Several studies, including those of Chang and Dasgupta (2009), Devos et al. (2017), Elsas and Florysiak (2011), Getzmann et al. (2014), Oino and Ukaegbu (2015), Qian et al. (2009) and Zhou et al. (2016), have identified firm size, profitability (PROF), asset tangibility (TAN), nondebt tax shields (NDTs) and growth opportunities as reliable firm-specific determinants of corporate leverage. The studies have used these as control variables for the estimation of firms’ SOAs. Smith et al. (2020) argued that firms can be classified as value and growth firms. Value firms are those firms whose value derives from their assets in place, while growth firms derive their value from their future profitable growth options.

Barclay and Smith (2020), Smith et al. (2020) and Titman and Wessels (1988) argued that value firms tend to be large, mature, financially flexible, profitable and well-diversified firms that generate excess free cash flows. These firms also have high stocks of tangibles, high credit ratings and low stocks of profitable growth options, and they are less financially constrained. These characteristics of value firms increase their debt capacities while reducing their overall financing costs (Dang 2013; Frank & Goyal 2009). On the other hand, growth firms tend to be small, young, less-profitable and less-diversified firms which generate limited cash flows. These firms have low stocks of tangibles, low credit ratings, low debt capacities and high stocks of profitable growth options, and they tend to be financially constrained (Dang et al. 2012; Smith et al. 2020).

The dynamic trade-off theory argues that debt interest tax shields are more valuable to value firms while NDTs in the form of capital allowances are more valuable to growth firms (Bessler, Drobetz & Kazemieh 2011). The theory argues that the reduced financial distress costs, increased taxable earnings and higher agency costs of free cash flows all push value firms to use more debt than growth firms. (Bessler et al. 2011). Growth firms, on the other hand, will rely more on equity financing to reduce their agency costs of underinvestment. This financing behaviour of both value and growth firms implies that value firms are highly leveraged, whereas growth firms are less leveraged. Thus, based on the trade-off hypothesis, firm leverage is positively related to firm PROF, size and TAN while it is negatively related to firm growth rate and nondebt interest tax shields.

In support of the dynamic trade-off theory, Chang and Dasgupta (2009), Dang et al. (2012), Dang (2013), Elsas and Florysiak (2011), Flannery and Hankins (2013), Moyo (2015), Oino and Ukaegbu (2015), Qian et al. (2009), Ramjee and Gwatidzo (2012) and Zhou et al. (2016) found a positive relationship between leverage and TAN and firm size. Consistent with the predictions of the dynamic trade-off theory, Chang and Dasgupta (2009), Dang (2013), Dang et al. (2012), Elsas and Florysiak (2011), Flannery and Hankins (2013), Flannery and Rangan (2006) and Qian et al. (2009) document a negative correlation between leverage and firm growth rate and NDTs. However, Chang and Dasgupta (2009), Dang et al. (2012), Dang (2013), Elsas and Florysiak (2011), Flannery and Hankins (2013), Moyo (2015), Oino and Ukaegbu (2015), Qian et al. (2009), Ramjee and Gwatidzo (2012) and Zhou et al. (2016) found a negative relationship between leverage and PROF and a positive relationship between leverage and TAN and firm size, which rejects the hypothesis of the dynamic trade-off theory.

The 2007–2008 GFC was a negative economic shock that affected firms across the globe. Campello et al. (2010) contend that this financial crisis financially constrained several firms across the globe as it affected their financing choices and hence their capital structures. This impact was severe on firms that were already financially-constrained prior to the crisis. According to Barclay and Smith (2020), Hadlock and Pierce (2010), Smith et al. (2020) and Titman and Wessels (1988), value firms tend to be large and financially-unconstrained, while growth firms are normally small and financially-constrained. Given their high credit ratings and lower external financing costs, value firms face lower target adjustment costs compared to growth firms. This implies that value firms should rebalance their capital structures more frequently than the growth firms.

This study therefore expects that, prior to, during and after the crisis, financially unconstrained firms will exhibit a persistently higher SOA than financially-constrained firms. Secondly, all firms’ SOAs will be at their lowest during the financial crisis, as they were all financially constrained. Thirdly, the effects of the financial constraints will encourage firms to always keep their target-deviation spreads to the minimum. This means frequent adjustments of capital structures; thus, the SOA after the GFC is expected to be higher than that of the precrisis period.

The partial adjustment model was initially conceptualised by Taggart (1977) and was later extended by Auerbach (1985), Fama and French (2002), Fischer et al. (1989), Flannery and Rangan (2006) and Jalilvand and Harris (1984). Following these studies, the single-step capital structure partial adjustment model is generally stated as follows: LEVi,t+1=(1−λ)LEVi,t+λ(Xi,tγ)+εi,t+1+εi,t+1[Eqn 1]

Where LEV_i,t+1 is the firm’s next year’s leverage ratio, measured either by the market-to-debt ratio (MDR) or the book-to-debt ratio (BDR), λ represents the firm’s SOA towards the optimal target leverage, where X_i,t represents a vector of firm-specific, γ represents a coefficient vector, i is the firm-fixed effect variable and ε_i,t+1 is an error term.

Model 1 assumes the validity of the dynamic trade-off theory, and hence the firm’s SOA can be estimated. This assumption can only hold if some of the elements of the coefficient vector are greater than zero, that is, γ ≠ 0. This means that λ must lie between 0 and 1, that is, 0 < λ < 1, in order to validate the dynamic trade-off theory. In situations where λ = 1, it implies that the firm has no target leverage to adjust towards, and hence its SOA is zero. Such a result invalidates the dynamic trade-off theory as a valid theory in explaining the financing decisions of firms. A result where λ = 1 means that the firm has an optimal leverage ratio which it never deviates from as it continuously adjusts its capital structure (Elsas & Florysiak 2015). A result where 0 < λ > 1 would result in a negative SOA, which means that the firm over-eliminates its target-deviation spread (Qian et al. 2009).

The present study used both the MDR and BDR to estimate firms’ SOA as previous studies such as those of Elsas and Florysiak (2015) and Moyo (2015) have shown that SOA depends on the dependent variable used. In line with the previous studies of Drobetz and Wanzenried (2006), Elsas and Florysiak (2011) and Frank and Goyal (2009), the study used PROF, TAN, size (SIZE), market-to-book (growth rate) (MTB) and NDTs as the vector of variables in model 1. The final models that were used in the current study were obtained by expanding model 1 and are specified as follows: MDRi,t+1=(1−λ)MDRi,t+β1PROFit+β2TANit+β3SIZEit  +β4MTBit+β5NDTSit+εi,t+1[Eqn 2]

The variables are defined as follows:

The study used both models to test the following hypotheses of the study:

The study’s population was made up of 208 JSE-listed nonfinancial firms that were listed during the period 2001–2018. In this study, all financial services firms were excluded, as their financing is regulated and supervised by the South African Reserve Bank. This regulation limits, for instance, banks’ flexibility to alter their capital, as they may breach their minimum capital requirements. A purposive sampling strategy was to construct the final sample of the study. The sample included only nonfinancial firms that were continuously listed for at least 16 years or more in the period 2001–2018. The sample excluded all nonfinancial firms that had missing data for two or more consecutive years. The final sample was made up of 104 nonfinancial firms, and the total number of firm years was 1872.

All the data that were used in the study were extracted from the firms’ financial statements, which were ethically obtained from the IRESS Research Domain database. The IRESS Research Domain database has all the published financial statements of all JSE-listed firms from 1972 to date. The collected data were used to construct a panel dataset, with the resulting panel being unbalanced, as some firms had missing observations. According to Baltagi (2009), the main benefits of using panel data are that they allow for the control of individual heterogeneity, and they result in reduced collinearity and more degrees of freedom, leading to increased estimation efficiency. Panel data are also compatible with several efficient modern estimators, such as difference generalised method of moments (GMM), the system GMM, the censored random effects tobit estimator (RE tobit) and the least squares dummy variable (LSDVC) estimators.

Following Almeida and Campello (2007), Denis and Sibilkov (2010) and Fazzari, Hubbard and Petersen (1988), the study used firm size (size was measured by the natural logarithm of total assets) to categorise the firms in the sample as either financially constrained or financially unconstrained. According to Hadlock and Pierce (2010) and Hovakimian and Hovakimian (2009), firm size is a good predictor of the level of financial constraints. The financially constrained firms were those whose average size was lower than the sample’s median size of 15.4026.

To enable the investigation of the effects of the GFC on the SOA of the nonfinancial firms listed on the JSE, the full sample (Panel 1, [2001–2018]) was then broken down into three subsamples: Panel 2: pre-financial crisis period (2001–2006); Panel 3: during financial crisis period (2007–2012); and Panel 4: post-financial crisis period (2013–2018). The full sample and subsamples were further subdivided into financially constrained and financially unconstrained firms to investigate if the SOA varied between these two categories of firms in these three subsample periods.

The study used Stata 15 to generate the full sample’s summary statistics and to fit the two partial adjustment models using the full sample and subsamples panel datasets described earlier. This econometric data analysis software supports all the leading dynamic panel estimators that include the GMM, system GMM, the censored RE Tobit estimator and the LSDVC. Using Stata 15, all the variables were winsorised at the 1st and 99th percentiles to remove outliers. Variance inflation factors (VIF) were used to detect multicollinearity between the variables. All the variables’ VIFs were less than 10; hence, no variables were dropped from the model.

The partial adjustment model is by nature an autoregressive-distributed lag model that is characterised by autocorrelation and heterogeneity among the individuals. The dynamic RE tobit maximum likelihood estimator was to fit models 2 and 3 using the full sample and subsamples panel datasets. The RE tobit estimator is equivalent to Elsas and Florysiak (2011) and Elsas and Florysiak (2015)’s double-censored dynamic panel data with a fractional dependent variable (DPF) estimator. According to Elsas and Florysiak (2015), the DPF estimator yields consistent and unbiased estimates when used to fit autoregressive-distributed lag models using unbalanced panel data that has a fractional dependent variable. This estimator is thus ideal for capital structure research as the partial adjustment model uses debt ratios which are fractional in nature and lie between 0 and 1. The estimator is implemented with the Gauss-Hermite quadrature integration option. The SOAs were calculated using marginal effects.

Table 2 contains the descriptive statistics for the study’s full sample.

The final full sample was made up of 104 nonfinancial firms listed on the JSE during the period 2001–2018. Only firms that were consecutively listed for at least 16 years during this period were included in the sample. The total observations for the unbalanced panel data set varied between 1658 and 1819.

All the variables were winsorised at the 1st and 99th percentiles to eliminate the outlier observations and most extreme recorded data.

From Table 2, firms’ mean MDR and BDR are 0.2400 and 0.1954, respectively. The South African nonfinancial firms are less leveraged than both the US and Nigerian firms whose MDR are 0.2412 and 0.629, respectively (Oino & Ukaegbu 2015; Zhou et al. 2016). Table 2 also shows that the sample’s maximum MDR is 0.6004, which is much lower than 0.9980 for Nigerian firms (Oino & Ukaegbu 2015). This implies that South African firms use debt conservatively and thus rely heavily on equity finance. This conservative use of debt may partly explain why few South African firms failed during the 2007–2008 GFC. The sample’s mean firm PROF and SIZE ratios are 0.1287 and 15.2647, respectively. Table 2 also shows that the mean asset TAN is 0.2923, while the mean firm MTB is 1.1182. This means that on average, South African nonfinancial firms are smaller, hold lower stocks of tangibles, have lower MTB and have lower PROF ratios than both Nigerian and US firms (Oino & Ukaegbu 2015; Zhou et al. 2016). The variations in both the MDR and BDR of firms between the years 2001 and 2018 are shown in Figure 1.

MDR, winsorised fraction 0.1: is the firms’ market-to debt ratio, and BDR, winsorised fraction 0.1: is the firms’ BDR. All the ratios are winsorised at the 1st and 99th percentiles to eliminate the outlier observations and most extreme recorded data.

As shown in Figure 1, firms’ average MDRs sharply declined from about 0.2953 in 2002 to 0.1728 in 2005 and then increased slightly to 0.1813 in 2006, just before the onset of the GFC in 2007. The BDRs also steadily declined from 0.2092 in 2002 to 0.1790 in 2005, after which they increased slightly to 0.1915 in 2006.

Figure 1 also shows that during the GFC, both average debt ratios increased sharply to 0.2463 for the MDR and 0.2180 for the BDR, respectively. This increase was to be expected as Campello et al. (2010) and Ivashina and Scharfstein (2010) stated that during the financial crisis, borrowers drew heavily from their bank credit lines in fear of future restricted access to the credit lines. The low returns on the stock market exacerbated the debt situation, and this led to the sharp increase in the MDRs of firms. Based on Figure 1, it is clear that both debt ratios continued to increase after the GFC peaking-off in 2009 at 0.2715 for the MDR and at 0.2251 for the BDR. These increases signalled the after-effects of the GFC. Thereafter, the ratios declined steadily until 2011, after which they began to increase again. The upward trend in both debt ratios has continued until 2018, although there was a slight drop in 2017. This increase can be attributed to the country’s current macroeconomic problems, which include low stock market returns, slow economic growth, the recent credit downgrade and the depreciation of the rand against the major currencies. The shapes of the two graphs differ in some years because the two measures are different, as they are calculated differently. The BDR is calculated using book values of debt and assets, while the MDR is calculated using the book value of debt and the market value of equity. Thus, the MDR is more volatile than the BDR because of the volatility of the market value of equity that is included in its calculation (Fama & French 2002). The MDRs and BDRs of both financially constrained and financially unconstrained firms are shown in Figure 2. As shown in Figure 2, financially constrained firms have higher MDRs than unconstrained firms, and the debt ratios follow the same pattern as those of the full sample shown in Figure 1.

MDR_CO, winsorised 0.1, is the market-to debt ratio of constrained firms; MDR_UN, winsorised 0.1, is the market-to debt ratio of unconstrained firms; BDR_CO, winsorised 0.1, is the book-to debt ratio of constrained firms, and BDR_UN, winsorised 0.1, is the book-to debt ratio of unconstrained firms. All the ratios are winsorised at the 1st and 99th percentiles to eliminate the outlier observations and most extreme recorded data.

The study’s main empirical results are contained in Table 3, Table 4 and Table 5. The Wald χ² model fit statistics confirms that all the models are properly fitted. The focus of the study is the impact of the 2007–2008 GFC on firms’ SOA towards the optimal target capital structure, and therefore the correlation results are only discussed in brief.