JEF Journal of Economic and Financial Sciences 1995-7076 2312-2803 AOSIS JEF-19-1094 10.4102/jef.v19i1.1094 Original Research The cost of debt in South Africa: A reassessment of commonly used control variables https://orcid.org/0000-0003-2806-4877 Boshoff-Knoetze Anet 1 https://orcid.org/0000-0001-5503-9883 Nel George F. 1 https://orcid.org/0000-0002-2597-2987 Erasmus Pierre D. 2 School of Accountancy, Faculty of Economics and Management Sciences, Stellenbosch University, Stellenbosch, South Africa Department of Business Management, Faculty of Economics and Management Sciences, Stellenbosch University, Stellenbosch, South Africa Corresponding author: Anet Boshoff-Knoetze, anetk@sun.ac.za 23072026 2026 19 1 1094 16102025 22052026 © 2026. The Authors 2026 Licensee: AOSIS. This work is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. Orientation

The cost of debt (COD) is incorporated into the weighted average cost of capital, which is used in valuations, capital budgeting and costing applications.

 Research purpose

The study investigated factors that are correlated with the COD of listed companies in South Africa.

 Motivation for the study

The results of previous studies are divergent as to which factors are reliably correlated with the COD, measured as interest divided by average borrowings.

 Research approach/design and method

Potential determinants were identified from existing literature. Panel data from 229 companies listed on the Johannesburg Stock Exchange (JSE) were analysed using regression techniques.

 Main findings

Only one of the 18 commonly included control variables for COD, as identified from the existing literature, provided robust support for the hypothesised directional relationship with the COD: A binary variable for loss-making entities. Leverage, return on assets (ROA), asset turnover and listing age were statistically significant in the opposite direction to what was hypothesised, whereas the other 13 potential determinants did not demonstrate a stable, statistically significant relationship with the COD across different model specifications.

 Practical/managerial implications

Existing perceptions of the impact of factors, such as leverage and ROA, on the COD might be misguided, which brings the results of previous studies into question.

 Contribution/value add

The study raises questions about whether researchers include the correct control variables when performing regression analyses on the COD. The results also indicate that measuring the COD as interest divided by average borrowings warrants further scrutiny.

 debt cost cost of debt cost of capital financing cost emerging markets Johannesburg Stock Exchange South Africa Funding information The authors received no financial support for the research, authorship, and/or publication of this article. Introduction

Debt is a major source of financing for most companies, with South African public and private companies reporting significant use of debt sources (Stats SA 2026). The cost of debt (COD) is one of the elements incorporated into the weighted average cost of capital (WACC), which is used in valuations, financial reporting, capital allocation decisions and costing applications. The COD is also often used as a risk proxy, indicating whether investors’ perceived risk is influenced by factors such as voluntary disclosure (Guidara, Khlif & Jarboui 2014), Internet investor relations (Nel, Smit & Brümmer 2019) or reporting quality (Muttakin et al. 2020). A level of understanding of the COD is therefore required to perform accurate and relevant research.

Nikolaev and Van Lent (2005) demonstrated that the inclusion or exclusion of a single independent variable (or control variable) in a multiple regression model can alter research findings, particularly in the context of COD. Li (2021) further warns against the unexpected consequences of including or excluding control variables in a multiple regression model, calling on scholars to carefully consider whether to include them. Despite this result, there is little consistency in the control variables being included in research on the COD.

The South African context provides a unique background for exploring the COD. The use of traded debt is uncommon in South Africa, so most companies utilise private placements. The South African financial system is considered one of the best in the world, providing accurate financial records; however, the country scores poorly in terms of corruption, which could impact private placements (Schwab 2019).

Even though some research has been performed on the COD in a South African context (see, for example, Guidara et al. 2014; Muttakin et al. 2020; Radier et al. 2016), a brief investigation into the factors expected to be correlated with the COD, as evidenced from previous research, provides divergent results. This complicates the task of researchers attempting to perform multiple regression analysis on the COD because it is unclear which control variables to consider. In addition, while researchers investigating developed markets have several different options for measuring the COD, such as credit spread, credit ratings and interest rates on private loans, data limitations in South Africa, like in many other emerging economies, limit researchers to the COD measured as average debt cost (ADC, interest divided by average borrowings).

Consequently, this study aims to: (1) discuss the theoretical framework of the relationships between commonly included control variables and the COD, measured as ADC, and (2) identify the determinants of the COD of Johannesburg Stock Exchange (JSE)-listed companies.

The research questions (RQs) derived from these aims are: (1) What are the commonly included control variables used in regression analysis for the COD, measured as ADC?, (2) What are the theoretical bases for commonly included control variables’ relationships with the COD?, (3) Is there a statistically significant relationship between the COD (measured as ADC) and each of the commonly included control variables?, and (4) Are there indications of ADC having measurement validity concerns?

 Literature review Measurement of the cost of debt

In academic textbooks, the COD is referred to as ‘the return that lenders require on new borrowing’ (Firer et al. 2012:437), ‘the opportunity cost of capital for the investors who hold the firm’s debt’ (Brealey, Myers & Allen 2017:216) or ‘the interest rate on the firm’s new debt’ (Brigham & Ehrhardt 2020:339). In peer-reviewed research, Magnanelli and Izzo (2017:251) proposed defining the COD as ‘the overall rate paid by a firm to use debt financing’. Even in these few definitions, there appears to be disagreement in what researchers are trying to measure when determining the COD.

Perhaps more practical is the measurement instruments used in previous research when measuring the COD, which can broadly be categorised as:

The bond yield spread (the difference between the risk-free rate and the bond yield rate) for traded bonds and debt instruments (Ahmed, Anderson & Zarutskie 2015; Bharath, Sunder & Sunder 2008; Ertugrul et al. 2017).

The company’s credit rating (Graham, Li & Qiu 2008; Lim, Mann & Mihov 2017; Sengupta 1998).

The interest rate on specific private loan issues or active loans, as disclosed by banks (see, for example, Berger & Udell 1990; Blackwell, Noland & Winters 1998).

Average debt cost (Ozkaya 2018; Pittman & Fortin 2004; Stanišić, Stefanović & Radojević 2016).

In the context of an emerging market such as South Africa, traded bonds are rarely used (Ahwireng-Obeng & Ahwireng-Obeng 2022), and consequently, companies generally do not obtain a credit rating, resulting in the first two measurement instruments being unusable. To address specific private loan issues, one would need such a database, which is typically not available in emerging economies such as South Africa. This leaves only the last measurement instrument, being ADC, as a viable alternative for research on the COD in South Africa.

 Identifying potential determinants of the cost of debt

In order to identify potential determinants of the COD in South Africa (RQ1), previous studies on the COD, measured as ADC, were reviewed. The lack of consistency in the independent variables included in previous studies is notable and discussed in further detail.

Guidara et al. (2014) examined the relationship between voluntary and timely disclosure and the COD for a sample of 20 firms over a 4-year period. They discovered a significant relationship between voluntary disclosure and COD, but no such link was found for timely disclosure. Additionally, none of the control variables – such as earnings variability, firm size (measured by the natural logarithm of total assets) or a dummy variable indicating whether the company was making a loss – showed a significant association with COD.

Nel et al. (2019) studied the impact of Internet investor relations on the COD, using a dataset covering just 1 year. They found a significant negative relationship between Internet investor relations and COD. Although several variables, including leverage, market-to-book equity ratio (MV/BV equity), earnings per share volatility, return on equity, interest coverage and dual listing, were incorporated in a stepwise regression analysis, only earnings per share volatility was significantly linked to COD.

Muttakin et al. (2020) found a negative relationship between financial reporting quality, the publication of an integrated report (measured by a dummy variable) and the COD. They also included several independent variables that did not show a statistically significant relationship with the COD, such as the percentage of black directors, the percentage of independent board members, the market value to book value (MV/BV) equity, tangibility (measured by the ratio of property, plant and equipment [PPE] to total assets), the current ratio, earnings volatility and dummy variables for the use of a ‘Big Four’ auditor and block shareholders. Apart from the variables of interest (financial reporting quality and integrated report publication), only a dummy variable for loss-making firms and company size showed a statistically significant correlation with COD.

Johnson (2020) examined the relationship between environmental, social and corporate governance (ESG) disclosure and the COD. The study found significant negative correlations between the COD and the social dimension, corporate governance and company size. However, no significant relationships were identified between the COD and the environmental dimension or leverage.

Medhioub and Boujelbene (2023) found that the relationship between tax avoidance and the COD is influenced by the presence of integrated report assurance (i.e. when an independent third party verifies the accuracy of the integrated report). A significant portion of their study focused on addressing endogeneity issues, with various robustness tests conducted. They found statistically significant relationships between the COD and factors like size, leverage and credit risk. However, there was no statistically significant relationship between COD and return on assets (ROA), Altman’s z-score or a dummy variable for loss-making in the previous year.

The results of the above-mentioned research studies highlight inconsistencies in, firstly, the control variables being included and secondly, the results of these control variables. It seems surprising that control variables such as leverage (Johnson 2020; Nel et al. 2019) and loss-making indicators (Guidara et al. 2014; Medhioub & Boujelbene 2023) would not yield statistically significant correlations with COD.

Considering the lack of consensus in the South African research, a review of international literature was performed to identify potential independent variables that could be determinants of COD. Additional research studies, focusing on COD research using ADC, conducted in the international context were perused. Potential determinants were identified based on previous studies that a statistically significant relationship with these independent variables and COD, measured as ADC. The focus was on identifying potential determinants to be used as control variables in future studies; consequently, only variables that were practicably obtainable were included in the results, as shown in Table 1.

Summary of control variables included in previous studies on the cost of debt and the expected association with the cost of debt.

Article or summary item Size Tan Cur R ROA Int Cov Age CF/Assets Tob Q MtB Econ G Asset G ESG Asset turn Lev E Var Loss Int rate Neg Eq
EXP - - - - - - - - - - - - - + + + + +
[1] No NI NI Yes Yes NI NI NI NI NI NI NI NI Yes Yes NI NI NI
[2] Yes NI NI NI Yes NI NI NI NI NI NI NI NI OP NI NI Yes NI
[3] No Yes Yes No Yes Yes NI NI NI NI NI NI NI OP NI NI NI Yes
[4] Yes NI NI No NI No Yes Yes NI Yes NI NI NI Yes NI NI NI NI
[5] No Yes NI NI NI Yes No NI NI NI NI NI NI Yes NI NI Yes Yes
[6] Yes OP NI NI NI Yes NI NI NI NI NI NI NI OP Yes NI NI Yes
[7] Yes No Yes No NI OP No NI Yes Yes NI NI Yes Yes NI NI NI NI
[8] Yes OP NI Yes NI NI NI No NI NI Yes NI NI Yes NI NI NI NI
[9] NI Yes No No NI No No NI OP NI NI NI NI Yes NI No NI NI
[10] Yes No Yes NI Yes NI OP No NI NI Yes Yes NI Yes NI NI NI NI
[11] Yes No No NI NI NI NI NI No NI NI NI NI NI No Yes NI NI
[12] Yes NI Yes No NI NI NI NI NI NI NI NI NI Yes NI No NI NI
[13] OP Yes OP NI Yes NI NI NI NI Yes NI NI NI NI No NI Yes NI
[14] Yes No NI Yes NI NI Yes NI Yes NI NI NI NI No Yes NI NI NI
[15] No Yes NI NI NI NI NI NI NI NI NI NI NI NI NI NI NI NI
[16] Yes NI NI NI NI NI NI NI NI NI NI Yes NI No NI NI NI NI
[17] No Yes Yes NI Yes NI NI Yes NI NI No NI NI No No NI NI NI
[18] No NI NI NI NI NI NI NI NI NI NI NI NI NI No No NI NI
[19] NI NI NI NI No NI NI NI NR NI NI NI NI NR NR NI NI NI
[20] NI NI Yes No NI NI NI NI NI NI NI NI NI Yes NI NI NI NI
Yes 10 6 6 3 6 3 2 2 2 3 2 2 1 9 3 1 3 3
No 6 4 2 6 1 2 3 2 1 0 1 0 0 3 4 3 0 0
OP 1 2 1 0 0 1 1 0 1 0 0 0 0 3 0 0 0 0
NI 3 8 11 11 13 14 14 16 15 17 17 18 19 4 12 16 17 17
NR 0 0 0 0 0 0 0 0 1 0 0 0 0 1 1 0 0 0
TOTAL 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20

Notes: YES indicates that the independent variable in the regression model was statistically significant in its correlation with COD in the direction hypothesised, at a minimum of 90% confidence interval. NO indicates that the independent variable in the regression model was not statistically significant in its correlation with COD at a minimum of 90% confidence interval. OP indicates that the independent variable was found to be statistically significant in its correlation with COD at a minimum of 90% confidence interval, but in the opposite direction to what was expected.

Size, size of the company; Tan, tangibility; Cur R, current ratio; ROA, return on assets; Int Cov, interest cover; Age, company’s age; CF; Cash flow; Tob Q, Tobin’s Q; MtB, market-to-book value; Econ G, economic growth; Asset G, total asset growth; ESG, ESG score; Lev, leverage; E Var, earnings variability; Loss, loss-making; Int rate, interest rate; Neg Eq, negative book equity; Asset Turn, inverse of asset turnover; NI, not included; NR, not reported.

After identifying the potential determinants of the COD from previous research, a brief investigation into the theoretical background of the relationship between the COD and each of the independent variables was conducted, addressing the second research question. Based on this, a hypothesis is formulated for each independent variable, with the null hypothesis that there is no relationship between the COD and the independent variable in question.

Size

A company’s size can be an indication of its risk level, as large businesses are assumed to be less risky than smaller entities. Petersen and Rajan (1994) explain that small companies tend to have a concentrated pool of lenders, whereas large companies have an extended pool of funders. This larger pool of funders provides more competitive pricing of debt finance. Large companies also have less return variability, resulting in lower risk, which could lead to lower required return by debt providers (Rutkowska-Ziarko 2015). Consequently, in line with traditional finance theory, size is expected to be negatively correlated with the COD. The following is hypothesised (H) in the current study:

H1: There is a negative relationship between a company’s size, measured as total assets, and its COD.

 Tangibility

John, Lynch and Puri (2003) hypothesise that certain assets, such as land and buildings, provide greater assurance to capital providers than others because their value is less susceptible to erosion. However, empirical evidence implies otherwise. Previous studies (John et al. 2003) have found that providing collateral for a specific financing arrangement is associated with higher COD. Researchers have theorised that this surprising result is because capital providers require riskier borrowers to provide collateral (Berger & Udell 1990; John et al. 2003). However, the result stands, even when controlling for other risk factors.

Khaw et al. (2019) and Tran (2022) investigated tangibility and also found a positive relationship between tangibility and the COD and assert that, despite this result being contrary to their expectation, the reason could be that firms with the capacity to borrow more would borrow more, which would result in higher risk.

In the studies analysed (Table 1), researchers expected a negative relationship between COD and tangibility, and most studies did find a statistically significant negative relationship. However, other empirical results have been leaning towards a positive relationship between COD and tangibility although the reasons for this are unclear. In line with the majority of studies, a negative correlation between COD and tangibility is hypothesised in the current study:

H2: There is a negative relationship between a company’s tangibility, measured as property, plant and equipment divided by total assets, and its COD.

 Current ratio

Following traditional finance theory, firms with more liquid assets available to service debt will reduce the risk for the debt provider. Lower risk typically results in a lower expected return (Carey et al. 2021; Medhioub & Boujelbene 2023). However, this relationship has not always been clear in the empirical literature, and there is a paucity of studies that explicitly investigate the relationship between the current ratio and the COD or even between the current ratio and the cost of capital.

In the context of financial distress and corporate failure prediction, Li (2023) investigated a non-linear relationship between liquidity ratios and corporate failure, identifying a statistically significant negative relationship between liquidity ratios and corporate failure at low current ratio levels. Although corporate failure and the COD are distinct concepts, these constructs are expected to be strongly related (Mansi, Maxwell & Zhang 2012). Consequently, a negative relationship between the COD and the current ratio is expected. The following is hypothesised in the current study:

H3: There is a negative relationship between a company’s current ratio, measured as current assets divided by current liabilities, and its COD.

 Return on assets

Santosuosso (2014) investigated the relationship between COD and profitability, measured through various proxies and argued that the reduction in COD due to higher profitability stems from corporate failure prediction literature. Lower profitability increases the probability of failure, which in turn increases COD. They also found a negative association between the COD and profitability.

It is therefore hypothesised that higher profitability indicates a lower risk of corporate failure and, consequently, a lower COD. Therefore, it is expected that the COD will have a negative association with profitability, proxied by ROA, measured as earnings before interest, tax, depreciation and amortisation (EBITDA) over total assets. The following is hypothesised in the current study:

H4: There is a negative relationship between a company’s return on assets, measured as earnings before interest, tax, depreciation and amortisation divided by total assets, and its COD.

 Interest cover

Ramsay and Sidhu (1998) posit that most lenders include interest cover limits as one of their loan covenant requirements. This implies that a higher interest cover provides greater negotiation power for the company seeking financing. Pandya (2016) investigated the impact of financial leverage on market value added and found that interest cover is the most significant predictor of market value added, compared with, for example, the debt-to-equity ratio or the debt ratio. Pandya (2016) also theorised that a higher interest cover would indicate that a company can more comfortably meet its debt obligations, thereby reducing debt providers’ risk and potentially lowering their required return. Interest cover is, therefore, expected to be a strong indicator of risk for debt providers, with required returns following suit. Therefore, the anticipated relationship between interest cover and the COD is negative. The following is hypothesised in the current study:

H5: There is a negative relationship between a company’s interest cover, measured as earnings before interest, tax, depreciation and amortisation divided by interest expense, and its COD.

 Age

Company age has been researched extensively in the context of risk (e.g. Fink et al. 2004), profitability (Pástor & Veronesi 2003) and corporate failure prediction (Kücher et al. 2020). Pástor and Veronesi (2003) found that younger companies have more volatile returns, whereas Loderer and Waelchli (2010) counter this claim by stating that companies’ average profitability declines with age. They theorise that older firms tend to stagnate due to older employees, older assets and ‘rent-seeking behaviour’ (i.e. employees and managers becoming more concerned with the distribution of wealth than with its creation), which stunts growth.

Coad et al. (2018) comment on the various effects at play during different life cycles and divide corporate life cycles into three stages. Initially, the firm is struggling against the ‘liability of newness’ – the company has no established routines, no experience and is in a fight for survival, as approximately half of all companies do not survive this stage. However, during this stage, the company benefits from high growth and energy. The second stage involves maturity in processes, relationships and employees, and change is more gradual. The final phase is when the company is fighting the ‘liabilities of ageing’, which includes an inclination to rigidity, lack of opportunity identification and a deficiency in keeping up with technological changes. Ideally, a company should strive to engage in strategic renewal to stave off the third stage and remain in the second stage indefinitely.

How do theories on company age relate to the COD? In a company’s initial phase, there is a high risk of failure. Consequently, debt providers might require a higher return. As companies age, innovation and profitability appear to dwindle, and debt providers might become cautious about the business’s sustainability. Given these diverse considerations, it is evident that the relationship between age and the COD would not be linear. Previous studies indicate that researchers focusing on COD have theorised that age is negatively correlated with the COD (Ni et al. 2022; Pittman & Fortin 2004). Consistent with these previous studies, a negative correlation with the COD is expected in the current study. The following is hypothesised in the current study:

H6: There is a negative relationship between a company’s age, measured as listing age, and its COD.

Cash flow from operations, scaled by total assets.

It would make financial sense that debt providers, more so than equity providers, would be interested in a company’s cash flow, considering that a debt provider has a right to repayment of pre-specified amounts at pre-specified dates. If there is no cash available to pay the debt provider, the company is in default. Equity providers would be more likely to focus on long-term growth, as they would benefit from capital appreciation.

The concept of the level of cash flow impacting debt providers’ risk has been investigated thoroughly in the corporate failure prediction literature. Beaver (1966) found that cash flow to total debt was the strongest indicator of impending failure. Consequently, it has been part of failure prediction models for decades, as evidenced by the works of Altman (1968), Altman and Sabato (2007) and Gupta et al. (2014), to name but a few.

In the context of research on the COD using traded debt indicators, the effect of cash flow levels on the COD is not evident; however, it has been found that cash flow variability is associated with larger credit spreads (Güntay & Hackbarth 2010).

Consequently, it is concluded that a negative association is expected between the COD and cash flow, based on previous research in the context of corporate failure prediction and prior studies on the COD, albeit a limited number of them. The following is hypothesised in the current study:

H7: There is a negative relationship between a company’s cash flow, measured as cash flow from operations divided by total assets, and its COD.

 Tobin’s Q and market-to-book value of equity

Originally named after James Tobin, Tobin’s equilibrium equation was developed as a macroeconomic approach to monetary theory (Tobin 1969). In their paper titled ‘The misuse of Tobin’s Q’, Bartlett and Partnoy (2020) argue that Tobin’s Q was never intended to be used on an individual company level and that the practice has become to use book values of debt, rather than market values, in the formula, which is a poor measure of value. Based on prior researchers’ expectations in regression models, the expected relationship between the COD and Tobin’s Q, or the COD and MV/BV equity, is negative. However, the theoretical foundation for its inclusion is weak. The following is hypothesised in the current study:

H8: There is a negative relationship between a company’s Tobin’s Q, measured as the sum of market capitalisation and the book value of long- and short-term debt divided by the book value of total assets, and its COD.

H9: There is a negative relationship between a company’s market-to-book value of equity, measured as the market-to-book value of equity, and its COD.

 Asset growth

Intuitively, having more assets in a company, ceteris paribus [all other things being equal], will provide more collateral for debt providers and thus reduce their risk, which in turn will lead to a lower required rate of return. Several researchers have investigated the relationship between asset growth and share performance. An increase in assets has been linked to a decline in performance; this relationship is pervasive and statistically significant (Cooper, Gulen & Schill 2008). There are two leading theories as to the driving force in this perceived anomaly. Firstly, the evidence of mispricing, as market participants overestimate the impact of growth (Cooper et al. 2008). Secondly, the evidence of the risk-return trade-off, as market participants would perceive higher growth signalling lower risk and, consequently, would require a lower return (Hou, Xue & Zhang 2015).

The phenomenon of higher asset growth being associated with lower share returns was investigated in the context of traded debt, and the same association was found, that is, when assets increase, bond returns decrease (Chen et al. 2021). These researchers utilised various models to conclude that, although some mispricing was evident, the predominant driver was the risk-return trade-off, and they asserted that debt providers require a lower return when assets increase.

Accordingly, the literature supports the hypothesis that there will be a negative association between the COD and asset growth. The following is hypothesised in the current study:

H10: There is a negative relationship between a company’s asset growth, measured as growth in total assets, and its COD.

 Economic growth

Fama and French (1989) found that poor economic conditions drive bond spreads up, and the opposite is true for strong economic conditions. They theorise two possibilities for this phenomenon: The first possibility is that poor economic conditions leave individuals with limited income. High investment returns divert funds from consumption and attract investors. The second possibility reverts to the risk–return concept – in poor economic conditions, the risk of low return is high, and investments are considered riskier, thereby increasing bond spreads. Gross domestic product growth is often used as a control for general economic conditions (Ni et al. 2022; Ozkaya 2018; Shailer & Wang 2015). Consequently, it is expected that the COD will have a negative association with GDP growth. The following is hypothesised in the current study:

H11: There is a negative relationship between economic growth, measured as GDP growth, and a company’s COD.

 Environmental, social and governance performance

A comprehensive systematic literature review by Bauer, Follert and Stöckl (2025) on the impact of ESG performance on COD provides a broad discussion of the theoretical basis for an expected negative relationship between ESG and the COD. They summarise that the argument for the relationship is two-fold. Firstly, from a risk-mitigation perspective, firms with stronger ESG practices are perceived as lower risk and, consequently, perceived as presenting a lower probability of default. Secondly, investor preferences may lead debt providers to accept lower returns from firms with stronger ESG performance. This theoretical expectation is further supported by their review of 64 empirical studies. Although the authors report mixed findings and identify the use of diverse COD proxies as a key challenge that complicates meaningful synthesis, the conclusion remains that the expected relationship between ESG performance and the COD is negative. The following is hypothesised in the current study:

H12: There is a negative relationship between a company’s ESG performance, measured as the Bloomberg-reported ESG score, and its COD.

 Leverage

Trade-off theory (Kraus & Litzenberger 1973) implies a trade-off between the advantage of incorporating cheap debt into the capital structure and the risk of financial distress. As this risk increases, the required return of all capital providers, both equity and debt, increases. An increase in debt in the capital structure would increase the COD. The theory that increased leverage increases the COD was empirically tested by Van Binsbergen, Graham and Yang (2010), who developed a mathematical function derived from total assets and other company characteristics, which measures the marginal benefit (or cost) of utilising debt. Consequently, a positive association between the COD and leverage is expected. The following is hypothesised in the current study:

H13: There is a positive relationship between a company’s leverage, measured as total debt divided by total assets, and its COD.

 Earnings variability

The capital asset pricing model (Lintner 1965; Mossin 1966; Sharpe 1964) utilised earnings volatility to calculate the risk premium, which determines capital providers’ required rate of return. Trueman and Titman (1988), through deductive reasoning, present a mathematical model demonstrating that smooth earnings impact value. Li and Richie (2016) used Trueman and Titman’s (1988) model and the traded bond market to empirically test the relationship between the COD and volatility in earnings and found a positive association. Consequently, a positive relationship is expected between earnings volatility and the COD. The following is hypothesised in the current study:

H14: There is a positive relationship between a company’s earnings variability, measured as the standard deviation of net income over a rolling prior three-year period, and its COD.

 Loss making

If ROA has already been included as an indicator of profitability, why would a firm’s reported loss be significant in determining the COD? In a seminal article, Burgstahler and Dichev (1997) found that managers are more likely to manipulate earnings when the objective is to avoid reporting a loss than when the objective is to avoid reporting a decrease in earnings. Burgstahler and Dichev (1997) mentioned the terms offered by loan providers as an example of an incentive for managers to manipulate earnings. They also argue that their findings correspond to the theory of Kahneman and Tversky (1979), who postulated that individuals perceive gains and losses differently. Specifically, Kahneman and Tversky (1979) theorised that individuals attach a higher value to moving from a loss to a gain than to moving from a gain to a higher gain. Consequently, a positive association is expected between the COD and a dummy variable to indicate whether a company is making a loss. The following is hypothesised in the current study:

H15: There is a positive relationship between a company being loss making, measured using an indicator variable for loss-making in the prior year, and its COD.

 Interest rates

An interest rate is a government-determined rate used by banks and central banks as a benchmark rate for floating rate instruments. In South Africa, most traded bonds are floating rate instruments (Naidoo, Nkuna & Steenkamp 2010), and it is expected that private placements will align with this despite a lack of empirical evidence. The most common reference rate for traded bonds in South Africa is the 3-month Johannesburg Interbank Agreed Rate (JIBAR) (Rushton 2022). Consequently, a positive association between the COD and interest rate is expected. The following is hypothesised in the current study:

H16: There is a positive relationship between the interest rate, measured as JIBAR, and a company’s COD.

 Negative book equity

If a company reports negative book equity, it theoretically means that its liabilities exceed its assets, and technically, the asset value has been corroded (Ang 2015). In academic research, firms with negative book equity have often simply been excluded from the sample because they are in financial distress (Brown, Lajbcygier & Li 2008) and, therefore, not considered representative of the population.

More in-depth investigations of the information content of negative book equity, however, do not necessarily support the conclusion that these firms are truly in financial distress. Ang (2015) found that many firms reporting excessive negative book equity are financially healthy, and the negative book equity is often a consequence of intentionally leveraging intangible assets that are inappropriately valued in the accounting records. They also indicate that numerous firms with consecutive years of reporting negative book equity continue to operate and do not display higher failure rates than the control firms.

Based on previous studies on the COD, it is expected that a dummy variable for negative book equity will be positively correlated with the COD although the theoretical foundation for this expectation is debatable. The following is hypothesised in the current study:

H17: There is a positive relationship between a company having negative book equity and its COD.

 Total asset turnover

The original Du Pont analysis examined efficiency (asset turnover) and profitability (net profit margin) to assess ROA maximisation (Liesz & Maranville 2008). This model was developed by an engineer named F. Donaldson Brown and used by the Du Pont company to analyse its financial interests. It became a dominant method of structured financial analysis until the 1970s (Liesz & Maranville 2008). Subsequently, researchers have continued to view asset turnover as a strong indicator of future performance (Bunea, Corbos & Popescu 2019).

Hu et al. (2011) consider asset turnover to be a proxy for financial performance and subsequently report a statistically significant negative relationship between the COD and asset turnover, using bank loan data. Wang, Zhou and Zhou (2013) justify the expected (and subsequently evidenced) negative relationship between credit default swap (CDS) spreads and asset turnover by arguing that firms with stronger future growth prospects tend to exhibit lower credit risk. Han, Kang and Shin (2016) support the inclusion of asset turnover in their evaluation of credit ratings, citing its inclusion in Moody’s credit rating methodology. They found a statistically significant negative relationship between asset turnover and the COD. Therefore, in the current research, a negative correlation between the COD and asset turnover is expected. The following is hypothesised in the current study:

H18: There is a negative relationship between a company’s total asset turnover, measured as revenue divided by total assets, and its COD.

 Research methods and design

To investigate the relationship between each independent variable and the COD, thereby addressing the third research question, companies listed on the JSE between 2010 and 2021 across all non-financial sectors were included in the study. Following previous studies on the COD, the capital structure of companies listed in the financial sector was considered inherently different and therefore not appropriate for this study. The sample was drawn by identifying all listed non-financial companies at each calendar year-end, which ensured that companies that delisted during the period were also included, reducing survivorship bias. Ultimately, a total of 2100 observations were obtained from the 229 companies in the unbalanced panel data. Data were extracted from Bloomberg.

Descriptive statistics

As noted by Pittman and Fortin (2004), the measurement instrument ADC is evidently noisy; yet, it is commonly used, as discussed in the literature review. One possible source for this measurement imprecision is that the numerator (interest expense) is measured throughout the year on the actual outstanding balances of various borrowings, whereas the denominator is measured at two points in time – the start and end of the year. This could potentially distort the measured value if the balance of borrowings changes significantly throughout the year. Following previous studies (Khaw et al. 2019; Pittman & Fortin 2004; Tran 2022), trimming was therefore applied at 5%.

After applying trimming, descriptive statistics are reported in Table 2.

Descriptive statistics.

Variables Mean Minimum 25th 50th 75th Maximum SD
Dependent variable: COD 0.10 0.02 0.07 0.09 0.12 0.32 0.05
Size 19000.80 2.22 892.59 3939.29 15813.50 794435.79 49419.99
Interest cover 34.84 −1742.68 3.44 6.95 14.31 10293.26 318.35
Tangibility 0.32 - 0.11 0.28 0.49 0.93 0.23
Earnings variability 857.77 0.09 23.20 93.11 370.04 72401.09 3949.49
Leverage 0.28 - 0.10 0.22 0.40 1.00 0.22
Current ratio 1.76 0.06 1.12 1.46 1.99 15.33 1.21
Listing age 24.21 - 8.00 17.00 30.00 126.00 22.53
ROA 0.07 −6.34 0.03 0.08 0.12 1.45 0.21
Tobin’s Q 1.10 0.05 0.56 0.81 1.30 45.14 1.34
MV:BV equity 2.01 −115.13 0.69 1.25 2.22 233.88 8.43
Cash flow 0.07 −5.28 0.03 0.07 0.13 0.97 0.20
ESG score 45.36 14.88 37.91 45.42 53.34 75.16 11.09
Asset growth 0.19 −0.95 0.02 0.06 0.16 84.39 2.23
Asset turnover 1.22 - 0.68 1.07 1.52 6.23 0.81
Economic growth 1.05 −17.50 0.50 1.30 2.40 19.10 n/a
Interest rate 0.06 0.04 0.05 0.06 0.07 0.11 n/a

COD, cost of debt; ESG, environmental, social and governance; MV/BV, market value to book value equity; ROA, return on assets; SD., standard deviation.

The descriptive statistics in Table 2 reveal that there is still notable variation in the measured COD, despite trimming. The descriptive statistics for interest rate and economic growth are reported over the period 2010–2021, not per observation.

Some notable observations from Table 2 are that size, reported in millions, indicates a median of R3939 million, but that companies with as little as R2.22m in reported assets are also included in the database. Tangibility reveals that the median percentage of property, plant and equipment (PPE) to total assets is 28%. Earnings variability indicates that the sample is heavily right-skewed, as demonstrated by the median of R93m, which is much lower than the average of R858m.

The average reported leverage is 28% (median of 22%). Companies included in the database have been listed for an average of 24.21 years, with a median of 17 years.

 Research method

Multiple regression was performed, using an F-test and a Hausman test to determine the most appropriate model. The results of the F-test and Hausman test are reported with the results of each model.

Potential model specification errors were considered as follows:

Autocorrelation was first addressed by selecting the most appropriate panel regression model. A Durbin–Watson test was performed for each regression to determine the extent of remaining autocorrelation. Unfortunately, the result was a test statistic outside the acceptable range of 1.5–2.5. Durbin–Watson test statistics are reported with each regression’s results. An additional potential method to address autocorrelation is to include a lagged dependent variable as an independent variable; the results are discussed later.

The lack of normality in errors was addressed by visual inspection of the normal probability plots (not presented). None of the normal probability plots indicated a problem with the normality of errors.

The Breusch–Pagan test for heteroscedasticity indicated that heteroscedasticity was significant, as evidenced by p-values of less than 0.01 for all the regressions performed. Consequently, the weighted least squares p-values are reported rather than the standard p-values for each regression.

Variance inflation factor (VIF) testing was performed to address multicollinearity. None of the VIF tests indicated multicollinearity in the regressions, as evidenced by VIF values below 5.2 in all cases.

 Ethical considerations

This article followed all ethical standards for research without direct contact with human or animal subjects.

 Results

Table 3 reports the regression results. Model [1] and Model [2] report the main regression results excluding and including ESG score, respectively. The reason for this differentiation is that ESG scores were not available for all the observations, and consequently, separate regressions were performed in which ESG performance was included and excluded to increase the number of observations for the other variables. Model [3], Model [4] and Model [5] report the results of including a lagged COD independent variable, a generalised method of moments (GMM) instrumental-variable specification and a robustness test, respectively, which is discussed in more detail later.

Main results.

Item D [1] [2] [3] [4] [5]
Sample size - 229 93 208 208 191
Observations - 2 100 933 1808 1808 1481
Test for fixed effects (F) – OLS versus fixed one-way (p-value) - 4.18 (< 0.01) 6.54 (< 0.01) 1.45 (< 0.01) N/Aa 1.6 (< 0.01)
Test for fixed effects (F) – fixed one-way vs fixed two-way (p-value) - 1.8 (0.06) 1.52 (0.13) 0.69 (0.72) N/Aa 1.46 (0.18)
Hausman-test – random one-way versus fixed one-way (p-value) - 32.02 (< 0.01) 40.22 (< 0.01) 129.74 (< 0.01) N/Aa 473.84 (< 0.01)
Preferred model - Fixed one-way Fixed one-way Fixed one-way Fixed one-way Random one-way
Breusch–Pagan test result (p < 0.01) - 181.35 100.84 243.05 56.98 201.44
Durbin–Watson test (p-value) - 1.4 (< 0.01) 1.25 (< 0.01) 1.96 (0.34) 1.96 (0.32) 1.94 (0.23)
R-Squared - 0.08 0.11 0.2 0.00 0.15
Adjusted R-squared - −0.04 −0.01 0.09 −0.14 0.01
Period - 2010–2021 2010–2021 2010–2021 2010–2021 2010–2019
Interest rate + 0.004* (1.7015) 0.002 (0.7475) 0.005** (2.3518) −0.017 (−0.5649) 0.003 (0.5473)
Listing age - 0.000 (0.1595) 0.001 (0.5047) 0.002* (1.8235) - 0.003 (1.3966)
GDP - -0.003** (−2.3441) −0.002 (−1.4236) −0.001 (−0.8803) −0.005 (−0.1152) 0.000 (0.0138)
Tangibility + −0.008 (−0.2246) −0.008 (−0.2219) −0.016 (−0.4663) −0.005 (−0.0065) 0.050 (1.5514)
Leverage + -0.119*** (−4.1418) -0.056** (−2.2443) -0.097*** (−4.1279) −0.541 (−0.5206) -0.113*** (−5.3428)
Loss making + 0.02*** (2.6698) 0.016** (2.1822) 0.019*** (2.6123) 0.464 (0.5944) 0.017** (2.1256)
Asset turnover - 0.028** (2.3011) 0.048* (1.7232) 0.02* (1.8997) 0.084 (0.6388) 0.014** (2.216)
log10 (size) - -0.035* (−1.8177) 0.019 (0.5687) −0.028 (−1.4409) −0.119 (−0.2237) -0.035** (−2.0864)
Interest cover - -0.000** (−2.1428) 0.000 (−0.332) -0.001** (−2.1502) −0.004 (−0.7965) -0.000*** (−4.0627)
log10 (earnings variability) + 0.006 (0.5622) −0.025 (−1.4769) 0.002 (0.2633) −0.177 (−0.6899) 0.001 (0.1836)
Current ratio - 0.000 (0.0826) 0.004 (0.7131) 0.002 (0.4288) 0.121 (0.6357) 0.003 (1.067)
ROA - 0.128*** (2.9703) 0.106*** (2.6504) 0.152*** (3.1234) −1.221 (−0.6161) 0.115*** (3.7815)
Tobin’s Q - -0.019*** (−2.7048) −0.015 (−1.6095) -0.018*** (−2.8564) 0.222 (0.1883) -0.016*** (−2.8716)
MV/BV equity - 0.001 (0.4009) 0.001 (0.2262) 0.000 (0.2183) 0.037 (0.0804) 0.000 (0.1025)
Cashflow - −0.057 (−1.4594) 0.018 (0.4483) −0.041 (−1.0023) 2.158 (0.4027) −0.048 (−1.5327)
Asset growth - 0.013 (1.1242) 0.020 (1.392) 0.014 (1.0346) −0.700 (−0.409) 0.004 (0.4394)
ESG score - - 0.000 (−0.3625) - 0.404 (1.1299) -
COD(t-1) + - - 0.355*** (6.3585) −0.017 (−0.5649) 0.308*** (11.4121)

Note: The non-standardised coefficients of the relationship between COD and each independent variable are reported first, followed by the t-statistic for each coefficient in brackets. Coefficients in bold indicate independent variables with a statistically significant relationship between the COD and the independent variable, either with a 1% (***), 5% (**) or 10% (*) level of significance. Negative equity was dropped in all models due to model over specification, with no relationship reported between the COD and negative equity. Listing age was dropped in Model [5] due to model over specification. The independent variables “Size” and “Earnings variability” were logged to normalise the distribution.

COD, cost of debt; ESG, environmental, social and governance; GDP, gross domestic product; MV/BV, market-to-book equity ratio; OLS, ordinary least squares; ROA, return on assets.

, The GMM estimation was performed using the dynamic panel-data procedure available in Statistica, which assumes a fixed one-way specification.

, level of significance = 10%;

, level of significance = 5%;

, level of significance = 1%.

As reported in Table 3, in Model [1], the interest rate, GDP, loss making, size, interest cover and Tobin’s Q have a statistically significant relationship with the COD in the direction as hypothesised. Leverage, asset turnover and ROA are statistically significantly correlated with COD in the opposite direction to what was hypothesised.

Comparing Model [1] and Model [2] many of the independent variables lost their statistical significance in Model [2]. Only four of the statistically significant independent variables in Model [1] are also statistically significant in Model [2]. This result could be attributed to the smaller sample size in Model [2], which resulted in reduced predictive ability. However, a higher R-squared (i.e. model fit) was achieved in Model [2] compared to Model [1], which may reflect that companies receiving ESG ratings are often larger and more established than firms without ESG ratings (Bikmetova & Pirinsky 2026). It is therefore plausible that the model fit is better for this smaller, but more homogeneous group of companies.

Considering the result of the Durbin–Watson test being outside the acceptable range of 1.5–2.5 for Model [1] and Model [2], autocorrelation needed to be addressed. Consequently, a regression with a lagged COD independent variable was performed (Model [3]). In this model, ESG was excluded, considering the limited number of companies with ESG ratings and the fact that ESG was not statistically significant in Model [2]. Model [3] revealed similar results to Model [1], with the following changes noted: Lagged COD is statistically significant, while listing age is statistically significant in the opposite direction to what was hypothesised. Gross domestic product and size are no longer statistically significant. The Durbin–Watson test for this regression was within the acceptable range.

Following previous studies (e.g. Medhioub & Boujelbene 2023), endogeneity is a concern in COD studies, as some independent variables, including interest cover and leverage, may be endogenous due to reverse causality, omitted variables or simultaneous determination with COD. In order to address this, Model [4] uses a GMM instrumental-variable specification, in which lagged values of the explanatory variables were used as instruments. Only the independent variables where a lagged variable would be meaningful were included as instrumental variables (all except interest rate, GDP and listing age). As reported in Table 3, all independent variables lost their significance in Model [4]. An untabulated result excluding the lagged COD (i.e. applying instrumental variables to Model [1]) delivered the same result, having no statistically significant independent variables.

The reported R2 and adjusted R2 statistics are very low for all models, with negative adjusted R2 statistics for Model [1], Model [2], and Model [4]. Although low R2 statistics are common in COD research (Guidara et al. 2014; Muttakin et al. 2020), the negative adjusted R2 should not be interpreted as indicating that the regressions are meaningless or incorrectly specified. Unlike the unadjusted R2, the adjusted R2 accounts for model complexity by penalising the inclusion of additional regressors. Consequently, where the explanatory contribution of the independent variables is modest relative to the number of variables included, the adjusted R2 may be negative. In the context of this study, this is not necessarily unexpected, given the inclusion of numerous firm-specific and macroeconomic controls.

Lastly, it is acknowledged that capital structures and the COD of companies could be divergent across different industries. However, industry fixed effects are not included separately in the fixed-effects specification, as firm fixed effects absorb all time-invariant firm characteristics, including industry membership, where firms do not change industry classification over the sample period. Disaggregation into industry sample sizes could address this; however, given the small sample sizes and the limited predictive power of the current models, this is noted as an opportunity for further research.

Robustness test

Two potential factors that could impact the findings were considered as follows:

Results reported for financial periods ending on or after 01 January 2020 could have been impacted by significant market reactions to the global pandemic resulting from the Coronavirus Disease 2019 (Baker et al. 2020).

The adoption of IFRS 16: Leases, which was issued on 01 January 2016 and became mandatory for financial reporting periods starting on or after 01 January 2019, could have significantly impacted several financial metrics, especially leverage, interest cover and total assets. The potential impact would differ from one industry to another (Morales-Díaz & Zamora-Ramírez 2018).

To address this potential concern, a robustness test was performed using only observations from 2019 and before (reported in Table 3 as Model [5]). Changes noted between Models [3] and Models [5] were that listing age and interest rate were no longer statistically significant, whereas size was statistically significant in Model [5].

 Discussion

The results indicated that many of the control variables commonly included when performing regression analysis on COD, measured as ADC, either:

delivered statistically significant results only in some models (interest rate, GDP, size, interest cover, Tobin’s Q, listing age);

were not statistically significantly related to the COD in any of the models (tangibility, earnings variability, current ratio, MV/BV equity, cashflow, asset growth, negative equity and ESG score); or

were statistically significantly related to the COD in the opposite direction to what was hypothesised by the existing literature (leverage, asset turnover, listing age and ROA).

The only independent variable that was statistically significantly related to the COD in the direction as hypothesised in four of the five model specifications was a dummy variable for loss-making entities. It is clear that our understanding or measurement of COD requires further investigation.

The strong statistical significance of the relationships between COD and leverage and COD and ROA, in the opposite direction to what was hypothesised, warranted further investigation.

Initially, bivariate regressions were performed between COD and leverage and between COD and ROA. Both leverage and ROA retained statistical significance in the opposite direction to what was hypothesised, indicating that the unexpected correlation was not caused by the inclusion of other independent variables.

In terms of the negative relationship between the COD and leverage, in which a positive relationship was predicted, a possible reason is that companies with low debt will not invest much time and effort in reducing their COD. They would also not be priority clients from a lender’s perspective. Companies with significant amounts of debt might have greater negotiation power and seek to reduce their COD, as it would represent a substantial expense to them.

It could thus be theorised that the relationship between the COD and leverage might not be linear – the risk of more debt might be acceptable to debt providers up to a certain point, beyond which they require higher returns. This is in line with the trade-off theory of Kraus and Litzenberger (1973), who theorised that there is a trade-off between the benefit of cheaper debt (compared to equity) and the risk of financial distress.

To test this theory, the sample was divided into a ‘low’ and ‘high’ leverage group (above and below the median) and bivariate regressions were performed. In addition, a regression with leverage and leverage squared was also performed. A brief overview of these results is reported in Table 4, which confirms that for low-leverage companies, leverage is negatively correlated with the COD, while the high-leverage companies report a positive relationship between COD and leverage, and when a quadratic leverage term is included, the coefficient for leverage is negative, but the coefficient for leverage squared is positive. All results exhibit statistical significance. These results confirm the U-shaped curve as theorised, and this improved understanding of the relationship between COD and leverage should be borne in mind by future researchers performing regression analysis on COD.

High and low leverage bivariate regression results.

Item Low leverage High leverage Leverage2
Sample size 129 142 230
Observations 937 970 2106
Preferred model Fixed one-way Fixed one-way Fixed one-way
Breusch–Pagan test result 103.81 0.46 180.35
Durbin–Watson test (p < 0.01) 1.67 1.6 1.42
R-squared 0.09 0.00 0.07
Adjusted R-squared −0.06 −0.17 −0.04
Period 2010–2021 2010–2021 2010–2021
Leverage coefficient (t-statistic) −0.791*** (−8.88) 0.057* (−1.768) −0.447*** (−11.46)
Leverage2 coefficient (t-statistic) - - 0.4426*** (9.56)

Note: The non-standardised coefficients of the relationship between COD and each independent variable are reported first, followed by the t-statistic for each coefficient in brackets. The non-standardized coefficients of the relationship between COD and each independent variable are reported first, followed by the t-statistic for each coefficient in brackets. Coefficients in bold indicate independent variables with a statistically significant relationship between the COD and the independent variable, either with a 1% (***) or 10% (*) level of significance.

, level of significane = 10%;

, level of significance = 1%.

Regarding ROA being positively correlated with COD rather than the hypothesised negative relationship, a potential explanation could be that higher ROA is associated with higher risk. Ceteris paribus, higher profitability should reduce the debt provider’s risk and, consequently, the required return. However, ensuring all other factors remain unchanged (ceteris paribus) is challenging. Thus, the question arises whether a higher ROA indicates a riskier investment, thereby increasing risk for the debt provider as well. Unlike leverage, there is no theoretical basis for a non-linear relationship. An improved understanding of the relationship between ROA and the COD is necessary to fully explain this outcome. However, this was beyond the scope of this study and is noted as an opportunity for future research.

Concerns regarding the measurement validity of average debt cost

The fourth research question, assessing whether there are indications that ADC has measurement validity concerns as a proxy for COD, was implicitly addressed through the assessment of prior literature and the regression analysis performed. There are several factors that could indicate a measurement validity problem:

Divergent results were reported by previous researchers, as evidenced in Table 1.

A few of the expected determinants of COD, according to traditional finance theory, have a statistically significant relationship with COD, as reported in Table 3.

Return on assets and asset turnover exhibit relationships in the opposite direction to those hypothesised, with ROA coefficients reported with high statistical significance.

None of the independent variables, including the lagged COD variable, were statistically significant in Model [4], which corrected for endogeneity using a dynamic panel GMM system.

Very low R2 statistics were reported for all models in Table 3.

Even though none of these factors on their own indicate validity concerns, taken together, they suggest that the measurement validity of COD using ADC requires further scrutiny.

 Limitations and opportunities for future research

A number of limitations and opportunities for further research are summarised:

Even though many previous studies using ADC were perused for potential determinants, not all potential determinants are publicly available, and a systematic literature review was not conducted to identify all potential determinants. A comprehensive literature review might reveal additional independent variables to be considered.

Return on assets and asset turnover were reported to have statistically significant positive relationships with the COD in four of the five regression models, when negative associations were hypothesised based on the existing literature. Listing age and interest cover also exhibited statistically significant relationships in the opposite direction to what was hypothesised in some models. These relationships should be investigated with greater rigour by considering non-linear relationships or different theories.

Endogeneity remains a challenge in COD research. In this study, a GMM dynamic panel model was used to address endogeneity; however, alternative methods using different instrumental variables could be considered.

Industry differences were not specifically investigated. However, considering the divergent results from the existing regression analyses, it is advisable to first address potential measurement validity and endogeneity concerns before making further inferences from smaller samples.

Environmental, social and corporate governance rating was found not to be statistically significantly related to COD, despite extensive evidence from previous studies, as evidenced by Bauer et al. (2025). Environmental, social and corporate governance ratings, however, were obtained from Bloomberg and were available for only 93 of the 229 companies included in the main regression. Obtaining ESG ratings from other, possibly more comprehensive, sources might impact the results specifically for ESG performance.

Addressing these limitations was beyond the scope of the current research and should be considered in future research.

 Conclusion

The study was conducted to reassess commonly used control variables in research on the COD of JSE-listed companies, considering the divergent results of previous studies. South Africa presents a unique context in which traded debt is illiquid, but the reliability of financial information is robust. In addition, previous studies in the South African context have often been unsuccessful in including control variables, which are correlated with the COD.

This study identified commonly included control variables used in regression analysis for the COD, based on the existing literature, addressing the first research question. The potential determinants were identified from previous South African and international studies, focusing on ADC as a proxy for the COD, as it is the most viable proxy in this context (considering that the debt market in South Africa is illiquid). In total, 18 such control variables were identified (size, interest cover, tangibility, earnings variability, leverage, current ratio, listing age, ROA, Tobin’s Q, MV/BV equity, cash flow, ESG score, asset growth, asset turnover, economic growth, interest rate, negative equity and a dummy variable for loss-making entities). However, the findings indicate that the control variables differ substantially between studies, and that the results for these variables are divergent. Often, control variables are not statistically significant in their relationship with the COD, and in some instances, control variables were even found to be statistically significant in the opposite direction to what was hypothesised (size, tangibility, current ratio, listing age, cash flow, MV/BV equity and leverage). The summary of previous control variables (Table 1) contributes to the existing literature by providing a comprehensive overview of control variables that should be considered in studies on COD.

The second research question, ‘What are the theoretical bases for commonly included control variables’ relationships with the COD?’, was answered by a detailed discussion on each independent variable, which resulted in the hypothesis statements. The theoretical bases for the relationships between the COD and each independent variable should be carefully contemplated by future researchers before including a control variable solely because it was previously included.

Addressing the third research question, the study used a large sample over a long period to identify independent variables that have a statistically significant relationship with the COD and should therefore be considered in regression analysis when working with the COD.

The findings of the regression analysis echo the divergent results found by previous researchers as presented in the literature review. In four of the five regression models, only one of the 18 potential determinants considered, namely loss-making entities, was statistically significantly associated with COD in the hypothesised direction. Four independent variables (ROA, leverage, listing age and asset turnover) were statistically significantly correlated with COD in the opposite direction as hypothesised. When endogeneity was addressed using a GMM dynamic system panel regression, none of the independent variables were statistically significantly correlated with the COD. A non-linear relationship between leverage and COD was investigated, and strong evidence was presented that leverage’s correlation with COD follows a U-curve – being negative at low leverage levels, and positive at high leverage levels.

The last research question, ‘Are there indications of ADC having measurement validity concerns?’, was addressed implicitly by the work performed for the first three research questions. The divergent results reported in this study, together with those reported by previous researchers, suggest that the validity of the measurement instrument ADC warrants greater scrutiny.

Potential opportunities for further research include quantifying the measurement instrument’s validity through a secondary proxy or perhaps finding other sources of information, such as company disclosures or lender information. Alternatively, since the quantitative approach generally followed to date has delivered such unsatisfactory results, a more qualitative approach should be considered to improve our understanding of the definition and potential determinants of the COD.

 Acknowledgements

This article is based on research originally conducted as part of Anet Boshoff-Knoetze’s master’s thesis titled ‘Determinants of the cost of debt for JSE-listed companies’, submitted to the Department of Business Management, Stellenbosch University in 2024. The thesis is currently unpublished and not publicly available. The thesis was supervised by Pierre D. Erasmus and George F. Nel. The manuscript has been revised and adapted for journal publication. The author confirms that the content has not been previously published or disseminated and complies with ethical standards for original publication.

Competing interests

The authors declare that they have no financial or personal relationships that may have inappropriately influenced them in writing this article. The authors, George F. Nel and Pierre D. Erasmus, serve as editorial board members of this journal. The peer review process for this submission was handled independently, and the authors had no involvement in the editorial decision-making process for this article. The authors have no other competing interests to declare.

 CRediT authorship contribution

Anet Boshoff-Knoetze: Conceptualisation, Data curation, Formal analysis, Investigation, Methodology, Project administration, Writing – original draft, Writing – review and editing. George F. Nel: Conceptualisation, Supervision, Validation, Writing – review and editing. Pierre D. Erasmus: Conceptualisation, Supervision, Validation, Writing – review and editing. All authors reviewed the article, contributed to the discussion of results, approved the final version for submission and publication and take responsibility for the integrity of its findings.

 Data availability

The data that support the findings of this study are not openly available and are available from the corresponding author, Anet Boshoff-Knoetze, upon reasonable request.

 Disclaimer

The views and opinions expressed in this article are those of the authors and are the product of professional research. They do not necessarily reflect the official policy or position of any affiliated institution, funder, agency or the publisher. The authors are responsible for the article’s results, findings and content.

 References Ahmed, J.I., Anderson, C. & Zarutskie, R.E., 2015, ‘Are the borrowing costs of large financial firms unusual?’, FEDS Working Paper No. 2015-024, viewed from https://www.federalreserve.gov/econres/feds/are-the-borrowing-costs-of-large-financial-firms-unusual.htm. Ahwireng-Obeng, A.S. & Ahwireng-Obeng, F., 2022, ‘Corporate bond issuance behaviour in African emerging markets’, Journal of African Business 23(1), 126–145. https://doi.org/10.1080/15228916.2020.1826853 Altman, E.I., 1968, ‘Financial ratios, discriminant analysis and the prediction of corporate bankruptcy’, The Journal of Finance 23(4), 589–609. https://doi.org/10.1111/j.1540-6261.1968.tb00843.x Altman, E.I. & Sabato, G., 2007, ‘Modelling credit risk for SMEs: Evidence from the U.S. market’, Abacus 43(3), 332–357. https://doi.org/10.1111/j.1467-6281.2007.00234.x Ang, T.C.C., 2015, ‘Are firms with negative book equity in financial distress?’, Review of Pacific Basin Financial Markets and Policies 18(3), 1550016. https://doi.org/10.1142/S0219091515500162 Baker, S.R., Bloom, N., Davis, S.J., Kost, K., Sammon, M. & Viratyosin, T., 2020, ‘The unprecedented stock market reaction to COVID-19’, The Review of Asset Pricing Studies 10(4), 742–758. https://doi.org/10.1093/rapstu/raaa008 Bartlett, R. & Partnoy, F., 2020, ‘The misuse of Tobin’s Q’, Vanderbilt Law Review 73(2), 353–424, viewed 10 June 2026, from https://scholarship.law.vanderbilt.edu/vlr/vol73/iss2/1. Bauer, L., Follert, F. & Stöckl, S., 2025, ‘The impact of environmental performance and disclosure on the cost of debt: A systematic literature review and a conceptual model’, Management Review Quarterly. https://doi.org/10.1007/s11301-025-00541-5 Beaver, W.H., 1966, ‘Financial ratios as predictors of failure’, Journal of Accounting Research 4, 71–111. https://doi.org/10.2307/2490171 Berger, A.N. & Udell, G.F., 1990, ‘Collateral, loan quality, and bank risk’, Journal of Monetary Economics 25(1), 21–42. https://doi.org/10.1016/0304-3932(90)90042-3 Bharath, S.T., Sunder, J. & Sunder, S.V., 2008, ‘Accounting quality and debt contracting’, The Accounting Review 83(1), 1–28. https://doi.org/10.2308/accr.2008.83.1.1 Bikmetova, N. & Pirinsky, C.A., 2026, ‘Do ESG rating agencies improve ESG performance?’, Journal of Business Ethics 204(2), 335–365. https://doi.org/10.1007/s10551-025-06063-0 Blackwell, D.W., Noland, T.R. & Winters, D.B., 1998, ‘The value of auditor assurance: Evidence from loan pricing’, Journal of Accounting Research 36(1), 57–70. https://doi.org/10.2307/2491320 Brealey, R.A., Myers, S.C. & Allen, F., 2017, Principles of corporate finance, 12th edn., McGraw-Hill Education, New York, NY. Brigham, E.F. & Ehrhardt, M.C., 2020, Financial management: Theory and practice, 19th edn., Cengage Learning, Boston, MA. Brown, S., Lajbcygier, P. & Li, B., 2008, ‘Going negative: What to do with negative book equity stocks’, The Journal of Portfolio Management 35(1), 95–102. https://doi.org/10.2139/ssrn.1142649 Bunea, O.I., Corbos, R.A. & Popescu, R.I., 2019, ‘Influence of some financial indicators on return on equity ratio in the Romanian energy sector – A competitive approach using a DuPont-based analysis’, Energy 189, 116251. https://doi.org/10.1016/j.energy.2019.116251 Burgstahler, D. & Dichev, I., 1997, ‘Earnings management to avoid earnings decreases and losses’, Journal of Accounting and Economics 24(1), 99–126. https://doi.org/10.1016/S0165-4101(97)00017-7 Carey, P., Khan, A., Mihret, D.G. & Muttakin, M.B., 2021, ‘Voluntary sustainability assurance, capital constraint and cost of debt: International evidence’, International Journal of Auditing 25(2), 351–372. https://doi.org/10.1111/ijau.12223 Chen, F., Li, Y., Wu, W. & Yu, T., 2021, ‘Asset growth anomaly of corporate bonds: A decomposition analysis’, The Review of Asset Pricing Studies 16(1), 50–94. https://doi.org/10.1093/rapstu/raaf007 Coad, A., Holm, J.R., Krafft, J. & Quatraro, F., 2018, ‘Firm age and performance’, Journal of Evolutionary Economics 28(1), 1–11. https://doi.org/10.1007/s00191-017-0532-6 Cooper, M.J., Gulen, H. & Schill, M.J., 2008, ‘Asset growth and the cross-section of stock returns’, The Journal of Finance 63(4), 1609–1651. https://doi.org/10.1111/j.1540-6261.2008.01370.x Dirman, A., 2020, ‘Cost of debt: The impact of financial factors and non-financial factors’, Dinasti International Journal of Economics, Finance and Accounting 1(4), 358–372. https://doi.org/10.38035/dijefa.v1i4.527 Ertugrul, M., Lei, J., Qiu, J. & Wan, C., 2017, ‘Annual report readability, tone ambiguity, and the cost of borrowing’, Journal of Financial and Quantitative Analysis 52(2), 811–836. https://doi.org/10.1017/S0022109017000187 Fama, E.F. & French, K.R., 1989, ‘Business conditions and expected returns on stocks and bonds’, Journal of Financial Economics 25(1), 23–49. https://doi.org/10.1016/0304-405X(89)90095-0 Fink, J., Fink, K.E., Grullon, G. & Weston, J.P., 2004, ‘Firm age and fluctuations in idiosyncratic risk’, SSRN Electronic Journal, 891173. https://doi.org/10.2139/ssrn.891173 Firer, C., Ross, S.A., Westerfield, R.W. & Jordan, B.D., 2012, Fundamentals of corporate finance, 5th South African edn., McGraw-Hill Education UK Ltd, Maidenhead. Francis, J., LaFond, R., Olsson, P. & Schipper, K., 2005, ‘The market pricing of accruals quality’, Journal of Accounting and Economics 39(2), 295–327. https://doi.org/10.1016/j.jacceco.2004.06.003 Graham, J.R., Li, S. & Qiu, J., 2008, ‘Corporate misreporting and bank loan contracting’ Journal of Financial Economics 89(1), 44–61. https://doi.org10.1016/j.jfineco.2007.08.005 Guidara, A., Khlif, H. & Jarboui, A., 2014, ‘Voluntary and timely disclosure and the cost of debt: South African evidence’, Meditari Accountancy Research 22(2), 149–164. https://doi.org/10.1108/MEDAR-09-2013-0042 Güntay, L. & Hackbarth, D., 2010, ‘Corporate bond credit spreads and forecast dispersion’, Journal of Banking & Finance 34(10), 2328–2345. https://doi.org/10.1016/j.jbankfin.2010.02.019 Gupta, J., Wilson, N., Gregoriou, A. & Healy, J., 2014, ‘The value of operating cash flow in modelling credit risk for SMEs’, Applied Financial Economics 24(9), 649–660. https://doi.org/10.1080/09603107.2014.896979 Han, S.-H., Kang, K. & Shin, Y.S., 2016, ‘Bond ratings, corporate governance, and cost of debt: The case of Korea’, Journal of Asian Finance 3(3), 5–15. https://doi.org/10.13106/jafeb.2016.vol3.no3.5 Hou, K., Xue, C. & Zhang, L., 2015, ‘Digesting anomalies: An investment approach’, Review of Financial Studies 28(3), 650–705. https://doi.org/10.1093/rfs/hhu068 Hu, Y., Li, S., Lin, T.W. & Xie, S., 2011, ‘Large creditors and corporate governance: The case of Chinese banks’, Review of Accounting and Finance 10(4), 332–367. https://doi.org/10.1108/14757701111185326 Huguet, D. & Gandía, J.L., 2014, ‘Cost of debt capital and audit in Spanish SMEs’, Spanish Journal of Finance 43(3), 266–289. https://doi.org/10.1080/02102412.2014.942154 John, K., Lynch, A.W. & Puri, M., 2003, ‘Credit ratings, collateral, and loan characteristics: Implications for yield’, The Journal of Business 76(3), 371–409. https://doi.org/10.1086/375252 Johnson, R., 2020, ‘The link between environmental, social and corporate governance disclosure and the cost of capital in South Africa’, Journal of Economic and Financial Sciences 13(1), 1–12. https://doi.org/10.4102/jef.v13i1.543 Kahneman, D. & Tversky, A., 1979, ‘Prospect theory: An analysis of decision under risk’, Econometrica 47(2), 263–292. https://doi.org/10.2307/1914185 Khaw, K.L.H., Zainudin, R. & Rashid, R.M., 2019, ‘Cost of debt financing: Does political connection matter?’, Emerging Markets Review 41, 100632. https://doi.org/10.1016/j.ememar.2019.100632 Kovermann, J.H., 2018, ‘Tax avoidance, tax risk and the cost of debt in a bank-dominated economy’, Managerial Auditing Journal 33(8/9), 683–699. https://doi.org/10.1108/MAJ-12-2017-1734 Kraus, A. & Litzenberger, R.H., 1973, ‘A state-preference model of optimal financial leverage’, The Journal of Finance 28(4), 911–922. https://doi.org/10.2307/2978343 Kücher, A., Mayr, S., Mitter, C., Duller, C. & Feldbauer-Durstmüller, B., 2020, ‘Firm age dynamics and causes of corporate bankruptcy: Age dependent explanations for business failure’, Review of Managerial Science 14(3), 633–661. https://doi.org/10.1007/s11846-018-0303-2 La Rosa, F., Liberatore, G., Mazzi, F. & Terzani, S., 2018, ‘The impact of corporate social performance on the cost of debt and access to debt financing for listed European non-financial firms’, European Management Journal 36(4), 519–529. https://doi.org/10.1016/j.emj.2017.09.007 Li, K., 2023, ‘Liquidity ratios and corporate failures’, Accounting & Finance 64(1), 1111–1134. https://doi.org/10.1111/acfi.13174 Li, M., 2021, ‘Uses and abuses of statistical control variables: Ruling out or creating alternative explanations?’, Journal of Business Research 126, 472–488. https://doi.org/10.1016/j.jbusres.2020.12.037 Li, S. & Richie, N., 2016, ‘Income smoothing and the cost of debt’, China Journal of Accounting Research 9(3), 175–190. https://doi.org/10.1016/j.cjar.2016.03.001 Liesz, T.J. & Maranville, S.J., 2008, ‘Ratio analysis featuring the Dupont method: An overlooked topic in the finance module of small business management and entrepreneurship courses,’ Small Business Institute Journal 1(1), 17–34, viewed 10 June 2026, from https://sbij.scholasticahq.com/article/26283-ratio-analysis-featuring-the-dupont-method-an-overlooked-topic-in-the-finance-module-of-small-business-management-and-entrepreneurship-courses. Lim, S.C., Mann, S.C. & Mihov, V.T., 2017, ‘Do operating leases expand credit capacity? Evidence from borrowing costs and credit ratings’, Journal of Corporate Finance 42, 100–114. https://doi.org/10.1016/j.jcorpfin.2016.10.015 Lintner, J., 1965, ‘Security prices, risk, and maximal gains from diversification’, The Journal of Finance 20(4), 587–615. https://doi.org/10.2307/2977249 Loderer, C. & Waelchli, U., 2010, ‘Firm age and performance’, Working Paper 26450, University of Bern, Bern. Lorca, C., Sánchez-Ballesta, J.P. & García-Meca, E., 2011, ‘Board effectiveness and cost of debt’, Journal of Business Ethics 100(4), 613–631. https://doi.org/10.1007/s10551-010-0699-3 Magnanelli, B.S. & Izzo, M.F., 2017, ‘Corporate social performance and cost of debt: The relationship’, Social Responsibility Journal 13(2), 250–265. https://doi.org/10.1108/SRJ-06-2016-0103 Mansi, S., Maxwell, W.F. & Zhang, A., 2012, ‘Bankruptcy prediction models and the cost of debt’, The Journal of Fixed Income 21(4), 25–42. https://doi.org/10.3905/jfi.2012.21.4.025 Medhioub, N. & Boujelbene, S., 2023, ‘Tax avoidance and cost of debt: Does integrated report assurance matter?’, Journal of Accounting in Emerging Economies 14(1), 75–98. https://doi.org/10.1108/JAEE-04-2022-0114 Morales-Díaz, J. & Zamora-Ramírez, C., 2018, ‘The impact of IFRS 16 on key financial ratios: A new methodological approach’, Accounting in Europe 15(1), 105–133. https://doi.org/10.1080/17449480.2018.1433307 Mossin, J., 1966, ‘Equilibrium in a capital asset market’, Econometrica 34(4), 768–783. https://doi.org/10.2307/1910098 Muttakin, M.B., Mihret, D., Lemma, T.T. & Khan, A., 2020, ‘Integrated reporting, financial reporting quality and cost of debt’, International Journal of Accounting and Information Management 28(3), 517–534. https://doi.org/10.1108/IJAIM-10-2019-0124 Naidoo, E., Nkuna, M. & Steenkamp, D., 2010, Developments in debt issuance costs of South African banks, WP/20/10, Economic Research and Statistics Department, South African Reserve Bank, Pretoria. Nel, G.F., Smit, E. & Brümmer, L.M., 2019, ‘The impact of internet investor relations on the cost of capital: Evidence from companies listed on the Johannesburg Stock Exchange’, Australian Accounting Review 29(1), 36–48. https://doi.org/10.1111/auar.12216 Ni, X., Jin, Q. & Huang, K., 2022, ‘Environmental regulation and the cost of debt: Evidence from the carbon emission trading system pilot in China’, Finance Research Letters 49, 103134. https://doi.org/10.1016/j.frl.2022.103134 Nikolaev, V. & Van Lent, L., 2005, ‘The endogeneity bias in the relation between cost-of-debt capital and corporate disclosure policy’, European Accounting Review 14(4), 677–724. https://doi.org/10.1080/09638180500204624 Ozkaya, H., 2018, ‘Effect of mandatory IFRS adoption on cost of debt in Turkey’, Business and Economics Research Journal 9(3), 579–588. https://doi.org/10.20409/berj.2018.124 Pandya, B., 2016, ‘Impact of financial leverage on market value added: Empirical evidence from India’, Journal of Entrepreneurship, Business and Economics 4(2), 40–58, viewed from http://scientificia.com/index.php/JEBE/article/view/42. Pástor, Ľ. & Veronesi, P., 2003, ‘Stock valuation and learning about profitability’, The Journal of Finance 58(5), 1749–1789. https://doi.org/10.1111/1540-6261.00587 Petersen, M.A. & Rajan, R.G., 1994, ‘The benefits of lending relationships: Evidence from small business data’, The Journal of Finance 49(1), 3–37. https://doi.org/10.1111/j.1540-6261.1994.tb04418.x Pittman, J.A. & Fortin, S., 2004, ‘Auditor choice and the cost of debt capital for newly public firms’, Journal of Accounting and Economics 37(1), 113–136. https://doi.org/10.1016/j.jacceco.2003.06.005 Radier, G., Majoni, A., Njanike, K. & Kwaramba, M., 2016, ‘Determinants of bond yield spread changes in South Africa’, African Review of Economics and Finance 8(2), 50–81. https://doi.org/10.2139/ssrn.2742963 Ramsay, I. & Sidhu, B.K., 1998, ‘Accounting and non-accounting based information in the market for debt: Evidence from Australian private debt contracts’, Accounting & Finance 38(2), 197–221. https://doi.org/10.1111/1467-629X.00010 Rushton, K., 2022, RMB SA credit monthly April 2022, RMB Markets Research, Sandton. Rutkowska-Ziarko, A., 2015, ‘The influence of profitability ratios and company size on profitability and investment risk in the capital market’, Folia Oeconomica Stetinensia 15(1), 151–161. https://doi.org/10.1515/foli-2015-0025 Sánchez-Ballesta, J.P. & García-Meca, E., 2010, ‘Ownership structure and the cost of debt’, European Accounting Review 20(2), 389–416. https://doi.org/10.1080/09638180903487834 Santosuosso, P., 2014, ‘Cost of debt and corporate profitability’, International Business Research 7(2), 13–21. https://doi.org/10.5539/ibr.v7n2p13 Schwab, K., 2019, The global competitiveness report 2019, viewed 25 July 2024, from http://www3.weforum.org/docs/GCR2017-2018/05FullReport/TheGlobalCompetitivenessReport2017–2018.pdf. Sengupta, P., 1998, ‘Corporate disclosure quality and the cost of debt’, The Accounting Review 73(4), 459–474. https://doi.org/10.2308/TAR-1325290 Shailer, G. & Wang, K., 2015, ‘Government ownership and the cost of debt for Chinese listed corporations’, Emerging Markets Review 22, 1–17. https://doi.org/10.1016/j.ememar.2014.11.002 Sharpe, W.F., 1964, ‘Capital asset prices: A theory of market equilibrium under conditions of risk’, The Journal of Finance 19(3), 425–442. https://doi.org/10.1111/j.1540-6261.1964.tb02865.x Stanišić, N., Stefanović, N. & Radojević, T., 2016, ‘Determinants of the cost of debt in the Republic of Serbia’, Teme XL(2), 869–882, viewed 10 June 2026, from https://www.ceeol.com/search/article-detail?id=441041. Stats SA, 2026, Debt in the formal business sector, viewed 06 May 2026, from https://www.statssa.gov.za/?p=19228. Tobin, J., 1969, ‘A general equilibrium approach to monetary theory’, Journal of Money, Credit and Banking 1(1), 15–29. https://doi.org/10.2307/1991374 Tran, Q.T., 2022, ‘Foreign ownership and cost of debt financing: Evidence from an emerging market’, International Journal of Emerging Markets 17(9), 2278–2289. https://doi.org/10.1108/IJOEM-09-2019-0750 Trueman, B. & Titman, S., 1988, ‘An explanation for accounting income smoothing’, Journal of Accounting Research 26, 127–139. https://doi.org/10.2307/2491184 Van Binsbergen, J.H., Graham, J.R. & Yang, J., 2010, ‘The cost of debt’, The Journal of Finance 65(6), 2089–2136. https://doi.org/10.1111/j.1540-6261.2010.01611.x Wang, H., Zhou, H. & Zhou, Y., 2013, ‘Credit default swap spreads and variance risk premia’, Journal of Banking & Finance 37(10), 3733–3746. https://doi.org/10.1016/j.jbankfin.2013.02.021

How to cite this article: Boshoff-Knoetze, A., Nel, G.F. & Erasmus, P.D., 2026, ‘The cost of debt in South Africa: A reassessment of commonly used control variables’, Journal of Economic and Financial Sciences 19(1), a1094. https://doi.org/10.4102/jef.v19i1.1094