Determinants of firm performance: How to choose the best model among panel data models?

Some determinants affect firm performance positively and some affect it negatively. Therefore, managers frequently look for determinants and factors that help them improve their firms’ performance. Many studies have investigated the firm performance of various sectors in the USA, Europe, the Middle East, and other parts of the world. Several prior studies have explored how a range of determinants impacts firm performance. The studies have contributed to the existing literature on the determinants of firm performance that negatively or positively affect financial performance by considering different sectors. The service industries play an important role in the economy. However, research in the service industries is at a nascent stage compared to other sectors (Ahmed & Bhuyan, 2020). Therefore, our understanding of how to improve financial firm performance in the service industries is still limited (Augustyn, Elshaer, & Akamavi, 2021). Concerning firm performance, many researchers used static panel data models. On the other hand, several researchers used dynamic panel data models. Selecting an appropriate panel model was also tricky in prior studies. Several recent studies have employed a two-step system GMM, e.g. Ghardallou (2023), Yousaf (2023), Tiwari, Sayyad, Azam, and Sudesh (2023), Yousaf and Bris (2021), Fernández-López, Rodeiro-Pazos, and Rey-Ares (2020), Sharma, Bakshi, and Chhabra (2020), Anser et al. (2020) and Saini and Singhania (2018). However, the researchers did not explain well why the two-step system GMM is better for the studies, as the model selection should be based on adequate justification. These problems motivated me to conduct this research.

There is limited empirical literature on determinants of firm performance, specifically for firms in the service industries (Ahmed & Bhuyan, 2020). I selected the service sector in the current study as the sector plays an essential role in the Czech economy. According to World Bank statistics, ^[1] the share of service industries in the gross domestic product (GDP) of the Czech Republic was 54.26% in 2017, which has increased to 58.31% in 2021. Hence, the share of the service sector in the country’s GDP is increasing with the passage of time. Therefore, I selected six sectors related to the service industries to examine a comprehensive review. Examining the determinants influencing service industry firms’ performance was the main goal of this study. Moreover, the study suggests how to select the best model from a variety of panel data models by applying an appropriate methodology step-by-step. Using Bond, Hoeffler, and Temple (2001) recommended criteria, the results of this study revealed that the two-step system GMM method is the best model among various panel data models. Leverage, tangibility, firm size, and firm age negatively impact firm performance in the service industries, according to the major results of the two-step system GMM method. On the other hand, physical capital intensity and the capital turnover ratio positively correlated with firm performance.

The present study extends the existing limited literature on the determinants of firm performance, particularly in service industries. The study contributes to the literature, as it explored various variables that affect firms’ performance positively or negatively. I introduced variables that are underrepresented in the existing literature but are critical to firms in the service industries. The current empirical study demonstrates that to increase profitability and enhance performance, service industry-related firms should employ capital-intensive technology rather than labor-intensive technology. To the best of our knowledge, this is the first study that answered the questions comprehensively, such as how and why to choose an appropriate model among dynamic panel models. The study examined the effect of the determinants on the firm performance of the service industries by employing comprehensive dynamic panel models step by step. Furthermore, I explained how each selected determinant affects the firm’s performance from the financial and economic perspectives. Consequently, the current study contributes to the body of knowledge and practical applications from various perspectives.

After the introduction section, the remainder of the article will include four sections. The second section will present a literature review analyzing the theoretical background and hypotheses developed for the selected variables that play an essential role in determining the firm’s performance. The third section will discuss the research methodology by presenting database information, sample selection, and calculation of the selected variables. The empirical results section will contain detailed step-by-step research procedures. It gives a comprehensive discussion on how to select an appropriate model among the dynamic panel models. The last section will focus on the conclusion of the research with concluding remarks, limitations of the current study, and future research directions.

If the firms are performing well, then they hire labor, pay taxes, and promote innovations. Therefore, firm performance is a significant source of sustainable economic growth. The reason for the improved firm performance is a vital area of consideration for managers, academics, researchers, and policymakers. The service sector is an essential contributor to any economy’s gross domestic product (GDP) (Bouranta, Psomas, Suárez-Barraza, & Jaca, 2019; Augustyn et al., 2021). There are various determinants that affect firm performance. With the help of existing literature, this section of the current study discussed the determinants that affect the firm’s performance. I will describe them one by one in the following sub-sections.

Firms’ capital or financial structure plays a significant role in firm performance. To take advantage of the tax shields, firms occasionally choose to use excessive leverage. According to the pecking order theory, organization size should be positively related to financial leverage. Small firms run a greater bankruptcy risk than large firms. Therefore, small firms should increase their equity rather than leverage (Adair & Adaskou, 2015). On the other hand, pecking order theory suggests there is a negative relationship between TDTA and firm performance. Consequently, under the condition that all other factors remain constant, firms that borrow less can earn more profits (Agyei, Sun, & Abrokwah, 2020). Haddad and Lotfaliei (2019) argue that the trade-off theory posits that firms make a strategic decision regarding their optimal leverage level by seeking to maximize the interest tax shield. According to Abuzayed (2012), a firm has to keep a balance between leverage and profitability. Many researchers have investigated a negative relationship between TDTA and ROA (Appiadjei, 2014; Nguyen & Nguyen, 2015; Tang & Chang, 2015; Khan, Shamim, & Goyal, 2018; Fernández-López et al., 2020; Ghardallou, 2022; Yousaf, 2022; Niazi, Othman, & Chandren, 2023). The negative relationship revealed that firms with high leverage earn less profit. Based on the pecking order theory and the findings of the previous studies, I hypothesized:

The tangibility (TANG) is measured as the ratio of fixed assets to total assets in the present research. Firms decide to invest in areas where they anticipate higher returns. As a result, firms prefer to make long-term investments, which results in a reduction in fixed assets. By themselves, fixed assets are insufficient to provide earnings. A firm must use fixed assets to cover short-term liabilities if it is unable to generate enough cash. Hence, a high ratio of TANG proposes to creditors a high level of security that assists in liquidating more assets in case of a firm’s bankruptcy (Baker & Martin, 2011). Chen and Chen (2023) argue that the higher the tangibility value, the lower the financial performance. Wattanawarangkoon, Sinthupundaja, Suppakitjarak, and Chiadamrong (2022), Ahmed and Bhuyan (2020), Nguyen and Nguyen (2015), and Maçãs, Serrasqueiro, and Sequeira (2009) investigated a negative relationship between TANG and ROA. On the other hand, Ghardallou (2022) and Birhan (2017) explored the positive impact of TANG on ROA. Hence, based on the mixed findings in the previous literature on the relationship between TANG and ROA, I hypothesized:

Firm size (FRS) may affect ROA due to economies of scale. The firms that can benefit from economies of scale become larger and dominate other firms. Chandrapala and Knápková (2013) argued that larger firms enjoy economies of scale, their production’s average cost is lower, and their operational activities are more efficient. Thus, larger firms produce larger returns on assets. According to the institutional theory perspective, large firms may be unable to respond faster to market changes. Therefore, firm size may have some disadvantages (Weinzimmer, Esken, Michel, McDowell, & Mahto, 2023). Earlier studies exposed that larger firms have relatively fewer adjustment costs, and it is easy for them to access the credit markets to gain more benefits (Haefner, Palmié, & Leppänen, 2023). Fernández-LópezPazos et al. (2020) examined the positive impact of FRS on ROA. We may note the same results in del Mar Fuentes‐Fuentes, Quintana‐García, Marchante‐Lara, and Benavides‐Chicón (2023), Nguyen, Hoang, and Tran (2022), Ghardallou (2022), Li, Zeng, Ye, and Huang (2021), Ahmed and Bhuyan (2020), Molodchik, Fernandez-Jardon, and Barajas (2016), and Maçãs et al. (2009). However, Shirodkar, Rajwani, Stadler, Hautz, and Mayer (2022), Ullah, Pinglu, Ullah, Zaman, and Hashmi (2020), Ullah, Kashif, and Ullah (2017), and Masnoon and Saeed (2014) explored the negative relationship between FRS and ROA. A study regarding Czech firms by Yousaf (2023a) also examined the negative relationship between the variables and found that small and medium enterprises (SMEs) earn more profits than large firms. On the other hand, the findings revealed that the bankruptcy risk for Czech SMEs is higher than the bankruptcy risk for large firms. Based on the prior studies’ findings, I hypothesized:

Kuntluru, Muppani, and Khan (2008) and Chandrapala and Knápková (2013) used the CTOR to measure firms’ capital intensity. In the present research, I computed the CTOR as the ratio of net fixed assets to total sales. Chandrapala and Knápková (2013) argued that the lower value of CTOR might imply better efficiency in capital utilization and result in higher firm performance. Kuntluru et al. (2008) and Chandrapala and Knápková (2013) revealed that CTOR negatively affects the firm’s performance. However, I formulated the following hypothesis to test the relationship between the variables.

Barbosa and Louri (2005) employed physical capital intensity (PHCI) to capture the impacts of labor intensity on the variability of the firm’s performance. I computed PHCI as the natural logarithm of the ratio of capital to the number of employees to find the average capital per employee. Chandrapala and Knápková (2013) also employed PHCI to examine the impacts of PHCI on the performance of Czech firms, but the authors did not find significant results. The study by Barbosa and Louri (2005) focused on Portugal and Greece-based firms. The researchers reported mixed results. The scholars recommended that Greece-based firms could improve performance by choosing labor-intensive technology. Conversely, Portugal-based firms should choose capital-intensive technology to improve the firm’s performance. I proposed the following hypothesis regarding the PHCI and ROA relationship.

In this study, I calculated firm age (FAGE) as the natural log of the year the sample data were collected minus the year of the firm’s registration. Moreover, FAGE is a vital determinant in defining firms’ performance variation. New firms frequently go through an initial period of developing skills in management, production, and marketing. The trade-off theory assumes that older firms have more experience and better reputation, which can reduce agency costs by signaling the quality of potential investments. Chhibber and Majumdar (1999) argued that older firms are more experienced and perform better than new firms. The relationship between FAGE and firm performance was the topic of studies by Batra and Tan (2003), Kuntluru et al. (2008), Tran, Grafton, and Kompas (2008), Park, Shin, and Kim (2010), Charoenrat and Harvie (2013), Li et al. (2021), and Tanaka (2021), investigated the positive impact of FAGE on ROA. On the other hand, Yousaf (2023a) reported a negative relationship between FAGE and the firm performance of Czech firms, which means that the new Czech firms earn more profits than the old firms. We may find data on the same relationship between the variables in Sinha, Mishra, Rl, and Prabhudesai (2022), Li et al. (2021), Charoenrat and Harvie (2013), Park et al. (2010), and Chhibber and Majumdar (1999). As studies report contradictory relationships between the variables, I hypothesized:

Summarizing, the findings are contradictory regarding the impact of the selected variables on firm performance. Some studies show a significant negative relationship between the selected determinants and firm performance, while others report the opposite. These diverse results call for further studies to investigate the relationships between the selected variables and firm performance. Hence, I find it essential to explore the determinants that impact firm performance, positively or negatively, in the context of service industries. This study responds to these issues by examining Czech firms.

I obtained secondary data on non-listed service industry firms from the CRIBIS database. ^[2] In recent studies, numerous scholars have obtained data from the CRIBIS database, such as Belas, Gavurova, Kubák, and Novotna (2023) or Civelek, Krajčík, and Fialova. (2023). The unbalanced panel data covered the period from 2017 to 2021. I obtained it from six sectors. I based the analysis on five years of data due to limited availability in the CRIBIS database. Table 1 describes Czech firms’ information according to the sector’s names, the number (No.) of firms, and their percentage.

I included six independent variables and one dependent variable. Many scholars have used return on asset (ROA) as a proxy to measure firm performance in their studies, e.g. Bamiatzi, Bozos, Cavusgil, and Hult (2016), Shirodkar et al. (2022), Wattanawarangkoon et al. (2022), Niazi et al. (2023), Ogoe and Suzuki (2023), Ullah (2023) and Ozili (2023). By comparing three proxies of performance measures, return on equity, return on capital employed, and ROA, Yousaf and Dey (2022) revealed that ROA is the best proxy to determine Czech firms’ performance. Therefore, I considered ROA a dependent variable to measure the firm’s performance. Table 2 shows the complete details of the selected variables in the current study.

My panel data combined time series and cross-section data. Therefore, I would prefer the panel data model. To quantify the impact of the selected determinants on firm performance (Section 2), I employed the following regression Model. Many authors used the same econometric model with different determinants in their recent research (Tiwari et al., 2023; Yousaf, 2023; Tran & Vo, 2020).

In Model 1, ROA_it is a dependent variable, α represents the intercept, from $β1$ to $β7$ are coefficients of the independent variables, i = 1, 2, 3,…., n are the number of firms, and t is the time from 2017 to 2021. Moreover, $ŋi$ and $μit$ are the unobserved firm-specific effects and error term for firm i at time t, respectively. ROA_it-1 is the lag value of firm performance. According to Wintoki, Linck, and Netter (2012), the theory suggests that the lagged variable should be uncorrelated with the error term in Model 1. As I had panel data, I employed the same model (Model 1) to estimate the pooled OLS, fixed effect, and GMM models’ results. All these models are popular panel data models. Thus, by estimating the models, I subsequently chose the best model among the panel data models.

I estimated pooled OLS, a fixed-effect model, and dynamic panel models to explain the impacts of different variables on firm performance. Table 3 presents the descriptive statistics of both dependent and independent variables. I obtained all empirical results by employing STATA 16.0 software.

Table 3 shows that the mean and standard deviation of ROA were 0.07 and 0.12, respectively. The mean and median of CTOR were slightly different, but the mean and median of other variables were almost identical. The value of the standard deviation (S.D.) of PHCI was the highest. On the other hand, the value of the S.D. of ROA was the lowest among the selected variables. The mean and median of ROA were positive, which means that the Czech service industry firms earned profits during 2017–2021.

Table 4 shows the correlation coefficients and variance inflation factor (VIF) of the selected variables of the Czech companies from the service industries.

As Table 4 shows, TDTA, TANG, CTOR, and PHCI were negatively correlated with ROA. Conversely, FRS and FAGE were positively correlated with ROA. All the correlation coefficients were significant at the 5% significance level. We may note that the highest value of the correlation coefficient between PHCI and TANG, which was 0.72. Conversely, the lowest value of the correlation coefficient was −0.34 between PHCI and TDTA. I used the VIF test to detect multicollinearity among independent variables. Nachane (2006) argued that multicollinearity might be a severe problem if the value of the VIF is more than 10. However, all values of VIF were lower than 10 in Table 4, so there was no multicollinearity issue at all.

Moreover, I checked the variables’ stationarity by employing the Fisher-type unit root test to avoid spurious regression results. To test the unit-roots in the panel data, the null hypothesis is that all panels contain unit-roots. Maddala and Wu (1999) argued that the Fisher-type unit root test does not need balanced panel data. Therefore, everyone can apply the individual augmented Dickey-Fuller (ADF) test. The results of the test revealed that all the chosen variables were stationary, as the p-value of all variables was zero. ^[3] Therefore, I could run the regression equation (Model 1) by pooled OLS, fixed-effect, one-step difference, two-step difference, one-step system, and two-step system GMM estimations.

I estimated Model 1 through the above-mentioned techniques to sort out the best technique among dynamic panel data estimations as proposed by Bond et al. (2001). First of all, I estimated Model 1 by pooled OLS and fixed effects. I considered the corresponding fixed effects estimate as a lower-bound estimate and the pooled OLS estimate as an upper-bound estimate. Then, I estimated the model using four different GMM techniques. To select the best estimate, I analyzed the coefficient of the lag-dependent variable. If the difference in the GMM estimate of the lagged dependent variable was lower or similar to the fixed effects estimate, it indicated bias due to weak instruments. Therefore, we should employ a system GMM estimator, as suggested in the previous literature that a system GMM is better in this situation. Hence, I tested Model 1 step by step and explained each estimation from the financial and economic perspectives.

I run the pooled regression to estimate Model 1. Table 5 presents the results.

The results of pooled OLS in Table 5 revealed that the coefficients of lag of ROA, TDTA, TANG, CTOR, PHCI, and FAGE were statistically significant. The lag of ROA (Lag1.ROA) positively affected firm performance (ROA). Conversely, TDTA, TANG, CTOR, PHCI, and FAGE negatively impacted ROA. It means that firms with low values of these variables perform better. Goddard, Tavakoli, and Wilson (2005) revealed that OLS estimation is inappropriate as the inclusion of individual firm effects in the disturbance term makes non-zero covariance between the lagged dependent variable and the disturbance term. Therefore, I estimated Model 1 by employing the fixed-effect model approach. Table 6 presents the results of the static panel data (fixed-effect model).

The TDTA, PHCI, and FAGE coefficients were statistically significant and had a negative impact on the firm’s performance in the service industries. On the other hand, the lag of ROA (Lag1.ROA) had a positive relationship with ROA. The TANG, FRS, and CTOR coefficients were statistically not significant. The overall Model was significant, as the p-value of the Model was zero. Hayakawa and Qi (2020) claim that the fixed effects estimator is inconsistent when T (time) is small and N (observations) is large.

After pooled OLS and fixed-effect models, I employed difference and system generalized method of moments (GMM) to avoid endogeneity bias and capture dynamic panel effects. I further estimated the regression model (Model 1) with the help of the GMM estimator suggested by Arellano and Bond (1991). I used system GMM and difference GMM estimators for the econometric investigation of dynamic economic relationships. The characteristic of such panel data is “small T, and large N,” where T is time, and N denotes observations, which means that the basic data covers a total of N×T individual observations. The observed individuals might be countries, firms, or consumers. The difference and system GMM estimators are prevailing tools to estimate dynamic panel data models with autoregressive processes. Both estimators employ instrumental variables, which are obtainable from within the system of equations. I obtained the results from a one-step difference GMM. Table 7 presents the results.

According to Table 7, three variables are statistically significant at the 0.01 and 0.10 significance levels. The significant results revealed that the lag of the dependent variable (Lag1.ROA) and FRS are positively correlated with ROA. However, TANG has a negative impact on the firm performance of the service industries.

According to Kayani et al. (2020), to confirm the GMM results, certain conditions need to be fulfilled, such as AR (1) and AR (2). Furthermore, all the instruments must be valid, as detailed in the Hansen test (Hansen, 1982). The AR (1) Arellano-Bond test (Arellano & Bond, 1991) found that the average autocorrelation in residuals of order 1 was 0. Therefore, the null hypothesis was:

The AR (2) Arellano-Bond test found that the average autocovariance in residuals of order 2 was 0. Therefore, the null hypothesis was:

• H0: no autocorrelation.

In Table 7, the p-value (0.253) of the AR (2) statistic was not significant, which confirmed the lack of second-order serial correlation in the residuals. I performed the Hansen test for instruments’ validity using the Hansen statistic. The null hypothesis of the Hansen test is that the overidentification restrictions are valid. Failure to reject the null hypothesis supported the instrument’s choice. The p-value of the Hansen test was 0.220, which was above the significant conventional value. I could not reject the null hypothesis. It means that the results of GMM reported in Table 7 were valid (Roodman, 2009; Kiviet, Pleus, & Poldermans, 2017). Table 8 presents the outcomes of the two-step difference GMM.

In Table 8, there are only two variables that are significant at the 0.01 level. The significant results revealed that the lag of the dependent variable (Lag1.ROA) had a positive relationship with ROA. However, TANG had a negative impact on ROA. Moreover, AR (1), AR (2), and Hansen test results were the same as reported in the one-step difference GMM. Table 9 presents the outcomes of the one-step system GMM.

In Table 9, three variables are statistically significant at a 0.01 significance level. The significant results show that the lag of ROA (Lag1.ROA) and the constant term were positively associated with ROA. On the contrary, TDTA had a negative impact on the firm’s performance. Moreover, AR (1), AR (2), and Hansen test results were the same as reported in the one-step difference GMM. Table 10 presents the outcomes of the two-step system GMM.

In Table 10, various variables are significant at the 0.05 and 0.10 significance levels. The significant results show that TDTA, TANG, FRS, and FAGE were negatively associated with the proxy of firm performance. On the other hand, CTOR, PHCI, and the lag term (Lag1.ROA) positively affected ROA. In the following parts, I will elaborate on the results of Table 10.

Table 11 shows all the required results of the lag-dependent variable to choose the appropriate model among the dynamic panel models.

I employed difference and system GMM to control for the endogeneity bias in the dynamic panel models. According to Ullah, Akhtar, and Zaefarian (2018), the dynamic GMM models serve to address dynamic endogeneity bias in panel data. Bond et al. (2001), suggest the following model as appropriate among the panel data models. First of all, pooled OLS and fixed-effects approach should serve to estimate the dynamic model. Then, the researcher should consider the corresponding fixed effects estimate as a lower-bound estimate and the pooled OLS estimate – as an upper-bound estimate. If the difference GMM estimate obtained is below or close to the fixed effects estimate, then the difference GMM estimate is biased due to weak instrumentation. Therefore, a system GMM estimator should be the preferred choice. According to the findings in Table 11, the difference estimators were lower than the fixed-effect estimator. Hence, the results of system GMM should be ideal in this case. Grohmann (2015) investigated the article by Daron, Johnson, Robinson, and Yared (2008) and found that the difference GMM estimator in the article performs poorly when the dependent variable is persistent. Similarly, using the system GMM estimator, Bobba and Coviello (2007) investigated the positive and significant outcomes and showed that education has no positive impact on democracy. Thus, they opposed the popular opinion that education has a positive effect on democracy. Many scholars, such as Grohmann (2015), Baltagi (2008), and Blundell and Bond (2000), showed that system GMM estimators can usefully overcome and give better outcomes than difference GMM estimators for dynamic panel models. Olaoye and Afolabi (2021) also confirmed that the two-step GMM produces reliable estimates that are more asymptotically efficient than one-step GMM estimates. Furthermore, I observed that recent studies by Ghardallou (2023), Tiwari et al. (2023), Yousaf and Bris (2021), Sharma et al. (2020), Kayani et al. (2020), or Saini and Singhania (2018) used two-step system GMM. All the above discussion revealed that scholars should prefer a two-step system of GMM results. Therefore, I will briefly discuss the findings of the two-step system GMM (Table 10) to conclude the determinants that affect the firm performance of the service industries.

Table 10 shows that TDTA and ROA had a negative relationship (supporting H1). It means that the capital structure consists of more debt, which decreases the firm performance of the service industries. The explanation is that the excess debts decline the firm’s value and increase the financial distress costs. It is true that if firms have overleveraged themselves, then performance will decrease. The findings resemble those of Fernández-López et al. (2020), Khan et al. (2018), Tang and Chang (2015).

The results of TANG revealed that the coefficient of TANG was statistically significant at the 0.05 level, as the coefficient of TANG was −1.680 (Table 10) with a p-value of 0.022 (supporting H2). The negative sign of the coefficient indicated that the relationship between TANG and firm performance of the service industries was negative. The negative relationship revealed that the service industry firms were not able to take advantage of the massive amount of fixed assets and were not using their assets efficiently to improve their performance. The outcomes of TANG are similar to the studies of Ahmed and Bhuyan (2020), Nguyen and Nguyen (2015), and Maçãs et al. (2009), as the scholars also examined the negative relationship between the variables.

I found a significant and negative relationship between the firm size of service industries and their performance (supporting H3). The negative coefficient of FRS exposed the fact that small Czech firms provide better services and earn more profits than larger firms. This relationship contradicts the findings of Shirodkar et al. (2022), Ullah et al. (2020), Ullah et al. (2017), Masnoon and Saeed (2014), who emphasized that larger firms had many advantages over smaller firms in terms of easy access to cheap funding sources, easy access to raw materials, human resources, and achieving economies of scale. My results about FRS are consistent with Yousaf (2023a), as the authors also stated that small Czech firms perform better than large firms.

Next, CTOR measures how efficiently net fixed assets are employed to generate profitability. The positive and statistically significant coefficient of CTOR revealed a significant and positive relationship between CTOR and firm performance, and it supports the H4. The efficient utilization of PHCI was also a factor that increased the firm’s performance. The coefficient of PHCI was, similarly to CTOR, statistically and positively related to firm performance (supporting H5). The significant and positive capital turnover and capital intensity ratios jointly showed that Czech service industry firms use capital-intensive technology efficiently to generate high profits. Chandrapala and Knápková (2013) reported the same findings about the Czech firms.

Similarly to FRS, the coefficient of FAGE was −0.217 with a p-value of 0.028, which was statistically significant at the 0.05 significance level (supporting H6). The significant and negative outcomes of FAGE indicated that new firms performed better than older firms. The reasons for this could be the advertising on social media, good management due to the small firm size (as already discussed above, small Czech firms earn more profits than large firms), the strength of demand, a popular tourist destination, etc. The relationship between firm performance and FAGE was consistent with studies by Li et al. (2021), Charoenrat and Harvie (2013), and Park et al. (2010).

Regarding the post-estimation results, the values of AR (1) were significant, and AR(2) – was not significant from Table 7 to Table 10. Roodman (2009) argues that the p-value of the Hansen Test should be between 0.10 and 0.25. The Hansen Test values reported in Tables 7 and 8 were within the suggested range. However, the Hansen Test values reported in Tables 9 and 10 were slightly greater than the suggested range. If the value is greater than 0.25, it means too many instrumental variables were employed. However, various researchers reported that the value of the Hansen Test greater than 0.30 (Wintoki et al., 2012; Yousaf, 2023b). Therefore, the results from Table 7 to Table 10 are valid.

To improve firms’ financial performance, managers and policymakers need to choose the right steps that can optimize financial returns on invested capital. The study investigated the determinants that affect firm performance and how to choose the best model among the dynamic panel data models. I obtained the secondary data of Czech service industries from the CRIBIS database from 2017 to 2021. I estimated the results with pooled OLS, fixed-effect model, one-step difference, two-step difference, one-step system, and two-step system GMM estimations. After employing the suggested criteria by Bond et al. (2001), I concluded that the outcomes of the two-step system GMM should be preferred among dynamic panel data models. The significant results of the two-step system GMM revealed that TDTA, TANG, FRS, and FAGE had a negative impact on the firm’s performance. Conversely, CTOR and PHCI had a positive relationship with firm performance.

The current research offers several contributions to the literature and practical knowledge. The research contributes to the limited existing literature on determinants of firm performance, specifically of service industries’ firms. The study will help academics, policymakers, and researchers, as it suggests several determinants that either positively or negatively impact the firm’s performance. The findings of the research show that businesses associated with the service industry should use capital-intensive technology instead of labor-intensive technology to boost profitability and improve performance. Additionally, the study’s results revealed that the new firms and small firms perform better than the old and large firms. Therefore, these findings will encourage investors, policymakers, and other stakeholders. The current research will contribute to practical knowledge for the first time, as it provides a proper methodology to choose the best model among several panel data models. Hence, the results of the current research will be helpful for the directors, policymakers, and other management of the firms in understanding, estimating, and interpreting the factors that affect (positively or negatively) the firm’s performance in the service industries. The empirical findings will help researchers and academics choose the best model among several panel data models by considering the lag of the firm performance proxies, as the methodology described step by step in the current research.

Most of the results of the chosen variables in the present research are consistent with those of prior studies. However, there are numerous limitations in the current research that future research can address. First, I considered the service sector of the Czech economy, but I did not include dummy variables such as year dummy (to focus more on time) or sector dummy (to focus more on sub-sectors of the service industries as described in Section 3.1) to study the impacts on firm performance. Secondly, in most studies, the scholars used the data for more than ten years in each study, such as Ahmed and Bhuyan (2020), and Khan et al. (2018), but I employed only a limited time period (5 years) in the current study because of the data unavailability of the selected variables. Third, I excluded the consequences of coronavirus disease (COVID-19).

The research identifies opportunities for future studies on factors influencing firm performance. Future research could add more explanatory variables and more than one explained variable. To determine the COVID-19 effects on the firm performance of any sector, it could be conceivable to include a year dummy or structural break test. I investigated only one country and a limited period. However, further research could investigate firm performance by extending time periods, macroeconomic variables, different countries, and sectors. The Russia-Ukraine and Middle East wars affect many countries. The services industry is one of the most affected industries. Therefore, I suggest that researchers and scholars incorporate the consequences of the wars into their forthcoming studies.