3479).

Table 2. Pearson and Spearman Correlation Coefficients, n = 4384.

	SYND	ACFOt−1	ADISCt−1	APRODt−1	DA_Kt−1	DA_Dt−1	RM1t−1	RM2t−1	SOX	LOSS	EXTRA	FOR	ROA	NOA	FINL	Z	MTB	BIG4	AT	ΔGDP	MF	LIT	MATUR	SIZE	HHI
1		−0.03 b	0.02	0.01	0.04 b	−0.03 b	0.00	0.04 b	0.01	−0.02	−0.03 b	0.00	−0.01	−0.02	−0.06 a	−0.05 a	−0.05 a	−0.07 a	−0.03 b	−0.05 a	0.21 a	−0.02	0.14 a	−0.49 a	-
2	−0.01		−0.12 a	−0.23 a	−0.53 a	−0.16 a	−0.06 a	−0.47 a	−0.30 a	0.13 a	−0.16 a	−0.02	0.17 a	0.26 a	0.08 a	−0.19 a	−0.23 a	0.27 a	0.06 a	0.10 a	0.01	0.07 a	0.16 a	−0.01	−0.10 a
3	0.00	−0.04 b		−0.31 a	0.00	−0.11 a	−0.81 a	−0.70 a	−0.74 a	0.02	−0.02	−0.07 a	0.00	0.04 b	0.11 a	−0.06 a	−0.06 a	0.14 a	0.02	0.00	0.04 a	−0.04 a	0.05 a	−0.03 c	−0.03
4	0.03	−0.04 b	−0.12 a		0.11 a	0.03 c	0.73 a	0.46 a	0.73 a	−0.01	0.05 a	0.03 b	0.00	−0.17 a	−0.23 a	0.16 a	0.18 a	−0.18 a	−0.02	0.01	−0.01	0.03 b	0.02	−0.05 a	0.10 a
5	0.03 b	−0.37 a	−0.03 c	−0.01		0.38 a	0.10 a	0.35 a	0.24 a	−0.16 a	0.04 a	−0.05 a	−0.08 a	−0.07 a	−0.01	0.05 a	0.06 a	−0.16 a	−0.03 c	−0.02	0.03 b	0.05 a	−0.16 a	0.04 b	0.03
6	−0.01	−0.23 a	−0.04 b	0.09 a	0.24 a		0.12 a	0.19 a	0.17 a	−0.01	0.03 c	−0.05 a	0.05 a	0.03 b	0.01	−0.08 a	−0.08 a	−0.02	−0.06 a	−0.06 a	0.04 b	−0.10 a	0.02	0.01	0.00
7	0.02	0.00	−0.75 a	0.75 a	0.01	0.09 a		0.73 a	0.93 a	−0.06 a	0.02	0.05 a	0.00	−0.10 a	−0.20 a	0.12 a	0.13 a	−0.20 a	−0.03 c	0.00	−0.02	0.05 a	−0.05 a	0.01	0.09 a
8	0.01	−0.79 a	−0.59 a	0.10 a	0.32 a	0.21 a	0.46 a		0.86 a	−0.11 a	0.11 a	0.08 a	−0.08 a	−0.17 a	−0.11 a	0.15 a	0.17 a	−0.27 a	−0.06 a	−0.04 b	−0.01	−0.04 b	−0.13 a	−0.01	0.07 a
9	0.02	−0.40 a	−0.67 a	0.70 a	0.16 a	0.17 a	0.92 a	0.74 a		−0.10 a	0.06 a	0.05 a	−0.05 a	−0.14 a	−0.20 a	0.15 a	0.16 a	−0.25 a	−0.04 a	−0.03 b	−0.10 a	0.00	−0.08 a	0.00	0.00
10	−0.02	0.13 a	0.04 b	−0.01	−0.19 a	−0.10 a	−0.04 b	−0.13 a	−0.08 a		0.00	0.13 a	0.17 a	0.04 b	−0.01	−0.06 a	−0.06 a	0.10 a	0.40 a	0.26 a	−0.03 c	0.06 a	−0.01	0.03 b	0.01
11	−0.03 b	−0.10 a	−0.03 b	0.02	−0.04 b	0.02	0.04	0.10 a	0.08 a	0.00		0.17 a	−0.05	−0.64 a	0.05 a	0.15 a	0.30 a	−0.22 a	−0.11 a	−0.06 a	−0.06 a	−0.20 a	0.02	0.00	−0.02
12	0.00	0.01	−0.06 a	0.03 c	−0.05 a	−0.05 a	0.06 a	0.03 b	0.05 a	0.13 a	0.17 a		0.05 a	−0.20 a	0.07 a	0.19 a	0.21 a	−0.11 a	−0.02	0.15 a	0.04 b	0.03 c	−0.05 a	0.01	−0.05 a
13	−0.01	0.10 a	0.03 c	0.04 b	−0.09 a	0.01	0.01	−0.09 a	−0.03 b	0.17 a	−0.05 a	0.05 a		0.02	0.05 a	0.02	0.02	0.13 a	0.10 a	0.33 a	0.12 a	0.03 b	0.09 a	−0.05 a	−0.04 b
14	0.02	0.14 a	0.02	−0.12 a	0.05 a	0.00	−0.10 a	−0.12 a	−0.14 a	0.01	−0.58 a	−0.13 a	0.02		−0.10 a	−0.33 a	−0.55 a	0.48 a	0.12 a	0.01	−0.03 b	0.14 a	0.00	0.04 b	−0.10 a
15	−0.07 a	0.00	0.00	−0.17 a	−0.01	−0.01	−0.11 a	0.00	−0.10 a	0.00	0.08 a	0.06 a	0.02	−0.09 a		−0.05 a	−0.02	−0.04 b	0.02	0.05 a	−0.10 a	0.03 c	−0.06 a	0.10 a	0.01
16	−0.02	−0.10 a	−0.08 a	0.10 a	−0.05 a	−0.02	0.12 a	0.13 a	0.15 a	−0.05 a	0.17 a	0.23 a	−0.01	−0.24 a	−0.07 a		0.95 a	−0.09 a	−0.01	0.20 a	−0.12 a	0.06 a	−0.13 a	0.07 a	0.01
17	−0.02	−0.13 a	−0.08 a	0.12 a	−0.06 a	−0.02	0.13 a	0.15 a	0.17 a	−0.04 a	0.32 a	0.24 a	−0.01	−0.50 a	−0.04 b	0.96 a		−0.20 a	−0.03 b	0.18 a	0.10 a	0.02	−0.11 a	0.05 a	−0.14 a
18	−0.03 c	0.10 a	0.13 a	−0.09 a	−0.03 c	−0.01	−0.15 a	−0.16 a	−0.18 a	0.03 b	−0.08 a	−0.07 a	0.07 a	0.21 a	−0.07 a	−0.07 a	−0.12 a		0.19 a	0.24 a	−0.02	0.08 a	0.18 a	0.00	0.00
19	−0.03 b	0.06 a	0.01	0.00	−0.03 b	−0.07 a	−0.01	−0.05 a	−0.03 c	0.40 a	−0.11 a	−0.02	0.10 a	0.09 a	0.01	−0.03 b	−0.06 a	0.09 a		0.29 a	0.03 c	0.10 a	0.03 c	0.08 a	−0.19 a
20	−0.05 a	0.09 a	0.00	0.06 a	−0.04 a	−0.06 a	0.04 a	−0.07 a	0.00	0.26 a	−0.06 a	0.16 a	0.33 a	0.00	0.00	0.13 a	0.11 a	0.09 a	0.29 a		−0.08 a	0.12 a	−0.04 a	−0.06 a	0.00
21	0.21 a	−0.03 c	0.04 b	−0.01	0.04 b	0.01	−0.03 c	0.00	−0.10 a	−0.03 b	−0.06 a	0.04 b	0.10 a	−0.05 a	−0.09 a	−0.11a	0.08 a	−0.02	0.02	−0.08 a		0.06 a	−0.17 a	−0.01	-
22	−0.01	0.13 a	−0.02	0.05 a	0.06 a	−0.10 a	0.04 a	−0.09 a	−0.01	−0.09 a	−0.20 a	−0.02	−0.02	0.13 a	0.01	0.07 a	0.02	0.02	0.07 a	0.07 a	0.06 a		−0.02	0.17 a	−0.04 b
23	0.14 a	0.11 a	0.11 a	0.02	−0.14 a	−0.05 a	−0.06 a	−0.15 a	−0.10 a	−0.01	0.02	−0.05 a	0.09 a	−0.02	−0.06 a	−0.11 a	−0.09 a	0.15 a	0.03 c	−0.03 b	−0.17 a	−0.02		−0.15 a	0.07 a
24	−0.48 a	−0.02	−0.05 a	−0.05 a	0.03 b	0.01	0.00	0.05 a	0.01	0.05 a	0.01	0.02	−0.05 a	0.02	0.09 a	0.07 a	0.06 a	−0.02	0.07 a	−0.08 a	0.01	0.15 a	−0.15 a		0.18 a
25	-	−0.02	−0.04 b	0.10 a	0.00	−0.01	0.08 a	0.04 b	0.02	0.01	−0.02	−0.02	0.00	−0.09 c	0.00	0.00	−0.09 a	0.03 c	−0.11 a	0.01	-	−0.04 c	0.07 a	−0.09 a	-

a, b, and c indicate significance levels at 1%, 5% and 10%, respectively. The upper diagonal reports are the Spearman correlations, and the lower diagonal reports the Pearson correlation. Correlations above the 0.80 threshold are shown in bold. Variables’ definitions and measurements are in Appendix A and Appendix B.

displays the Pearson (lower diagonal) and Spearman (upper diagonal) correlation coefficients. The Spearman coefficients of correlations suggest a positive, significant correlation between DA_K_t−1, RM2 and SYND, but a negative, significant correlation between ACFO_t−1, DA_D_t−1 and SYND. The Pearson coefficients of correlations show a positive, significant correlation between DA_K_t−1 and SYND. The Pearson correlation and Spearman coefficients of correlations show a positive, significant correlation between SYND and MATUR, but a negative, significant correlation between SYND and EXTRA, FINL, Z, MTB, BIG4, AT, LIT, ΔGDP and SIZE.13

3.4. Univariate Analysis

Table 3. Univariate analysis.

	Full Sample		SYND = 1		SYND = 0		Difference Tests
	n = 4384		n = 2192		n = 2192		t Test	Wilcox.
	Mean	Median	Mean	Median	Mean	Median	t Value	z Value
ACFOt−1	0.1099	0.0547	0.1041	0.0425	0.1158	0.0652	0.95	−2.059 **
ADISCt−1	0.0161	−0.0246	0.0174	−0.0195	0.0148	−0.0293	−0.26	1.37 *
APRODt−1	−0.0346	−0.0489	−0.0253	−0.0542	−0.0440	−0.0411	−1.96 *	0.88
DA_Kt−1	−0.1157	−0.0430	−0.1022	−0.0357	−0.1292	−0.0487	−2.19 **	2.33 **
DA_Dt−1	0.0031	0.0192	−0.0071	0.0190	0.0134	0.0204	0.56	−2.24 **
RM1t−1	−0.0507	−0.0293	−0.0426	−0.0306	−0.0589	−0.0244	−1.14	0.08
RM2t−1	−0.1260	−0.0290	−0.1214	−0.0175	−0.1307	−0.0473	−0.60	2.44 ***
SOX	0.4701	0.0000	0.4599	0.0000	0.4804	0.0000	1.36	−1.36 *
LOSS	0.1914	0.0000	0.1793	0.0000	0.2035	0.0000	2.04 **	−2.03 **
EXTRA	0.3141	0.0000	0.3139	0.0000	0.3143	0.0000	0.03	−0.03
FOR	0.6202	1.0000	0.6145	1.0000	0.6259	1.0000	0.78	−0.78
ROA	0.0519	0.0511	0.0535	0.0470	0.0503	0.0550	−1.05	−1.13
NOA	0.7013	0.5750	0.6541	0.5622	0.7484	0.5827	4.92 ***	−4.15 ***
FINL	0.6016	0.5816	0.5979	0.5684	0.6054	0.5884	1.41 ***	−3.21 ***
Z	−1.1449	−1.2849	−1.1740	−1.3960	−1.1158	−1.2091	1.22	−3.23 ***
MTB	2.7522	1.9763	2.6472	1.8982	2.8573	2.0961	1.73 *	−4.95 ***
BIG4	0.7005	1.0000	0.6848	1.0000	0.7162	1.0000	2.28 **	−2.27 **
AT	6.7155	6.6690	6.6211	6.5637	6.8098	6.7481	3.41 ***	−3.61 ***
ΔGDP	5.3812	5.7000	5.3719	5.7000	5.3905	5.7000	0.49	−1.44 *
MF	0.0417	0.0000	0.0835	0.0000	0.0000	0.0000	−14.13 ***	13.82 ***
LIT	0.1620	0.0000	0.2135	0.0000	0.1104	0.0000	−9.35	9.26 ***
MATUR	1.4855	1.6094	1.2462	1.3863	1.7247	1.7938	36.01 ***	−32.32 ***
SIZE	18.5059	18.6438	18.3911	18.5160	18.6208	18.8261	4.69 ***	−4.90 ***
HHI	-	-	0.2601	0.3539	-	-	-	-

***, **, and * indicate significance levels at 1%, 5% and 10%, respectively. This table reports the univariate analysis of differences in means and medians of AEM, RAM, loan- and firms-specific characteristics that engage in syndicated borrowing (SYND = 1) versus firms that do not engage in syndicated borrowing (SYND = 0). Variable definitions and measurements are in Appendix A and Appendix B.

shows the univariate analysis of characteristics of syndicate borrowers versus non-syndicate borrowers. Tests of differences suggest that syndicate borrowers exhibit higher, significant medians ADISC_t−1, DA_K_t−1, and RM2_t−1 but significantly lower median ACFO_t−1 and DA_D_t−1 than bilateral (non-syndicate) borrowers. Tests of differences also suggest that syndicate borrowers of firms are smaller in size, more likely to issue management forecasts, less audited by Big4, have less debt to assets and engage more in shorter-term maturity loans than non-syndicate borrowers. Overall, the results in Table 3 suggest that syndicate borrowers are significantly different from non-syndicate borrowers. Notably, they also engage in earnings management practices.

4. Results and Conclusions

4.1. Earnings Management Around Syndicated Lending

Table 4. Regression of syndication on determinants of syndication (n = 11,743). SYND_it = β₀ + β₁ SYND_it−1 + β₂ BIG4_it + β₃ REST_it + β₄ FOR_it + β₅ ROA_it + β₆ FINL_it + β₇ LIQ_it + β₈ LOSS_it + β₉ AT_it + β₁₀ BUSY_it + Σβ_i IND_CAT_it+ Σβ_t YEAR_it.

Parameter	Predicted Sign	Coeff.
Intercept	?	1.0084
SYNDt−1	+	4.8234 ***
BIG4t	+	0.3236 ***
RESTt	−	−0.3020 ***
FORt	+	−0.0182
ROAt	+	−2.3127 ***
FINLt	+	−0.1570
LIQt	+	0.0124
LOSSt	+	0.7406 ***
ATt	+	−0.3801 ***
BUSYt	+	0.3244
IND_CAT		Included
YAER		Included
#Obs.		11,743
Likelihood Ratio χ2		3609.08 ***

***, **, and * indicate significance levels at 1%, 5% and 10%, respectively. This table provides the results of the logistic regression of the relationship between the probabilities of the current year’s syndication on determinants of syndication in order to calculate the propensity score matched for syndication. p-values are based on two-tailed tests. SYND_t (SYND_t−1) is an indicator variable that equals 1 if the firm engaged in syndicate lending at the end of the year (in prior year), 0 otherwise. Variable definitions and measurements are in Appendix A and Appendix B.

displays the results of the logistic regression model (Model 1) used to calculate the propensity scores of the full sample (n = 11,743) of both the treatment sample (syndicate borrowers) and the control sample (bilateral “non-syndicate” private debt borrowers). As predicted, the results show that the likelihood of syndication is based on the previous year’s syndication, restatement, audit quality, firm performance, losses, and borrower size. Specifically, the probability of syndication is an increasing function of the prior year’s syndication, Big4, and losses, and a decreasing function of restatement, ROA, and firm size. The model is statistically significant at 1%, and the likelihood ratio is 3609.08. After obtaining the propensity scores from running Model (1), we then matched the treatment and control samples based on the propensity scores within the range of 0.10. The final sample is composed of 4384 facilities of matched pairs of syndicated and bilateral loans that will be used to test our first hypothesis.

Table 5. Regression of prior-year’s AEM on syndication (n = 4384). DA_it−1 = β₀ + β₁ SYND_it + β₂ SOX_it + β₃ SUB_it + β₄ LOSS_it + β₅ NOA_it + β₆ EXTRA_it + Β₇ FOR_it + β₈ ROA_it + β₉ FINL_it + β₁₀ Z_it + β₁₁ MTB_it + β₁₂ BIG4_it + β₁₃ AT_it + β₁₄ ΔGDP_it + β₁₅ MF_it + β₁₆ LIT_it + β₁₇ MATUR_it + β₁₈ SIZE_it + Σβ_i IND_CAT_it + Σβ_t YEAR_it.

Parameter	Predicted Sign	DA_Dit−1		DAD_INCit−1		DA_Kit−1		DAK_INCit−1
		Coeff.	t-Value	Coeff.	t-Value	Coeff.	t-Value	Coeff.	t-Value
Intercept	?	−13.0870	−1.56	−10.6457	−4.33 ***	−14.7374	−5.66 ***	1.0357	1.26
Variable of interest:
SYNDt	+	−0.0037	−0.09	0.0559	5.27 ***	0.0462	3.60 ***	−0.0107	−2.97 ***
Control variables:
SOX	−	2.3879	2.12 **	1.3363	4.26 ***	1.8067	5.59 ***	−0.2016	−2.03 **
ACFOt−1	−	−0.6025	−3.91 ***	−0.1794	−6.18 ***	−0.3595	−10.19 ***	−0.0682	−4.54 ***
ADISCt−1	−	−0.0955	−0.96	0.0828	5.45 ***	−0.0260	−1.31	0.0353	4.54 ***
APRODt−1	+	0.3941	2.80 ***	0.0425	2.65 ***	0.0117	0.59	−0.0086	−1.09
LOSS	+	0.0685	1.05	0.0251	1.33	0.0135	0.61	0.0167	2.38 **
NOA	−	−0.0619	−1.91 *	−0.0193	−2.21 **	−0.0209	−2.14 **	−0.0107	−3.39 ***
EXTRA	+	−0.1016	−2.24 **	0.0092	0.83	0.0035	0.27	0.0022	0.59
FOR	+	0.1393	3.56 ***	−0.0223	−2.77 ***	−0.0164	−1.53	0.0034	0.87
ROA	+	−0.1930	−0.21	0.4710	0.95	0.3177	0.44	−0.1240	−0.54
FINL	+	0.6474	0.48	−0.3595	−0.53	−0.2621	−0.27	0.2013	0.63
Z	+	−0.1589	−0.67	0.0538	0.45	0.0125	0.07	−0.0369	−0.66
MTB	+	0.0056	0.96	0.0036	2.03 **	0.0007	0.29	0.0022	1.41
BIG4	−	−0.0127	−0.22	0.0172	1.24	0.0531	2.90 ***	0.0061	0.96
AT	+	−0.0603	−2.88 **	0.0098	1.67 *	0.0025	0.40	−0.0050	−2.06 **
ΔGDP	+	1.3764	1.29	1.4471	4.67 ***	1.9588	6.15 ***	−0.1454	−1.51
MF	+/−	−0.1234	−1.94 *	−0.0010	−0.06	0.0084	0.46	0.0278	2.60 ***
LIT	−	−0.1030	−1.69 *	−0.0193	−1.14	−0.1003	−4.82 ***	0.0208	3.43 ***
MATUR	+	0.0510	1.25	0.0175	1.41	0.0218	1.54	0.0015	0.31
SIZE	+	0.0517	2.59 ***	−0.0042	−0.84	0.0067	1.15	0.0005	0.24
IND_CAT	?	Included		Included		Included		Included
YEAR	?	Included		Included		Included		Included
#Obs.		4384		4384		4384		4384
F Value		13.48		14.31		31.25		15.49
Pr > F		<0.0001		<0.0001		<0.0001		<0.0001
Adjusted R2		10.45%		11.07%		22.05%		11.94%

***, **, and * indicate significance levels at 1%, 5% and 10%, respectively. This table provides the results of Heteroscedasticity-Consistent Standard Error regression on the relationship between discretionary accruals and syndication and firm-specific characteristics. All independent variables are calculated at the beginning of the year to take into consideration reverse causality between earnings management and syndication. Table 5 reports results of four models; the dependent variables in all four models are measures of AEM, as follows. (1) DA_D_t−1: Signed discretionary accruals as in Dechow et al. (1995) at year t − 1. (2) DAD_INC_t−1: a variable that equals DA_D_t−1 if it is greater than zero, zero otherwise. (3) DA_K_t−1: signed discretionary accruals as in Kothari et al. (2005). (2) DAK_INC_t−1: a variable that equals DA_K_t−1 if it is greater than zero, zero otherwise. The independent variable of interest is SYND: an indicator variable equals 1 if the firm engaged in syndicate lending at year t, 0 otherwise. Variable definitions and measurements are in Appendix A and Appendix B.

presents the results of the Heteroscedasticity-Consistent Standard Error Regression results of lagged AEM models, with syndication as the independent variable of interest and other control variables. Lagged AEM is measured using four models, DA_D_it−1, DAD_INC_it−1, DA_K_it−1, and DAK_INC_it−1. The results suggests that syndicate borrowers engage in AEM, as evidenced by the significant positive association between SYND and DAD_INC_it−1 (β₁ = 0.0559, significant at 1%), as well as between SYND and DA_K_it−1 (β₁ = 0.0462, significant at 1%) and SYND. These findings indicate that syndicate borrowers engage in income-increasing discretionary accruals prior to syndication. The association between SYND and DAK_INC_t−1 is significant (β₁ = −0.0107, significant at 1%) but negative in sign. All four models of AEM are statistically significant at 1% and explain 10–22% of the variations in lagged AEM measures.

Table 5 also highlights several significant control variables influencing AEM. For example, SOX is positively and significantly associated with DA_K_it−1, suggesting increased AEM post-SOX due to heightened scrutiny by auditors and regulators. NOA consistently exhibits a negative association across models, indicating that firms with higher NOA engage less in AEM, possibly due to their stronger balance sheet positions. MF shows mixed results; in DA_K_it−1, MF’s coefficient is positive, suggesting that forecast-issuing firms are slightly more likely to use AEM. This could reflect their need to meet forecasted benchmarks through accrual adjustments.

Table 6. Regression of prior-year’s RAM on syndication (n = 4384). RAM_it−1 = β₀ + β₁ SYND_it + β₂ SOX_it + β₃ SUB_it + β₄ LOSS_it + β₅ NOA_it + β₆ EXTRA_it + Β₇ FOR_it + β₈ ROA_it + β₉ FINL_it + β₁₀ Z_it + β₁₁ MTB_it + β₁₂ BIG4_it + β₁₃ AT_it + β₁₄ ΔGDP_it + β₁₅ MF_it + β₁₆ LIT_it + β₁₇ MATUR_it + β₁₈ SIZE_it + Σβ_i IND_CAT_it + Σβ_t YEAR_it.

Parameter	Predicted Sign	ACFOit−1		ADISCit−1		APRODit−1
		Coeff.	t-Value	Coeff.	t-Value	Coeff.	t-Value
Intercept	?	5.0390	1.93 *	12.0803	5.86 ***	15.2477	5.97 ***
Variable of interest:
SYNDt−1	−/+	−0.0146	−1.15	−0.0188	−1.77 *	−0.0028	−0.27
Control variables:
SOX	−/+	−0.6208	−2.26 **	−1.4454	−5.31 ***	0.24944	−6.80 ***
DA_Kt−1	−/+	−0.3430	−9.66 ***	−0.0180	−1.30	0.01246	0.60
ADISCt−1	−/+	−0.0886	−2.79 ***	−0.0645	−2.78 ***	0.01483	−5.32 ***
APRODt−1	+	−0.0353	−0.86	−0.1312	−5.10 ***	0.01275	−0.84 *
LOSS	+	−0.0105	−0.58	−0.0341	−2.43 **	0.01471	−2.58 **
NOA	−/+	−0.0160	−1.39	0.0353	3.49 ***	0.0083	−6.70 ***
EXTRA	+	0.0022	0.18	−0.0215	−2.00 **	0.01044	0.89
FOR	+	0.0246	2.03 **	0.0016	0.16	0.01042	0.86
ROA	+	0.7551	0.55	1.2749	3.13 ***	0.6836	−1.16
FINL	+	−0.6375	−0.37	−1.7833	−3.42 **	0.86801	0.85
Z	+	0.0809	0.27	0.3081	3.35 ***	0.15304	−0.80
MTB	+	0.0043	2.77 ***	0.0089	5.87 ***	0.0012	−2.74 ***
BIG4	+	−0.0143	−0.72	−0.0136	−1.00	0.01378	0.63
AT	+	0.0011	0.17	0.0045	0.86	0.00502	3.47 ***
ΔGDP	+	−0.6166	−2.30 **	−1.3443	−5.12 ***	0.21449	−6.86 ***
MF	−/+	−0.0223	−0.74	0.0244	0.95	0.024	−0.70
LIT	−/+	0.0547	2.63 ***	0.0695	4.06 ***	0.0135	1.64
MATUR	−/+	−0.0314	−2.48 **	−0.0272	−2.38 **	0.01148	−1.28
SIZE	−/+	0.0183	3.00 ***	−0.0084	−1.63	0.0053	−2.91 ***
IND_CAT	?	Included		Included		Included
YEAR	?	Included		Included		Included
#Obs.		4384		4384		4384
F Value		38.38		11.03		16.01
Pr > F		<0.0001		<0.0001		<0.0001
Adjusted R2		25.91%		8.58%		12.31%

***, **, and * indicate significance levels at 1%, 5% and 10%, respectively. This table provides the results of the Heteroscedasticity-Consistent Standard Error regression of the prior year’s real earnings management on syndication and firm-specific characteristics. All independent variables are calculated at the beginning of the year to take into consideration reverse causality between earnings management and syndication. Table 6 reports results of three models; the dependent variables in all three models are measures of RAM, as follows. (1) ACFO_t−1 is the abnormal level of cash flow from operations calculated as the difference between the normal and estimated level of cash flow from operations, as in Roychowdhury (2006). (2) ADISC_t−1 is the abnormal level of discretionary expenses calculated as the difference between the normal and estimated level of discretionary expenses, as in Roychowdhury (2006). (3) APROD_t−1 is the abnormal level of production cost calculated as the difference between the normal and estimated level of production cost, as in Roychowdhury (2006). The independent variable of interest is SYND_it: an indicator variable equals 1 if the firm engaged in syndicate lending, 0 otherwise.

presents the results of the Heteroscedasticity-Consistent Standard Error Regression analysis, in which lagged measures of RAM are regressed on syndication and other control variables. The coefficient for ADISC is negative and statistically significant at the 10% level (β₁ = −0.0188), while the coefficients for other RAM variables are not statistically significant. These results are consistent with our prediction and suggest that syndicated borrowers engage less in, or have no association with, RAM. Taken together, the results in Table 5 and Table 6 provide evidence that syndicated borrowers rely more on AEM and less on RAM, supporting our first hypothesis that they manipulate earnings differently than non-syndicated borrowers.

The results for the control variable association with RAM indicate that SOX has a positive impact on some RAM measures, suggesting shifts toward operational manipulation post-regulation. NOA exhibits a negative relationship, reinforcing that firms with strong asset bases avoid real activities manipulation. MF generally shows inconsistent relationships with RAM and is often insignificant. LOSS displays mixed effects; it increases RAM in certain cases, possibly reflecting distressed firms’ need to manipulate operational metrics to maintain appearances. Overall, these results emphasize that RAM is less prevalent among syndicated loan borrowers, with monitoring variables exerting varying degrees of influence.

We also ran an OLS regression model with the aggregate measures of lagged RAM as the dependent variable (RM1 and RM2) and syndication and control variables as the independent variables. In line with our expectation, the results (untabulated) suggest a positive association at the 10% level between only RM2 and syndication. However, given the magnitude of the statistical significance and the lack of consistency with the RM1 model, these results may not support higher RAM than AEM.14

4.2. The Association Between Earnings Management and Lender’s Monitoring Mechanisms

Table 7. The differential effect of lender’s monitoring mechanisms on future changes in AEM and RAM (n = 2192). FC_EM_it+1 = β₀ + β₁ LENDER_MONITOR_it + β₂ SUB_it + β₃ SOX_it + β₄ LOSS_it + β₅ NOA_it + β₆ EXTRA_it + β₇ FOR_it + Β₈ ROA_it + β₉ FINL_it + β₁₀ Z_it + β₁₁ MTB_it + β₁₂ BIG4_it + β₁₃ AT_it + β₁₄ ΔGDP_it + β₁₅ MF_it + β₁₆ LIT_it + Σβ_i IND_CAT_it + Σβ_t YEAR_it.

Parameter	Pred.	FC_DADt+1		FC_DAKt+1		FC_ACFOt+1		FC_ADISCt+1		FC_APRODt+1
		Model (1)		Model (2)		Model (3)		Model (4)		Model (5)
	Sign	Coeff.	t-Value	Coeff.	t-Value	Coeff.	t-Value	Coeff.	t-Value	Coeff.	t-Value
Panel A: Lender’s Reputation
Intercept	?	124.7424	2.49 **	4.2312	0.18	−9.5022	−0.64	−0.9167	−0.06	7.4641	0.96
HHI	−/+	0.4094	−0.19	−3.6881	−2.06 **	0.6516	0.40	−3.4732	−1.38	−1.9895	−4.13 ***
Control variables		Included		Included		Included		Included		Included
IND_CAT	?	Included		Included		Included		Included		Included
YEAR	?	Included		Included		Included		Included		Included
F Value		2.09		6.90		3.17		2.46		1.71
Pr > F		0.0013		<0.0001		<0.0001		<0.0001		0.0049
Adjusted R2		1.04%		7.15%		2.75%		1.87%		0.09%
Panel B: Loan Size
Intercept	?	152.0350	2.61 **	−8.8007	−0.37	−14.4740	−1.02	−17.2976	−1.22	−7.5403	−1.14
SIZE	−/+	−1.7862	−2.30 **	1.3414	3.04 ***	0.6064	3.53 ***	1.7953	0.96	1.6183	3.79 ***
Control variables:		Included		Included		Included		Included		Included
IND_CAT	?	Included		Included		Included		Included		Included
YEAR	?	Included		Included		Included		Included		Included
F Value		2.38		6.90		3.21		2.58		1.81
Pr > F		<0.0001		<0.0001		<0.0001		<0.0001		0.0016
Adjusted R2		1.76%		7.15%		2.80%		2.03%		1.05%
Panel C: Number of Syndicates
Intercept	?	127.3685	2.45 **	3.1015	0.14	−10.8457	0.69	−5.2653	0.7107	6.4917	0.81
SYNDICATES	−/+	−1.0509	−1.15	−0.1702	−0.31	0.6761	0.96	1.3137	1.84 *	0.1166	0.39
Control variables:		Included		Included		Included		Included		Included
IND_CAT	?	Included		Included		Included		Included		Included
YEAR	?	Included		Included		Included		Included		Included
F Value		2.11		6.80		3.19		2.42		1.69
Pr > F		<0.0001		<0.0001		<0.0001		<0.0001		0.0048
Adjusted R2		1.43%		7.03%		2.78%		1.82%		0.90%
Panel D: Maturity
Intercept	?	125.6745	2.47 **	1.4207	0.06	−9.1926	−0.64	−3.8937	−0.25	9.4401	0.29
MATUR	−/+	0.9577	0.57	−2.1588	−2.19 **	0.0768	0.15	−2.2068	−2.24 **	5.0135	1.77 *
Control variables:		Included		Included		Included		Included		Included
IND_CAT	?	Included		Included		Included		Included		Included
YEAR	?	Included		Included		Included		Included		Included
F Value		2.10		6.86		3.16		2.44		1.96
Pr > F		<0.0001		<0.0001		<0.0001		<0.0001		0.0551
Adjusted R2		1.41%		7.10%		2.74%		1.84%		1.24%

***, **, and * indicate significance levels at 1%, 5% and 10%, respectively. This table provides the results of the Heteroscedasticity-Consistent Standard Error regression of future changes in earnings management on lender’s monitoring mechanisms and firm-specific characteristics. All independent variables are calculated at the beginning of the year to take into consideration reverse causality between dependent and independent variables. Table 7, Panel A reports results of five models; the dependent variables are measures of AEM and RAM, as follows. (1) FC_DAD_t+1: (2) FC_DAK_t+1: (3) FC_ACFO_t+1: (4) FC_ADISC_t+1: (5) FC_APROD_t+1: The independent variable of interest in Panel A is HHI: the lender’s Herfindahl–Hirschman Index (HHI) per two-digit sic code/industry. Other independent variables are included in the model as control variables, as described below. The independent variable of interest in Panel B is SIZE, measured as the loan size in millions. The independent variable of interest in Panel C is SYNDICATES, which is the log of number of syndications. The independent variable of interest in Panel D is MATUR, which is the log of loan maturity measured in years. A list of control variables is included in all models. Variable definitions and measurements are in Appendix A and Appendix B.

summarizes the empirical testing of the second research hypothesis on the association between the lender’s monitoring mechanisms and future changes in AEM and RAM. It is hypothesized that syndicate lenders are informed traders who possess private information about syndicate borrowers. Therefore, lenders can see through the mere accounting numbers and consequently could curb earnings management, mostly income-increasing AEM. Since RAM is indistinguishable from optimal business operations, lenders may or may not be able to curb such managerial opportunism. Table 7 presents the results of the Heteroscedasticity-Consistent Standard Error Regression results of future changes in AEM and RAM models on syndicate lenders’ monitoring mechanisms, namely, lenders’ reputation (HHI), loan size (SIZE), number of syndicates (SYNDICATES), and loan maturity (MATUR), Herfindahl–Hirschman Index (HHI), and other control variables. Panel A of Table 7 shows a significant negative association between FC_DAK_t+1 and HHI (β₁ = −3.6881, significant at 5%), suggesting that reputable lenders are associated with lower future discretionary accruals. The results also suggest that lenders’ monitoring mechanisms, facilitated by market concentration, enable them to curb RAM—though only those arising from abnormal production expenses (β₁ = −1.9895, significant at the 1% level for FC_APROD_t+1). These findings suggest that market concentration may serve as a proxy for lender monitoring mechanisms, with a more pronounced effect on real earnings management (RAM) through abnormal production costs than on accrual-based earnings management (AEM) through abnormal discretionary expenses.

Panel B of Table 7 displays the results of the association between earnings management (AEM and RAM) and loan size (SIZE). As shown in Panel B, SIZE is significantly associated with AEM as well as RAM measures. There is a negative (positive) significant association between FC_DAD_t+1 (FC_DAK_t+1) and SIZE (β₁ = −1.7862 and 1.3414, significant at 5% and 1%, respectively). There is also a positive association between SIZE and FC_ACFO_t+1 and FC_APROD_t+1, both significant at 1%. This latter result suggests that larger loan sizes are associated with lower (higher) future change in ACFO (APROD). These results are puzzling because they indicate an asymmetric effect of the loan size, as a proxy of the amount of governance the lender can exercise as part of his/her monitoring mechanisms, on future earnings management.

Panel C shows a significant positive association between SYNDICATES and FC_ADISC_t+1 (β₁ = 1.3137, significant at 10%). This provides marginal evidence that the number of syndications is associated with lower ADISC. Panel D of Table 7 shows a negative significant association between MATUR and FC_DAK_t+1 and FC_ADISC_t+1, both at 5% significance level, indicating that loans with longer maturity allow lenders to reduce future discretionary accruals as well as discretionary expenses.

Overall, the results in Table 7 support the second research hypothesis that earnings management through accruals are significantly associated with syndicate lenders’ monitoring mechanisms. It also suggests that the lender’s monitoring mechanisms are asymmetric when it comes to monitoring future RAM.

4.3. Additional Analysis

4.3.1. The Association Between RAM and Cost of Debt

Table 8. Regression of the cost of debt on prior year’s AEM and RAM (n = 2192). COD_it = β₀ + β₁ EM_it−1*HHI_it + β₂ EM_it−1 + β₃ HHI_it + Σβ₄ LENDER_MONITOR_it + Σβ₅ LOAN_CH_it + β₆ SOX_it + β₇ LOSS_it + β₈NOA_it + β₉ EXTRA_it + β₁₀ FOR_it + β₁₁ ROA_it + β₁₂ FINL_it + β₁₃ Z_it + β₁₄ MTB_it + β₁₅ BIG4_it + β₁₆ AT_it + β₁₇ ΔGDP_it + β₁₈ LIT_it + Σβ_i IND_CAT_it + Σβ_t YEAR_it.

Parameter	Pred.	COD
		Model (1)		Model (2)		Model (3)		Model (4)		Model (5)
	Sign	Coeff.	t-Value	Coeff.	t-Value	Coeff.	t-Value	Coeff.	t-Value	Coeff.	t-Value
Intercept	?	−25.314	−3.67 ***	−25.758	−3.75 ***	−24.776	−3.56 ***	−25.208	−3.75 ***	−23.917	−3.53 ***
DA_kit−1*HHI	−	−0.333	−1.60 *
DAK_INCit−1*HHI	−			−0.763	−1.05
ACFOit−1*HHI	+					0.924	2.73 ***
ADISCit−1*HHI	+							0.435	1.94 **
APRODit−1*HHI	+									−1.105	−4.30 ***
LENDER_MONITOR	−/+	Included		Included		Included		Included		Included
LOAN_CH	−/+	Included		Included		Included		Included		Included
CONTROL VARIABLES	−/+	Included		Included		Included		Included		Included
IND_CAT	−/+	Included		Included		Included		Included		Included
YEAR	?	Included		Included		Included		Included		Included
F Value		24.05		23.98		24.51		24.11		25.04
Pr > F		<0.0001		<0.0001		<0.0001		<0.0001		<0.0001
Adjusted R2		28.79%		24.74%		25.16%		24.84%		25.59%

***, **, and * indicate significance levels at 1%, 5% and 10%, respectively. This table provides the results of the Heteroscedasticity-Consistent Standard Error regression on the association between RAM and cost of debt (COD) among syndicate loan borrowers. All independent variables are calculated at the beginning of the year to take into consideration reverse causality between dependent and independent variables. The dependent variable in all models is the cost of debt (COD), which is the log of all-in-Spread Drawn and measured as the annual spread paid over LIBOR for each dollar drawn down from the loan. The independent variables of interest are the interaction terms of HHI with measures of AEM and RAM, such as the following: (1) DA_K_it−1: signed discretionary accruals as in Kothari et al. (2005), (2) DAK_INC_it−1: Income-increasing DA_K_t−1, Which equals DA_K_t−1 if it is a positive number, zero otherwise), (3) ACFO_it−1 is the abnormal level of cash flow from operations calculated as the difference between the normal and estimated level of cash flow from operations, as in Roychowdhury (2006), (4) ADISC_it−1 is the abnormal level of discretionary expenses calculated as the difference between the normal and estimated level of discretionary expenses, as in Roychowdhury (2006), and (5) APROD_it−1 is the abnormal level of production cost calculated as the difference between the normal and estimated level of production cost, as in Roychowdhury (2006). We also include lender’s monitoring mechanisms, loan characteristics (e.g., covenants, loan ratings and information asymmetry) and firm-specific characteristics as control variables.

presents the results of the empirical analysis examining the relationship between the interaction terms of the Herfindahl–Hirschman Index (HHI) and lagged accrual-based earnings management (AEM) and real earnings management (RAM) on the cost of debt (COD)15 for syndicated borrowers. In this context, the cost of debt, our dependent variable, is represented as the natural logarithm of the all-in-drawn spread. This spread is defined as the annualized margin paid over LIBOR for each dollar drawn down from the loan.

The findings in Table 8 indicate a statistically significant negative association between the interaction term DA_K*HHI and COD (β₁ = −0.333, significant at 10%). This suggests that higher levels of AEM are linked to a reduced cost of debt, implying that syndicated borrowers’ use of accrual-based earnings manipulation and the presence of reputable lenders with market concentration may be viewed favorably by lenders in certain contexts.

In contrast, results concerning RAM reveal a different pattern. The interaction terms involving real earnings management proxies, ACFO*HHI and ADISC*HHI, exhibit significant positive coefficients (β₁ = 0.924, and 0.435, significant at 5%, and 10%, respectively). However, the association between the interaction term APROD*HHI and COD shows a significant negative association (β₁ = −1.105, significant at 5%).

The findings in this section suggest that in markets with higher lender concentration, real earnings management (RAM) through reduced cash flows from operations and cuts in discretionary spending is associated with a higher cost of debt. This indicates that lenders in more concentrated markets may possess stronger monitoring capabilities and greater pricing power, making them more sensitive to earnings manipulation that undermines cash flow or long-term investment. As a result, firms engaging in RAM by offering lenient credit terms or reducing R&D and advertising may be viewed as higher risk, leading lenders to charge a risk premium.

In contrast, the negative association between abnormal production costs and the cost of debt suggests that lenders may view RAM through overproduction more favorably in concentrated markets. Since overproduction does not immediately harm liquidity and may be interpreted as a cost-management strategy, it is less likely to be penalized. Overall, lenders appear to distinguish between types of RAM, penalizing those that harm financial flexibility or future performance while being more tolerant of those perceived as less damaging or more transparent.

These results are consistent with our research hypotheses, which posit that syndicated borrowers are more inclined to engage in accrual-based earnings management compared to real earnings management, given the monitoring mechanisms imposed by lenders. The analysis in this section advances our understanding of the motivations and potential consequences of earnings management practices within the context of syndicated lending.

4.3.2. The Association Between RAM and Cash Flow from Operations

To examine whether RAM has an impact on future cash flow from operations, we run OLS regression models with current and future cash flow from operations as dependent variables and lagged RAM as independent variables of interest.

Table 9. Regression of current and future cash flow from operations on RAM (n = 2192). CFO_it = β₀ + β₁ RAM_it−1 + β₂ LENDER_MONITOR_it + β₃ SOX_it + β₄ LOSS_it + β₅ NOA_it + β₆ EXTRA_it + β₇ FOR_it + Β₈ ROA_it + β₉ FINL_it + β₁₀ Z_it + β₁₁ MTB_it + β₁₂ BIG4_it + β₁₃ AT_it + β₁₄ ΔGDP_it + β₁₅ LIT_it + Σβ_i IND_CAT_it + Σβ_t YEAR_it.

Panel A: Regression of Current Cash Flow from Operations on RAM:
	CFOit
	Pred. Sign	Coeff.	t-value	Coeff.	t-value	Coeff.	t-value
Panel A: Regression of Current Cash Flow from Operations on RAM:
Intercept	?	25.7710	3.16 ***	27.9180	3.45 ***	24.9950	3.15 ***
ACFOit−1	+	0.0020	0.06
ADISCit−1	+			−0.1199	−3.38 ***
APRODit−1	−					0.1673	4.74 ***
CONTROL VARIABLES		Included		Included		Included
IND_CAT	?	Included		Included		Included
YEAR	?	Included		Included		Included
F Value		69.22		69.94		70.37
Pr > F		<0.0001		<0.0001		<0.0001
Adjusted R2		54.26%		54.52%		54.68%
Panel B: Regression of Future Cash Flow from Operations on RAM:
	CFOit+1
Parameter	Pred. Sign	Coeff.	t-value	Coeff.	t-value	Coeff.	t-value
Intercept	?	−19.2800	−3.68 ***	−14.3510	−2.78 ***	−17.7450	−3.43 ***
ACFOit−1	+	0.1077	2.94 ***
ADISCit−1	+			−0.1868	−5.06 ***
APRODit−1	−					0.0202	0.06
CONTROL VARIABLES	+/−	Included		Included		Included
IND_CAT		Included		Included		Included
YEAR	?	Included		Included		Included
F Value		158.30		160.31		157.33
Pr > F		<0.0001		<0.0001		<0.0001
Adjusted R2		73.23%		73.48%		73.11%

***, **, and * indicate significance levels at 1%, 5% and 10%, respectively. This table provides the results of the Heteroscedasticity-Consistent Standard Error regression of current and future cash flow from operations (CFO) on RAM, lender’s monitoring mechanisms and firm-specific characteristics. All independent variables are calculated at the beginning of the year to take into consideration reverse causality between dependent and independent variables. Table 9 reports results of six models; the dependent variables in all models are measures of current and future cash flow from operations. Panel A of Table 9 summarizes the results of the association between current CFO_it and the prior year’s RAM measures as independent variables of interest (ACFO_it−1, ADISC_it−1 and APROD_it−1). Panel B of Table 9 summarizes the results of the association between future CFO_it+1 and prior year’s RAM measures as independent variables of interest (ACFO_it−1, ADISC_it−1 and APROD_it−1). Other control variables are also included. Variable definitions and measurements are in Appendix A and Appendix B.

, Panel A summarizes the results. The regression of current cash flow from operations (CFO) on lagged RAM shows a negative (positive), statistically significant association at 1% between CFO and ADISC_t−1 (APROD_t−1), suggesting that manipulating RAM using APROD_t−1 is associated with better short-term performance. The results in Table 9, Panel B, show that future cash flow from operations is negative (positive), significantly associated with lagged ADISC_t−1 (ACFO_t−1). The results in this section suggest that manipulating ADISC is associated with lower current and future CFO but manipulating APROD_t−1 (ACFO_t−1) is significantly associated with higher current (future) CFO. Overall, the results support our earlier finding that syndicated borrowers manipulate ADISC less than bilateral borrowers, possibly due to its negative impact on cash flow from operations.

5. Conclusions and Recommendations for Future Research

This study examines the trade-off between accrual-based earnings management (AEM) and real activities manipulation (RAM) among syndicated loan borrowers in the period preceding loan origination. The primary objective was to assess whether weaker lender monitoring in syndicated loans encourages firms to favor AEM, and whether lender oversight mechanisms constrain post-origination earnings management. Drawing on a propensity score-matched sample of 2192 syndicated, and 2192 bilateral loans issued between 1989 and 2005, we utilized data from Loanware, COMPUSTAT, and Bloomberg. We applied multiple earnings management models (Dechow et al., 1995; Kothari et al., 2005; Roychowdhury, 2006) and examined key lender monitoring proxies—loan size, maturity, lender reputation, and number of syndicate participants—within a robust regression framework.

The results reveal that syndicated loan borrowers engage significantly more in AEM than RAM prior to loan issuance, consistent with the notion that AEM is less costly and more difficult to detect (Cohen & Zarowin, 2010; Zang, 2012). Post-origination, stronger monitoring mechanisms—especially lender reputation and loan maturity—are associated with lower levels of AEM (Ahn & Choi, 2009; Chung et al., 2005). These findings confirm and extend previous research on the role of creditor oversight in shaping managerial reporting behavior (Denis & Mihov, 2003; Bharath et al., 2008b; Wittenberg-Moerman, 2008). Additional analyses show that AEM is positively associated with a higher cost of debt, while RAM has mixed implications for short- and long-term cash flows (Ge & Kim, 2014; Gunny, 2010; Taylor & Xu, 2010). Notably, these findings hold even after controlling for voluntary disclosure (Gao et al., 2024), supporting the robustness of the results.

Our findings provide meaningful real-world insights into how firms strategically engage in earnings management prior to syndicated loan origination. Specifically, the increased reliance on accrual-based earnings management (AEM) over real activities manipulation (RAM) highlights a preference for financial reporting strategies that do not disrupt actual operations or future cash flows, an important consideration in maintaining lender trust and minimizing borrowing costs. This behavior aligns with the theoretical model proposed by Cohen and Zarowin (2010) and is consistent with the results of Ahn and Choi (2009), who found that earnings management tends to decrease as bank monitoring improves. Unexpectedly, RAM usage among syndicated borrowers is less pronounced despite its post-SOX popularity, likely due to its more observable operational effects and long-term cash flow implications (Roychowdhury, 2006; Ge & Kim, 2014).

Furthermore, our results show that enhanced lender monitoring, through factors like syndicate size, lender reputation (HHI), and loan maturity, is associated with reduced future AEM. This supports the monitoring hypothesis and underscores the effectiveness of institutional oversight in curbing opportunistic reporting. In comparing our findings to studies in other private debt markets such as those in Europe (e.g., Mafrolla & D’Amico, 2017), we observe similar patterns where earnings management is strategically employed to optimize borrowing terms, though regulatory and institutional differences may affect the intensity and form of such behavior. These insights are particularly relevant for financial regulators and institutional lenders seeking to design loan structures that minimize agency conflicts and promote transparent borrower behavior.

This study contributes to the extant literature in accounting and finance in several ways. Theoretically, it enhances understanding of how the structure of syndicated lending affects firms’ financial reporting incentives. Methodologically, it applies a rigorous matching design and multiple model specifications to address selection bias and measurement concerns. From a financial and managerial standpoint, it highlights the importance of lender monitoring in deterring earnings manipulation, especially through accruals, and offers practical implications for lenders, auditors, and regulators involved in private debt contracting (Bushman & Wittenberg-Moerman, 2012; Francis et al., 2005; Sufi, 2007).

Nevertheless, the study has limitations. The use of discretionary accrual and RAM models may be subject to measurement error (Dechow et al., 1995; Roychowdhury, 2006), and the focus on U.S. syndicated loans may limit generalizability. Future research could extend this work by exploring the role of internal governance mechanisms, investigating cross-country loan markets, or analyzing behavioral factors such as managerial overconfidence (Adam et al., 2020). These avenues would further clarify how earnings management evolves under different institutional and regulatory environments, particularly around key debt financing events.