A Linear Mixed Model Approach for Determining the Effect of Financial Inclusion on Bank Stability: Comparative Empirical Evidence for Islamic and Conventional Banks in Kuwait

Abstract

Recently, financial inclusion and bank stability have gained attention among researchers, particularly since the 2008 global financial crisis. This study investigates how financial inclusion may have influenced bank stability given differences in banks’ structure based on operating principles (Islamic and conventional banks) during the period of 2003–2017, using Kuwait as a high-income economy case. The current paper assesses how bank stability responds to financial inclusion. This work adopts a Linear Mixed Model (LMM), which tracks variables over time while considering other time-invariant variables. The findings show that the adopted measures of financial inclusion, access and depth, are both significant and negatively related to bank stability. Furthermore, the results unveil a slight difference between the response of Islamic and conventional banks’ stability to the dimensions of financial inclusion. Additionally, the study concludes that the financial crisis had an inverse and significant impact on bank stability. However, the extent of the impact appears to have been greater on Islamic banks compared to their conventional counterparts. Based on this study, banking with more financial inclusion can improve stability if institutional quality in Kuwait is improved so that these banks can operate more efficiently.

1. Introduction

In the past twenty years, both financial inclusion and financial stability have received significant attention from policymakers, economists, and practitioners. However, this issue has attracted deeper attention since the 2008–2009 global financial crisis, with most countries acting urgently to maintain stability in their banking systems. The aim is to meet the worldwide sovereign credit risk crisis and protect bank stakeholders’ interests while controlling for banking exclusion, which can cause deterioration of a country’s entire financial system. As one action, regulatory officials have responded to this challenge by promoting a financial inclusion process as part of their own development and financial initiative strategies for controlling banking instability. Such actions raise questions about the potential influence of financial inclusion on bank stability, as prior research reveals ambiguous evidence in this regard.

Overall, the first phase of financial inclusion starts by owning an account at a financial service provider (most likely a bank) to deposit and/or save funds or engage in fund transactions (receiving or making payments). In its second phase, financial inclusion can include receiving credit from a bank or holding insurance coverage from an insurance firm to eliminate financial risks (see [1,2]). (Financial inclusion represents usage of formal financial services, and means that all or a large portion of the adults in a population (particularly low-income households and small firms) have the chance to be granted access to various financial services that satisfy their financial necessities at a reasonable (competitive) cost). Both stages aim at providing financial services to a large portion of (if not all) community members so that they will have the chance to benefit from various financial services, including savings deposits, payment transactions, and transfers, and access to affordable credit and insurance. Financial inclusion increases saving rates among local community households and corporate categories in both rural and non-rural regions, which leads to improved productive investment opportunities for local businesses, lowers levels of poverty (poverty alleviation), improves community standards of living, stimulates financial development, boosts and sustains economic development, and so on (see, [1,3]). (As a global concern, financial inclusion has been counted as one of the key pillars of global development (see the G20 Summit held in Seoul, South Korea, November 2010). Also, in 2015, the United Nations introduced the role of financial inclusion as a main dimension of achieving the Sustainable Development Goals (SDGs). Financial inclusion has also been discussed in OECD, AFI, APEC, World Bank, IMF and ASEAN forums (see, [2,4]). Economic recessions and global financial and credit crises illustrate how the banking system has become a vital issue of concern. These crises show that when the banking system is not functioning efficiently, fund allocation and economic growth are negatively affected. Furthermore, a growing amount of literature argues that broader access to multiple financial and banking services can increase financial inclusion. Therefore, inadequate financial inclusion is likely to cause banking instability.

While the literature generally describes the significant advantages of financial inclusion, less attention has been focused on examining whether financial inclusion can impact a bank’s stability. We believe that testing for the impact of financial inclusion on bank stability is crucial in order to understand banks’ failure to expand their financial services, particularly after the 2008–2009 global financial crises. Researchers argue that such concern is crucial, since bank stability contributes positively and significantly to economic development.

To this end, this study has the following objectives. First, to examine the influence of financial inclusion on bank stability in Kuwait. Second, to investigate whether the effect of financial inclusion on stability is different between Islamic banks and conventional banks. Third, to examine the impact of the financial crisis on bank stability.

To address these objectives, this study asks the following questions. First, what is the influence of financial inclusion on bank stability? Second, is there an effect of financial inclusion on stability, and is it different between Islamic banks and conventional banks? Third, what is the impact of the financial crisis on bank stability?

This paper contributes to the related literature in the following ways. First, the paper is the first attempt to examine financial inclusion in Kuwait and its impact on bank stability. Second, the paper differentiates between the effect of financial inclusion and the financial crisis on the stability of Islamic banks and conventional banks. Third, the paper adds to the literature by providing evidence on the impact of financial crises on bank stability in the case of Kuwait. Fourth, the paper uses country-based data (Kuwait in our case), which can serve as a reference study for policymakers from Kuwait. Fifth, in terms of methodology, the paper applies the Linear Mixed Model (LMM), which allows tracking variables over time.

Our empirical findings reveal that the two proxies for financial inclusion (access and depth) are significant, and negatively influence bank stability. Additionally, the results highlight the low variation between the effect of financial inclusion on the stability of Islamic and conventional banks. In terms of time, the results reveal that the banks’ stability increased from the beginning of 2003 until 2006, and then decreased until 2011, possibly due to the financial crisis. The rest of the paper is organized as follows: Section 2 presents a summary of previous research. Section 3 provides the data and econometric methodology. Section 4 highlights the empirical findings. Finally, Section 5 concludes the paper and presents its policy implications.

2. Literature Review

Overall, too few studies have examined the connection between financial inclusion and financial stability, given the limitations on financial inclusion due to cross-sectional and/or time-series data availability. Also, many regulatory officials began to pay serious attention to the financial inclusion process only after the financial crisis in 2008 (see, among others, [1,5,6,7,8]. Nevertheless, there is no guarantee that financial inclusion will positively impact financial stability. A negative impact is also a possibility. This depends on the measure used as a proxy for financial inclusion. Several studies have proposed that financial inclusion can affect financial stability. In contrast, other studies contend that the effects of financial inclusion on financial stability are conditional on some specific micro and macro factors.

In the literature, the positive effect of financial inclusion is identified as bank stability, due to many attributes diversifying a bank’s credit base. Expanding lending to a wide range of individual borrowers and granting/extending credit to SMEs, particularly labor-intensive ones, lowers bank default risks by lowering loan portfolio volatility and avoiding positive correlation among individual borrowers, in addition to improving employment rates (see [5]). Also, expanding a bank’s base of small and low-income depositors and savers reduces its reliance on wholesale and international financing choices and helps to avoid cyclical economic and financial risks.

Some researchers reveal that expanding a bank’s depositor base by expanding access to the banking system leads to funding stability and reduces the bank’s pro-cyclical risks and total asset volatility [9]. Furthermore, by switching from cash transactions to electronic transfers, individuals will be more capable of following a consumption budget, since their payment history can be analyzed at any time by them or by the lending parties. Meanwhile, others conclude that consumption volatility is less for economies with greater levels of financial inclusion [10].

In their work, Mehrotra and Yetman [11] assert that more financial inclusion can lead to more economic activity, resulting in a higher relevance of interest rates, which could improve the effectiveness of monetary policy, and hence, lead to more financial sustainability. Furthermore, Ratana et al. [12] reveal that more access to bank accounts, availability of digital payment systems, and diversification of the banking deposit base contribute positively to financial stability by stabilizing the banking system. Moreover, more usage of debit cards by adults can lead to more careful monitoring of their accounts, which can result in higher savings rates, indicating more trust toward the banking system, which improves the stability of the banking sector [13].

Negative impacts may result from extending credit and expanding the borrowing base without relevant loan approval standards, in conjunction with the lack of a reliable credit follow-up process and loan supervision mechanisms, which can cause deterioration in loan quality and increase credit loan portfolio risks. In his work, Khan lists the negative contributions of financial inclusion to financial stability [14]. These include: a reduction in bank lending standards when banks attempt to expand their credit market share by increasing their borrower pool; increasing bank reputation risks when relying on external bodies for credit assessment in order to offer smaller loans; and expanding lending to microfinance businesses without proper lending standards. De la Torre et al. and Cihak et al. argue that the negative contribution of financial inclusion to financial stability is linked to the extent to which a country has weak lending standards and an inadequate supervision process [15,16]. Moreover, Cihak et al. show that more financial inclusion can lead to extensive borrowing by household (individual) borrowers, enlarging the possibility of credit risk and increasing loan charge-offs, thus increasing bank instability [16].

A third area of the literature considers the effect of financial inclusion on financial stability to be conditional and related to some specific micro and macro factors, such as the number of existing financial intermediaries. Expanding financial inclusion in a particular economy means a greater number of banking participants, indicating that more financial and banking transactions will be implemented by a limited number of financial intermediaries. This can lead to higher transaction costs. Consequently, there will be an increase in moral and social hazard, negatively affecting financial stability (see [17]).

Some studies focus on the bank type (rural banks vs. non-rural banks) and size. Some scholars state that local institutions (i.e., cooperatives or rural banks) have less capability to face market financial and non-financial risks, causing them to be defenseless against natural disasters, recessions, and financial crises [18]. Other studies linked financial inclusion to a country’s growth (see [5,19,20,21]). For Nigeria, scholars identify the effect of financial inclusion on economic growth [3]. The former is a result of the later, but financial inclusion accelerates economic growth. For India, others reveal that a bidirectional and unidirectional causal relationship has been observed between financial inclusion and economic growth [22]. For the Organization of Islamic Cooperation, researchers conclude that financial inclusion makes a vital contribution towards stimulating economic growth, and both variables are intercorrelated [23].

To this end, the above findings motivated us to go further in our test to evaluate the influence of financial inclusion on bank stability, particularly when examining whether deep financial services are mandatorily inclusive when applying high-income country base data (Kuwait, in our case) rather than a cross-country base data set. We also believe that our further examination of the bank structure (Islamic bank or conventional bank) will contribute to the financial inclusion and bank stability literature, given the significant differences among the two groups in terms of specific bank variables. The current paper also accounts for the effect of financial crises on bank stability in aggregate databases, adding valuable input to the long debate about how bank stability behaves before and after a financial crisis.

3. Research Methodology

3.1. Sample and Data

This research adopts annual panel data for the banks listed in the Kuwait Stock Exchange (KSE) from 2003–2017. The reason for the selection of this period of time is to capture the financial crisis period, as well as because of the unavailability of more recent data. There are ten banks listed on the KSE, five of which operate according to Shari’ah principles. While the banks’ specific variables were collected from Refinitiv Eikon DataStream, the World Bank’s Global Financial Development is the main source of financial inclusion data. Moreover, GDP data was collected from the World Bank database.

3.2. Variables Construction and Measurement

3.2.1. Dependent Variable

Various proxies are used to measure the stability of banks. In this paper, we choose the Z-score, as it is widely used in the literature of finance (see, e.g., [24,25,26,27,28,29]). This value refers to the banks’ riskiness, and signifies the deviation of the mean return below the average return before the equity is depleted [30]. The rationale for using this measure is that the greater return on assets and capitalization, the better the stability of the banks, and vice versa. The score is calculated based on the return of assets, volatility, and leverage as follows:

((ROA + Equity)/Assets))/Standard Deviation (ROA).

ROA is the return on assets. Equity refers to the average stockholder equity to total assets ratio, and Standard Deviation (ROA) is the standard deviation of the return on total assets.

3.2.2. Independent Variable

As this study investigates the effect of financial inclusion on bank stability, the independent variable is the financial inclusion of banks in Kuwait. Initially, researchers have considered several indicators for measuring financial inclusion, and have mainly focused on two constructs: depth and access [2,26,29], (See

Table 1. Financial Inclusion Main Constructs.

Index	Variables
Financial inclusion– access	Bank accounts per 1000 adults (demographic penetration) ATMs per 100,000 adults (demographic penetration) Bank branches per 100,000 adults (demographic penetration)
Financial inclusion– depth	Bank deposits (usage) Private sector credit (usage) Liquid assets (usage)

). In this study, the constructs were gauged using a combination of various measures, and a comprehensive proxy indicator was developed for each using Principal Component Analysis (PCA). Some researchers utilized loans and deposits as the main usage indicators [31,32]. According to [32], the penetration of the financial system can be measured using transaction points, such as branches and ATMs. Moreover, others opine that the regularity and frequency of customers (number of accounts) using the financial services indicate penetration of the financial system [33].

PCA is used to construct an index that adequately deals with the problems of multicollinearity and over-parameterization as an overall indicator of the extent of financial inclusion [26,34,35,36].

Table 2. Variables definition and measurement.

Variables *	Measurement	Source
Dependent:
Z score	((ROA + Equity)/Assets))/Standard Deviation (ROA)	Authors’ calculations
Independent:
Financial inclusion– access	Index using PCA	World Bank
Financial inclusion–depth	Index using PCA	World Bank
Control:
TA	Total assets of the banks	DataStream
LLP	Loan/finance loss provisions to total loans	DataStream
BL	Bank loans to total assets	DataStream
EA	Earnings assets to total assets	DataStream
EQ	Equity to total assets	DataStream
NII	Non-interest income	DataStream
GDP	Gross domestic per capita	DataStream
ROA	Net income to total assets–return on assets	DataStream
Eff	Overhead costs to total assets	DataStream

* Variables that are significant and reported on the outputs; other insignificant variables are excluded.

shows the variables used to develop each index.

3.2.3. Control Variables

To consider all factors that might have effect on bank stability, researchers have derived several variables from the Definitive Eikon DataStream. These variables were commonly used proxies that are likely to affect bank stability [26,29]. Total assets of the banks (TA) indicate the size effect on the stability of the banks [29]. The notion of ‘too big to fail’ is of serious concern to bank stability [37]. The ratio of loan/finance loss provision to total loans (loan/finance loss provision) is used to account for individual bank loan/finance portfolio risk [26]. Bank loans to total assets is another striking factor that may influence stability via liquidity risks [25,26]. Earnings to total assets is another factor used in the model to account for excessive risks. Researchers argue that excessive risk-taking may indicate bad management [26]. Therefore, we use earnings assets to total assets to control for the quality of management in managing risk. Equity to assets ratio is used to control for risk aversion [38]. Specifically, well-capitalized banks are expected to perform better than thinly capitalized banks. Bank capitalization is expected to influence stability, as it boosts depositors’ confidence and indicates solvency [39]. Non-interest income indicates income diversification, influencing bank stability [37,40]. Researchers also utilize the overhead costs to total assets as an indicator of the efficiency of the banks, with a plausible link between them being observed in the literature [30]. Lastly, GDP is used to control for the economic cycle [30]. As such, GDP growth leads to greater stability because of reduced defaults, improved asset quality, and lower risks. The list of all variables is provided in Table 2, and significant variables are reported in the results.

3.3. Linear Mixed Model (LMM)

As a generalization of OLS, LMM has proven to be more accurate than OLS models, as it models error terms resulting from the repeated measures of bank stability over time [41,42,43,44,45,46,47,48]. In cross-bank studies, observations may not be independent, as the data for the same bank could be replicated, leading to in-cluster (bank stability) correlation. Repeated measures lead to a clustered data structure, and failing to consider such a structure leads to biases in the model estimates and potentially higher values of Type I error [49,50]. LMM is used in this study to model bank stability, as it takes into account the correlation of measures (bank stability) over time for all banks studied, controlling (incorporating) for modeled factors that do not change over time [51]. LMM has the advantage of including time and treating it as a fixed or random effect. The time factor included in the model can be linear, quadratic or cubic for more accuracy in tracking changes of bank stability over time. This leads to an improvement in randomized effect modeling and less bias and type I error in the estimation of the error terms [48,50,52,53].

To measure the impact of financial inclusion (FI) on bank stability and to track changes in bank stability across time including all control variables (only significant variables included), the general form of the LMM is proposed as:

$$\begin{gathered} {\mathrm{S}\mathrm{T}}_{\mathrm{i}\mathrm{j}}=\underset{\mathrm{f}\mathrm{i}\mathrm{x}\mathrm{e}\mathrm{d}\mathrm{e}\mathrm{f}\mathrm{f}\mathrm{e}\mathrm{c}\mathrm{t}}{\underbrace{{{{(\mathsf{\beta }}_{\mathrm{o}\mathrm{o}}+\mathsf{\beta }}_1\mathrm{t}}_{\mathrm{i}\mathrm{j}}+{{\mathsf{\beta }}_2\mathrm{t}}_{\mathrm{i}\mathrm{j}}^2+{{\mathsf{\beta }}_3\mathrm{t}}_{\mathrm{i}\mathrm{j}}^3)}}+\underset{\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{d}\mathrm{o}\mathrm{m}\mathrm{e}\mathrm{f}\mathrm{f}\mathrm{e}\mathrm{c}\mathrm{t}}{\underbrace{\left({\mathrm{b}}_{0\mathrm{i}}+{\mathrm{b}}_{1\mathrm{i}}{\mathrm{t}}_{\mathrm{i}\mathrm{j}}\right)}}+{\mathsf{\beta }}_4{\mathrm{B}\mathrm{I}\_\mathrm{A}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\beta }}_5{\mathrm{B}\mathrm{I}\_\mathrm{D}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\beta }}_6{\mathrm{E}\mathrm{f}\mathrm{f}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\beta }}_7{\mathrm{T}\mathrm{A}}_{\mathrm{i}\mathrm{j}}+ \\ {\mathsf{\beta }}_8{\mathrm{B}\mathrm{L}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\beta }}_9{\mathrm{E}\mathrm{Q}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\beta }}_{10}{\mathrm{N}\mathrm{I}\mathrm{I}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\beta }}_{11}{\mathrm{R}\mathrm{O}\mathrm{A}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{j}} \end{gathered}$$

(1)

where ${\mathrm{S}\mathrm{T}}_{\mathrm{i}\mathrm{j}}$ is the bank stability for Bank $\mathrm{i}$ at Time $\mathrm{j}$, ${\mathsf{\beta }}_{\mathrm{o}\mathrm{o}}$ is the mean bank stability, and ${\mathrm{b}}_{0\mathrm{i}}$ is the bank deviation from this mean, ${\mathrm{b}}_{0\mathrm{i}}~\mathrm{N}\left(0,{\mathsf{\sigma }}_0^2\right)$. Similarly, ${\mathsf{\beta }}_1$ is the mean growth/decline rate in ST and ${\mathrm{b}}_{1\mathrm{i}}$ is the bank deviation from this mean, ${\mathrm{b}}_{1\mathrm{i}}~\mathrm{N}\left(0,{\mathsf{\sigma }}_1^2\right)$. ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{j}}$ are independent random errors for each bank $\mathrm{i}$ and time $\mathrm{j}$, ${\mathsf{ϵ}}_{\mathrm{i}\mathrm{j}}~\mathrm{N}\left(0,{\mathsf{\sigma }}_{\mathsf{\epsilon }}^2\right)$. BI_A (Financial inclusion–access) and BI_D (Financial inclusion–depth) are the covariates in the model to evaluate their effect on ST. Eff (efficiency), TA (total assets), BL (bank loan%), EQ (equity to total assets), NII (non-interest income), and ROA (return on assets) are the control variables.

Quadratic $\left({\mathsf{\beta }}_2\right)$, and cubic (${\mathsf{\beta }}_3$) terms are added to enhance the accuracy of the proposed model and track bank stability more accurately over time.

3.4. Robustness of the Results

Quadratic and cubic terms of time are added to the model for the abovementioned reason. Several error structures (variance-covariance matrices) are also tested. Fixed and random effects are also included and tested in the model. All components and terms added to the model are tested using a log-likelihood ratio test (based on AIC values). Some scholars argued that the random-effect-within-between (REWB) model (LMM), is the strongest random effect model for panel data, as it allows modeling of random and fixed effects [54]. The results revealed that the LMM is more accurate and solves the variable bias and endogeneity problems. The present article also highlights the inclusion of random slopes as a solution to heterogeneity. In addition, the hypothetical model also includes the most used control variables in the literature to avoid any bias in the estimation (endogeneity).

Our results showed that the minimum AIC value for the proposed model, modelling the intercept as a random component, was AIC = 81.53, with a p-value < 0.0001. Components and terms are added to the model progressively (building the model) using the random coefficient modeling framework, with independent and control variables for different variance-covariance structures. Terms are included in the model, and a backward stepwise regression is used to select the best subset of variables related to bank stability [55]. The best model (after excluding all insignificant variables) is estimated using restricted maximum likelihood (REML). The model is then validated and diagnosed using residual analysis [47].

Furthermore, this study included a number of analyses to ascertain the robustness of the results. First, data winsorisation (see [56,57]) was applied to check the sensitivity of the findings. The model is refitted after winsorising the data at the 1% and 99% levels, and the results (not shown) are almost the same. Second, the data was split into Islamic and non-Islamic banks, resulting in almost the same results, with slight differences, as reported in the results sections. This suggests that our evidence is rigorous.

4. Empirical Results

4.1. Descriptive Statistics and Correlations

Descriptive statistics for the variables included are shown in

Table 3. Descriptive statistics for the total sample as well as for Islamic and conventional banks.

Variable	Category	N	Mean	StDev	Median	Min	Max
Bank Stability	Islamic	65	0.92	0.73	0.92	−0.99	2.76
	Conventional	74	1.46	1.43	1.11	−5.22	4.98
	Total	139	1.21	1.18	0.99	−5.22	4.98
FI_Access	Islamic	65	25.02	6.37	26.57	0	31.06
	Conventional	74	24.45	6.75	26.46	0	31.06
	Total	139	24.71	6.56	26.46	0	31.06
FI_Dep	Islamic	65	69.88	18.03	61.42	48.36	103.06
	Conventional	74	68.72	17.5	61.42	48.36	103.06
	Total	139	69.26	17.7	61.42	48.36	103.06
TA		139	17,645,803.1	17,991,941.03	11,946,303	435,970	86,266,514
LLP		139	0.01	0.01	0.01	-0.01	0.08
BL		139	0.74	0.1	0.74	0.4	0.93
EA		133	83.27	26.5	92.71	5.56	97.45
EQ		139	0.17	0.09	0.15	0.02	0.8
NII		139	0.01	0.01	0.01	−0.01	0.03
GDP		139	39,472.4	9760	38,577.5	22,148.38	55,494.95
ROA		139	1.42	1.23	1.3	−6.87	4.21
Eff		139	1.13	0.08	1.14	0.95	1.24

. The results show that for the ten banks in Kuwait, five are Islamic and the other five are conventional throughout the study from 2003 to 2017. As illustrated in the table, the overall Z-Score of bank stability is 1.21 (SD = 1.18), with a minimum of -5.22 (Gulf Bank in 2008) and a maximum of 4.98 (National Bank of Kuwait in 2006). The stability average for conventional banks is higher than for the Islamic banks, at 0.92 and 1.46, respectively. However, there is less variability (almost half) in the Islamic banks’ stability values than for the conventional banks. The overall mean value for Financial Inclusion–Access (FI_A) is 24.7% (SD = 6.56). The FI_A ranges from 0 (min) to 31.06 (max), with a median of 26.46, meaning most FI_A values are above 26%. The FI_A average and standard deviations are almost the same for Islamic and conventional banks. For Financial Inclusion–Depth (FI_D), the overall average is much higher than FI_A, with a value of 69.26 (SD = 17.7), ranging from 48.36 to 103.06. The FI_D average and standard deviations are almost the same for Islamic and conventional banks. Table 3 also contains descriptive statistics for the control variables.

Further analysis of bank stability over the period of study (as illustrated in Figure 1) shows that bank stability increased from 2003 to 2006, then decreased until 2011, and then started to increase again at a slower rate. We believe that these steep changes in bank stability rates may be attributable to the effect of the global financial crisis effect of 2008. Financial Inclusion–Access increased overall, although in 2008, it declined at a slow rate and then started to increase again in 2014. On the other hand, the FI_D showed a different trend, as it was below average (with slight fluctuations) before starting to increase dramatically after 2013.

Pearson pairwise correlation coefficients between the variables are given in

Table 4. Pearson’s correlation matrix for all variables.

	1	2	3	4	5	6	7	8	9	10	11
1.  Z	1
2.  FI_ACC	−0.221 **	1
3.  FI_DEP	−0.236 **	0.339 **	1
4.  TA	0.229 **	0.191 *	0.209 *	1
5.  LLP	−0.572 **	0.179 *	0.075	0.104	1
6.  BL	−0.079	0.200 *	0.023	−0.504 **	−0.116	1
7.  EA	0.012	−0.021	−0.027	−0.489 **	−0.159	0.710 **	1
8.  EQ	−0.003	−0.072	−0.044	−0.148	−0.164	−0.033	0.009	1
9.  NII	0.387 **	−0.119	−0.373 **	0.166	−0.09	−0.460 **	−0.420 **	−0.05	1
10.  GDP	0.007	0.284 **	−0.608 **	0.041	0.216 *	0.186 *	−0.047	0.005	0.136	1
11.  ROA	0.880 **	−0.273 **	−0.317 **	0.001	−0.645 **	−0.115	−0.026	0.039	0.511 **	−0.085	1
12.  EFF	−0.123	0.432 **	0.252 **	0.176 *	0.239 **	0.131	−0.074	0.044	−0.168 *	0.336 **	−0.269 **

Note: * significant at the 0.05 level, ** significant at the 0.01 level.

. Financial Inclusion (Access and Depth) are both significantly and negatively correlated with stability. For the control variables, TA (0.23), LLP (−0.57), NII (0.39), and ROA (0.88) are significantly correlated with performance, and the rest are not significant, as illustrated in Table 4.

A correlation matrix shows significant correlation among variables included in the model, which may cause a problem of multicollinearity in the regression analysis. VIF values and stepwise regression analysis are used to avoid multicollinearity in the estimation process. The results indicated that all VIF values were low (all less than 4; please see

Table 5. Linear Mixed Model baseline results.

Fixed Effects	Model 1	VIF
(Intercept)	0.27 (0.66)
Time	3.19 **** (0.35)
$\mathbf{T}\mathbf{i}\mathbf{m}{\mathbf{e}}^2$	0.62 *** (0.2)
$\mathbf{T}\mathbf{i}\mathbf{m}{\mathbf{e}}^3$	2.07 **** (0.23)
FI_ACC	−0.25 **** (0.04)	1.8
FI_Dep	−0.07 (0.04)	3.9
Eff	−0.74 (0.46)	2.4
TA_Z	−0.15 ** (0.06)	2.2
BL	−0.89 * (0.53)	1.9
EQ	−0.61 * (0.35)	1.2
NII	−15.24 ** (7.29)	2.0
ROA	0.84 **** (0.03)	1.8
Random component standard deviation
Intercept	0.49 ***
Residuals	0.25
AIC	77.22

* p < 0.1, ** p < 0.05, *** p < 0.01, **** p < 0.001; standard errors of estimates are in parentheses.

), and multicollinearity does not appear to be an issue for the data analyses.

4.2. Main Findings

Table 5 presents the results of modeling of Financial Inclusion (Access and Depth) of bank stability along with control variables for 10 banks in Kuwait over 15 years (2003 to 2017). The results reveal that Financial Inclusion (Access and Depth) are significant and negatively associated with bank stability (−0.25 p-value < 0.01) and (−0.073). The results of this investigation are not consistent with prior research; for example, refs. [8,57,58] show that advancements in financial inclusion through more account ownership and digital payments have a stabilizing effect on the banking industry. However, in their findings, they indicate that stronger governmental power and a looser financial environment may promote higher risk-taking or lower bank stability during the financial inclusion process. Moreover, this part of the research findings answers the first research question regarding the influence of financial inclusion on bank stability.

In addition, five control variables were also significant in relation to bank stability, as illustrated in Table 5; see Figure 2 for a significant effect plot.

The quadratic and cubic time variables in Table 5 are significant. This indicates that bank stability behavior is not linear (constant) over time. The positive effect of the quadratic term of time suggests that bank stability increased slowly and then started to slow down before increasing significantly at the end of the study period.

Furthermore, the random intercept term is significant. This indicates that the initial level (at 2003) of bank stability is different at the bank level.

4.3. Robustness of the Model

To check the robustness of the results and whether the relationships between bank stability and financial inclusion are the same for Islamic and conventional banks, the model was run after dividing the sample into two parts, Islamic and conventional banks.

Table 6. Results of Linear Mixed Model for Islamic and conventional bank stability.

Fixed Effects	Islamic Banks	Conventional Banks
(Intercept)	0.04 (0.64)	−0.92 ** (0.35)
Time	2.19 **** (0.42)	3.43 **** (0.32)
$\mathbf{T}\mathbf{i}\mathbf{m}{\mathbf{e}}^2$	0.55 ** (0.25)	0.66 *** (0.22)
$\mathbf{T}\mathbf{i}\mathbf{m}{\mathbf{e}}^3$	1.67 **** (0.23)	1.82 **** (0.19)
FI_ACC	−0.24 **** (0.06)	−0.27 **** (0.05)
FI_Dep	NS	NS
Eff	−1.13 * (0.62)	NS
TA_Z	NS	−0.22 *** (0.06)
BL	NS	NS
EQ	NS	NS
NII	NS	−36.21 ** (13.94)
ROA	0.79 **** (0.05)	0.89 **** (0.03)
Random component standard deviation
Intercept	0.199 ***	0.63
Residuals	0.21	0.23
AIC	16.5	35.65

* p < 0.1, ** p < 0.05, *** p < 0.01, **** p < 0.001, NS: Not Significant; standard errors of estimates are in parentheses.

presents the results of the examination of bank stability for Islamic and conventional banks. The main objective is to determine whether there is a significant difference in the effect of financial inclusion on the level of bank stability between the two groups over our period of study (2003–2017). There is a slight difference in the results when modeling each set of banks alone (Islamic and non-Islamic banks). The growth rate of conventional banks is slightly faster than that of Islamic banks (time-based coefficient: 3.43 vs. 2.2). Financial inclusion for both banks was found to be not significant, which is similar to previous results. In terms of control variables, efficiency is significant for Islamic bank stability but is not significant for conventional. At the same time, TA and NII are significant for conventional banks but not significant for Islamic banks (see Figure 3 and Figure 4). Therefore, this part of the research findings responds to the study’s second research question related to the difference in the effect of financial inclusion on Islamic and conventional banks.

4.4. Financial Crisis

To test if there is a significant impact of the financial crisis (2008) on financial stability, a dummy variable was added to our primary model (for both Islamic and conventional banks) for each group of banks. The results of the three models along with their plots (full, Islamic, and conventional banks) are summarized in

Table 7. Results of Linear Mixed Model for Islamic and conventional bank stability.

Fixed Effects	Full Model—All Banks	Islamic Banks	Conventional Banks
(Intercept)	0.28 (0.65)	0.19 (0.61)	−0.92 ** (0.35)
Time	3.43 **** (0.35)	2.53 **** (0.42)	3.43 **** (0.32)
$\mathbf{T}\mathbf{i}\mathbf{m}{\mathbf{e}}^2$	1.13 **** (0.27)	1.2 **** (0.34)	0.66 *** (0.22)
$\mathbf{T}\mathbf{i}\mathbf{m}{\mathbf{e}}^3$	2.21 **** (0.23)	2.1 **** (0.27)	1.82 **** (0.19)
FI_ACC	−0.24 **** (0.04)	−0.22 **** (0.06)	−0.27 **** (0.05)
FI_Dep	NS	NS	NS
Financial Crisis	−0.31 ** (0.13)	−0.43 ** (0.16)	NS
Eff	NS	NS	NS
TA_Z	−0.16 *** (0.06)	NS	−0.22 **** (0.06)
BL	−0.98 * (0.53)	NS	NS
EQ	−0.59 * (0.35)	NS	NS
NII	−14.22 ** (7.17)	NS	−36.21 ** (13.94)
ROA	0.82 **** (0.03)	0.74 **** (0.05)	0.89 **** (0.03)
Random component standard deviation
Intercept	0.5	0.22 ***	0.63
Residuals	0.25	0.21	0.23
AIC	73.7	18.2	35.65

* p < 0.1, ** p < 0.05, *** p < 0.01, **** p < 0.001, NS: Not Significant; standard errors of estimates are in parentheses.

and Figure 5, Figure 6 and Figure 7.

The results reveal that the financial crisis had a negative and significant impact on all Kuwait banks’ stability. However, this impact is significant and has a higher magnitude for Islamic banks, and is not significant for conventional banks. This result is surprising for Islamic banks as well as conventional banks, since the previous literature suggested that Islamic banks showed more stability during the financial crisis than conventional banks [30]. However, others argued that diversification may play a role in this relationship [59]. Previous results suggest that bank diversification decreased the variance of bank stability prior to the financial crisis but increased its variance during the crisis. Therefore, banks should reduce diversification during a crisis and focus on the main intermediation functions of deposits and loans, which may not be relevant for Islamic banks compared to conventional banks.

This study found that the effect of the financial crisis reduced (on average) the Islamic banks’ stability by 0.43. Again, Financial Inclusion–Access is significant but Financial Inclusion–Depth is not. In terms of control variables, most control variables are significant to all banks, except for Efficiency. For Islamic banks, only ROA is positively associated with bank stability. However, ROA, NII, and TA are significantly associated with conventional banks. This section of the results addresses the third research question of the study, which is focused on the impact of the financial crisis on bank stability.

5. Conclusions and Policy Recommendations

This paper studies the impact of financial inclusion on the stability of banks in Kuwait. The study also comparatively differentiates between Islamic and conventional banks. The results reveal that financial inclusion indices are related significantly and negatively to bank stability. The banking structure shows minimal variation between Islamic banks and conventional banks in the effect of financial inclusion on stability. It is clearly shown that the association between financial inclusion and stability is negative, indicating that more inclusiveness of the financial system is related to lower banking stability. The results are not aligned with the findings of previous studies [26]. However, others supported our results, indicating that financial inclusion could compromise financial stability [16]. According to the study, greater financial inclusion is linked to more individual borrowing, and therefore may expose banks to the risk of abnormal events. Surprisingly, Islamic banks show a higher magnitude of negative and significant impact on stability during the financial crisis than their conventional counterparts. These findings are not in line with other literature, which substantiates the argument that Islamic banks are less prone to risk during the financial crisis [30]. The results of this paper show the connection between financial inclusion and stability in a rich country for the first time. The results empirically highlight that financial inclusion is a barrier to financial stability. This might indicate that reducing the marginal cost of producing outputs is not optimal, contributing to pricing power and making banks less stable. The essence of financial inclusion is that it stimulates social and economic development, and therefore banking with more financial inclusion and countries with better institutions can improve stability. However, the results may point in the direction of the need to improve levels of institutional quality in Kuwait to allow banks to operate more efficiently. This study offers some policy recommendations for regulators to look further into the liberalization of the Kuwaiti banking system, align it with international practices, increase inclusiveness and promote stability. Increasing the access of households and SMEs to financial services will ensure achieving inclusive economic growth. Additionally, some factors of financial inclusion–access can be considered as limitations for this study and should be considered as recommendations for future research. Other future research recommendations include considering a cross-country base data set to compare income levels, poverty levels, domestic savings rates, investment cycles, financial development level, and social and political stability levels, together with problems related to financial inclusion factors across countries over time.