لخّصلي

1. Introduction
   As a result of the deregulation in the late 90's, the banking landscape changed significantly, especially in terms of business models.

The supervisory concern on restoring profitability, while improving capital and liquidity, led to several calls for income diversifi-
cation, from traditional to non-interest-bearing activities, such as trading, advice, underwriting or the distribution of third-party

products. This phenomenon has been further reinstated after global financial crises and shows a global increasing trend (BIS, 2018).
Several studies examine the impact of income diversification on profitability and its volatility, with evidence for both a positive
and a negative relationship (Chiorazzo, Milani, & Salvini, 2008; DeYoung & Rice, 2004; Maudos, 2017; Stiroh & Rumble, 2006). The

same issue involves studies on diversification and bank stability, with recent studies showing greater evidence of a negative re-
lationship (Abuzayed, Al-Fayoumi, & Molyneux, 2018; Ahamed, 2017; Köhler, 2015; Sanya & Wolfe, 2011).

The effect of income diversification varies across banks and depends on both business models and the economic environment. On
the one hand, traditional activities (i.e. deposits and loans) are considered to be stable, despite exposing to significant credit, liquidity
and interest-rate risks. Non-interest-bearing operations, on the other hand, are prone to market, operational and reputational risks

https://doi.org/10.1016/j.gfj.2020.100517
Received 26 July 2018; Received in revised form 3 September 2019; Accepted 27 February 2020
⁎ Corresponding author.
E-mail addresses: [email protected] (A. Paltrinieri), [email protected] (A. Dreassi), [email protected] (S. Rossi),
[email protected] (A. Khan).

Global Finance Journal xxx (xxxx) xxxx

1044-0283/ © 2020 Elsevier Inc. All rights reserved.

Please cite this article as: Andrea Paltrinieri, et al., Global Finance Journal, https://doi.org/10.1016/j.gfj.2020.100517

and show greater volatility, but at the same time involve greater expected returns.
The global financial crisis of 2008–2009 and the following trends of commodity prices forced the Organization of Islamic
Cooperation (OIC) countries, which are dependent on the oil sector, to review their economic policies, mainly with the purpose of
increasing the contribution to GDP by the non-oil sector. Due to lower oil prices, government deposits shrunk, pushing banks to raise
funds through other more expensive channels, with impacts on their profitability (KPMG, 2017).
Although Islamic banks offer products similar to those of their conventional counterparts, compliance to Sharia (prohibition of
interest-based transactions, excessive uncertainty and gambling, among other requirements) significantly influences their business

models. Previous studies show that Islamic banks are comparatively less diversified and, therefore, may benefit more from di-
versification, also in enhancing financial stability. At the same time, interest-bearing activities are not present in Islamic banks and

the distinction impacting diversification is between income from financing and non-financing activities (Abuzayed, Al-Fayoumi, &
Molyneux, 2018).
Additionally, conventional banks face agency issues towards both depositors and borrowers. This problem may be lower in

Islamic banks for two reasons. Firstly, equity-based funding models (Mudarabah and Musharakah), as opposed to those that are debt-
based (Murabahah and Ijarah), should increase the incentive of depositors to monitor and exert discipline on banks' management.

Secondly, an important monitoring role is played by the Sharia supervision boards of each institution: assuring compliance with
religious requirements of each operation may avoid excessive risk-taking or poor-quality lending.
In this paper, we investigate the impact of income diversification on risk-adjusted profitability and stability of Islamic and
conventional banks within the OIC region. More specifically, we test the aforementioned link under different measures of profitability
and firm-risk, including both accounting and market-based specifications.
We contribute to the existing literature by adding our findings to the few existing studies on income diversity, bank profitability
and stability in emerging economies. Unlike previous studies, with which we share motivation and some comparable results, we also

extend the geographical scope of analysis beyond GCC countries, to include other economies in which Islamic banking is a well-
established phenomenon. Moreover, our evidence is grounded on a large number of banking institutions, on a more recent time frame

and with utmost care in ensuring data quality. Finally, to the best of our knowledge, this study is the first attempt to investigate this
matter by considering both accounting and market-based measures of firm-risk and to investigate jointly on the same dataset all
implications of income diversity on risk-adjusted profitability and stability.
Our analysis shows that diversification provides lower rewards for Islamic banks than conventional banks, with stronger results
associated with accounting-based measures. While shares of non-interest income positively contribute to profitability regardless of
the business model, income diversification shows a not significant effect on the risk-adjusted profitability of Islamic banks. Moreover,
we do not find any relationship between income diversification and stability for both conventional and Islamic banks.
The remainder of the paper is organized as follows. Section 2 reviews the relevant literature. Section 3 describes our dataset,
variables and econometric strategy. Section 4 discusses our findings and provides a series of robustness tests. Finally, Section 5
concludes with our policy implications.
2. Literature review
2.1. Income diversification and bank profitability
Diversification should be beneficial, with the leading examples of firms' income sources and investors' asset allocation. In the
specific case of banking, non-interest-bearing activities less than perfectly correlated with traditional operations should provide profit
smoothing, resilience to downside risks and a reduction of the overall riskiness of the firm (Chiorazzo et al., 2008).

However, DeYoung and Roland (2001) argue that non-interest income may be more volatile for three reasons. Firsly, relationship-
based loans have high switching costs when compared to fee-based activities, and notwithstanding higher credit and interest-rate

risks, they generate more stable revenues. Secondly, non-interest activities require significant fixed costs (technology and human
capital), while the marginal cost of interest-bearing operations is relatively low. Lastly, several non-interest activities incur limited
effects on regulatory capital and may incentivize leveraging and, as a result, higher earnings volatility.
A significant number of studies focus on revenue diversity in conventional banking with a positive (Ahamed, 2017; Al-Obaidan,
1999; Elsas, Hackethal, & Holzhäuser, 2010; Meslier, Tacneng, & Tarazi, 2014; Roengpitya, Tarashev, Tsatsaronis, & Villegas, 2017)
and negative relationship (Acharya, Hasan, & Saund, 2006; DeYoung & Rice, 2004; Maudos, 2017; Stiroh, 2004). Al-Obaidan (1999)
finds diversification in commercial banks to increase allocative and scale efficiency and an overall positive economic gain, however,
the results are negative for technical efficiency. Likewise, Elsas et al. (2010) find a positive effect of diversification on profitability
and market valuation. Ahamed (2017) finds that a higher share of non-interest income increases profits and risk-adjusted profitability
in Indian banks, especially when trading is involved.
Roengpitya et al. (2017) conclude that commercial banking models show lower cost-to-income ratios and more stable profitability

than the trading model. Moreover, they measure an average 2.5 percentage points improvement on the return-on-equity of deposit-
funded banks. Meslier et al. (2014) investigate diversification dynamics in emerging countries' banks and find that non-interest

income has positive effects on profits and risk-adjusted profitability.
To begin with negative nexus of diversification-profitability, DeYoung and Rice (2004) investigate the impact of non-interest

income on the performance of US commercial banks over the period 1989–2001. They find that well-managed banks are less de-
pendent on non-interest income, while banks with good service quality and customer relationships are likely to produce more non-
interest income. Stiroh (2004), while examining the diversification benefits in US banks, finds non-interest income to be remarkably

A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

2

volatile and correlated with net interest income; banks relying heavily on non-interest income show also lower risk-adjusted prof-
itability.

Acharya et al. (2006) argue that diversification does not assure improvements in performance or greater stability. By analyzing
105 Italian banks for the period of 1993–1999, they find that diversification reduces earnings for high-risk banks, while generating
riskier loans and having no or marginal effects on risk-return profiles for low-risk banks. Using data for a panel of European banks
over the period of 2002–2012, Maudos (2017) shows a negative impact of increases in non-interest income on profitability during the
crisis.
Banks, traditionally, borrow funds from depositors and lend to borrowers, building profitability on the net interest margin.
However, profit may also arise from non-interest activities, such as fees and commissions or trading. The literature provides mixed
results on the effects of income diversification.
Banks can diversify their sources of income through technological advancement and managerial skills. This may increase revenues
and reduce the costs of running non-interest activities (Ahamed, 2017; Meslier et al., 2014; Sanya & Wolfe, 2011). In contrasts, banks
with more aggressive diversification may lose its competitive managerial advantage and generate more volatile returns (Stiroh, 2004:
Acharya et al., 2006; DeYoung & Roland, 2001).

Islamic banks are relatively young and experiencing significant growth. Therefore, they could better reap the benefits of di-
versification (Chen, Liang, & Yu, 2018; Molyneux & Yip, 2013). Moreover, specific operations (f.i. sukuk underwriting) and religious

affairs (f.i. Hajj) require the services of Islamic banks, increasing their profitability. At the same time, complying with the Shariah on

top of other regulatory requirements may generate higher costs (f.i. the Shariah board) and exclude some prohibited business ac-
tivities. Thus, the net benefit of diversification could be affected downwards for Islamic banks.

The above arguments lead us to the following research questions:
RQ1. : “How does diversification affect the profitability of banks?”
RQ2. : “Is there any difference in diversification effects on profitability between Islamic and conventional banks?”
2.2. Income diversification and bank stability

Traditional banking (i.e. deposit-funded loans) involves exposure to interest-rate, credit and liquidity risks, that may show sig-
nificant correlations. By participating to non-interest-bearing activities, banks should be able to reduce the impact of the worsening

quality of their loan portfolio, as well as to offset credit losses with fee-based revenues. Non-traditional operations seem to be
negatively associated with the asset quality of credit institutions (Ahamed, 2017).
The existing literature on the link between income diversification and bank stability in conventional banks focuses on advanced
economies, with mixed results and non-conclusive explanations.
Lepetit, Nys, Rous, and Tarazi (2008) investigate bank risk and product diversification in Europe for the period 1996–2002 and
show that higher insolvency probabilities are attributed to firms switching to non-interest-bearing activities, as compared to those
involved in traditional banking.
De Jonghe (2010) explores divergent strategies within the context of specialization and diversification of financial activities and

their impact on bank stability in Europe. The author finds that non-interest-bearing activities increase bank's systematic risk, sug-
gesting that diversification of financial activities does not contribute to stability. Köhler (2015) studies the impact of business models

on stability of EU countries over the period 2002–2011 and finds that banks with an increasing share of non-interest income are more
stable and profitable.
DeYoung and Torna (2013) test if non-traditional banking activities contributed to failures in the US banking industry: they find
that the probability of failure increases with asset-based non-traditional activities (f.i. venture capital or securitizations), while
decreases with pure fee-based operations (f.i. securities brokerage and insurance distribution).
Williams (2016) examines the relationship of Australian banks' income structure and risk, finding that a lower non-interest
income and higher revenue concentration is associated with lower volatility. Although non-interest income is typically found to be
risk increasing, some of its sources are risk decreasing, after controlling for bank specialization effects. Using quarterly data of almost
7000 US commercial banks for the period 2007–2016, Abedifar, Molyneux, and Tarazi (2018) find no adverse effect of non-interest
income on credit risk, while cross-subsidization between non-interest activities and lending is observed for larger banks.
Additionally, there is a growing body of literature investigating the relationship between income diversification and bank stability
in emerging economies.
Sanya and Wolfe (2011) find that diversification across and within both interest and non-interest-bearing activities has positive
impacts on stability and profitability. Using a broad dataset of almost 1000 banks in 55 emerging countries, Amidu and Wolfe (2013)
identify diversification across and within non-interest-bearing activities as a way by which competition improves bank stability.

Similarly, Moudud-Ul-Huq, Ashraf, Gupta, and Zheng (2018) posit that diversification depends on the riskiness of the related ac-
tivities. They find that diversification in the banking sector of Thailand, Vietnam, the Philippines, and Malaysia improves the risk-
return tradeoff.

While investigating the impact of income diversification on profitability and asset quality of Indian banks, Ahamed (2017) finds
that entities with lower asset quality benefit more from diversification than those with higher asset quality.
Finally, to the best of our knowledge, Abuzayed, Al-Fayoumi, & Molyneux, 2018 and AlKhouri and Arouri (2019) are the only
authors that focus on the comparative impact of diversification on the stability of Islamic and conventional banks, yet with
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

3

inconclusive results, focusing only on GCC countries, and for the periods 2001–2014 and 2003–2015 respectively. These authors find
diversification to adversely influence the stability of both banking systems, but the results are more pronounced for conventional
banks. They also find that asset diversification improves the stability of Islamic banks. Despite similarities in the purpose of our
research, our study differs both by considering the impacts of diversification on risk-adjusted profitability jointly with stability, as
well as extending the geographical scope significantly to the OIC region.
Islamic banks are generally risk-averse in their investment-based operations, for instance by not exposing themselves to CDOs or
CDS instruments. This might increase their stability (Hassan, Khan, & Paltrinieri, 2019) on one hand, while it can limit diversification
opportunities.
Therefore, the inconclusive literature on diversification in emerging economies further intensifies our motivation and leads us to
the following additional research questions.
RQ3. : “How does diversification affect the stability of banks?”
RQ4. : “Is there any difference in diversification effects on the stability of Islamic and conventional bank?”
3. Data and methodology
3.1. Data
We collect data in the 2007–2016 period for banks in eleven Members of the Organization of Islamic Cooperation (OIC) for which
Islamic banks represent an important and well-established type of financial institutions. These countries include the GCC area
(namely United Arab Emirates, Saudi Arabia, Qatar, Bahrain, Oman, Kuwait), as well as Malaysia, Indonesia, Pakistan, Bangladesh
and Turkey.
Financial data were obtained from different sources: Bankscope (2007–2010) and Orbis Bank Focus (2011–2016) for accounting
information, Bloomberg Professional Services for the Distance-to-Default (DD) measures and World Bank for macroeconomic figures.1
Since we are interested in the effect of diversification on the risk-return profile of a specific entity, we use consolidated data, where
available, and individual data for the remaining banks.
Matching data from different sources is usually a tricky task to fulfill; since data quality is a major concern in empirical analysis,
we put great efforts in ensuring the reliability of our sample. Bankscope and Orbis Bank Focus share a common set of variables' names
and identification codes; however, some items are calculated with different formulas and identification codes are sometimes the same

for unconsolidated and consolidated financial statements. Then, we started collecting data from both the datasets, checking accu-
rately the consistency of the time series of each entity, matching the unique bank identifying code and financial statement type

(consolidated vs. unconsolidated) in Bankscope and Orbis Bank Focus. Several banks showed overlapping time series: we controlled
that observations coming from different datasets for the same year have the same value. When we found different values (since, as
previously said, some variables are computed with different equations), we re-calculated the corresponding figures in Bankscope
using Orbis Bank Focus standards. For several banks overlapping observations were not available; in that case we checked the
percentage variation of each variable between 2010 and 2011 (i.e. the matching point between the different data sources), starting
from total assets figures. Where we found abnormal values of increase or decrease, hand-made further checks have been made to
explain this event; if nothing seemed to justify these extreme variations, we dropped the corresponding entities from the dataset. This
selection of banks has been matched with Bloomberg Professional Services data using the ISIN code as a unique identifier of each
bank.
The resulting sample is made of 201 banks (47 Islamic and 154 conventional banks) and it is unbalanced, due to the scattered

pattern of available data. However, the sample dataset remains highly representative of the related countries and only few ob-
servations required to be dropped. Furthermore, for the computation of distance to default as proxy of bank stability, the banks are

required to be listed banks. Due to some data unavailability, we are forced to reduce the sample of this analysis to 169 banks (34
Islamic banks and 135 Commercial banks).
We pay great attention to sample composition, excluding banks that were not mainly operating in the traditional “commercial
banking framework” (e.g. several Islamic banks which showed a clear orientation towards investment banking or corporate finance
services). Table 1 describes the sample in greater detail.
3.2. Target variables
Since we are interested in the effects of diversification on banks' risk-adjusted profitability and stability, we use two variables that
are widely used in the literature (Chiorazzo et al., 2008; Maudos, 2017; Stiroh & Rumble, 2006), namely the Risk Adjusted Return On
Average Assets (RAROAA) and the Risk Adjusted Return On Average Equity (RAROAE). Both these measures are calculated dividing

the return on average assets and return on average equity for their respective standard deviation. More specifically, standard de-
viation has been calculated as the variability of ROAA and ROAE for the whole period under consideration. Volatility figures are

1 Specifically, for the variable capturing regulatory and supervisory conditions of each country, we used World Bank's BRSS database, available at
the following website: www.worldbank.org/en/research/brief/BRSS
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

4

affected by the number of data used for the computation: a small number of items can lead to extremely low or high levels of standard
deviation (with the extreme case of zero-volatility that makes impossible to calculate the corresponding risk-return variables). In
order to reduce these drawbacks and improve the quality of data, we performed further robustness checks including only banks with
at least 5 years of consecutive values for ROAA and ROAE; main results are unaffected by this stricter data filtering.
Additionally, bank stability is investigated with both an accounting and a market-based variable, accordingly to the literature

(Abuzayed, Al-Fayoumi, & Molyneux, 2018; Beck, Demirgüç-Kunt, & Merrouche, 2013; Čihák & Hesse, 2010): respectively, the Z-
Score and the distance-to-default.

Z-score is calculated as the sum of the ROAA and the equity-to-asset ratio, divided by the standard deviation of the ROAA. Higher
values of Z-score signal higher resilience and, therefore, more stability. Čihák and Hesse (2010) argue that Islamic banks, by having
large portions of investment account holders (IAH), sharing similarities with equity capital, are not fully reflected in this traditional
measure. Therefore, measures can be biased and lead Islamic banks to be perceived as less stable. In order to control for this issue, we

adopt also a market-based measure of stability, i.e. Merton's Distance to Default (DD). Consistently with the literature (Abuzayed, Al-
Fayoumi, & Molyneux, 2018; Kabir, Andrew, & Gupta, 2015), this measure should be more efficient in predicting bank stability.

A traditional measure of DD is the difference between the market value of assets and a default point, defined as the sum of short
term and half of long term liabilities, divided by the product of the market value of assets and their volatility. Therefore, the higher
the DD, the higher the stability. For this study, we collect default probabilities from Bloomberg Professional Services and measure the
DD by the inverse cumulative distribution function as follows.
Let DD be a standard normal variable, where D~N(0, 1) . The probability of default (Pdefault) is defined as:
Pdefault = − = − =− CDF DD DD DD ( ) ɸ( ) 1 ɸ( )
= ∫−∞
− DD π ɸ( ) e dt 1
2
DD t
2
2

or, equivalently:
= ⎡
⎣
⎢ + ⎛
⎝
⎞
⎠
⎤
⎦
⎥ DD erf DD ɸ( ) 1
2
1
2 (1)

Eq. 1 allows us to define DD from the probability of default, as follows:
ɸ ( ) 2 (2 1), (0, 1) − − P e default = −∈ rfP P default default 1 1 (2)
To measure income diversification, we build a variable (DIV) based on the shares of operating revenues represented by financing
and non-financing streams of income. In order to calculate DIV, we firstly collect data on NONsh, which is the share of operating
revenues attributable to non-interest income (or non-financing income, in the case of Islamic banks). Higher values of NONsh indicate
a greater exposure to non-traditional sources of revenues. According to the literature, observations with values outside the [0;1]
range are excluded. Following recent literature (Abuzayed, Al-Fayoumi, & Molyneux, 2018; Sanya & Wolfe, 2011), we include the
squared value of this variable among the control covariates in our estimations.
DIV, instead, is built accordingly to the Herfindahl–Hirschman Index (HHI), as follows:
DIV 1 {(NONsh) (1 NONsh) } =− +− 2 2 (3)
By definition, DIV values can range between 0 and 0.5, with lower values indicating less diversification. It is worth noting that an
increase in non-traditional activities (NONsh) does not necessarily lead to a greater diversification (DIV): the final effect depends on
the initial level of NONsh. For instance, consider a bank with operating revenues composed by 40% financing/interest-bearing
activities, and 60% for the remaining: its level of diversification would be 0.48. If non-traditional activities increase to 70%,
Table 1
Sample description.
Country Sample dataset Listed banks Of which Islamic Total

Islamic banks Conventional banks

Bahrain 8 9 9 6 17
Oman 3 6 5 9
Kuwait 5 5 9 4 10
Qatar 5 6 8 3 11
Saudi Arabia 4 9 11 4 13
UAE 8 18 16 4 26
Pakistan 2 18 20 2 20
Bangladesh 7 22 29 7 29
Malaysia 1 10 11 1 11
Indonesia 2 41 41 2 43
Turkey 2 10 12 2 12
Total 47 154 171 35 201
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

5

diversification would fall to 0.42. We control for this issue in the robustness checks section, where we test our estimations on a
subsample of banks with NONsh lower than 50%.
Finally, in our estimation DIV is interacted with the dummy variable capturing Islamic banks. This choice aims at testing the
existence of a link between diversification and performance or stability that is contingent on the type of financial institution.
Therefore, this interaction term is the key variable of our analysis, allowing us to check whether Islamic banks per se are different in
terms of profitability or resilience through the pursuit of specific diversification strategies.
3.3. Control variables
Following the literature on income diversification, we add firm-specific variables to control for other effects on profitability and
stability.
We use the natural logarithm of total assets (SIZE) to capture the impact of each bank's dimension. Larger banks might be more
profitable due to economies of scale or scope, or greater investments in technology. We also consider the level and squared value of

total assets growth (Asset Growth) to reflect potential non-linear relationship between bank expansion and the risk-adjusted per-
formance: a greater focus on growth could encompass more relaxed credit screening criteria and lower, in the longer run, profitability

(Chiorazzo et al., 2008).
To control for leverage effects, we use the ratio of tangible equity to total assets (Equity Ratio). Higher values should indicate that
the bank faces less financial fragility.
The ratio between loans and total assets (Loans/TA) is considered to assess the bank's lending strategy. Higher values could
encompass greater profitability but also a greater exposure to credit risks.
Lastly, we also include the cost-to-income ratio (Cost income) to control for bank efficiency; this variable is used only in stability
measures estimations, since it is a relevant component of profitability ratios and this can lead to biased estimates.
Macroeconomic conditions are usually crucial for banks' profitability and overall soundness; we account for this effect including
two variables (GDP and INF) that respectively measure the annual growth of gross domestic product and the level of inflation at
country level.

Moreover, following existing literature (Abuzayed, Al-Fayoumi, & Molyneux, 2018; Barth, Caprio Jr., & Levine, 2004), we in-
troduce a variable (REG) that measures the level of regulatory restrictions in a specific country with reference to several services, such

as brokerage or trading. REG can assume 4 values: 1 (no restrictions); 2 (allowed); 3 (restricted); 4 (prohibited). Since this kind of
regulation can affect both diversification strategies and financial results, the variable is an effective exogenous instrument in our
econometric estimation.
Finally we control for the impact of the financial crisis by adding a dummy variable (Crisis) for years 2008 and 2009.
Table 2 summarizes and briefly describes the variables used in the econometric estimations; the correlation matrix is provided in
Table 3.
3.4. Methodology
Following Stiroh and Rumble (2006), we first estimate the mean values of all variables over the whole sample period. This allows
us to run a first OLS regression to investigate the cross-sectional nature of our sample using the following model:
Table 2
Description of variables.
Type Variable Measure
Dependent variables ROAA Net Income/Average Total Assets
ROAE Net Income/Average Total Equity
σROAA Standard deviation of ROAA
σROAE Standard deviation of ROAE
RAROAE ROAE/σROAE
RAROAA ROAA/σROAA
Z-SCORE (ROAA + Equity/TA)/σROAA
DD Market value of assets Default Point −
(Market value of assets)(Volatility of assets)
Independent variables NONsh2 (Non-interest revenues/Total income)2
DIV 1 - Herfindahl-Hirschman index (built on NONsh)
Size Natural Logarithm of total assets
Equity Ratio Equity/Total Assets
Loans/TA Net Loans/Total assets
Cost income Operating Expenses/Total Revenue
Asset growth Annual growth of total assets (level and squared value)

Macroeconomic variables GDP GDP Growth

INF Inflation (Consumer price index)
REG Regulatory restriction which can take 4 values: 1(no restriction), 2(allowed), 3(restricted), 4(prohibited)
Crisis Dummy variable takes value 1 for the year 2008 and 2009 and 0 otherwise.
This table summarizes our dependent and independent variables, together with the explanation of their measure.
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

6

Table 3
Correlation matrix.
RAROAA RAROAE Z-score DD DIV NONsh2 Size Equity ratio Loans/TA Asset growth Asset growth2 Cost income Crisis GDP INF REG
RAROAA 1
RAROAE 0.7870* 1
Z-score 0.8287* 0.5562* 1
DD 0.2871* 0.2528* 0.2134* 1
DIV −0.0188 −0.0251 −0.1098* 0.1089* 1
NONsh2 0.0708* 0.1086* 0.0407 −0.1123* −0.4617* 1
Size 0.2270* 0.2376* −0.0263 0.3965* 0.3059* −0.1329* 1
Equity Ratio −0.031 −0.1403* 0.1830* 0.1785* −0.1526* −0.0497 −0.2850* 1
Loans/TA −0.0846* −0.1256* −0.1053* 0.0732* 0.1158* −0.3053* 0.0372 −0.1292* 1
asset_g1 −0.0292 0.0145 −0.0031 −0.0049 −0.0695* 0.0374 −0.1775* 0.0427 −0.0271 1
asset_g2 −0.1063* −0.0920* −0.0083 −0.0914* −0.0909* 0.0276 −0.2199* 0.1249* −0.047 0.8071* 1
costincome1 −0.3373* −0.3264* −0.1026* −0.4007* −0.1768* 0.1467* −0.4621* 0.0610* −0.1454* 0.0551* 0.1422* 1
Crisis −0.0679* −0.0495* −0.0869* −0.2481* −0.0047 0.0943* 0.0039 0.0012 0.0029 0.0274 0.0799* −0.0493 1
Gdp 0.0775* 0.1004* 0.0427 0.0592* 0.0247 0.0626* −0.1093* 0.0244 0.0322 0.1769* 0.1237* −0.0156 −0.1887* 1
Inf 0.0148 0.0653* −0.0502* −0.2931* −0.0621* 0.2681* −0.1501* −0.2044* 0.0227 0.0131 0.0285 0.1057* 0.2212* 0.0952* 1
Reg 0.1534* 0.1674* 0.1129* −0.0316 −0.0843* 0.2260* −0.1223* −0.1959* 0.0458 0.0493 0.0038 0.1315* −0.0618* 0.2990* 0.3922* 1
Stars indicate statistical significance at 5% level.

A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx 7

Y α β DIV β DIV Islamic γCV ε i ii i =+ + × + + 0 1 2 i (4)
where Yi is the alternative measure of profitability, profit volatility or stability that we adopt, CVi are firm-specific control variables,
α is the intercept and εi is the error term. DIV x Islamic is an interaction term between our measure of revenue diversification and a
dummy that identifies Islamic banks. We also include country dummies which in these first regressions account for country-specific

macroeconomic conditions and regulatory framework. In order to disentangle the effect of profit level and volatility on our risk-
adjusted profitability measures, we run 4 additional regressions using ROAA and ROAE, and their standard deviation, as dependent

variables.
In the second part of the econometric estimation, we introduce a two-step system GMM model (Arellano & Bover, 1995; Blundell
& Bond, 1998). This approach is particularly effective in dealing with endogeneity problems, which are typical drawbacks of the
analyses of the effect of diversification on financial outcomes in the banking sector.
More specifically our estimations are based on the following equation:
Yit it , ,1 = + + × + + + ++ Y − β DIV β DIV Islamic γCV θM β REG α ε 1 i t, 2 i t, ,it it, 3 it i it , , (5)
Year dummies are included in the level equation. DIV and CV assume the same meaning of Eq. 4; M is a vector of macroeconomic
data (GDP growth and INF) and REG is the level of regulatory restrictions.
3.5. Descriptive statistics

Since bank-specific figures may include borderline observations, we improve the data quality through a light winsorizing ap-
proach (1% each tail). Tests on data excluding the extreme 2 percentiles (one each tail) or more intensive winsorizing (2.5% each tail)

lead our estimations to the same results. Additionally, results are confirmed also when excluding observations falling outside a three
standard deviation range from the mean (for RAROAA, RAROAE and Z-Score figures).
Table 4
Descriptive statistics (original data).
Variable Whole sample Islamic banks Conventional banks
Obs. Mean St. Dev. Min Max Obs. Mean St. Dev. Obs. Mean St. Dev.
RAROAA 1574 3.561 4.380 −3.434 42.032 342 2.115 2.989 1232 3.962 4.615
RAROAE 1575 3.424 3.680 −4.027 27.261 342 3.174 4.612 1233 3.494 3.375
Z-score 1574 32.791 32.220 −2.134 284.912 342 26.522 30.762 1232 34.531 32.411
DD 1288 3.154 0.488 1.051 5.998 263 3.278 0.518 1025 3.122 0.475
NONsh2 1471 0.234 0.235 0.004 1.000 315 0.201 0.205 1156 0.243 0.242
DIV 1471 0.400 0.119 0.000 0.500 315 0.401 0.100 1156 0.399 0.124
Size 1579 15.478 1.690 10.230 19.102 343 15.393 1.359 1236 15.502 1.771
Equity ratio 1579 0.133 0.121 −1.143 0.998 343 0.156 0.208 1236 0.126 0.081
Loans/TA 1575 0.595 0.132 0.033 0.964 339 0.591 0.149 1236 0.596 0.127
Cost income 1572 0.543 0.377 0.067 7.238 338 0.637 0.607 1234 0.517 0.278
Asset growth 1371 0.142 0.378 −0.669 10.533 294 0.215 0.374 1077 0.123 0.376
Asset growth2 1371 0.163 3.040 0.000 110.935 294 0.186 0.987 1077 0.156 3.391
This table summarizes the descriptive statistics for our variables, for the whole sample as well as for our two sub-samples, considering the original
data on our bank population.
Table 5
Descriptive statistics (after winsorization).
Variable Whole sample Islamic banks Conventional banks
Obs. Mean St. Dev. Min Max Obs. Mean St. Dev. Obs. Mean St. Dev.
RAROAA 1574 3.459 3.661 −2.106 23.199 342 2.129 2.967 1232 3.828 3.749
RAROAE 1575 3.411 3.546 −1.926 22.697 342 3.143 4.341 1233 3.485 3.290
Z-score 1574 32.534 30.573 0.744 192.200 342 26.302 29.144 1232 34.264 30.746
DD 1288 3.152 0.463 1.992 4.419 263 3.268 0.487 1025 3.123 0.452
NONsh2 1471 0.234 0.235 0.004 1.000 315 0.201 0.205 1156 0.243 0.242
DIV 1471 0.400 0.119 0.000 0.500 315 0.401 0.100 1156 0.399 0.124
Size 1579 15.478 1.690 10.230 19.102 343 15.393 1.359 1236 15.502 1.771
Equity ratio 1579 0.135 0.094 0.016 0.711 343 0.167 0.136 1236 0.126 0.076
Loans/TA 1575 0.595 0.129 0.189 0.801 339 0.592 0.144 1236 0.596 0.125
Cost income 1572 0.530 0.259 0.186 1.885 338 0.595 0.330 1234 0.511 0.233
Asset growth 1371 0.129 0.191 −0.241 1.081 294 0.191 0.230 1077 0.112 0.176
Asset growth2 1371 0.054 0.146 0.000 1.168 294 0.089 0.206 1077 0.045 0.123
This table summarizes the descriptive statistics for our variables, for the whole sample as well as for our two sub-samples, considering the data
resulting from the winsorization procedure.
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

8

Table 6
Cross section on mean values: Profitability and profit volatility measures.
Variables (1) (2) (3) (4)
ROAA ROAE ROAA (St.Dev.) ROAE (St.Dev.)
DIV 0.00 −0.03 −0.01 0.02
(0.014) (0.065) (0.008) (0.047)
DIV x Islamic −0.02*** −0.07*** 0.01*** 0.03
(0.005) (0.027) (0.003) (0.022)
NONsh2 −0.00 −0.07* 0.01 0.04
(0.007) (0.040) (0.004) (0.027)
Size 0.00*** 0.03*** −0.00** −0.01***
(0.001) (0.004) (0.000) (0.003)
Equity ratio 0.08*** 0.18** 0.01* −0.12**
(0.017) (0.074) (0.007) (0.046)
Loans/TA 0.03** 0.02 −0.00 0.06
(0.011) (0.064) (0.007) (0.036)
Asset growth 0.02 0.21* −0.04*** −0.30***
(0.018) (0.106) (0.012) (0.082)
Asset growth2 −0.03 −0.26** 0.05*** 0.30***
(0.020) (0.108) (0.012) (0.085)
Constant −0.07*** −0.40*** 0.02** 0.17***
(0.016) (0.090) (0.009) (0.057)
Observations 198 198 198 198
R-squared 0.46 0.47 0.43 0.24
This table presents the cross-section effects of income diversification on profitability ratios and their volatility. The squared share of non-interest
income (NONsh2

) and DIV are income diversification variables. The natural log of total assets (SIZE), the equity to total assets (Equity/TA), loans to

total assets (Loans/TA), the annual growth of assets (Asset growth) and its squared value (Asset Growth2

) are the bank-specific control variables.
Islamic is dummy variable to control the specialization effect of a bank being Islamic: it is used in the interaction term (DIV x Islamic). Robust
standard errors in parentheses. Significance codes: *** indicate statistical significance at 1%, ** at 5% and * at 10%, respectively.
Table 7
Cross section on mean values – Risk adjusted and stability measures.
Variables (1) (2) (3) (4)
RAROAA RAROAE Z-score DD
DIV −6.51 −8.52* −59.71 −0.11
(4.895) (4.379) (41.035) (0.363)
DIV x Islamic −3.20** 0.49 −11.46 0.37***
(1.508) (1.550) (14.222) (0.142)
NONsh2 −3.08 −1.65 −23.30 0.15
(2.881) (2.336) (23.105) (0.165)
Size 0.97*** 1.04*** −0.57 0.07***
(0.217) (0.214) (2.070) (0.020)
Equity ratio 6.46 1.41 40.02 1.18***
(3.939) (2.912) (34.860) (0.442)
Loans/TA 0.10 −1.13 −29.28 0.14
(3.958) (2.677) (34.291) (0.315)
Cost income −28.54*** −0.54***
(9.083) (0.162)
Asset growth 9.86** 9.15** 126.67*** 0.39
(3.882) (4.239) (45.798) (0.594)
Asset growth2 −9.92** −9.38** −119.86*** −0.79
(3.866) (4.118) (39.624) (0.773)
Constant −11.07** −10.63*** 78.97* 1.73***
(4.270) (3.630) (44.898) (0.443)
Observations 198 198 198 169
R-squared 0.34 0.30 0.27 0.64
This table presents the cross-section effects of income diversification on risk-adjusted profitability ratios and profit stability measures. The squared
share of non-interest income (NONsh2

) and DIV are income diversification variables. The natural log of total assets (SIZE), the equity to total assets

(Equity/TA), loans to total assets (Loans/TA), the annual growth of assets (Asset growth) and its squared value (Asset Growth2

) are the bank-specific
control variables. Islamic is dummy variable to control the specialization effect of a bank being Islamic: it is used in the interaction term (DIV x
Islamic). Robust standard errors in parentheses. Significance codes: *** indicate statistical significance at 1%, ** at 5% and * at 10%, respectively.
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

9

Tables 4 and 5 present the summary statistics of our data – respectively before and after the winsorizing process – with evidence of
mean values and volatility of each variable for the sub-samples of Islamic and conventional banks.
Mean values of risk-adjusted profitability measures and Z-Score are higher for conventional banks, especially in the case of
RAROAA, while the opposite is true for DD. Differences of Islamic banking with impacts on stability can be explained by the effects of
Sharia compliance, namely higher liquidity and capital ratios (Abedifar, Molyneux, & Tarazi, 2013), a better asset quality (Beck et al.,
2013) and the lack of exposure to derivatives (Ahmed, 2009). Interestingly, Islamic banks show levels of diversification similar to
conventional banks (DIV around 0.4 for both sub-samples), but at the same time experience a slightly lower value of NONsh2 (0.20
versus 0.24). The main reason for this finding is a lower exposure to non-financing income. These results are consistent with the
previous literature (Abuzayed, Al-Fayoumi, & Molyneux, 2018; Chen, Liang, & Yu, 2018; Molyneux & Yip, 2013).
We also find that Islamic banks in our sample are only slightly smaller in size and show marginally higher equity ratios. Finally,
cost-to-income ratios are significantly higher (0.59 versus 0.51), consistently with potential unexploited scale economies and greater
monitoring costs.
Table 8
Baseline estimation (dynamic panel model).
Variables (1) (2) (3) (4)
RAROAA RAROAE Z-score DD
Dependentt-1 0.79*** 0.80*** 0.88*** 0.29***
(0.041) (0.040) (0.035) (0.048)
DIV 0.27 0.44 −3.03 0.10
(0.563) (0.591) (4.705) (0.122)
DIV x Islamic −1.27** −0.60 −11.49*** 0.06
(0.520) (0.517) (3.226) (0.218)
NONsh2 0.79*** 0.57* −2.04 0.20*
(0.304) (0.316) (2.715) (0.105)
Size 0.12*** 0.08*** −0.06 0.06***
(0.035) (0.027) (0.238) (0.019)
Equity ratio 0.91 −0.42 13.20** 0.99**
(0.804) (0.458) (6.089) (0.466)
Loans/TA 0.09 −0.61 −2.22 0.06
(0.554) (0.697) (3.453) (0.168)
Asset growth 1.07*** 1.65*** −11.92*** 0.51***
(0.361) (0.328) (3.290) (0.142)
Asset growth2 −1.05* −1.66*** 5.94 −0.52**
(0.557) (0.542) (4.444) (0.254)
Cost income −4.88* −0.37**
(2.728) (0.166)
Crisis −0.46*** −0.38*** −1.29*** −0.25***
(0.093) (0.102) (0.463) (0.057)
GDP 0.02 0.02* −0.08 −0.01***
(0.012) (0.013) (0.055) (0.004)
INF −0.05*** −0.04** −0.16* −0.03***
(0.016) (0.016) (0.093) (0.007)
REG 0.15 0.12 1.73** 0.15**
(0.123) (0.101) (0.819) (0.065)
Constant −1.75** −0.76 9.11 1.21***
(0.776) (0.739) (6.605) (0.424)
Observations 1281 1279 1279 1024
Number of banks 199 199 199 165
AR1 0.000 0.000 0.000 0.000
AR2 0.569 0.212 0.753 0.570
Hansen test 0.489 0.480 0.377 0.967
No of instruments 198 197 198 198
F Test F(13, 198) = 110.90 F(13, 198) = 108.26 F(14, 198) = 127.58 F(14, 164) = 68.72
Prob > F 0.000 0.000 0.000 0.000
This table presents the impact of diversification on profitability and stability measures using a two-steps system GMM approach. Each of the four
models includes the lagged dependent variable (Dependentt-1). Bank profitability measures are the risk-adjusted return on average assets (RAROAA)
and the risk-adjusted return on average equity (RAROAE); stability measures are the Z-Score and the distance to default (DD). The squared share of
non-interest income (NONsh2

) and DIV are income diversification variables. The natural log of total assets (SIZE), the equity to total assets (Equity/
TA), loans to total assets (Loans/TA), the cost to income ratio (Cost income), Annual growth of assets (Asset growth) and its square (Asset Growth^2) are
the bank-specific control variables. Macroeconomic variables are GDP growth (GDP) and inflation (INF). REG measures the level of regulatory
restrictions and Crisis is a dummy equal to 1 for years 2008 and 2009. Islamic is dummy variable to control the specialization effect of a bank being
Islamic, used in the interaction term (DIV x Islamic). Robust standard errors in parentheses. Significance codes: *** indicate statistical significance at
1%, ** at 5% and * at 10%, respectively.
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

10

4. Discussion of findings
   4.1. Income diversification, bank profitability and stability
   Table 6 shows the results of our OLS-based analysis of the relationship between diversification and profitability (mean and
   volatility values).
   We observe a general lack of statistical significance of the coefficients associated with our measures of diversification. However,
   we find that a significant role in explaining the pattern of financial outcomes is played by the interaction term (DIV x Islamic). More
   specifically, for Islamic banks diversification lowers profitability ratios, especially in the case of the ROAE, and increases their
   volatility, however with statistical significance only in the case of the ROAA.
   With reference to firm-specific variables, we observe that size, regulatory capital and growth enhance profits level and reduce
   their variance, while the impact of loan share is scarcely significant from a statistical point of view, except for the ROAA. These
   outcomes are consistent with previous literature, which argues that banks with higher capital ratios are associated with greater
   profitability, because they are less dependent from borrowing and are more prudent while lending (Tan, 2016).
   Table 9
   Only GCC banks
   Variables (1) (2) (3) (4)
   RAROAA RAROAE Z-score DD
   Dependentt-1 0.40*** 0.54*** 0.78*** 0.15*
   (0.076) (0.083) (0.149) (0.074)
   DIV 5.67* 6.10** −1.94 0.79
   (3.340) (2.544) (10.235) (0.702)
   DIV x Islamic −4.89** −2.39* −3.17 −0.72
   (2.114) (1.331) (7.874) (0.595)
   NONsh2 −1.25 −0.17 −2.49 −0.65
   (1.343) (1.174) (4.663) (0.422)
   Size 0.03 −0.07 0.56 0.01
   (0.251) (0.157) (0.600) (0.061)
   Equity ratio 2.61 0.28 21.55* 1.32**
   (1.772) (1.264) (12.797) (0.657)
   Loans/TA −0.42 0.47 −2.39 −0.23
   (2.481) (1.521) (6.092) (0.352)
   Asset growth 2.15*** 1.74*** −21.10** −0.44
   (0.556) (0.545) (8.007) (0.333)
   Asset growth2 −2.57*** −2.01*** 16.39* 0.33
   (0.693) (0.683) (8.829) (0.450)
   Cost income −7.29 −0.19
   (7.330) (0.325)
   Crisis −0.34 −0.43* −1.79** −0.38***
   (0.240) (0.232) (0.714) (0.084)
   GDP 0.04 0.05** 0.07 −0.00
   (0.022) (0.021) (0.077) (0.006)
   INF −0.00 −0.02 0.07 −0.01
   (0.028) (0.024) (0.136) (0.009)
   REG 0.45 0.31 −0.09 0.31***
   (0.422) (0.382) (1.366) (0.075)
   Constant −1.36 −0.79 1.32 2.32**
   (3.632) (2.362) (12.674) (1.145)
   Observations 626 626 626 479
   Number of banks 85 85 85 58
   AR1 0.000 0.000 0.005 0.002
   AR2 0.566 0.576 0.196 0.085
   Hansen test 0.244 0.130 0.188 0.679
   No of instruments 72 72 72 72
   F Test F(13, 84) = 10.92 F(13, 84) = 13.05 F(14, 84) = 30.41 F(14, 57) = 15.68
   Prob > F 0.000 0.000 0.000 0.000
   This table presents the impact of diversification on profitability and stability measures using a two-steps system GMM approach, with our sample

limited to GCC banks. Each of the four models includes the lagged dependent variable (Dependentt-1). Bank profitability measures are the risk-
adjusted return on average assets (RAROAA) and the risk-adjusted return on average equity (RAROAE); stability measures are the Z-Score and the

distance to default (DD). The squared share of non-interest income (NONsh2

) and DIV are income diversification variables. The natural log of total
assets (SIZE), the equity to total assets (Equity/TA), loans to total assets (Loans/TA), the cost to income ratio (Cost income), Annual growth of assets
(Asset growth) and its square (Asset Growth^2) are the bank-specific control variables. Macroeconomic variables are GDP growth (GDP) and inflation
(INF). REG measures the level of regulatory restrictions and Crisis is a dummy equal to 1 for years 2008 and 2009. Islamic is dummy variable to
control the specialization effect of a bank being Islamic, used in the interaction term (DIV x Islamic). Robust standard errors in parentheses.
Significance codes: *** indicate statistical significance at 1%, ** at 5% and * at 10%, respectively.
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

11

We extend this analysis by employing our set of risk-adjusted profitability measures and stability indicators as dependent vari-
ables (Table 7).

Once more, the interaction term shows a negative and significant coefficient for the RAROAA. Moreover, strong statistical sig-
nificance is present also in the case of the DD, with a positive coefficient.

Size and growth maintain a positive effect on our dependent variables across most estimations; however, the negative coefficient
associated to the squared value of growth suggests a pattern of diminishing marginal returns from asset expansion.
After analyzing the cross-sectional nature of our data, we are interested in increasing the robustness and depth of our findings by
exploring the panel dimension of our sample. We therefore implement a GMM panel regression, whose results are provided in
Table 8.
In line with earlier findings, the interaction term shows negative coefficients for risk-adjusted profitability measures. The same
occurs for the Z-Score estimation, while a positive coefficient is found for the DD, despite the lack of statistical significance. These

results are in line with the literature (Chen, Liang, & Yu, 2018; Molyneux & Yip, 2013) and indicate that for Islamic banks di-
versification strategies are likely to negatively affect profitability and stability measures and reveal differences with conventional

banks. We argue that Islamic banks, being smaller and having a limited client base may incur higher fixed costs (DeYoung & Roland,
Table 10
Banks with average non-interest income lower than 50%.
Variables (1) (2) (3) (4)
RAROAA RAROAE Z-score DD
Dependentt-1 0.71*** 0.66*** 0.89*** 0.21***
(0.070) (0.066) (0.065) (0.055)
DIV 2.35*** 3.06*** −2.79 0.75*
(0.852) (0.991) (8.603) (0.428)
DIV x Islamic −3.03* −3.23** −7.92 −0.25
(1.563) (1.575) (7.299) (0.588)
NONsh2 −0.60 −1.02 −4.64 −0.40
(0.628) (0.631) (3.861) (0.283)
Size 0.10** 0.05 −0.05 0.08***
(0.048) (0.050) (0.405) (0.032)
Equity ratio 1.09 −0.21 13.61 1.33***
(1.014) (0.823) (11.886) (0.476)
Loans/TA 0.32 −0.58 −1.31 −0.22
(1.126) (0.908) (4.527) (0.289)
Asset growth 0.73* 1.08** −11.59** 0.40**
(0.414) (0.434) (4.450) (0.172)
Asset growth2 −0.88 −1.40** 4.22 −0.73**
(0.532) (0.586) (5.701) (0.346)
Cost income −5.04 −0.53**
(3.922) (0.236)
Crisis −0.45*** −0.34** −1.31** −0.17***
(0.163) (0.164) (0.603) (0.065)
GDP 0.02 0.05** −0.01 −0.00
(0.016) (0.017) (0.070) (0.006)
INF −0.04 −0.02 −0.03 −0.01
(0.027) (0.025) (0.099) (0.013)
REG 0.22 0.09 1.52* 0.13*
(0.201) (0.211) (0.829) (0.079)
Constant −1.99 −0.67 6.88 1.40**
(1.242) (1.123) (11.379) (0.620)
Observations 910 910 910 626
Number of banks 134 134 134 105
AR1 0.000 0.000 0.000 0.000
AR2 0.180 0.192 0.428 0.773
Hansen test 0.177 0.360 0.299 0.136
No of instruments 72 72 72 71
F Test F(13, 133) = 21.16 F(13, 133) = 32.78 F(14, 133) = 68.65 F(14, 133) = 11.54
Prob > F 0.000 0.000 0.000 0.000
This table presents the impact of diversification on profitability and stability measures using a two-steps system GMM approach, with our sample
limited to banks with an average level of non-interest income lower than 50%. Each of the four models includes the lagged dependent variable
(Dependentt-1). Bank profitability measures are the risk-adjusted return on average assets (RAROAA) and the risk-adjusted return on average equity
(RAROAE); stability measures are the Z-Score and the distance to default (DD). The squared share of non-interest income (NONsh2
) and DIV are
income diversification variables. The natural log of total assets (SIZE), the equity to total assets (Equity/TA), loans to total assets (Loans/TA), the cost
to income ratio (Cost income), Annual growth of assets (Asset growth) and its square (Asset Growth^2) are the bank-specific control variables.
Macroeconomic variables are GDP growth (GDP) and inflation (INF). REG measures the level of regulatory restrictions and Crisis is a dummy equal to
1 for years 2008 and 2009. Islamic is dummy variable to control the specialization effect of a bank being Islamic, used in the interaction term (DIV x
Islamic). Robust standard errors in parentheses. Significance codes: *** indicate statistical significance at 1%, ** at 5% and * at 10%, respectively.
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

12

2001): this translates in lower profitability and, eventually, influences the bank's stability. Moreover, Islamic banks are subject to
restrictions in terms of admissible non-Islamic financial services under the Shariah. Furthermore, more mature conventional banks
may have already achieved the desired level of diversification, hence explaining the related non-significant results.
Size, regulatory capital and growth assume the same positive sign previously described. We also find positive and statistically
significant coefficients for NONsh2

: hence, a greater share of non-financing revenues seems to increase the level of profitability/

stability.
Crisis dummies show a statistically significant negative impact on risk-adjusted profitability for both Islamic and conventional
banks. One could expect the impact of the subprime financial crisis to be limited in OIC countries, due to their marginal exposure to
the main asset classes involved (IMF, 2010). However, emerging countries and their banking systems suffered indirectly from the
changing global economic and financial landscape, consistently with their broad reliance on foreign capitals and export of natural
resources.
Negative and significant coefficients are found also for inflation, while GDP has a scattered pattern of results. Interestingly,
stricter regulations have a positive impact on the measures of stability.
Table 11
Banks with average cost-income ratio below the median value.
Variables (1) (2) (3) (4)
RAROAA RAROAE Z-score DD
Dependentt-1 0.64*** 0.64*** 0.91*** 0.15**
(0.104) (0.100) (0.092) (0.073)
DIV 2.47** 4.38*** 5.54 0.01
(1.082) (1.299) (8.489) (0.659)
DIV x Islamic −3.39*** −4.35** 1.26 −1.04
(1.185) (2.022) (8.775) (1.143)
NONsh2 1.27** 1.58** 4.71 0.10
(0.633) (0.617) (4.105) (0.324)
Size 0.22*** 0.08 0.31 0.06
(0.072) (0.089) (0.312) (0.039)
Equity ratio 4.17* 1.42 21.82 0.94
(2.449) (1.620) (17.988) (0.874)
Loans/TA −1.28 −1.66 −0.20 0.12
(1.531) (1.504) (4.261) (0.335)
Asset growth 2.15*** 2.01*** −15.19** 0.46
(0.747) (0.633) (7.482) (0.296)
Asset growth2 −2.11* −1.54 11.55 0.10
(1.102) (0.966) (11.983) (0.471)
Cost income −14.54 −0.96
(10.432) (0.821)
Crisis −0.50** −0.40* −2.38** −0.38***
(0.207) (0.229) (0.918) (0.069)
GDP 0.01 0.03* −0.07 −0.01**
(0.019) (0.020) (0.073) (0.004)
INF −0.05* −0.04 −0.03 −0.03***
(0.028) (0.029) (0.102) (0.009)
REG 0.21 0.00 0.67 0.21**
(0.296) (0.331) (0.735) (0.089)
Constant −3.36 −1.21 −1.81 1.90
(2.035) (1.802) (10.646) (1.281)
Observations 694 694 694 574
Number of banks 100 100 100 87
AR1 0.000 0.000 0.000 0.000
AR2 0.240 0.746 0.321 0.496
Hansen test 0.120 0.233 0.460 0.315
No of instruments 72 72 72 72
F Test F(13, 99) = 23.02 F(13, 99) = 19.46 F(14, 99) = 101.98 F(14, 86) = 20.76
Prob > F 0.000 0.000 0.000 0.000
This table presents the impact of diversification on profitability and stability measures using a two-steps system GMM approach, with our sample
limited to banks with an average cost-income ratio below the sample median value. Each of the four models includes the lagged dependent variable
(Dependentt-1). Bank profitability measures are the risk-adjusted return on average assets (RAROAA) and the risk-adjusted return on average equity
(RAROAE); stability measures are the Z-Score and the distance to default (DD). The squared share of non-interest income (NONsh2
) and DIV are
income diversification variables. The natural log of total assets (SIZE), the equity to total assets (Equity/TA), loans to total assets (Loans/TA), the cost
to income ratio (Cost income), Annual growth of assets (Asset growth) and its square (Asset Growth^2) are the bank-specific control variables.
Macroeconomic variables are GDP growth (GDP) and inflation (INF). REG measures the level of regulatory restrictions and Crisis is a dummy equal to
1 for years 2008 and 2009. Islamic is dummy variable to control the specialization effect of a bank being Islamic, used in the interaction term (DIV x
Islamic). Robust standard errors in parentheses. Significance codes: *** indicate statistical significance at 1%, ** at 5% and * at 10%, respectively.
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

13

4.2. Robustness checks
We apply a series of robustness checks to assess the validity of our findings, presenting in this section the results of the four
leading alternative settings.
Firstly, we run our analysis on GCC countries only, since these nations have been largely investigated in previous research as

leading hubs for Islamic finance (Table 9). Our results remain consistent: the interaction term that accounts for the effect of di-
versification for Islamic banks on profitability and stability measures is negative and strongly significant for RAROAA and RAROAE,

while not statistically significant in the other regressions.
In a second test (Table 10), we consider the impact of income diversification on banks with an average share of non-financing
income lower than 50%. This allows us to obtain a sample strictly composed by banks focused on interest or financing income. Our
results are largely unaffected by this alternative setting: the interaction term is still associated with negative coefficients. Interestingly
DIV shows positive and strongly significant coefficients. These outcomes seem to support a picture in which banks focused on the
traditional borrowing and lending activities enjoy a greater effect of diversification; however, for Islamic banks, this benefit is lower.
Lastly, since diversification is a strategy that is likely to increase the level of complexity and costs of a bank, we split our sample
Table 12
Banks with average cost-income ratio above the median value.
Variables (1) (2) (3) (4)
RAROAA RAROAE Z-score DD
Dependentt-1 0.67*** 0.67*** 0.76*** −0.00
(0.077) (0.080) (0.046) (0.115)
DIV 1.24 1.35 −3.08 0.12
(1.118) (1.056) (7.008) (0.334)
DIV x Islamic −2.37 −2.19* −12.80 0.41
(1.464) (1.226) (9.448) (0.551)
NONsh2 −0.01 −0.36 −4.87 0.18
(0.692) (0.727) (3.151) (0.187)
Size 0.10 0.15** −0.50 0.08***
(0.070) (0.063) (0.716) (0.029)
Equity ratio 0.78 −0.39 15.94** 1.34***
(1.062) (0.820) (7.383) (0.496)
Loans/TA −0.08 −0.60 −1.34 0.18
(1.029) (0.864) (5.794) (0.280)
Asset growth 0.71 0.75 −5.40** 0.65***
(0.526) (0.462) (2.578) (0.189)
Asset growth2 −1.01 −0.94 0.94 −0.64*
(0.646) (0.614) (3.471) (0.350)
Cost income −5.35 −0.34*
(5.437) (0.196)
Crisis −0.33* −0.33** −0.71 −0.27***
(0.183) (0.138) (0.771) (0.062)
GDP 0.02 0.02 −0.16 −0.01
(0.030) (0.028) (0.173) (0.012)
INF −0.10** −0.06* −0.51*** 0.00
(0.047) (0.035) (0.165) (0.011)
REG 0.44 0.23 4.70*** 0.06
(0.432) (0.315) (1.634) (0.187)
Constant −1.41 −1.42 15.36 1.57**
(1.389) (1.321) (17.786) (0.707)
Observations 587 585 585 450
Number of banks 99 99 99 78
AR1 0.000 0.000 0.001 0.020
AR2 0.135 0.132 0.254 0.837
Hansen test 0.238 0.525 0.481 0.251
No of instruments 72 72 72 72
F Test F(13, 98) = 20.97 F(13, 98) = 27.88 F(14, 98) = 57.75 F(14, 77) = 11.39
Prob > F 0.000 0.000 0.000 0.000
This table presents the impact of diversification on profitability and stability measures using a two-steps system GMM approach, with our sample
limited to banks with an average cost-income ratio above the sample median value. Each of the four models includes the lagged dependent variable
(Dependentt-1). Bank profitability measures are the risk-adjusted return on average assets (RAROAA) and the risk-adjusted return on average equity
(RAROAE); stability measures are the Z-Score and the distance to default (DD). The squared share of non-interest income (NONsh2
) and DIV are
income diversification variables. The natural log of total assets (SIZE), the equity to total assets (Equity/TA), loans to total assets (Loans/TA), the cost
to income ratio (Cost income), Annual growth of assets (Asset growth) and its square (Asset Growth^2) are the bank-specific control variables.
Macroeconomic variables are GDP growth (GDP) and inflation (INF). REG measures the level of regulatory restrictions and Crisis is a dummy equal to
1 for years 2008 and 2009. Islam;8ic is dummy variable to control the specialization effect of a bank being Islamic, used in the interaction term (DIV
x Islamic). Robust standard errors in parentheses. Significance codes: *** indicate statistical significance at 1%, ** at 5% and * at 10%, respectively.
A. Paltrinieri, et al. Global Finance Journal xxx (xxxx) xxxx

14

into two parts, respectively showing a level of the cost income ratio below or above the median value. Our results (Tables 11 and 12)

remain consistent with previous findings, while some differences emerge between the sub-samples. More specifically, banks char-
acterized by higher cost efficiency can gain from a greater level of diversification and the same occurs increasing the share of non-
financing revenues. However, for Islamic banks, these diversification strategies are less effective in boosting risk-adjusted profits. All

these relationships are strongly statistically significant for RAROAA and RAROAE estimations. For less efficient banks this statistical

significance disappears: this outcome is coherent with a framework in which the costs linked to the pursuit of diversification stra-
tegies can offset its benefits. This can be one of the “dark sides” of diversification.

5. Conclusions
   Several studies focused on income diversification and its impact on profitability and risk in conventional banks, both in developed
   and emerging economies. This paper extends this literature by investigating these issues, in a comparative and extended framework,
   using a comprehensive high-quality dataset of 47 Islamic and 154 conventional banks from 11 countries in the OIC region.
   Our main results suggest that diversification provides a different outcome for Islamic banks than conventional institutions: for the
   former, both the profitability and the stability are reduced, adding additional empirical evidence to the existing literature and
   supporting the conclusion that revenue diversity should not be considered as a rewarding strategy per se. In particular, our results are
   stronger when we consider accounting-based measures (the ROAA or ROAE, their standard deviation and the Z-Score) rather than
   market-based ones (the distance-to-default).
   Additionally, we find that an increase in the share of non-financing income is associated with both an increased profitability and
   stability, regardless of the banking business model.
   Finally, our robustness checks underline that when cost-income ratios are above median values, the inefficiency proxied by this
   variable leads the significance of our independent variables to disappear almost entirely.
   These findings are consistent with the existing literature, especially considering the growing body of research on the limitations
   and undesired effects of diversification. We show that, contingent on the alternative measure used for profitability and stability,
   Islamic banks show differences from their conventional counterparts in the same geographical area, but the partial instability of
   results across different settings seems to underline that revenue diversity is not always beneficial, but contingent on firm- and
   environment-specific conditions.
   Regulators and bank managers should consider the implications of these results, as well as the need to explore the link between
   diversification and performance further.
   Declaration of competing interest
   The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
   influence the work reported in this paper.