Financial and corporate social performance in the UK listed firms: the relevance of non-linearity and lag effects

Abstract

Using environmental, social and governance scores compiled by Reuters Datastream for each company’s corporate social performance (CSP), we examine the relationship between CSP and corporate financial performance (CFP) of 314 UK listed companies over the period 2002–2015. We further evaluate the relationship between prior and subsequent CFP and prior and subsequent CSP. Based on the system-GMM estimation method, we provide direct evidence that suggests that while CFP and CSP can be linked linearly; however, when we examine the impact of CSP on CFP, the association is more non-linear (cubic) than linear. Our results suggest that firms periodically adjust their level of commitment to society, in order to meet their target CSP. The primary contributions of this paper are testing (1) the non-monotonous relationship between CSP and CFP, (2) the lagged relationship between the two and the optimality of CSP levels, and (3) the presence of a virtuous circle. Our results further suggest that CSP contributes to CFP better during post-crisis years. Our findings are robust to year-on-year changes in CFP and CSP, financial versus non-financial firms, and the intensity of corporate social responsibility (CSR) engagement across industries.

1 Introduction

The extant literature on corporate social responsibility (CSR, hereafter) presents a broader view of the firm as an entity that should consider its relationships with stakeholders and not just with shareholders. Freeman (1984) notes that if stakeholders are able to voice their concerns, the socially responsible behaviour of a firm may minimise externalities and maximize synergies in their relationships with stakeholders. As a result of this pressure from stakeholders, including government and the public, several firms now report on their ethical, social and environmental conduct. Firms are beginning to take ‘green’ issues very seriously and are minimizing their negative impact on the environment (see Martín-de Castro et al. 2015).¹ This, coupled with pressure on firms from various stakeholders (see Fieseler 2011) to invest more in socially responsible projects, highlights the need to investigate whether a relationship exists between corporate social performance (CSP, hereafter) and corporate financial performance (CFP, hereafter), given that the latter remains an important goal for the firm. If a positive association is found to exist between CSP and CFP, it will increase the need for corporations to commit more resources to improve CSR. However, if a negative relationship exists between CSP and CFP, corporations will be less receptive to calls from stakeholders to invest in socially responsible activities. Therefore, this paper examines whether CSP has any varying impact on CFP by also considering the direction of causation, and the non-linear and dynamic associations.

Does it really pay for a company to be green? This is a very important question, yet unresolved issue, despite previous scholarly attention (see Wang et al. 2015). For over three decades, academics have empirically investigated the potential link between CSP and CFP (see Cochran and Wood 1984; Aupperle et al. 1985; Stanwick and Stanwick 1998; Barnett and Salomon 2006, 2012; Galbreath 2016; Wu et al. 2017). For instance, Tosun (2017) finds that at fund-level, socially responsible firms underperform the market when there is more investment in high CSR firms. However, Filbeck et al. (2013) show that CSR constructed portfolios experience better performance. The arguments for and against corporate social initiatives have motivated researchers to examine the CSP–CFP in different countries, beyond the USA, where studies on the subject have been traditionally contextualized (Ullmann 1985). Also, a vast body of literature has examined the CSP–CFP relationship in different business sectors and in different countries. For instance, Simpson and Kohers (2002) investigated the CSP–CFP relationship in the banking industry; Gregory and Whittaker (2007) also analyzed the CSP of ‘ethical’ unit trusts in the UK, while in more recent times Li et al. (2013) examined whether firms’ performance affects CSR disclosure in China.

Despite the very useful contributions of these papers, a general consensus has not yet been reached. Friedman (1970) argues that CSR actions incur costs with no returns. Preston and O’Bannon (1997) suggest that a negative or neutral relationship exists between CSP and CFP, and that a link exists between past CFP and resulting CSP. There seems to be a general disagreement in the literature on the question of whether CSP adds value to firm performance. Academics in recent times have taken a contrasting view to suggest that firms do have other responsibilities than maximising shareholders profits (see Flammer 2015 and Wang et al. 2015). Thus, they point out that firms can benefit financially from investing in CSR projects which can be demonstrated to their stakeholders through effective communication, for example, CSR reporting. Indeed, proponents of a positive CSP–CFP link suggest that the term ‘socially responsible’ does not necessarily mean that firms have to reduce profits when adopting CSR policies. They argue that firms with better CFP can meet their social responsibilities, and the greater their profits, the greater their ability to be socially responsible (Donaldson and Preston 1995; Freeman 1984). This argument is generally anchored on the slack resources theory, which predicts that information intensity regarding ethical and moral issues influence consumers’ brand attitude and buying intentions (Schuler and Cording 2006). This may imply that companies improve their CSP to sustain future sales. Brammer and Millington (2008), Nelling and Webb (2009) and Scholtens (2008b) have also examined the causality of any link between CSR and CFP. Simpson and Kohers (2002) and Cochran and Wood (1984) argue that whilst the CSP–CFP link is ambiguous and difficult to measure, both companies and stakeholders would benefit from a better understanding of this relationship. In fact, if a relationship between CSP and CFP could be found, clarification surrounding the exact purpose, nature, role and responsibilities of a firm may be reached. Given that this is an unresolved issue in the literature, this study seeks to empirically address this important issue.

In doing this, we specifically differentiate our study from the existing studies with regard to data and methods: first, the study utilizes a robust and well-established index, ESG (environmental, social and governance) scores, provided by Thomson Reuters, which takes account of the multidimensional aspects of CSP. We are not aware of any other published work that has used this data in the CSP/CFP context. Furthermore, we use the system-GMM estimation technique, which accounts for the endogeneity problem as a result of the random shocks influencing both CSP and CFP and their determinants simultaneously, using recent panel data of 314 UK firms for the period 2002–2015.² GMM also controls for unobserved firm heterogeneity, and provides the short-term and long-term relationship between the two factors. This technique also enables us to address the presence of a possible ‘virtuous circle’, which suggests that a higher CSP leads to higher CFP via the strategic use of CSR, and vice versa (Waddock and Graves 1997). Moreover, Short et al. (2015) argue that CSP—being a major aspect of CFP—has been under studied empirically. We thus aim to fill the gap in the literature as we study the determinants of CSP.

The uniqueness and contributions of this paper therefore come from two main aspects. First, relying on perspectives from the slack resources theory and the optimality of the CSP literature, we empirically examine the presence of non-linear (parabolic and cubic) relationships between CSP and CFP. This is an important, yet unresolved issue in the literature (Lankoski 2008; Elsayed and Paton 2009; Barnea and Rubin 2010; Barnett and Salomon 2006, 2012). Second, we introduce a partial adjustment process to investigate the possibility of companies adjusting the intensity of CSR activities for internal and external developments. This is important, given the large and growing CSP–CFP literature that refers to such a possibility as ‘social responsiveness’ (McWilliams and Siegel 2001; Brammer and Millington 2008). Through our methodology, we consider this possibility in an estimation model by adopting the partial adjustment process. To the best of our knowledge, no empirical study has examined this aspect of the CSP–CFP nexus.

Our study also contributes to the CSP and CFP literature in five unique ways. First, we hypothesize, and find direct evidence consistent with the linear relation between CFP and CSP. Second, by incorporating two strands of literature on finance and CSR, we find a non-linear (cubic) association between CSP and CFP, suggesting that firms regularly adjust their levels of commitment to society to meet their CSR targets. Third, we find that the speed of a firm’s adjustment to the targeted CSP level is higher in the non-financial UK firms. Fourth, we split our sample into two sub-periods, pre-crisis (2002–2008) and post-crisis (2009–2015) to analyse whether any potential CSR practice change stemming from the global financial crisis influences the CSP and CFP link: our additional analyses reveal that the positive impact of CSP on CFP is more salient during the post-crisis years and for both periods we report the existence of optimal CSP levels and non-linear association between CSP and CFP.³ Finally, we clarify the implications of wide variations in the degree of CSP intensity across industries. We find evidence supporting the notion that firms in industries with low CSR engagement find it beneficial to increase their current CSP so as to enhance current CFP.

The remainder of the paper is structured as follows. Section 2 reviews relevant literature and hypothesis development. Section 3 describes the data. Section 4 presents the empirical analysis and results. Section 5 concludes the paper and offer suggestions for future studies.

2 Literature review and hypothesis development

CSR has become a global phenomenon, which continues to shape and influence discourse, policies and practices (Scherer and Palazzo 2011; Amaeshi et al. 2016). Existing studies offer several definitions of CSR, which leaves the construct ambiguous (Henderson 2001; Windsor 2001; van Marrewijk 2003). Summarising prior studies, we argue that CSR-responsive firms will advance social good (McWilliams and Siegel 2001), use legal and ethical means to earn profit (Carroll, 1991), minimise adverse environmental and social impact (European Commission 2011; Amaeshi et al. 2016). This broad understanding of CSR enables firms to balance the needs of stakeholders. Thus the issue of commitment to financially rewarding shareholders and notions of equity and fairness to other stakeholders (Adegbite and Nakajima 2011; Deakin and Whittaker 2007) may offer financial benefits to firms. It is from this CSR understanding that we explore the relation between CSP and CFP.

2.1 The non-linearity between CSP and CFP

Do firms that are socially responsible experience better financial performance (a positive association) relative to their competitors who are non-responsive (a negative association)? Robinson et al. (2011) find a significant increase in the market share of firms that are added to the Dow Jones Sustainability Index. A large body of literature on CSP and CFP mainly assumes a causal relationship between CFP and CSP (see McWilliams and Siegel 2000 and Waddock and Graves 1997). However, the association may be non-linear. For instance, CFP can improve with higher CSP up to a certain point, and then deteriorates as a result of the diminishing benefits of excessive commitment to CSR resulting in a reverse U-shaped relationship (Barnea and Rubin 2010). Another viewpoint is that it may be irrational for companies to engage in CSR as they may have to sacrifice financial resources earmarked for projects with positive net present values to pursue CSR, whilst their competitors, who argue that it does not pay to engage in CSR, will have sufficient funds to undertake projects with positive net present values. Existing studies show a U-shaped relation (see Barnett and Salomon 2006; Brammer and Millington 2008). Furthermore, Lankoski’s (2008) model demonstrates an inverted-U relationship between CSR outcomes and economic performance, such that as the marginal costs of CSR activities increase, the marginal revenues decline.⁴ Porter’s (1980) and Porter and Kramer’s (2002) competitive advantage arguments suggest that corporations following differentiation or low-cost policies are more likely to perform better than their counterparts. This implies a cubic link for firms with moderate (low or high) CSP would have lower (higher) CFP. A recent study by Barnett and Salomon (2012) highlights the importance of considering a U-shaped link between CSP and CFP since some firms may not adequately generate positive returns, despite having significant investment in CSR.

We anchor our paper on the slack resources theory, which implies that prior high levels of CFP may allow managers a greater amount of slack resources to invest in CSR activities (Ullmann 1985; Waddock and Graves 1997). As CSP depends to some extent on a manager’s individual discretion, the initiation or cancellation of environmental policies may depend greatly on the amount of resources available to managers (McGuire et al. 1988). Waddock and Graves (1997) argue that firms with better CFP history tend to have higher current levels of CSR, and that raising CSR levels in turn results in stronger CFP. This is termed a ‘virtuous circle’. Orlitzky (1998) concluded that better CSP is both a predictor and the consequence of a stronger CFP, which is consistent with Waddock and Graves’ (1997) findings of a virtuous circle.

A firm with a strong CSP may implement implicit contracts which may improve CFP and reduce variability in performance measures. However, if CSR is viewed as a considerable cost, firms with strong past CFP may be more willing to incur these costs in the future. Conversely, firms with poor past CFP may be less willing to incur these costs in the future. This time lagged analysis is consistent with Waddock and Graves (1997) who used time lags to test between prior and subsequent CSP and prior and subsequent CFP. Jo et al. (2015) report that the reducing environmental costs take about 2 years before they improve profitability. Furthermore, adopting both CSP and CFP interchangeably as dependent and explanatory variables is consistent with Scholtens (2008b) who examined the causality between the CSP–CFP nexus and showed that the direction of causation runs from CFP to CSP.

The combination of the two hypotheses implies that CFP_t−2 improves CSP_t−1, and subsequently enhances CFP_t. Examining these interrelationships help to test for the presence of a virtuous circle. This extends the work of Shahzad and Sharfman (2015), who considered a lagged analysis by showing that the direction of causality runs from CSP to CFP.

2.2 The optimality of CSP

Firms aim to maximise CFP but not necessarily CSP. Guo et al. (2016) find that corporate culture disclosure improves financial performance. On the other hand, a firm’s innovation strategy could influence its CSP performance by reducing environmental impacts and improving health and safety (Pavelin and Porter 2008). As the literature on the optimality of CSP implies (see e.g., Barnea and Rubin 2010; Elsayed and Paton 2009; Fernandez-Kranz and Santalo 2010; Salzmann 2008), there are costs and benefits of being socially responsible in a competitive business environment. Brammer et al. (2006) argue that investment in social activities destroy shareholders’ wealth. Nevertheless, Scholtens (2008a)—using an alternative framework for assessing CSR—shows that CSP within the banking sector improved significantly between 2000 and 2005. A recent study by Nollet et al. (2016) finds a positive association between CSP and CFP after investment up to a certain threshold has been met. McWilliams and Siegel (2001) propose that managers trade-off the demand for CSR against the cost of CSR activities. They suggest that an optimal CSR level can be identified. Lankoski (2000) argues that firms that deviate from optimal CSP level may experience a lower CFP. Salzmann (2008) finds this relationship intuitively appealing given that excessively improving the CSP (for example, aiming for carbon neutrality) is extremely costly and would certainly reduce a firm’s CFP. This may explain why various empirical studies have failed to find either a positive or negative association between CSP and CFP. Lankoski (2008) argues that some exogenous factors (e.g., technology, definition of stakeholderism) may evolve over time and thereby change the CSR-related costs. Wang and Choi (2013) emphasize the relevance of consistency in social performance over time when examining the CSP–CFP link. Therefore, as discussed in Aupperle et al. (1985) and Ullmann (1985), it may be feasible for firms to optimize the intensity of their CSR by trading-off the benefits against the costs. Gregory and Whittaker (2007) examine the performance of ethical unit trusts in the UK and report that the findings are sensitive to whether static or time-varying models are adopted. Therefore, to empirically test for the presence of a possible optimal CSR level, we adopt the following partial adjustment process: we assume that a company $i$ has a desired level of CSR for time t ($CSP_{it}^{*}$), which is determined by x explanatory variables.

$$CSP_{it}^{*} = \mathop \sum \limits_{k = 1} \delta_{k} x_{kit} + \pi_{it}$$

(1)

where x is a vector of k explanatory variables; π_it is a serially correlated disturbance term with a mean of zero and possibly heteroscedastic; and δ_k’s are unknown estimable parameters. The model assumes that companies adjust their current CSR structure (CSP_it) according to the degree of adjustment coefficient ‘α’, to obtain the target CSR structure:

$$CSP_{it} - CSP_{it - 1} =\upalpha\left( {CSP_{it}^{*} - CSP_{it - 1} } \right)$$

(2)

The actual change will be equal to the desired change when α = 1. No adjustments are made in the case of α = 0, suggesting that either the lagged level is the target level, or the adjustment cost is higher than the cost of remaining off target. By combining (1) and (2) we obtain:

$$CSP_{it} = \left( {1 - \alpha } \right)CSP_{it - 1} + \mathop \sum \limits_{k = 1} \alpha \delta_{k} x_{kit} + \alpha \pi_{it}$$

(3)

Equation (3) assumes that α lies between 0 and 1. If the cost of being in disequilibrium is higher (lower) than the cost of adjustment, then α converges to one (zero).⁵ We examine the presence of adjusting the level of CSR activities to achieve the target CSP level. Clearly, drawing on evidence of CFP and a time-varying degree of CSR, we test the following three hypotheses.

H1

CSP is positively associated with CFP.

H2

Prior CFP (CSP) will have a positive impact on subsequent CSP (CFP).

H3

Firms dynamically adjust the level of CSR activities to maintain their target CSP.

3 Data

We use an unbalanced panel data of 314 UK firms over the period 2002-2015. All company financials and share price data were collected from Thomson Reuters Datastream⁶ and CSP data were collected from Thomson Reuters “Asset4” module. After the standard data filtering (e.g., deleting firms with missing data, as well as inconsistent and extreme values of variables), we restricted our sample size to 314 companies with 3240 firm-years between them. Given our adopted adjustment process, this large sample size helps us to provide a robust analysis.

3.1 Corporate social performance

The measures used in prior related empirical studies have frequently been one-dimensional, lacked clarity and have been applied to small samples of companies. As highlighted by Sheehy (2015) and Siegel and Vitaliano (2007), among others, there is a clear need for a multidimensional measure applied across a wide range of industries and larger samples. An overall measure of CSP is extremely difficult due to its complexity and because just one CSP measurement provides a limited perspective on how well a firm is socially performing.

We construct our CSP measure by utilizing detailed social performance data from Thomson Reuters Asset4 in Datastream. We measure CSP for each firm-year by using seven equally weighted dimensions which consist of employment quality, health and safety, training and development, diversity, human rights, community, and product responsibility. The variables are normalized on a scale of 100. Therefore, CSP is between 0% (lower commitment to CSR activities) and 100% (higher commitment to CSR activities). Moreover, the emphasis of our study is on social performance of companies and hence we did not include ESG’s corporate governance components (i.e., board structure, board function, compensation policy, shareholder rights and vision and strategy) for our CSP construct, which are available from Thomson Reuters Datastream currently for over 4300 global companies. Similarly, Liang and Renneboog (2017) used only the environmental (E) component of the ESG scores when their empirical focus is corporate environmental responsibility.⁷

3.2 Corporate financial performance

Studies using accounting-based measures for CFP have generally found a positive relationship between CSP and CFP (see Cochran and Wood 1984). Studies such as these are influenced by performance measurement types, as each type focuses on different aspects with their own biases. Another limitation is that they do not control for differences in risk. Ullmann (1985) argues that accounting-based measures should be adjusted for risk and industry characteristics. However, other studies have used market-based performance measures to examine the relationship between CFP and CSP which reflect investors’ perceptions of firms’ ability to generate future profits rather than using past CFP (see Ullmann 1985). Market-based CFP measures are less likely to be affected by differences in accounting procedures and managerial manipulation. Our study, therefore, uses two accounting-based measures (return on assets—ROA and return on equity—ROE). We also use a market-based measure (share price performance—SPP) to ensure that CSP is not sensitive to a particular performance measure. ROA is operating income over total assets; ROE is net income over common equity; and SPP is the annual change in adjusted share prices.

3.3 Control variables

We use several variables mainly drawn from prior literature, which have been shown to have effects on CFP as well as CSP. First, as in Pava and Krausz (1996) and Waddock and Graves (1997), we control for both firm size and risk effects. A company size does have a significant effect on both CSP and CFP. For instance, a firm’s CSR activities depend on its size, diversification level, consumer income, labour market conditions, stage in the industry life cycle, country of origin, and country of operation (see Adegbite and Nakajima 2011; McWilliams and Siegel 2001). We therefore use the natural logarithm of total assets (SIZE) to control for firm size.⁸

Porter and Kramer (2002) argue that firms with a well-directed CSR strategy have a better chance of surviving hard times. This is because firms require strong relationships with employees, suppliers and customers, which can be effectively managed through stakeholder relationships. Therefore, as both leverage and firm risk can induce ‘hard times’, they should be related to CSP.

We use company beta figures based on the capital asset pricing model (CAPM) theory (BETA) to control for time-varying firm risk relative to market risk. These figures capture market risk and depict the relationship between individual stock return volatility and market return volatility. We compute the annual beta for each firm using rolling time series regressions of excess company returns with respect to the FTSE All-Share Index returns using monthly data for the past 5 years. Low levels of CSR may result in greater exposure to financial risk as investors may believe that firms with less CSR are more risky due to the perception that the management of those firms possesses poor skills (Alexander and Buchholz 1978). Investors will demand high returns from firms that show less commitment to CSR as they believe lack of CSR may result in increased financial risk as a result of heavy fines and lawsuits. Also, low debt implies that firms can meet their obligations relatively easily.

McWilliams and Siegel (2000) argue that a generic model in the literature is inadequate as the direction of the relationship keeps changing when new variables, such as research and development (RD), investment and industry advertising intensity, are included. They note that past studies had generated spurious results due to model misspecification. Therefore, following Surroca et al. (2010), we include in our model RD, measured as the research and development expenses divided by total sales. If a company is highly leveraged, it would be under great pressure to meet its loan repayments from creditors and ensure satisfactory economic performance, which might lower CSR. We calculate leverage (LEVER) as total debt divided by total assets. We use current ratio (CUR) to capture a firm’s liquidity and short-term financial strength; we calculate CUR as current assets divided by current liabilities. We use GROWTH to control for current growth rate; we calculate GROWTH as the percentage change on the previous year’s sales. Following existing literature, we employ market-to-book ratio (MBR) to capture future growth rate; we compute MBR as the ratio of ‘total assets plus market value of equity less book value of equity’ to total assets. Short et al. (2015) show that CSP is associated with industry characteristics. Therefore, to account for this, we employ nine industry dummies (INDUSTRY) based on the classification provided in Table 13.⁹ Furthermore, the literature related to the managerial viewpoint contends that the financial decisions of corporations are largely influenced by their managers’ preferences, desires and objectives. For instance, Barnea and Rubin (2010) argue that entrenched managerial ownership reduces incentives to allocate substantial financial resources to CSR expenditure. Following Sun et al. (2016), amongst others, we use managerial ownership (MANOW) to capture managerial entrenchment.

4 Empirical analyses

4.1 Model specifications

We adopt the system-GMM model for our regression analysis. This is as a result of the dynamic nature of the model which accounts for other unobservable factors. Nelling and Webb (2009) found that CSP is determined more by firm-specific factors than by CFP. This necessitates the need to control for the effects of certain fixed factors, such as capital intensity and managerial reputation, among others. Furthermore, as it is difficult to maintain exogeneity in firm-level data, the direction of causation between variables could be problematic because of the endogeneity issue, i.e., the correlation between regressors and the error term. Therefore, using contemporaneous data for CFP or CSP and their determinants may generate spurious results. Thus, to account for these econometric problems, we use the system-GMM specification (see e.g., Duanmu and Guney 2013; Wintoki et al. 2012). Below, we illustrate our models without considering non-linearity and lagged analysis.

$$LNCSP_{i,t} = \alpha + \beta_{1} CFP_{i,t} + \varSigma\upbeta_{\text{s}} CONTROLS_{s, i,t} + \varSigma\upbeta_{\text{j}} INDUSTRY_{j} + \varSigma\upbeta_{\text{k}} TIME_{k} +\upmu_{i} +\upmu_{t} + \varepsilon_{i,t}$$

(4)

$$CFP_{i,t} =\uptau + \gamma_{1} LNCSP_{i,t} + \varSigma \gamma_{\text{s}} CONTROLS_{s, i,t} + \varSigma \gamma_{\text{j}} INDUSTRY_{j} + \varSigma \gamma_{\text{k}} TIME_{k} +\upmu_{i} +\upmu_{t} + \varPsi_{i,t}$$

(5)

where LNCSP is the logarithmic transformation of the social performance (CSP) measure; CFP is corporate financial performance based on ROA, ROE or SPP (without the log transformation as they are in decimal values); TIME (INDUSTRY) is for yearly (industry) dummy variables, respectively. The term µ_i represents unobservable time-invariant firm-specific effects, such as company reputation and µ_t represents time-variant effects common to all firms, such as an economic downturn; ε_i,t and Ψ_i,t are the time-varying disturbance terms that are serially uncorrelated with mean zero and standard deviation δ. The subscripts: i = 1 to 314 (firms); t = 2002 to 2015 (years) and hence k represents 13 yearly dummies; j =1 to 8 (industry dummies) and s = 1 to 8 as CONTROLS, represent these eight variables (i.e., SIZE, RD, BETA, CUR, LEVER, GROWTH, MBR and MANOW). Finally, α, τ, γ’s and β’s are estimable coefficients. To consider the non-linearity issue, we employ the following setting:

$$\begin{aligned} LNCSP_{i,t} & = \alpha + \beta_{1} CFP_{i,t} + \beta_{2} CFP_{i,t}^{2} + \beta_{3} CFP_{i,t}^{3} + {{\Sigma \beta }}_{\text{s}} CONTROLS_{s, i,t} \\ & \quad + {{\Sigma \beta }}_{\text{j}} INDUSTRY_{j} + {{\Sigma \beta }}_{\text{k}} TIME_{k} + \mu_{i} + \mu_{t} + \varepsilon_{i,t} \\ \end{aligned}$$

(6)

$$\begin{aligned} CFP_{i,t} & = {{\uptau }} + {{\upgamma }}_{1} LNCSP_{i,t} + \gamma_{2} LNCSP_{i,t}^{2} + \gamma_{3} LNCSP_{i,t}^{3} + {{\Sigma \gamma }}_{\text{s}} CONTROLS_{s, i,t} \\ & \quad + {{\Sigma \gamma }}_{\text{j}} INDUSTRY_{j} + {{\Sigma \gamma }}_{\text{k}} TIME_{k} + \mu_{i} + \mu_{t} + \varPsi_{i,t} \\ \end{aligned}$$

(7)

where CFP² and CFP³ are the squared and cubed terms of CFP, respectively; and LNCSP² and LNCSP³ are the squared and cubed terms of LNCSP, respectively.

Furthermore, to consider the lag effects of CSP (CFP) on CFP (CSP), we include the corresponding lagged variables in Eqs. 4 and 5, respectively. In the next section, we employ a set of combinations in the models by including all the parabolic and cubic terms, and similarly some of the factors are lagged by one and two periods, which is to ensure comparative robustness as in Nelling and Webb (2009). Following the implications of Eq. (3), we further include lagged LNCSP in Eqs. (4) and (6).

4.2 The system-GMM estimations

Under the two-step system-GMM setting, the model is estimated at both levels and first differences; i.e., in the stacked regressions level, equations are simultaneously estimated using differenced lagged regressors as instruments. Regarding the consideration of a virtuous circle for example, it is expected that CFP_t−2 enhances CSP_t−1, which in turn increases CFP_t. The system-GMM method accounts for such potential endogeneity issues by using appropriate instrument sets.¹⁰ This estimation technique, hence, controls for this econometric issue that may arise from random shocks affecting both CFP and CSP, and their determinants simultaneously. As explained in Arellano and Bover (1995), and Blundell and Bond (1998), among others, pooled OLS, fixed effects, instrumental variables and even traditional difference-GMM methods would produce biased results for dynamic models.

4.3 Univariate analysis

In Table 1, we divide the sample into four quartiles by sorting firms according to their CSP, which is based on the minimum value of 4.49% and maximum value of 98.83%. The results (statistically different mean and median values) show that the characteristics of firms with high CSP (quartile 4), in general, differ significantly from low CSP firms (quartile 1). As CSP increases on average from quartile 1 to quartile 4, SPP reduces. However, the other financial performance figures (i.e., ROA and ROE) seem to suggest the absence of a linear relationship between CSP and CFP.

Table 1 The CSP quartiles

Table 2 shows the descriptive statistics for our variables, including year-on-year change effects in CSP and CFP. The firms show an average CSP score of 63.071 which suggests that the average firm has less CSP concern. ∆CSP_t, ∆CSP_t−1, and ∆CSP_t−2 are 1.333, 1.315, and 1.310 respectively. Similarly, the average market-based SPP is 0.048, whilst ∆SPP_t, ∆SPP_t−1, and ∆SPP_t−2 are − 0.013, − 0.001, and 0.017 respectively. The year-on-year change effects for the accounting-based performance measures are ∆ROA_t, (− 0.001), ∆ROA_t−1, (0.000), and ∆ROA_t−2 (0.000), ∆ROE_t, (− 0.001), ∆ROE_t−1, (0.005), and ∆ROE_t−2 (0.003). CSP exhibits considerable volatility with a standard deviation of 25.836. CFP is fairly stable given a standard deviation of 0.129, 0.566, and 0.437 for ROA, ROE, and SPP respectively.

Table 2 Descriptive statistics

4.4 Bivariate correlation analysis

Correlation analysis is reported in Table 3. Not surprisingly, the CFP measures are positively and significantly correlated with each other. The highest correlation amongst CFP measures is observed between ROA and ROE. CSP is highly and positively correlated with SIZE and leverage ratio. On the other hand, CUR, MBR and GROWTH are inversely and significantly correlated with CSP. Furthermore, BETA, ROA, ROE, SPP and RD do not significantly correlate with CSP. All CFP measures are strongly and negatively correlated with SIZE. The (unreported) variance inflation factors (VIFs) are far below the threshold value of 10 (minimum  = 1.02; maximum = 2.72), which implies the absence of multicollinearity problems among the explanatory variables. MANOW is negatively and significantly correlated with LNCSP, suggesting that entrenched managerial ownership reduces the incentive of firms to commit financial resources to enhance CSP. This is consistent with the negative correlation between MANOW and SPP.

Table 3 Pearson’s pairwise correlation coefficient between the dependent and independent variables

4.5 Main regression results and discussion

Table 4 provides the GMM results for the model analyzing CFP determinants.¹¹ When assuming a linear relationship between CFP and CSP in model 1, the current CSP has a very significant and positive influence on ROA. In model 2, we include the lagged values of CSP at time t and t − 1; although these lagged effects are statistically insignificant, the current CSP continues to have a strong and positive link with the current CFP. In model 3, we consider the possibility of a non-linear association between CSP and CFP; the respective coefficients transpire to be significant, although the squared and cubed terms are significant at the 10% level. When CFP is proxied by ROA, our results do not follow closely hypothesis 1. In models 4 and 5, current CSP significantly affects current ROE. In model 6, we test for the presence of a non-linear correlation between CSP and ROE but the results are insignificant. These findings rather suggest a linear link between CSP and CFP, and hence are in favour of hypothesis 1 when CFP is measured by ROE. In model 5, our results lend some support to hypothesis 1 with respect to the positive coefficient on CSP lagged one period and are statistically significant at the 10% level.

Table 4 The determining factors for corporate financial performance

On the other hand, when a non-linear relationship between CFP and CSP is assumed in model 9, we obtain a cubic link between CSP and SPP as the coefficient estimates are all statistically significant at the 1 and 5% levels. This means that CFP improves with low and high levels of CSR activities but declines at the medium CSP levels.¹² This result may suggest an optimal CSR intensity. Also, the CSP coefficients in models 7 and 8 are insignificant. This means that hypothesis 1 is not supported when CFP is measured by SPP.

Regarding other factors, the significant and negative coefficients in models 1–3 for SIZE suggest that smaller firms have higher profitability ratios. Firm beta (BETA) negatively affects ROE but does not influence ROA or SPP. Higher firm liquidity (CUR) has a negative and statistically significant effect on SPP only in model 8. Debt ratio is positively (negatively) associated with ROE (SPP). Surprisingly, higher RD activities reduce ROA and ROE. Future growth options (MBR) positively impact on CFP, whereas current growth rate’s effect on CFP depends on how we measure CFP. Finally, managerial ownership (MANOW⁾ exerts a statistically significantly positive influence on ROE and ROA although this effect is insignificant on SPP.¹³

Table 5 examines the determinants of current CSP levels. Following the implications of Eq. (3) as a dynamic model, including the lagged dependent variable [CSP₋₁] in the model as one of the explanatory variables may capture the presence of such an optimality. The results show that the coefficient on CSP₋₁ is always between 0 and 1, and is statistically significant at the 1% level. These findings imply that companies find it rational to dynamically adjust the level of their CSR activities because of the varying costs and benefits that are associated with the process. This thereby supports hypothesis 3. Given that the speed of adjustment [α = 1 minus coefficient estimate on CSP₋₁] ranges between 0.326 and 0.410, one can say that once the company hugely deviates from the optimal CSR activities, the adjustment process is not slow as α is far away from zero. Therefore, the adjustment costs that need to be allocated for the purpose of the deviation from the optimal CSR levels are not too deterring.

Table 5 The determining factors for corporate social performance

When we made the initial assumption that CFP affects CSP monotonously, there is strong evidence that higher CFP leads to higher CSP (models 1, 2, 4, 5 and 8). Being consistent with hypothesis 1, these results suggest that financial affordability plays a key role in a company undertaking CSR activities. As in Nelling and Webb (2009), our analysis in Table 5 can be considered to be testing the presence of Granger causality from CFP to CSP. The coefficients on lagged CFP in models (5) and (8) are significant at both at the 5 and 10% level, which supports the hypothesis that the presence of this is causality. Regarding the consideration of the virtuous circle between CSP and CFP, we find that CFP lagged two periods positively affects CSP lagged one period (model 5 of Table 5) and then this lagged CSP exerts a direct influence on current CFP (model 5 of Table 4).¹⁴ In other words, when CFP is measured by ROE, our study finds the presence of a virtuous circle for the UK firms.

When we assume that CFP affects CSP non-monotonously in Table 5, the link between CSP and CFP based on ROA or ROE shows a cubic pattern in models 3 and 6. These findings suggest that CFP negatively affects CSP at low and high levels of CFP but for the medium CFP levels the effect is positive.¹⁵ Therefore, these results do not support hypothesis 1 when we proxy CFP by ROA and ROE. The results in Table 5 imply both a linear and non-linear relationship between CSP and CFP. This suggests that one needs to be cautious with regard to whether a linear or non-linear association is more appropriate when investigating the effect of financial performance on social performance. To address this issue, we employed Ramsey’s RESET specification test in which the null hypothesis suggests a linear association against the alternative hypothesis of non-linear association. The test suggests a strong rejection of the null hypothesis (p value = 0.00), which means that the non-linear form is more appropriate.

It should be noted that the significant coefficient estimate on CFP (see models 3 and 6 of Table 5) does not necessarily mean that the relationship between financial performance and social performance is linear; this is because in models with quadratic equations, the variable with a polynomial degree of one (CFP) should be considered together with the parabolic (CFP²) and cubic (CFP³) terms. Table 5 shows that the causal link between CSP and CFP runs from one direction of CSP, as LNCSP(−1) is statistically significant at the 1% level from models 1–9. Table 5 reports the results of the control variables. SIZE has a positive and statistically significant effect on CSP. However, CSP reduces significantly when firm liquidity or leverage ratios increase. The remaining control variables generally do not exert any significant influence on CSP.

We further examine the associations between CSP and CFP within UK firms for the period 2002–2015. Our focus is to determine whether any linear link exists, and the causality of any such relationship by considering the endogeneity problem. We also analyze if any non-monotonous relationship exists between CSP and CFP, and if companies will have optimal or target CSR activity levels. We find certain degrees of non-linear links between CSP and CFP. In addition, current and past CFP values seem to impact linearly on the current CSP but the presence of a cubic association between these two variables is more apparent. Our results suggest that CFP and CSP are neither strictly positively nor negatively correlated but the association is rather non-linear. This confirms Brammer and Millington’s (2008) findings, and therefore suggests that the disagreement on the CSP–CFP link debate is due to the fact that the literature has ignored the non-linearity and target CSR issues (see also Barnett and Salomon 2006). Our study differs from that of Barnett and Salomon (2006) in several ways: we extend their parabolic setting, by considering a cubic link between CSP and CFP. They focused on socially responsible investing in the U.S., whereas our CSP measure for the UK firms is more comprehensive, and we also include several explanatory variables. We further conduct a dynamic analysis and use a robust system-GMM estimation method that is efficient for panel data analyses.

Our analysis reveals that concurrent CFP (ROA, ROE or SPP) linearly and significantly affects concurrent CSP in the sense that higher financial performance suggests higher social performance. The results suggest that financially stable companies can afford to be socially responsible. Similarly, concurrent CSP linearly, significantly and positively affects CFP proxy by ROA and ROE, and past CSP positively affects current ROE, although they are statistically significant at the 10% level. Moreover, CFP lagged one and two periods (SPP and ROE definitions) have a positive and statistically significant influence on current CSP. On the other hand, we report a cubic link between CSP and CFP (based on SPP) which suggests that at low and high levels of CSR activities, CFP improves but CFP reduces at the medium levels. Hence, the issue of whether firms should differentiate themselves with high commitment to CSR to impress stakeholders or save the resources seems to matter. Furthermore, when we examine the effect of ROA and ROE on CSP, we find another cubic association, which implies that firms with low and high financial performance negatively affect CSP but medium financial performance is associated with an improved CSP.

On the virtuous circle of the CSP-CFP relationship, a more integrated relationship receives support from our empirical analysis. Thus, it seems firms can be socially responsible and financially successful at the same time, and companies can have a competitive advantage if they invest in CSR activities (see also Gregory et al. 2016). Our findings further lend some support to the presence of the virtuous circle, which suggests that past CFP improves present CSP which then improves future CFP. Our partial adjustment process reveals that firms are prone to having target CSR structures and they periodically revise the intensity of their CSR activities in order to be at their optimal CSP levels. This is consistent with our hypothesis 3. Furthermore, the diagnostic tests for the system-GMM estimates highlight the relevance of the endogeneity problem when running regressions. It is also important to note that the regression results are not independent from econometric specifications and the proxies of CFP.

4.6 Additional tests

In this section, we perform additional analyses to provide robustness to our primary findings. First, we analyse year-on-year changes in CFP and CSP to underline our main results. Second, we perform our analysis based on financial and non-financial firms. Third, we consider our analysis based on the intensity of CSR engagement across industries. Finally, we assess the relevance of the global financial crisis by running the models across two time periods.

4.6.1 Year-on-year change effects

In our main analyses, we considered the association between CSP and CFP in levels. In this sub-section, we focus on the effects of year-on-year changes. Table 6 reports the regression results when CSP is first-differenced at time t, t − 1 and t − 2 (i.e., ∆CSP_t, ∆CSP_t−1, and ∆CSP_t−2, respectively) as a set of explanatory variables and CFP is first-differenced at time t (i.e., ∆CFP_t) as the dependent variable. When Tables 4 and 6 are compared, it seems that CSP and CFP have more significant links when they are measured in levels compared to when they are in first-differences. However, our analysis obtained a few significant links regarding the year-on-year changes: in model 3, lagged CSP at time t − 2 impacts positively on first-differenced ROA at time t. On the other hand, in models 9–12 when we proxy CFP by market base variable SPP, we find that a reduction in CSP negatively affects SPP. This is consistent with Cheung (2016) who argues that CSP has a negative effect on idiosyncratic risk. An increase in idiosyncratic risk perceived by investors as a result of reduction in financial resources committed by firms to CSP will negatively affect the firm’s share price.

Table 6 The determining factors for corporate financial performance: year-on-year change effects

Table 7 reports the regression results when CFP is first-differenced at time t, t₋₁ and t₋₂ (i.e., ∆CFP_t, ∆CFP_t−1, and ∆CFP_t−2, respectively) as a set of explanatory variables and CSP is first-differenced at time t (i.e., ∆CSP_t) as the dependent variable. When we compare the findings of Table 7 with Table 5, the lagged first differenced CSP is significant in all models with coefficients taking values between values 0 and 1 in absolute terms; the corresponding signs are negative because the analyses are done by taking into account year-on-year changes. Overall, these findings suggest the presence of optimal CSR activities. As for the CSP–CFP nexus, the hypothesized positive link is apparent in model 4 where all dimensions of transformed ROA are statistically significant. Concerning ROE, its effect on ∆CSP is positive and statistically significant only for the first-differenced values at t₋₂. Furthermore, with regard to SPP, the expected positive is reported only when SPP has its first-differenced transformation at t₋₁.

Table 7 The determining factors for corporate social performance: year-on-year change effects

4.6.2 Financial versus non-financial firms

The importance or implications of being a socially responsible company may change depending on the business sector or industry group the company is operating in (see e.g., Jenkins 2004; Scholtens 2008b; Waddock and Graves 1997). In Table 8, we investigate the CSP and CFP determinants by splitting our sample into financial (industry groups 8 and 9) and non-financial firms (industry groups 1–7 and 10) based on the industry classification in Table 13 in the Appendix.¹⁶ Panel A of Table 8 reports the results when we regress CFP on CSP: the only significant result in panel A is the positive link between CSP and ROA in model 1. In Panel B, we conduct the same regressions for non-financial firms; the only significant and positive link between CSP and CFP is observed when CFP is measured by ROE in model 4. Panel B further reveals that CSP and CFP have a cubic relationship but the nature of this relationship depends on the CFP proxy; in model 3, higher or lower CFP improves ROA but at the medium CSP level, ROA decreases. In model 9, on the other hand, higher or lower CFP reduces ROA but at the medium level CSP improves ROA.

Table 8 The determining factors for corporate financial and social performance: financial versus non-financial firms

In Panels C and D of Table 8, we regress CSP on CFP for financial and non-financial firms, respectively. Although both business sectors seem to adopt optimal CSR levels, the speed of adjustment to the optimal CSP level is higher for non-financial firms. Moreover, the positive effects of CFP on CSP are more apparent for non-financial firms. With regard to financial firms, current CFP does not impact on current CSP. This relationship is consistent and statistically significant for non-financial firms. Although it is not very persistent, there is some evidence to suggest that financial firms’ past ROA or past SPP have a positive and statistically significant effect on current CSP. An interesting finding in Panel D shows that CFP and CSP have a cubic link: CSP and ROA are negatively (positively) linked for lower and higher (medium) values of ROA.

4.6.3 Intensity of CSR engagement across industries

In this section, we divide our sample industries into three broad groupings based on the intensity with which the industry that firms belong to engages in CSR; (1) based on industries where CSR engagement is at high-levels; (2) industries where CSR participation is at medium-levels; and (3) industries where CSR is below the average level or is low. Using the mean CSP score for each industry in Table 13, we classify groups 7, 8 and 9 as those industries with low CSR engagement; groups 1, 5, 6 and 10 are industries with medium CSR engagement, and industries with high CSR engagement are groups 2, 3 and 4. Table 9 reports the regression results for these classifications. Panel A shows that firms in industries with low CSR engagement find it beneficial to increase their current CSP in order to improve their current CFP, whereas Panel C suggests that such an advantage is virtually non-existent. These findings imply that increasing the existing high level of CSR activities does not correspond to an increase in profitability; however, significantly low or high CSP can influence firms to optimize their CSR activities. Panels A, B and C show an interesting pattern to suggest a cubic relationship between CSP and SPP. For firms operating in medium-to-high CSR intensity industries (Panels B and C), the effect of CSP on SPP is negative for the low and high levels of CSP but this effect is positive for medium CSP levels. However, for firms operating in low-CSR intensity industries (Panel A), the impact of CSP on SPP is positive for the low and high levels of CSP but negative for medium CSP levels. This is consistent with Lins et al. (2017) who show that high CSP intensity firms experience higher profitability than low CSP firms as a result of high social capital and trust from investors, and Cheung (2016) who argues that high-CSR intensity firms experience lower idiosyncratic and systematic risk with stakeholders and investors respectively.

Table 9 The determining factors for corporate financial and social performance: the level of CSR engagement

Panels D, E and F of Table 9 show the effects of CFP on CSP based on the level of CSR engagement. With respect to the coefficient estimates on lagged CSP, the results indicate that all sub-groups adopt optimal CSR policies and the speed of adjustment to the desired CSP level is highest (lowest) in Panel F (Panel D). This suggests that it is easier for firms in high-CSR intensity industries to be on their target CSP levels relative to their peers operating in low-CSR intensity industries. Panel D reveals that current ROA and ROE positively affect current CSP levels. Although we observe similar correlations in Panels E and F, the mentioned association is less convincing. Therefore, one can assert that firms operating in low-CSR intensity industries experience the positive effect of current CFP on current CSP.

In Panels D, E and F of Table 9 we observe a cubic association between CSP and CFP in three models. Model 3 of Panel D suggests that for firms in low-CSR intensity industries, CSP increases with low or high ROA figures but it decreases with medium ROA figures. However, model 9 of Panel D shows that firms in low-CSR intensity industries experience a decrease in CSP with low or high SPP figures but it increases with medium SPP figures. Finally, model 9 of Panel F shows that CSP decreases with low or high SPP figures but it increases with medium SPP figures in firms operating within high-CSR intensity industries.

4.6.4 The relevance of the 2007–2008 global financial crises

Some studies attribute the occurrence of 2007–2008 global financial crisis partly to poor CSR commitment as a result of corporate greed and unethical behaviour (see Argandoña 2009). For instance, Karaibrahimoglu (2010) argues that during the global financial crisis, CSP was low as firms engaged in cost cutting activities. However, other recent studies show that CSP has improved following the global financial crisis (see Kemper and Martin 2010). Lopatta and Kaspereit (2014) note that there is a perceived increase in CSP around the world post the global financial crisis. Similarly, Lins et al. (2017) state (and then empirically show) that as the public trust in firms following the financial crisis went down, the value of being socially responsible is bound to be rising during post-crisis times. We therefore split our sample into two periods [i.e., pre-crisis (2002–2008), and post-crisis (2009–2015)] to provide additional robustness to main results.¹⁷

Table 10 reports the CSP and CFP results after splitting the sample. An interesting finding emerges in Panels A and B. We observe that when CFP is based on ROE, the impact of CSP on CFP is negative during the pre-crisis period whereas the same association is positive during the post-crisis period. This finding is in line with the main findings of Lins et al. (2017). Furthermore, the non-linear link obtained in Table 4 (model 3) for the whole sample is reported again for the post-crisis period only in Table 10 (Panel B, model 3). Splitting the sample leads to the presence of another significant non-linear association that we did not observe in the previous analyses: the effect of CSP on ROE is negative when CSR activities are low or high but at medium level the relationship is positive. Overall, we fail to report any linearly positive effect of CSP on CFP for the pre-crisis period.

Table 10 The determining factors for corporate financial and social performance: the relevance of the financial crisis

Panels C and D show that in both periods, the UK firms continue to maintain optimal CSP and the speed of adjustment to this optimality is higher during the pre-crisis period. When both periods are compared, the linearly positive effect of CFP on CSP is clearly more apparent for the post-crisis period as this connection is virtually non-existent for the pre-crisis times. The only non-linear link is reported in Panel D (model 6) for the post-crisis years: the effect of ROE on CSP is negative (positive) if the intensity of CSR activities is low or high (medium).

In Table 11 we consider again the pre-crisis and post-crisis years but only for financial firms. Regarding the effect of CSP on CFP, the results do not change across both time periods for these firms. With the exception of the positive effect of CSP on ROA in model 1, the results appear to be generally statistically insignificant. Yet, they are comparable to the related analyses reported in Table 8. Moreover, in Panels C and D, it is observed that financial firms adjust the level of their CSR activities towards target CSP levels faster in the post-crisis years than in pre-crisis years. Similarly, the positive effects of financial strength on CSP are more pronounced for the post-crisis period.

Table 11 The determining factors for corporate financial and social performance: the financial crisis and financial firms

In Table 12, we repeat our analyses attached to the ones in Table 11 but only focused on non-financial firms. For both periods, the only significant and positive effect of CSP on CFP is observed when CFP is proxied by ROE. Yet again, for both periods we report a cubic association (model 3), which confirms the corresponding findings in Table 8 (Panel B, model 3). However, when CFP is based on SPP, the same cubic association that we report in Table 8 (Panel B, model 9) is obtained only during the pre-crisis times for the non-financial firms in Table 12 (Panel A, model 9).

Table 12 The determining factors for corporate financial and social performance: the financial crisis and non-financial firms

In Panels C and D of Table 12, the presence of the optimal CSP continues to hold for these non-financial firms. Further, the speed of adjustment is higher for the pre-crisis years than for the post-crisis years, which is in contrast with the case of financial firms in Table 11. Finally, reminiscent of the previous results in this sub-section, the positive effects of CFP on CSP is more salient for the post-crisis period for the non-financial firms.

5 Conclusions

This study has examined the existence of non-linear link between CSP and CFP and, more importantly, how this shapes the optimal level of commitment to CSR activities. The presence of non-linearity between CSP and CFP is an important addition to the extant literature on CSR. Indeed, our empirical analysis indicates that (1) medium levels of CSR activities reduce financial performance while low and high CSR levels increase financial performance, and (2) firms with low and high financial performance are less committed to CSR activities, while firms with medium financial performance engage more in CSR commitments. Evidence from the intensity of CSR engagement across industries shows a cubic link between CSP and SPP. We find a strong support for CSP not having a significant impact on CFP in financial firms; however, CSP does have a significant impact on CFP in non-financial firms. We also find a significant impact of CFP on CSP in both financial and non-financial firms. Our paper provides additional evidence as to how CSR activities and CFP interact during financial crisis and tranquil times. We report that the positive effect of CSP on CFP is more apparent during the post-financial crisis years. Furthermore, it appears that optimal CSR activities and the existence of non-monotonous relationships are relevant for both periods and the speed of adjustment to desired CSP levels depends on the time periods and industrial background of the firms.

These findings have implications for slack resources theory, which suggests that CSR activities should increase with higher firm financial resources. This may not always be so, as the findings of this study suggest. Indeed, explaining why there are such non-linear links through other theoretical perspectives, such as institutional theory and legitimacy theory considerations may highlight firms’ motivations in engaging in CSR activities. Despite the theoretical relevance of our findings, it further implies that stakeholders (including government, customers and the wider public) may need to curtail their expectations of higher CSP from higher resourced firms. This has significant implications for CSR advocacy and policy. Methodologically, our study utilizes a robust and well-established index, ESG scores, which takes into account the multidimensional aspects of CSR. This is an important addition to the CSP/CFP discourse. Furthermore, the estimation model and partial adjustment process which we adopted in the study are useful empirical contributions to the CSP–CFP nexus. Our examination of the presence of non-linear (parabolic and cubic) relationships between CSP and CFP suggests that companies might adjust the intensity of their CSR activities because of internal and external developments. This includes practical implications with regard to the drivers of CSR, the level of commitment to CSR activities, and the determinants of CSR activities, which are not always based on economic rationales.

While this study shows that CFP and CSP are linked both monotonously and non-linearly, it remains to be seen why this occurs, although the non-linearity effect is more apparent. Future research can address this limitation through more advanced methods that have more robust testing for the presence of a virtuous circle, Furthermore, it is likely that the benefits and costs associated with CSP activities vary over time and managers need to respond appropriately to these changes. Future research could consider including an advertising metric in the analysis, based on the amount a company spends advertising its CSR actions to its stakeholders. This study is also related to corporate reputation and branding (Neville et al. 2005). Moreover, it is important to further investigate the timing associated with the relationship between CSP and CFP (Brammer and Millington 2008) as the interactions between these would become clearer if we know how long it would take for the impact of CSR on CFP to be shown. Data analysis with longer panels in emerging markets would be another future research opportunity. Also, future studies could methodologically scrutinize and identify the specific situations whereby it is unbeneficial to be socially responsible. Future studies that can further provide detailed empirical and theoretical analyses of stakeholder decision making processes might improve our understanding of how CSP interacts with CFP.

Appendix

See Tables 13 and 14.

Table 13 Industry classification

Table 14 Definitions of Variables