Entities’ Performance and Human Resource Costs Derecognition in the Statement of Financial Position (SOFP): GMM Evidence from the NGX

Abstract

This study explored Entities’ Performance as an explained function of Human Resource Costs (HRC) to further justify recognition of the Labour Costs proxies in the Statement of Financial Position (SOFP). This has been investigated to provide robust empirical evidence from the Nigerian Exchange Group (NGX) to spur the International Accounting Standard Board (IASB) to release an Exposure Draft (ED) for public discussion and have a standard to recognize proxies of HRC as assets in the SOFP. To provide grounds for inclusion of HRC in the SOFP by the IASB, unlike most other empirical studies reviewed, which deployed limited methods and years of time series data, this study expanded the scope and methods using Pooled Cross-Sectional (PCS) time series data of 27 quoted companies from 1992 to 2023 in the NGX. While most studies employed inefficient Ordinary Least Squares (OLS), this current study progressed from Descriptive Statistics to OLS, Pooled OLS, and Rodman’s Xtabond2 Generalized Method of Moments (GMM) to resolve the conundrums of endogeneity, reversed causality, and stationarity common to unbalanced PCS time series data. The results revealed from the GMM showed that LSW (18.40), positive, and LTD (−22.63), inverse, and Wald ^2 = 66.35 with p-value (0.002), obviously validated the strong joint significance of the regressors on ROA (performance) of 27 sampled firms in the NGX. It is recommended that IASB align with the momentum from the output of research from academia by issuing standards to recognize HRC as assets in the SOFP.

1. Introduction

Over the past four decades, academia and Professional Accounting Bodies have devoted great attention to continued investigation of issues around the recognition of Human Resource Cost (HRC) elements in the Statement of Financial Position (SOFP). The serious concern shown by these bodies appeared reasonable because animal, physical, and intangible items are included in the SOFP as assets under International Accounting Standards (IAS) 16 on Property, Plant and Equipment (PPE), Intangible Assets (IAS 38), IAS 41 (Agricultural Assets), and IAS 2 (Inventory). As of now, the International Accounting Standards Board (IASB) has issued Accounting Standards to recognize physical, intangible, and agricultural assets in the SOFP, but under IAS 1 on Presentations of General-Purpose Financial Statements, the elements of Human Resource Costs (Labour Costs) are expensed annually in the Income Statements. Contrary to the prescriptions in IAS 1, Ionel et al. (2010) denoted Labour Costs as the most essential asset of entities. I. R. Akintoye (2012) pronounced that investment in Human Capital Cost (HCC) meets the criteria prescribed in IAS 16, IAS 38, and IAS 41 under PPE, Intangibles, and Agricultural Assets. Despite the positive inferences from the large number of research and conceptual definitions of assets in the Accounting Standards, HCC elements are still currently expensed annually in the Income Statements. Notwithstanding the robust and positive empirical evidence available from academia and accounting practitioners on the treatment of Labour Costs as assets in the SOFP, proxies of HCC are still annually expensed in the Income Statements. This is unlike PPE, non-human physical agricultural assets, and intangible assets that had been prescribed in IAS 16, IAS 2, IAS 38, and IAS 41. Furthermore, the derecognition of proxies of HRC, such as Salaries and Wages, Training and Development Cost, Medical, Gratuity and Pension Benefits, among other related costs, equally reduces the reliability and faithful representation of the information reported as contained in IAS 1 on Presentation of Financial Statements. The derecognition of the elements of personnel costs in the SOFP has both direct and indirect effects on the myriad users of the Annual Report (Kankpang et al., 2024; Anthun et al., 2024; I. Akintoye et al., 2018).

As of now, despite the positive momentum on the matter, the IASB has yet to issue Exposure Drafts (ED) and Accounting Standards on the treatment of HCC as assets in the SOFP (Yikarebogha & Adesola, 2025; Okorie & Ebere, 2025; Li et al., 2024; Ibrahim et al., 2024; Kankpang et al., 2024; Anthun et al., 2024; I. Akintoye et al., 2018; I. R. Akintoye, 2012; Schultz, 1961; Schmidt, 1997). HRC, just like other assets, has a useful life of more than a year, 35 years in the NGX entities, yet the cost of labour is expensed in the Income Statement, whereas physical, intangible, and agricultural assets are capitalized in the SOFP. The users of Financial Statements bear the consequences of using understated assets in the SOFP to make daily investment decisions. To address this quandary, this study aims to examine Entities’ Performance as an explained function of HRC elements in the Financial Statements with a view to questioning why Human Resource, which uses other assets to generate wealth in entities, is derecognized in the SOFP.

2. Literature Review

Conceptual, Theoretical, and Empirical Review

HRC refers to the cost of labour engaged in producing goods or rendering services in entities. The American Accounting Association (2003) defined HRC as a process of identifying and measuring Labour Costs and related amounts incurred to compensate for mental and physical efforts. This is to create wealth in the short, medium, and long terms. This study is grounded in the Human Resource Conceptual Framework and Cost and Earnings Models, which capture constructs such as Return on Assets (ROA), Equity (EQ), Salaries and Wages (SW), and Training and Development, as premises for analysis in the Materials and Methods section. Labour Costs, Human Resource Costs (HRC), and Human Capital Costs (HCC) are used interchangeably in the study to signify the quantum of expenditures incurred by employers on salaries, wages, hiring, training, medical, and other payroll-related costs. This conceptual definition is invaluable to measure variables engaged in Section 3 of the study. A significant amount of the extant literature reviewed in this study shows serious concerns about the continuous derecognition of HRC in the SOFP; for example, I. Akintoye et al. (2018), Enyi and Akindehinde (2014), presented the seamless usage of employees taking major investment decisions in entities on a daily basis over the years, and were puzzled about why the cost elements of labour are expensed annually. Besides, it was argued that expensing HRC annually distorts and underreports the totality of the assets engaged to generate income, profits, or losses of entities in a financial year.

The theory of Human Resource Costs (HRC), according to Schultz (1961), Becker (1964), Hekimian and Jones (1967), Likert (1967), Lev and Schwartz (1971), E. Flamholtz (1971), E. G. Flamholtz et al. (2002) and American Accounting Association (2003), consists of a quantum of costs and the associated cost of hiring, training, retraining, renumeration, gratuity, medical and other related benefits to employees in entities. Based on data availability from the NGX firms, the HRC theory provided a rationale for the choice of explained and explanatory variables (ROA, LEW, LSW, and LTD), denoted in Section 3 of this study.

The effects of the cost elements of Human Resource on income or value generation in entities have been extensively studied by researchers and professionals, with similar results across different countries showing that HRC creates value over the years in entities. In the developed economy, for example, Canada, Bontis (2003) asserted that many Scandinavian entities report HRC in their Income Statements. However, Bontis (2003) concluded that a negligible percentage of the entities in Canada (68 out of 10,000) disclosed and used the terminology HRC in their Financial Statements but did not report this as assets in the SOFP (Bontis, 1998; also endorsed this position). Also, Ansari and Flamholtz (1978) and Houghton (1980) revealed that the elements of Human Resource Costs contributed to the productivity, wealth, and performance of entities. The study supported HRC being viewed as a total concept, measured as assets in the SOFP, and represented in terms of efficiency, effectiveness, productivity, and wealth creation in entities.

Furthermore, the empirical study of Chen et al. (2005), using Descriptive Statistics, Correlation, and Multiple Regression with lagged values of explained and explanatory variables such as Return on Equity (ROE), Return on Assets (ROA), and Employee Productivity (EP) recognized that the coefficient of HRC positively and significantly influenced performance measured with profitability ratios. HRC was measured with the proxies of Labour Costs as earlier indicated. As a result of the non-availability of HRC as assets in the SOFP, Chen et al. (2005) asserted that investors use unknown and derecognized values of Labour Costs to make daily major economic decisions about potential values of the sampled entities.

The evidence for inclusion of HRC in the SOFP is also corroborated by the American Accounting Association (2003), Anam et al. (2011), Kankpang and Nkiri (2019), Anthun et al. (2024), Asutay and Ubaidillah (2023), among other publications from academia in the developed economies. Also, Arseneault and Gagnon (2024) certified how HRC portended values with a sample size of 186 entities. The study, with a survey, regressed Return on Assets (ROA) as an endogenous variable on Sales and Employee Growth and labour cost elements (Salaries, Training Costs, etc.). It revealed that HRC exerted shock on corporate performance. Similarly, the study of Vithana et al. (2023), using the stock price as the explained variable and proxies of investment in Human Resource Cost (HRC) in the United Kingdom Stock Exchange of 100 FSTE over five years from 2005 to 2009, with Descriptive Statistics and Cross-Sectional Regression authenticated that investment in HRC positively and significantly ignited shock on earnings and stock price. Besides, investors in the Capital Market trusted that additional investments in HRC attracted higher returns and provided a scarce opportunity to create wealth and progress entities’ growth trajectories over the years (Vithana et al., 2023).

Also, in the empirical study of Asutay and Ubaidillah (2023), where HRC elements and performance were explored to measure wealth creation in Islamic Banking, the results revealed that performance metrics were positively and significantly high. Similarly, the work of Rong et al. (2025) using Descriptive Statistics, Correlation Matrix to test Multi-Collinearity of variables, Regression and Pooled OLS to measure Fixed Effect and Random Effect using Valued Added Intellectual Coefficient (Value added by the entity—Human Cost), Capital Employed (Tangible and Financial Capital) and Employees’ Cost further gave credence to the fact that elements of HRC were positively and significantly correlated and affected growth in the values of Commercial Banks in the European Countries. This evidence was earlier corroborated in the developed and emerging economies in the studies by Benston (1968), Birnberg and Nath (1968), and further supported by Becker (1964), E. Flamholtz (1971), Elias (1972), Schwan (1976), Schultz (2001), and Haddock-Millar et al. (2016) in the United States and United Kingdom and later by Vithana et al. (2023), Okorie and Ebere (2025), etcetera. To sustain the momentum, Raghavendra (2022) utilized the responses from 50 Human Resource practitioners who were part of the workforce of reputable companies in Bengaluru and Karnataka, India, and employed a one-sample t-test to validate a strong relationship with the proxies of HRC and performance. This result was corroborated in China by Wang et al. (2024) using the Structural Equation Model, and Anam et al. (2011) in Malaysia, who employed OLS.

Furthermore, Brummet et al. (1969) at R.G. Barry Corporation of Columbus, Ohio, with the support of Michigan University in 1967, using five variables (Salaries, Recruitment, Acquisition, Training Cost, Familiarization, and Development Costs), revealed that the elements of HRC should be capitalized in the SOFP and not reported as expenses in the annual Income Statements. To renew the empirical evidence, Arseneault and Gagnon (2024) regressed a firm’s performance on HRC elements and percentage change in sales, and 5-year employees’ growth confirmed the extent to which HRC portended values in the sampled entities. The results further endorsed earlier empirical inferences that HRC influenced corporate performance significantly. In Norway, Anthun et al. (2024), in the study of the influence of HRC on the financial performance of the Indonesian Banking Sector using descriptive qualitative review, confirmed the positive influence of HRC elements on the performance of the sampled entities. To further authenticate the momentum, Rong et al. (2025), using Pooled OLS (Hausman Test), HRC elements, and performance metrics, asserted that financial and human capital costs positively and significantly correlated with values of the sampled entities). From the foregoing, it can be inferred that the empirical studies from the emerging and developing economies validated HRC as an important driver of profit and performance of entities and should be classified as assets in the SOFP.

In the context of developing nations such as Nigeria, the research of Kankpang et al. (2024), showed the effect of HRC using Profit After Tax was regressed on Staff Remuneration Cost (SRC), Staff Size, and Staff Training and Development Cost (STDC) in 15 Oil and Gas companies. Just like this current study, Staff Remuneration was positively significant, whereas Staff Training and Development Cost were inversely significant on the performance. Also, a related study of Emeka-Nwokeji and Agubata (2019), Okoye et al. (2019) and Balogun et al. (2020) and the variables deployed using the Return on Asset (ROA) and the Return on Capital Employed (ROCE) as explained functions of the Board Size, Training and Development Cost, Directors’ Renumeration and Log of Total Assets among other data obtained from 93 out of 122 nonfinancial firms and five manufacturing companies from 2006–2015 and 2015–2019 in the NGX, validated with Pooled OLS, Correlation Matrix and Multiple Regression, demonstrated that proxies of HRC have significant impact on performance. The results are similar to those of Bonsu et al. (2019) studying Ghana from China University using OLS to confirm that HRC significantly influenced performance. Adegbayibi et al. (2024b), employing Pooled OLS with Return as a function of Employee Cost to Revenue and Revenue per Employee, revealed that HRC is significant on the sampled entities. Akinlo and Olayiwola (2017), using data from 50 listed entities from 2007–2014, with Pooled OLS, asserted that earnings were positively associated with Salaries and Wages and Labour Turnover. This evidence was also authenticated in Ghana by Bonsu et al. (2019) and Ogunbiyi-Davies et al. (2023), and Okorie and Ebere (2025) in Nigeria.

The result of Omole et al. (2017a) and Ogunbiyi-Davies et al. (2023) with the Return on Assets (ROA) as a function of the Training Cost, the Staff Welfare Cost, the Staff Hiring Cost, the Staff Safety Cost, Salaries and Wages, Retirement Benefits, the Log of Staff Employee, and the Labour Turnover ratio among other similar variables, with the same Pooled OLS, OLS, and Correlation Matrix, is not different. However, the OLS and Pooled methods, as contained in Olusanya et al. (2016), appeared inappropriate, and the sample size was limited. In Kenya, a questionnaire employed by Mbithi (2019) with Cronbach’s Alpha coefficient recognized a significant positive association between HRC and the performance of the entities. However, questionnaires and categorical analyses appeared subjective to draw inferences (Olusanya et al., 2016).

Similar evidence was earlier pronounced by Leyira et al. (2012) using 52 companies from data obtained from the Nigeria Exchange Group (NGX) with Descriptive Statistics, Regression, and Correlation analyses, as earlier cited, to be spurious and inefficient to account for the significant contribution, 75.9% of HRC on Return on Equity (ROE). Similarly, the study of Adegbayibi et al. (2024a), V. A. Akinjare and Ologunde (2021), Ndum and Oranefo (2021), Onyekwelu and Akanni (2021), Onyekwelu and Ironkwe (2021), Bankole (2020), Abdulateef et al. (2018), Olowolaju and Oluwasesin (2016), and I. R. Akintoye and Adidu (2008) used methods such as Descriptive Statistics, OLS, Variance Inflation Factor (VIF) and Breauch Pagan/Cook-Weisberg/Woodridge tests, among other methods, to measure multicollinearity. The results showed that Employee Remuneration Cost (ERC) and retirement benefits exerted a positive and significant shock on profit, whereas Safety and Health Costs were positively insignificant.

The results were similar in the earlier study of Enyi and Akindehinde (2014), where the likely effect of HRC was measured on the decision-making process by regressing data obtained from questionnaires distributed in 16 quoted banks in the NGX. The results supported the need to report HRC elements as Intangible Assets. Ajayi (2003) recognized that outputs, Q of entities in an economy, can only be produced with the least combination of Labour (L) and Capital (K). The study questioned why the traditional accounting practitioners recognized Capital (K), Ordinary Share Capital, Retained Earnings, Premium, Preference Shares, Redeemable and Irredeemable Loan Notes, items of Current Liabilities (Trade Payables, Accruals etc.), Tangible Assets in IAS 16, Intangible Assets in IAS 38, Inventory in IAS 2, receivables and other Current Assets in the Statement of Financial Position, whereas the quantum of the reward for labours (Salaries and Wages, Placement and Familiarization Cost, Training and Development Costs, inter alia), which produce the wealth over years are derecognized as assets and reported as expenses yearly.

For emphasis, Enyi and Akindehinde (2014) employed secondary data in 16 quoted banks in the NGX with an ex-post factor design, 6 Points Likert Scale, and a Cronbach’s Alpha pilot test. The empirical results revealed that it is important to value the Human Asset (HA) like other intangible assets in the SOFP. The study is similar to the study of Reeta (2015), where Recruitment Cost, Salaries and Wages, Training and Development Costs, Retention Cost, Pension and Gratuity were deployed to measure Human Resource Cost (HRC). The results of the OLS with 15 years of data obtained with ex-post designed factor supported that elements of HRC have a significant positive effect on the values of entities. The evidence was also supported by Kankpang et al. (2024) and I. R. Akintoye (2012), among others, in the developing economies. The empirical study of Edom et al. (2015) using OLS of only Access Bank Plc. from 2003–2012 in Nigeria asserted that the Training Cost (TC) provided a significant positive influence on the development of the bank. However, the sample size is limited, and the regression appeared spurious because the Best Linear Unbiased Estimator (BLUE) and robust autocorrelation test were not conducted (Olusanya et al., 2016; Hansen, 1982; Chamberlain, 1987; Russel & MacKinnon, 1993).

From the above, it appears there exist material Accounting Standard and methodological gaps because most of the studies reviewed employed OLS and Pooled OLS. To the best of the researcher’s knowledge, this is the first time GMM would be employed to measure the elements of Human Capital Costs and Performance of Entities quoted in the NGX. Despite the deceptive nature of Ordinary Least Squares (OLS) in addressing reverse causality, autocorrelation, heterogeneity, and endogeneity-related econometric issues in time series data, most of the empirical studies reviewed engaged OLS and Pooled OLS. Olusanya et al. (2016), Hansen (1982), Chamberlain (1987), Russel and MacKinnon (1993), and Ruud (2000) revealed the spurious nature of the outcome of OLS and Pooled OLS, where unbalanced or balanced pool time series data were employed, with suspicion of dissimilar accounting policies of entities in different sectors. The study of Barak and Sharma (2024) is novel. In the study of Public Sector Banks in India, GMM was engaged to confirm that elements of HRC had an impact on Return on Earnings, Equity (ROE), Return on Asset (ROA), and Return on Capital Employed (ROCE). Besides, there seems to be an Accounting Standard gap in the non-issuance of standards to treat HRC as assets in the SOFP. To fill the gap in the frontier of accounting knowledge and possibly further push IASB to issue Standard on the disclosure of the proxies of HRC in the SOFP, this study addressed the methodological and literature gaps on the non-issuance of Accounting Standards by progressing from descriptive analyses, which measures behaviour, to pooled OLS, which measures relationship, to the deployment of the unique and dynamic Generalized Method of Moments (GMM) to address the problem of endogeneity, heterogeneity and reversed causality common in an unbalanced and dissimilar Pooled Data.

3. Materials and Methods

This study adopted secondary multivariate dynamic panel data using a quantitative longitudinal design. This is supported in Barak and Sharma (2024), Rong et al. (2025), etc. This design is necessary because the dataset cuts across all sectors in the NGX. It consists of 27 out of 150 (18%) firms quoted as at 2023 on the floor of the Nigerian Exchange Group (NGX). The choice of NGX is attractive to provide robust evidence to further corroborate the inclusion of HRC elements in the SOFP. NGX has a larger number of listed companies in Africa after the Egyptian and South African Exchanges. The study engaged random and judgmental sampling to select the firms based on the available yearly Annual Reports online and hard copies at the NGX Library. Random Sampling appeared efficient because it reduced the bias of selection, and every firm out of the 150 has an equal likelihood of selection to make it representative. The Judgemental Sampling technique presented opportunities to select firms with a higher number of datasets (variables). The Annual Reports were downloaded from the database of the companies, NGX, and the libraries of the entities, Registrars of the companies, the Security and Exchange Commission, etc. The sampled firms include Manufacturing, Brewery, Oil and Gas, Confectionery, Banking, and Insurance, among other entities. The dataset covered the time period from 1990 to 2023. The Annual Reports of the Companies in 2024 and 2025 had not yet been published at the time of the data compilation used for the research output. Also, due to paucity of data at the NGX, some of the firms have unbalanced panel data from 1992 to 2023, 1994 to 2023, and 2012 to 2023, respectively. This further justifies the use of Arellano xtabond2 GMM, as uniquely and uncommonly deployed in Barak and Sharma (2024).

However, most of the works reviewed by the study adopted OLS and Pooled OLS, which had been largely de-weighted by Olusanya et al. (2016) and Hansen (1982). Chamberlain (1987); Russel and MacKinnon (1993), and Ruud (2000). OLS and Pooled OLS have been confirmed to be inappropriate and inefficient in Econometrics when the structure of time series data is unbalanced, pooled, cross-sectional in nature, covariance is unknown, and when estimators are inconsistent, and the Best Linear Unbiased Estimator (BLUE) is violated. Similarly, the study progressed from Pooled OLS as a result of the assumptions of no difference among entities (note that entities in the study are dissimilar); that all observations are not independent (autocorrelation), and that no distinction exists between the trends in the time series. These resulted in noise in the dissimilar figures obtained from the financial statements of entities in unrelated sectors. Consequently, OLS and Pooled OLS are inefficient to resolve endogeneity, heterogeneity, reverse causality, and autocorrelation. A large amount of literature cited in the study deployed similar responses and explained variables (Yikarebogha & Adesola, 2025; Beida, 2024; Adegbayibi et al., 2024b; Ogunbiyi-Davies et al., 2023; Raghavendra, 2022; Ogundajo et al., 2022; Ovedje & Iserien, 2021; Y. S. Akinjare et al., 2019; Leyira et al., 2012; Enyi & Akindehinde, 2014).

The response variable (ROA) represented the firm’s performance indicator, while the explanatory variables captured elements of HRC to influence the response outcomes. The three major inferential estimation techniques were employed to ensure robustness and capture different model assumptions: OLS and Pooled OLS, which assumed homogeneity across firms and time, and Arellano GMM xtabond2, which addressed endogeneity. OLS measured relationships, and Pooled OLS captured fixed effects (FE) and random effects (RE) models to control unobserved time-invariant heterogeneity within firms. The Hausman test was employed to determine the appropriate model between FE and RE, and VIF for multicollinearity. In addition, the Breusch–Pagan/White test was conducted to measure heteroskedasticity. Also, the Wooldridge test was deployed for serial correlation. This was utilized alongside goodness-of-fit statistics such as R², adjusted R², and F-statistics. E-view 9 was used to run Descriptive Statistics and co-integration tests between the variables. As a result of the limitations of E-View 9 in running a mathematical method, the Stata 13 software package was engaged to run Arellano Generalized Method of Moments (GMM) xtabond2 to address endogeneity-related issues. The methods provided a robust framework for analysing both within-firm and between-firm variations over time, ensuring consistent and efficient parameter estimates in the presence of firm-level heterogeneity.

Unlike many empirical studies in developed, emerging, and developing economies in Nigeria and other sub-Saharan Africa countries, the study steadily progressed from Descriptive Statistics, OLS, Pooled OLS, to Arellano Generalized Method of Moments (GMM) xtabond2. From the empirical evidence reviewed, it appeared that none of the studies combined OLS, Pooled OLS, and GMM xtabond2 in Nigeria and other developing nations. The common methods deployed were Descriptive Statistics, OLS, and Pooled OLS, which appeared inappropriate and deceptive as a result of reverse causality, autocorrelation, heterogeneity, and endogeneity-related balanced and unbalanced datasets of firms in dissimilar sectors with different accounting policies, (Olusanya et al. (2016); Hansen (1982); Chamberlain (1987); Russel and MacKinnon (1993); Ruud (2000)). However, in India (an emerging economy), Barak and Sharma (2024) deployed Arellano GMM xtabond2 to resolve the noise and inappropriateness of OLS and Pooled OLS with cross-sectional data, where the nature of business and accounting policies were not similar. The classical OLS analysis relied on the pillar of non-endogeneity of the variables. This portends that none of the explanatory variables should explain the other explained variables, where this occurs; it is suggestive of multicollinearity of the variables. There is also the probability of reverse causality between ${ROA}_{it}$ and ${LEQ}_{it}$. As previously denoted as endogenous and exogenous variables, either of ROA or LEQ can be used as an explanatory variable. Thus, the study rigorously addressed econometrics issues earlier highlighted with the use of Arellano GMM xtabond2. According to Olusanya et al. (2016), Hansen (1982). Chamberlain (1987); Russel and MacKinnon (1993), and Ruud (2000), the Arellano Generalized Method of Moments (GMM xtabond2) employed in the investigation is a novel non-static model. In fact, Olusanya et al. (2016) revealed that it is the most efficient technique in estimating panel data. The method is dynamic, flexible, and reliable for small and large sample sizes and unbalanced panel data that were used in the study. In addition, the model is simple and highly predictive in both the short and the long run. Also, the most flexible and appropriate among the non-static and instrumental techniques are the Two (2) Least Squares Approach and Generalized Method of Moments (GMM) (Olusanya et al., 2016). However, the Two (2) Least Squares Method was not adopted in the study because of its mathematical complexity, inflexibility, and lack of robustness. The GMM provided a plausible alternative to automatically determine instrumentations; unlike the former, which calculates instruments using difficult processes.

3.1. Model Specification and Variable Definitions

The variables of this study are well defined and specified in Equations (1)–(4).

$${ROA}_{it}={\beta }_0+\sum _{j=1}^6{\beta }_j{Y}^{\prime }+\sum _{j=7}^9{\beta }_j{Q}^{\prime }+\sum _{j=10}^{13}{\beta }_j{T}^{\prime }+{\epsilon }_t$$

(1)

$${ROA}_{it}={{\beta }_1∆ROA}_{it-1}+\sum _{j=2}^7{\beta }_j{∆Y}^{\prime }+\sum _{j=8}^{10}{\beta }_j{∆Q}^{\prime }+\sum _{j=11}^{14}{\beta }_j{∆T}^{\prime }+{∆\epsilon }_t$$

(2)

Expanding Equations (1) and (2) to capture a priori expectations of exogenous variables resulted in Equation (3) as:

$${∆ROA}_{it}=+\sum _{j=0}^P{\beta }_1∆{ROA}_{it-j} +\sum _{j=0}^P{\beta }_2∆{LEQ}_{it-j} -\sum _{j=0}^P{\beta }_3∆{LSW}_{it-j}-\sum _{j=0}^P{\beta }_4∆{LTD}_{it-j}+{∆\epsilon }_t$$

(3)

Simplifying Equation (3) to have Equation (4)

$${ROA}_{it}=K_0+K_1\ast {LEQ}_{it}-K_2\ast {LSW}_{it}-K_3\ast {LTD}_{it}+m_{it}$$

(4)

3.2. Empirical Inferential Models Deploying Arellano GMM xtabond2

The empirical inferential models, deploying Arellano GMM xtabond2, are defined in Equation (5).

$${∆ROA}_{it}=+\sum _{j=0}^P{\beta }_1∆{ROA}_{it-j}+\sum _{j=0}^P{\beta }_2∆{LEQ7.30}_{it-j}+\sum _{j=0}^P{\beta }_3∆{18.40LSW}_{it-j}-\sum _{j=0}^P{\beta }_4∆22.63{LTD}_{it-j}+{∆\epsilon }_t$$

(5)

• i = 1, 2, 3, …, N denotes the cross-sectional units as earlier defined in the study.
• t = 1, 2, 3, …, T denotes the time periods (years). $m_{it}$ refers to the composite error term that can be expressed as $m_{it}$ = $u_i$ + $r_{it}$, where $u_i$ captures unobserved firm-specific effects and ${ris}_{it}$ was the idiosyncratic error term. Equations (1)–(3) captured the rate of responses of the lagged values of the exogenous variables to the endogenous variables. The rate of responses and degree of significance of one (1) of the three (3) exogenous variables-${LEQ}_{it}$, ${LSW}_{it}$ and ${LTD}_{it}$ in firms I at time t measured the contributions of HRC to the wealth (${ROA}_{it}$) of the sampled firms. A priori, it is expected that the series could be positive or inverse; this is supported by the theories and empirical studies (Barak & Sharma, 2024). A priori, it is technically expected that: $K_1$ > 0, the higher equity (${LEQ}_{it}$) the more the levels of wealth (${ROA}_{it}$). This implies more equity funds. However, $K_2$ < 0, $K_3$ < 0 are the coefficients of expenses in the Income Statement as contained in IAS 1 on Presentation of Financial Statements. The more the variables, ${LSW}_{it}$ and ${LTD}_{it}$, the less the wealth (${ROA}_{it}$) to equity holders of the selected firms.

The dataset consisted of one (1) explained ${ROA}_{it}$. It refers to the Return on Assets of firm I at time t and three (3) explanatory variables defined as: ${LEQ}_{it}$, which is denoted as the Log of Equity (firm’s equity size capitalization or wealth), ${LSW}_{it}$, which measures the log of Salaries and Wages, and ${LTD}_{it}$, which captures the log of Training and Development Cost (a labour related cost). ROA was deployed as a dependent variable, in compliance with Ekwe (2012) because it is a measure of overall efficiency and LSW and LTD, which denote remuneration for the productivity of Human Resource (HR) of entities. The variables (${ROA}_{it}$, ${LEQ}_{it}$, ${LSW}_{it}$ and ${LTD}_{it}$) were operationalized in line with empirical studies such as Yikarebogha and Adesola (2025), Okorie and Ebere (2025), Enyi and Akindehinde (2014), Beida (2024), Adegbayibi et al. (2024a) among others. ${ROA}_{it}$ was measured with Profit After Tax as a ratio of Total Assets in the SOFP, ${LEQ}_{it}$ denoted Log of Shareholders’ Funds reported in the SOFP, ${LSW}_{it}$ measured Log of annual Salaries and Wages and ${LTD}_{it}$ captured Training and Development Cost of the employees reported in the Income Statements. Logs were introduced for the numbers to be on the same basis as the others to avert outliers. A large number of studies engaged variables such as Salaries and Wages, Directors’ Salaries, Training and Development Costs, Medical Bills, Pension Cost, Gratuity Cost, Medical, Labour Productivity, and Labour Turnover, among other payroll costs. These explanatory variables are highly correlated, and their effects are inseparable. Thus, a Replacement Method using Correlation Matrix was engaged, and the explained variables with low matrices were eliminated to eliminate the consequences of multicollinearity of explanatory variables. In addition, the figures for ${LEQ}_{it}$, ${LSW}_{it}$ and ${LTD}_{it}$ in the sampled firms were higher than those for the other variables. This is to eliminate outliers in the performance metrics (ROA) of the sampled entities.

4. Results

Table 1. Descriptive Statistics of the Dataset.

	ROA	LEQ	LSW	LTD
Mean	68.65313	7.866962	10.20179	5.880944
Median	8.819345	7.509148	6.829678	5.697552
Maximum	12,763.64	12.25678	30.38617	10.25916
Minimum	−75.17682	2.687560	4.117309	2.579923
Std. Dev.	522.3251	1.729826	7.974908	1.614862
Skewness	22.45920	0.789058	1.640744	0.407582
Kurtosis	543.7546	3.584492	3.851070	2.877979
Jarque–Bera	7,900,625	75.99411	308.3816	18.23007
Probability	0.000000	0.000000	0.000000	0.000110
Sum	44,212.61	5066.324	6569.951	3787.328
Sum Sq. Dev.	1.75 × 108	1924.048	40,894.26	1676.801
N	644	644	644	644

Source: Author’s output from E-View 9 (2025).

reports the Descriptive Statistics (behaviour) of ${ROA}_{it}$, ${LEQ}_{it}$, ${LSW}_{it}$ and ${LTD}_{it}$ in firms, I at time t. The sample size was based on 644 observations. The table measured the spread and checked the centre of the sample size of 644 with Kurtosis, skewness, and examined the normality of the distribution. All the variables showed Kurtosis of 3 or approximately 3, suggesting that the data had higher tails, which clustered towards the mean with a likelihood of outliers. The Skewness is a measure of the extent of asymmetry in data distributions. A dataset is skewed to the right (positively skewed) when the Mean exceeds the Median; that is, most of the observations moved nearer to the lower values, and a few of the extremely large values skewed the distribution to the right. Where it is in opposite directions (Mean value less than Median), it means the data is negatively skewed, implying that there are extremely low values that dragged the distribution to the left.

The Mean values of variables (ROA, LEQ, LSW, and LTD) are higher than the Median, indicating the data are positively skewed to the right. This implies that the distributions are characterized by relatively small values, with a few large values having a significant effect. In situations where conventional procedures are deployed, the skewness of the distribution has implications of biased statistical inference, inefficient parameter estimates, and unreliable hypothesis testing. The skewness of the data towards the right is high in Table 1, especially the ROA; thus, some remedial actions are needed to redress the situation. To address the outlier indicated, robust Econometrics tools, Pooled OLS, and Arellano xtabond2 Generalized Method of Moments (GMM) were deployed in the study. In Table 1, the Central tendency metrics showed that LEQ has a symmetric distribution with a Mean of approximately 7.87 and a Median of 7.51. The p-value is less than 5%. It means that LTD has a symmetric distribution with a Mean of 5.88 and a Median of 5.70. Thus, significant differences exist between the Means of LSW and ROA (10.20 and 68.65) and Medians (6.83 and 8.82), respectively. The ROA skewness is notably significant. This is indicative that there are most likely firms with extremely high profitability, which skewed the total distribution. This portended that LSW influenced performance values (ROA) as authenticated by Barak and Sharma (2024) and Yikarebogha and Adesola (2025), where Salaries and Wages significantly improved the performance. A symmetric distribution is one whose Mean, Median, and Mode are very similar behaviourally. That is, both sides of the distributions are equal in terms of deviations from the Mean. In this situation, the skewness is near zero, and classical methods of estimation are known to work efficiently in this zone.

LEQ and LTD have minor positive skewness of 0.79 and 0.41. ROA has high skewness (22.46) due to huge outliers, but LSW has considerably lower positive skewness (1.64). LSW (3.85) and LEQ (3.58) imply that the distribution is leptokurtic. This means that the distribution is in the extreme values, whereas LTD (2.88) is approaching a normal distribution. The data appeared not symmetrical on account of the fact that the Mean and Median values differed, and the skewness of the coefficients was positive. The standard deviation determines the degree to which the observations are away from or around the Mean values. In a situation where the deviation is high, it indicates the likelihood of outliers. Except for ROA with a high standard deviation of 522.33 and LSW with a greater standard deviation (7.97), which indicated dispersion and strong outliers, LTD (1.61) and LEQ (1.73) showed minimal variability, indicating closeness to their averages.

Kurtosis indicates the extent of concentration of the observations around the Mean and the thickness of the tails of the distributions. The threshold of kurtosis is 3, ROA (543.75), LEQ (3.58), LSW (3.85), and LTD (3) were greater than 3 or approximately 3, indicating a heavy-tailed distribution with outliers. This portended leptokurtosis distributions with a sharp peak and heavy tails. In a nutshell, it means that there is a very high concentration around the mean with more chances of having extreme values. The Jarque–Bera statistic gives an indication of normality and goodness-of-fit. It is a joint test of both skewness and kurtosis. All variables reported in Table 1 have a p-value of virtually zero. This is less than 0.005. Hence, the null hypothesis of normality is statistically rejected. It suggests the distributions were largely not normal because they are asymmetrical and have excess kurtosis. Consequently, further robust influential Statistics (Pooled OLS, Correlation Statistics, and GMM) were engaged to remedy the outliers reported in Table 1.

The Correlation Statistics in

Table 2. Correlation between the variables.

Correlation
t-Statistic
Probability	LEQ	LSW	LTD	ROA
LEQ	1.000000
LSW	0.873014	1.000000
	45.35667
	0.0000
LTD	0.760490	0.841169	1.000000
	29.67439	39.41266
	0.0000	0.0000
ROA	0.204132	0.230338	0.170120	1.000000
	5.283497	5.997504	4.374204
	0.0000	0.0000	0.0000

Source: Author’s output from E-View 9 (2025).

measured the intensity and the direction of the linear relationship between the variables, where −1 to +1 are the values. As evidenced in the reported matrix, the explanatory variables of LEQ (0.8730), LSW (0.7605), and LTD (0.8412) showed very strong correlations with one another. These are all statistically significant at p-values of 0.000. The values are greater than the threshold of 0.70. This implies that the regressors are very interdependent. In addition, the correlation (0.1701, 0.2303, and 0.204132) between each explanatory and the explained variables (ROA) appeared weak and statistically significant at p-value (1.00000) higher than 0.05. Thus, the joint impacts among endogenous and exogenous variables might be significant in the explanation of the variation in performance. The extreme levels of correlations between the regressors and weak correlation with ROA indicate the existence of strong multicollinearity. This justifies why further tests were employed to obtain reliable and unbiased estimates of the regression parameters.

This study employed the Group Unit Root test, Levin, Lin and Chu (LLC), Im, Pesaran and Shin (IPS), Fisher type tests, or Hadri tests to detect whether the variables were stationary or have a unit root. The non-stationary variables have the possibility of giving spurious regression results and weak predictions. The null hypothesis in Panel data is whether a variable has a unit root or is stationary, and it is tested. The threshold of p-value ≤ 0.05 causes a null hypothesis to be rejected, indicating that the variable is stationary. On the other hand, p-value > 0.05 indicates an inability to reject the null hypothesis. It means that the variable is non-stationary, and it may be necessary to differentiate or transform it. As revealed in

Table 3. Group Unit Root Tests.

			Cross-
Method	Statistic	Prob. **	Sections	Obs
Null: Unit root (assumes common unit root process)
Levin, Lin & Chu t *	1.96762	0.9754	27	590
Null: Unit root (assumes individual unit root process)
Im, Pesaran and Shin W-stat	2.39876	0.9918	27	590
ADF—Fisher Chi-square	69.4890	0.0763	27	590
PP—Fisher Chi-square	120.677	0.0000	27	617

Note: t * refers to the adjusted t-statistics, while Prob. ** implies that the test is significant at 5%. Source: Author’s output from E-View 9 (2025).

, the Group Unit Root tests were employed to determine whether or not the variables in the Panel were stationary or had a unit root. This is important in ensuring that a valid, effective, and efficient regression results. Both the Levin, Lin, and Chu (LLC) test and Im, Pesaran, and Shin (IPS) test were based on the assumption of common and individual unit root processes, respectively, and the p-values are very high (0.9754 and 0.9918), showing a failure to reject the null hypothesis of a unit root, suggesting that the datasets were not stationary. Thus, the study progressed from OLS to Pooled OLS and Arellano GMM xtabond2 to further address the nonstationary-related issues.

The stationarity test (

Table 4. Stationary test.

Variables	Order of Integration
LEQ	I(1)
LTD	I(0)
LSW	I(0)
ROA	I(0)

Source: Author’s output from E-View 9 (2025).

) determines whether a variable’s statistical properties, like the Mean, Variance, and Autocovariance, are constant over time to ensure reliable regression analysis. Variables with p ≤ 0.05 are considered stationary at I(0), while those with p > 0.05 contain a unit root and require differencing. In this study, LEQ is I(1), meaning it is stationary after first differencing; however, LTD, LSW, and ROA are I(0), meaning stationary at levels. This ensures adequate model specification and avoids spurious results.

Table 5. Cross-section dependence test.

Test	Statistic	d.f.	Prob.
Breusch–Pagan LM	685.6819	351	0.0000
Pesaran scaled LM	12.63175		0.0000
Pesaran CD	2.066172		0.0388

Source: Author’s output from E-View 9 (2025).

reports the Residual Cross-Section Dependence Test. The Breusch–Pagan LM Test is commonly used in panel data to determine whether pooled OLS is appropriate or whether panel effects exist. It revealed a substantial correlation in the residuals of the Panel model. The Breusch–Pagan LM test statistic of 685.6819 with a p-value of 0.0000 rejects the null hypothesis of no cross-sectional dependency. This indicates that there is a panel effect and significant cross-sectional residual variance, and these are linked and related. The Pesaran scaled LM statistic of 12.63175 (p-value = 0.0000) rejects the null hypothesis. It confirmed the existence of cross-sectional of the datasets. The Pesaran CD test results are 2.066. The p-value of 0.0388 is below the 0.05 level of significance. This also affirmed significant cross-sectional dependence. As a result of entities in dissimilar sectors and different presentations in the Financial Statements (Accounting Policies), the model progressed from Descriptives (behavioural), Ordinary Least Squares (OLS), Pooled OLS, Hausman Test, and Arellano xtabond2 GMM to further scrutinize the outliers in unbalanced time series and cross-sectional datasets output reported in

Table 6. The parameter estimation progressing from the OLS, the Pooled OLS, and the DGM xtabond2.

Variables	OLS	RE	DGMM xtabond2
ROA: L1			−0.0446972
			(0.0495239)
t-Stats			−0.90253797
Z-Stats			−0.90
p-value			0.368
LEQ	6.610913	8.993237	7.299119
SE	(23.90792)	(29.40043)	(27.67398)
t-Stats	0.2765156		0.2637539
Z-Stats			0.26
p-value			0.792
LSW ***	18.39456	17.46306	18.39779
SE	(6.227057)	(40.02479)	(6.791579)
t-Stats	2.95397328		2.7089120
Z-Stats			2.71
p-value			0.007
LTD	−26.77267	−117.1410	−22.627
SE	(23.09628)	(58.69031)	(23.87116)
t-Stats	−1.1591767		0.9478802
Z-Stats			−0.95
p-value			0.343
C	−13.56346	−508.6490	-
SE	(166.3183)	(456.3083)
R-Squared	0.055078	0.133560
Adjusted R-Squared	0.050649	0.092637
Durbin-Watson Stats.	2.058577	2.241370
F-statistics	12.43494	3.263686
Prob(F-statistics)	0.000000	0.000000
MSE	508.9256	497.5440
AIC	15.30867	15.30271
Wald Chi-Square (36)			66.35
Prob > Chi-Square			0.002
Arellano-Bond test for AR(1) in first differences			z = −7.56 Pr > z = 0.000
Arellano-Bond test for AR(2) in first differences			z = −0.70 Pr > z = 0.483
Sargan test of overidentifying restrictions			Chi2(32) = 6.22 Prob > Chi2 = 1.000
Sargan test excluding group			Chi2(31) = 4.02 Prob > Chi2 = 1.000
Difference (null H = exogenous)			Chi2(1) = 2.19 Prob > Chi2 = 0.139

Source: Author’s output from E-View 9 and Stata 13 (2025). *** implies that the test is significant at 1%.

and

Table 7. Hausman test.

Test Summary		Chi-Sq. Statistic	Chi-Sq. d.f.	Prob.
Cross-section random		2.479587	3	0.4790
Cross-section random effects test comparisons:
Variable	Fixed	Random	Var (Diff.)	Prob.
LEQ	8.993237	8.303527	168.252475	0.9576
LSW	17.463056	21.506004	1541.802682	0.9180
LTD	−117.141008	−48.214784	2345.916313	0.1547

Source: Author’s output from E-View 9 and Stata 13 (2025).

. This is to present reliable and predictive empirical models to avoid misleading conclusions.

The OLS regression was reported in Table 6 in the first column. As a result of inherent limitations of OLS, as previously indicated in Olusanya et al. (2016) and Ruud (2000), among others, the coefficients showed minimal and inefficient explanatory (positive and negative) powers of LEQ (6.61), LSW (18.39), and LTD (−26.77) on ROA from 1990 to 2023. Notice that LSW was significantly positive with a t-statistic of 2.95, greater than the threshold of 1.96; LEQ, with a lower t-statistic of 0.28, and LTD (−1.16) were insignificant. The significance and positivity LSW portended that for every one naira increase in performance proxied with ROA, there is a corresponding increase of 18.39, whereas LTD (−22.77) is inversely insignificant over 34 years across 27 firms in different sectors. This evidence is supported by Rong et al. (2025). Ogunbiyi-Davies et al. (2023) and I. R. Akintoye (2012), among other empirical studies cited in the study, where OLS was deployed with similar results. Also, the low F-statistic (12.43) with a p-value of 0.0000 lower than 0.05 was significantly not a good fit. Similarly, the R-Squared and Adjusted R-Squared (0.055 and 0.050) were very low. This suggests the model’s proportion of the variance was not reliable. The Durbin–Watson statistic value of 2.24 was approximately 2, affirming no autocorrelation. The study is also corroborated in other similar studies; for example, Salaries and Wages were positively significant on the Return on Assets in the empirical studies of Omole et al. (2017b), Akinlo and Olayiwola (2017), Ogunbiyi-Davies et al. (2023), Adegbayibi et al. (2024a), and the training cost was significantly higher in Adegbayibi et al. (2024b), Olowolaju and Oluwasesin (2016), among other studies. However, in the OLS results, cross-sectional heterogeneity was detected, the overall linear relationships were doubtful, necessitating the progression from OLS to Pooled OLS and Arellano GMM xtabond2.

In addition, the Random Effect (RE) parameters reported in column 2 of Table 7 indicated mixed reactions where LEQ (8.99) and LSW (17.46) were significantly positive, whereas LTD (−117.14) was inversely significant on ROA. It could be inferred that the higher the LEQ and LSW, the more the LTD was significantly inverse to ROA. The F-Stats (3.26) with p-value (0.000000) less than 0.05 is statistically significant, suggesting the model is a good fit and linear. The Durbin–Watson statistic value of 2.24 is close to the threshold of 2, authenticating the absence of autocorrelation. This evidence was endorsed by Omole et al. (2017a) and Balogun et al. (2020), who also engaged in Pooled OLS. However, this is not sufficient because of the endogeneity and outliers in the Pooled datasets. Hence, the study progressed to a more reliable Arellano GGM xtabond2 as used in Barak and Sharma (2024) to resolve endogeneity, heterogeneity, and other related noise in unbalanced and balanced panel data. The Hausman test reported in Table 7 was used in the panel data analysis to decide between FE and RE estimators. This is to test the key assumption of the model with respect to the fact that the unobserved individual-specific effects are uncorrelated with the regressors. If the p-value of the Chi-square statistic ≤ 0.05, the null hypothesis is rejected, affirming that the regressors are correlated with the individual effects and that the FE model should be utilized; otherwise, the RE estimator is appropriate. The results from Table 7 revealed that the Hausman test with a p-value of 0.4790 is higher than a 5% significance level. This indicates the RE model is preferred over the FE model for the dataset. This portended that LSW (18.40) was positively significant and LTD (−117.14) was negatively significant. Thus, LSW exerted a positive influence on performance, justifying the inclusion of HCC in the SOFP. The relationships among the variables are mixed. The good news is that OLS and Pooled OLS aligned with the same relationships, and LSW exerted a positive shock on ROA (performance), supporting earlier inferences on the definition of assets and why HRC elements are expensed annually in the Income Statements. This was earlier certified by Ovedje and Iserien (2021) and Okorie and Ebere (2025), where OLS and Pooled OLS were employed with similar elements of HCCs and ROA as variables. The Pooled OLS ignores unobserved individual heterogeneity, endogeneity, and reverse causality as earlier indicated in Olusanya et al. (2016). Consequently, Arellano GMM xtabond2 was engaged to further examine time series properties of the datasets to enhance efficiency and predictability. This evidence is confirmed and ratified in the empirical studies of Asutay and Ubaidillah (2023), Adegbayibi et al. (2024a), Barak and Sharma (2024), Rong et al. (2025), Yikarebogha and Adesola (2025), Enyi and Akindehinde (2014), where elements of HRCs were jointly and individually statistically significant in creating wealth in the sampled entities.

To resolve the limitations of OLS and Pooled OLS, the study adopted a parsimonious dynamic panel specification consistent with prior GMM-based corporate performance studies. This study did not consider the inclusion of lagged dependent variables and firm-specific effects in the system because the GMM framework helps to control the consequences of unobserved heterogeneity and endogeneity. The Arellano GMM xtabond2 model certifies the conditions such as correct dynamic specification, valid moment conditions, absence of second-order serial correlation, and controlled instrument count, among other robust assumptions. Stata 13 was deployed to run the model Arellano GMM xtabond2 one-step system equation with collapse options, the equation without collapse options, and one-step difference with and without collapse options. The results were similar. Thus, the study settled for only the system equation with collapse options presented in Table 7 under column 3 to make inferential decisions for predictions.

In the table, Wald ${\chi }^2$ = 66.35 with a p-value (0.002) of less than 0.05 clearly authenticated strong joint significance of the regressors on ROA. This gives credence to the fact that LEQ, LSW, and LTD significantly and jointly influenced ROA (wealth). Hence, it could be inferred that the explanatory variables significantly influenced ROA (performance). Similarly, all models satisfy AR (2), implying no misspecification. The Sargant test reported is p = 1.00. This indicates that the models (both the collapsed and non-collapsed) are good. However, the collapsed models reduced the instruments to preserve orthogonality conditions and restore power to over-identification tests. Hence, the collapse condition is preferred. In addition, the difference-in-Sargan test (exogeneity tests) showed all p-values > 0.10. This implies that the Arellano system GMM xtabond2 is preferred. Furthermore, the joint significance of the crucial explanatory variables was further scrutinized individually to validate inferential decisions and assertions made in OLS and Pooled OLS. LSW (18.40) with Z-Stats (2.70) > than the threshold of 2 and p-value (0.007) less than 0.05 exhibited a positive, strong, and meaningful effect on ROA. This clearly aligned with earlier questions in the treatment of rewards to labour as expenses rather than as assets in the Financial Statements. This positive association is endorsed in the empirical evidence of Barak and Sharma (2024), where GMM was employed, Okorie and Ebere (2025), Adegbayibi et al. (2024a), and Enyi and Akindehinde (2014), where OLS and Pooled OLS with Salaries and Wages and Training costs among other HCC elements reported were employed. However, it is not surprising that LEQ (7.30) is positive but insignificant, and LTD has a negative, insignificant coefficient. It conforms with the a priori expectations. An expense has an inverse relationship with the performance (profits). If it is positive, it implies that labour productivity as reported in Ajayi (2003) is >than the cost, benefits of labour > amount expended on human resources. This further validates the recognition of the Human Resource Cost elements in the SOFP rather than as an expense in the Income Statements. The Arellano–Bond tests further give credence to the model reported in the study. It showed insignificant AR(2) (p = 0.483 > 0.05). This confirmed the correct dynamic specification and absence of second-order serial correlation. Besides, the Sargan and Difference-in-Sargan test results, with p ≥ 0.05, indicated that the instrument set is valid and that the additional level moment conditions are exogenous. Hence, all the diagnostic tests confirmed that the Arellano System GMM xbond2 estimator with collapsed instruments is appropriate. It provided reliable estimates for the dynamic panel model (Arellano & Bond, 1991). The study is novel as a result of the progression from OLS, Pooled OLS, and Arellano GMM xtabond2 models, with the results aligning with previous empirical studies cited to confirm joint positive significance and individually mixed significant and insignificant reactions to spur IASB to issue standards on the treatment of proxies of Human Resource Costs (HCC) as assets in the SOFP.

According to IASB (2012), assets denote resources entities controlled as a result of events of the past, which generate future cash inflows over the years, usually more than a year. Human Capital, just like animal, physical, and nonphysical (PPE in IAS 10, Inventories in IAS 2, Animal and Agricultural assets in IAS 41, Goodwill, Patents, Software in IAS 38), fits into this definition because it also generates cash inflows over the years, thus should not be expensed annually. Similarly, in the developed economies, such as the United Kingdom, Spain, France, Germany, and Italy, among other developed nations, acquisition fees, contract extensions, and sign-on bonuses of club players, in the Football Economy, are treated as Intangible Assets in the SOFP, but the wages of the players are expensed in the Income Statements annually. The contributions of Human Resource Capital in any entity cannot be overemphasized in wealth and profit generation. Factually, other assets cannot function without human efforts. Even in an automated environment, human efforts still trigger the use of Artificial Intelligence (AI). Thus, it questioned why Human Resource, which also fits into definitions of assets in IAS 2, 10,41 and 38, does not have a generally accepted Accounting Standard to prescribe the treatment in the SOFP like other assets.

5. Conclusions

The empirical study had painstakingly examined the effect of Human Resource Costs (HRC) on the performance of 27 sampled entities across all sectors based on secondary data availability in the Nigerian Exchange Group (NGX). This progressed from OLS, Pooled OLS, to Arellano GMM xtabond2. OLS assumes that regressors are exogenous; GMM explicitly handles endogeneity by using internal instruments (typically lagged values of the variables), thereby producing consistent estimates even when regressors are correlated with the error term. Unlike Pooled OLS, which ignores individual-specific effects, GMM eliminates unobserved heterogeneity through transformations such as first differencing or orthogonal deviations. A lagged dependent variable causes bias in OLS (Nickell bias); however, GMM provides consistent estimation by instrumenting the lagged dependent variable with its deeper lags. Additionally, GMM is robust to heteroskedasticity and autocorrelation, as it does not rely on strict classical assumptions about the error term. This makes the GMM model predictive and efficient for use by policy makers and Accounting Standard setters. The results from the Pooled OLS with Hausman tests (0.4790) greater than 0.05 accepted that Random Effect (RE) with proxies of reward for labour, LSW (21.51), exerting positive shocks on ROA, whereas LTD (−117.14) was inverse on the explained variable. To enhance reliability and predictability, the results from the OLS and Polled OLS were further scrutinized with Arellano GMM xtabond2. The results from the GMM showed that LSW (18.40), positive, and LTD (−22.63), inverse, and Wald ${\chi }^2$ = 66.35 with p-value (0.002) less than 0.05 clearly endorsed strong joint significance of the regressors on ROA (performance). The result further poses a challenge to the International Accounting Standard Board (IASB) on why elements of HCCs are expensed annually, whereas animal, physical, and non-physical assets are recognized in the Statement of Financial Position (SOFP).

Based on the mammoth and substantial empirical evidence from many developed, emerging, and developing economies and the robustness of the evidence from this study, IASB should be encouraged to put in place machinery to issue an Exposure Draft on the treatment of HCCs as assets in the SOFP. The IASB should issue Accounting Standards to treat HCCs as assets in the SOFP. In addition, the IASB initiates debates among academia and Professional Accounting Bodies on how to classify Human Resource Costs as either physical or intangible assets. Really, Human Resource Cost (HRC) elements, just like physical, animal, and intangible assets, should be recognized in the SOFP. Capitalizing the Human Resource Cost is useful across the world because, as of now, assets are presently understated in the SOFP by expensing Labour Costs, which have more than one year. Economically useful life, like other assets, is defined in the IAS issued by the IASB. The empirical evidence is unanimous on the joint significant impact of HRC elements on the Return on Assets and other performance of entities across countries, but the individual impact is mixed. Consequently, as a springboard to future research, academia should consider combining control variables and HRC elements across borders using dynamic models such as GMM or 2SLS methods to sufficiently address outliers in a cross-border cross-sectional time series dataset. The unavailability of datasets is a challenge in developing nations; however, it does not affect the outcome of this study because GMM was employed to address the consequences.