The Impact of Non-Performing Loans on Credit Growth of Commercial Banks in Cambodia

Abstract

This study investigated how banks’ balance sheet fundamentals shape their credit growth using panel co-integration methods and two estimation methods—pooled mean group (PMG) and dynamic fixed effects (DFE). Both approaches yielded consistent core results. First, weaker asset quality, proxied by higher non-performing loans (NPLs), was strongly and negatively related to credit growth: PMG produced a large negative long-run coefficient, and DFE’s error-correction form confirmed a significant adverse effect, consistent with higher provisioning, thinner capital buffers, and lower risk-taking. Second, capitalization (equity to assets) supported long-run growth under PMG, while DFE—imposing common slopes—did not, suggesting heterogeneous capitalization effects across banks that PMG captured but DFE muted. Third, operating expense intensity showed a positive long-run association with credit growth in both models, consistent with expansionary spending accompanying durable lending rather than costs causing lending. Long-run effects for liquidity and market-risk sensitivity were weaker or mixed: liquidity’s role was imprecise, and market-risk sensitivity was positive in PMG but not significant in DFE, again pointing to cross-sectional heterogeneity. Error-correction terms were large, negative, and highly significant in both models, indicating rapid convergence—near full adjustment within one period, with slight overshooting in DFE. Short-run results showed that higher liquidity and temporary cost spikes dampened contemporaneous growth. Policy implications emphasize sustained oversight of asset quality and prudent capital planning to support long-run credit supply.

1. Introduction

Commercial banks fund most of their activities with customer deposits, which appear as liabilities and provide a relatively stable, low-cost source of funds for lending. These funds are not free: banks pay interest and incur servicing costs, so the deposit mix directly influences the marginal cost of extending new credit and the pricing of loans. On the asset side, loans constitute banks’ principal earning assets and the main driver of net interest income. When loans are performing, contractual interest continues to accrue and cash flows arrive as scheduled, supporting profitability and capital formation (Mishkin & Eakins, 2018; Saunders & Cornett, 2019).

It is a different picture as non-performing loans (NPLs) become more prevalent. Once loans become non-performing, interest accruals are generally suspended or reversed, expected credit losses must be recognized through provisions, and further capital could be needed to absorb losses (Basel Committee, 2016; Montanaro, 2019). These changes do squeeze interest income and restrict lending ability, even in the face of continuing problem loans. Cross-country evidence indicates that higher NPL ratios undermine bank profitability, increase funding costs, and slow credit formation through lower earnings, thinner capital cushions, and lower risk taking. In short, deposits finance lending but have explicit costs; loans create income if and only if the quality of the asset is preserved. Thus, proactive credit risk management that reduces NPL exposure is critical for protecting interest income as well as maintaining the capital base and continuing to lend money to the real sector (Klein, 2013).

Banks typically lend more slowly and selectively when non-performing loans increase in prevalence. The intermediate term channel is balance-sheet pressure: interest income halts on impaired debts as provisions rise, squeezing capital ratios. Confronted with these limits, banks typically retrench—whether rationing new credit, tightening terms, or redeploying limited balance-sheet capacity to safer assets. This behavior is supported by supervisory oversight and prudential standards: enhanced loan loss provisions that absorb more of current period earnings and capital both in absolute terms as well as relative to NPLs subside, discouraging rapid loan growth before addressing legacy stock NPLs. Empirically, this can be seen in two forms. First, credit supply significantly improves when regulators lean on banks to sort out NPLs—support for the view that NPL resolution releases balance-sheet and management capacity for lending (Baskaya et al., 2023). Secondly, in times of asset quality deterioration and increased risk-taking, European survey and risk assessment evidence indicates that European credit standards tighten due to an added scrutiny on loan growth, particularly for riskier credits; the most recent supervisory reviews link the 2023–2024 tightening cycle to sourer loan flows and greater selectivity among banks (European Banking Authority, 2024). With the exception of the European Banking Authority, aggregate NPL ratios in advanced banking systems were still low in 2024, but official review reports signal a cyclical increase from troughs with the warning that it would provide headwinds to lending through higher cost of risk and balance-sheet costs (European Central Bank, 2024).

A decrease in the quality of assets (i.e., an increase in NPLs) leads, as previously noted, to banks being unwilling to expand lending. High-NPL-ratio institutions generally turn back to balance-sheet repair to comply with central bank requirements and may seek further capital for risk reduction. Empirically, little is known about the effect of NPLs in Cambodia on lending by banks and research that comprehensively examines both short-run dynamics as well as long-run relationships between balance-sheet variables and loan growth is scarce. Moreover, few studies in Cambodia have explored the role of non-performing loans on credit expansion in commercial banks. In this study, static panel estimation methods characterized by pooled OLS, random effect, and fixed effect models were utilized to assess the effect of the five CAMEL components, namely capital adequacy, asset quality, management efficiency, earnings performance, and liquidity, with asset quality proxied by the NPL ratio. The results showed statistically significant effects for these components, and the estimated coefficient for the effect of asset quality on loan growth was retained. The coefficient for the effect of asset quality was statistically significant at 5% and remained stable at the 5% level of significance across the three models (Lim, 2023). To fill this gap, we used a panel ARDL model to examine the influence of NPLs on credit growth in Cambodian commercial banks.

Section 1 of this paper introduces the study, its background, and the research gap. The literature review is covered in Section 2, which includes the relevant theories and empirical evidence on the research topic. Section 3 presents the model specification—short- and long-run panel ARDL models—data collection, methods of analysis, and hypothesis testing. Section 4 and Section 5 present the empirical findings and conclusion, respectively.

2. Literature Reviews

Analysis has found that banks with a high share of non-performing loans in their portfolio appear to behave in two characteristic ways. First, they are unwilling to lend even when offered creditworthy new customers. Secondly, prospective losses and the need to satisfy requirements regarding capital established by the central bank may prompt banks’ owners to drag their feet on new capital placement. The rationale for these actions is to prevent bank failure, which is more probable when the proportion of NPLs is higher. Moreover, if there is mandatory disposal of NPLs or prudential rules are enforced to write down bank capital in line with losses associated with legacy assets, only the owners of banks need face those losses. However, practically, some bank managers may decide not to acknowledge those losses and hold onto the non-performing loan assets on their balance sheet (Segura & Suarez, 2019).

The question of whether non-performing loans affect performing loans was explored using data for European banks during the period covering December 2014 to June 2018. The date range of this research was derived from the European Banking Authority, contributed by M’esonnier and Monks (2015) and Gropp et al. (2019). During this period, 75 banks were studied using two different econometric models. Both models had the same goal—to predict changes in performing loans. The first was based on leveled data, while the second followed a first-difference specification. The independent variables were represented by two indicators: gross performing loans (in thousand euros) and growth of gross performing loans. These independent variables were further categorized into bank-specific (using NPLs as the leading variable) and country-level variables. The models were estimated with panel least squares and the generalized method of moments (GMM). The results indicated a detrimental effect of NPLs on gross performing loan growth. Finally, all of the econometric models consistently provided the same outcome, with greater persistence found in banks showing less growth in gross performing loans (Serrano, 2021).

A range of factors influence the financial condition of banks and their ability to continue lending, with capital adequacy being a key one. Many studies have empirically investigated the relationship between bank capital and lending behavior and indicated that financial health is conducive to credit supply. Empirical analysis of the determinants of bank profitability has also revealed that capital adequacy has a positive effect on lending. Moreover, banks with higher capital buffers can lend to the public even under poor economic conditions (Athanasoglou et al., 2008). In the case of the U.S. financial crisis, for example, evidence showed that banks with higher capital ratios fared better in retaining—or even growing—their loan portfolios during periods of stress. During that time, capital was not only a cushion against loss, but also a sign of commitment, enhancing the confidence between borrowers and holders of debt claims (Berger & Bouwman, 2013). Using an alternative sample of Asian commercial banks, the analyses revealed a non-linear relationship between capital adequacy and loan growth (banks with very high level of capital ratios were not incentivized to take credit risks), implying the threshold where excess capitalization would hinder loan expansion (Lee & Hsieh, 2013).

Increases in the capital of commercial banks contributed positively to quality loan portfolios in Bangladesh, following the CAMELS framework. Importantly, well-capitalized banks not only provided additional loans but also reported better loan quality, supporting the two functions of capital to both expand and protect banks’ loan books (Ali et al., 2021). These findings were supported by the experience of emerging markets presented in this study, which revealed that stronger capital adequacy increased banks’ lending volume capacity. Larger capital adequacy ratios lead to greater growth of the loan portfolio, particularly in countries with more rigorous regulatory control (Akhtar et al., 2019). It is well documented that increased NPL ratios—which signal weaker asset quality—constrain commercial bank lending by increasing provisioning, depleting capital buffers, and depressing risk appetite. This is the sobering effect that usually manifests itself whenever regulation becomes lax or during times when the economy is not performing well. A systematic review revealed that high NPL ratios contracted banks growth by undermining banks’ balance sheet strength (Ozili, 2025). In addition, empirical evidence has shown that elevated NPL ratios negatively affect interest income and generate higher provisioning costs while also inhibiting bank liquidity and dampening new loans disbursements—these effects are more pronounced in developing countries with weak regulatory equipment (Khairi et al., 2021). Moreover, the increasing level of NPLs undermines investor confidence; this further contributes to increased financial unsustainability and reduced bank-based credit formation (Naili & Lahrichi, 2022).

A panel vector autoregressive (PVAR) model was employed to investigate the impact of NPLs on loan growth. The results also demonstrated that a higher NPL ratio suppresses loan growth rate, especially in the short run, suggesting that loan growth influences and is affected by movements of NPLs. This indicates that NPLs tarnish asset quality and erode capital and increase risk aversion, constraining banks’ ability to lend to the public (Katuka et al., 2023; Ozili, 2025). Overall, banks use the ratio of operating costs to assets as an index for managerial efficiency which is associated with lending ability and operational strength. When operating costs are high, they are a drag on profits and therefore weaken the build-up of capital due to there not being enough credit.

In addition, an analysis of empirical data found that institutions with the largest cost burdens have the smallest profit margins, especially in times of economic downturn. Undercutting profitability with higher operating expenses undermined banks’ ability to expand lending, meaning that because of these increased costs, they could not lend in hypotrophy (Mehzabin et al., 2023). Empirically, the results of this study were consistent with evidence from China indicating that a greater operating expenses to assets ratio in listed banks is correlated with weaker lending capacity and efficiency scores and institutions which experience strong loan and stock market growth typically have apparently leaner cost structures (Liao, 2024). Taking a cross-national approach, other studies have considered management quality separate from conventional drivers of credit-growth and employed non-interest expense ratios in their measurements; these researchers found that more intense cost oversight increased the efficiency of managing supplies of credit (Shahriar et al., 2022). Such results are also robust in emerging market contexts with a volatile exchange rate: excessive operating expenses have an economically and statistically significant negative influence on credit growth, but banks that enforce stricter cost discipline show more stable lending during currency turmoil. Over the longer term, management discipline and cost control prove to be particularly valuable. Higher operating expense ratios are not only associated with weaker performance in subsequent periods; they also correspond to slower loan growth, highlighting the importance of cost control for maintaining credit provision in good times and through the cycle (Basu et al., 2023; Mamonov et al., 2024).

Return on assets (ROA) is an efficient proxy of a bank’s profitability and efficiency, which affect the credit supplied through accumulation of capital, risk bearing ability, and statutory compliances (Mehzabin et al., 2023). By employing a panel dataset based on 26 banks in Vietnam between 2006 and 2016, Nguyen (2023) demonstrated that an ROA effect is positively related to future loan growth as it facilitates profitability, contributing to funding internal capital and relaxing balance-sheet constraints, particularly during expansion. Basu et al. (2023) showed that U.S. commercial bank lending growth is not as sensitive to business cycles when return on equity (ROE) grows, and this association holds even after controlling for size, capitalization, and macroeconomic conditions, with evidence confirming that profitability has the most independent effect among other factors affecting loan market. This evidence is broadly consistent with the hypotheses that reduced profitability leads to more aggressive regulation and tougher lending standards, which in turn discourages corporate borrowing, dampening aggregate real activity (Federal Reserve Bank of San Francisco, 2024). Moreover, these studies showed that profit has improved banks’ ability to borrow by increasing their capital buffers and risk-taking capacity, while reduced earnings compressed constraints with slower credit growth.

Liquidity management—usually represented by the ratio of a bank’s liquid assets to its total assets—is an important factor in how commercial banks determine their lending decisions. Recent empirical work found a trade-off: an optimal amount of liquidity provides support to the lending capacity, while an excess pushes down credit supply. Based on the data from 2010 to 2020, Stulz et al. (2022) observed that banks with fewer profitable lending opportunities have higher liquidity ratios. With deposit inflows coming in without any attractive loan propositions, institutions transact in liquid instruments and not loans, which could suffocate credit growth. This structure is consistent with portfolio theory, which views liquidity as a substitute for extending loans. Regulatory changes reinforced these effects. Adherence to the introduction of a liquidity balance-sheet rule in the Netherlands led banks to increase their stock of high-quality liquid assets (HQLAs), while banks exceeding the higher thresholds reduced their loan deliveries and experienced lower new originations (Mehzabin et al., 2023). Globally, under Basel III Liquidity Control Regulation (LCR), banks need to hold enough HQLAs to withstand 30 days of liquidity stress. It was found that the LCR had led to a reduction in the range of assets banks were prepared to hold during times of market stress and made it more difficult for them to increase lending (Bank for International Settlements, 2024). Liquidity buffers as a whole boosted resilience, but when inflated beyond what was needed for strategy, curtailed the willingness and ability of banks to lend.

Few studies in Cambodia have examined the impact of NPLs on credit growth in commercial banks. One study used static panel models—pooled OLS, random effects, and fixed effects—to assess the influence of the five CAMEL components: capital adequacy, asset quality, management quality, earnings quality, and liquidity. Asset quality was proxied by the NPL ratio. The results showed that asset quality had a statistically significant effect on loan growth; in all three models, the coefficient for the effect of asset quality was significant at the 5% level (Lim, 2023).

The drawback of static panel data models (i.e., pooled OLS, random effects and fixed effects) is that they cannot address the dynamic linkage between dependent variables (i.e., growth in credit provided by banks) and independent factors like the NPL ratio and other variables. While such dynamic relationships between endogenous and exogenous variables can potentially be modeled using a PVAR model, the PVAR is not able to address both long- and short-run linkages, specifically how fast credit growth returns to its long-run relationship after a short-run disturbance occurs. Therefore, in an attempt to plug this gap, a panel ARDL model design to investigate the effect of NPLs on the credit growth rate of Cambodian commercial banks will be applied in this research.

3. Methodology

The following model represents the levels of the panel ARDL model:

$${CG}_{it}={\alpha }_i+\sum _{k=0}^q{\beta }_{ik}^{\prime }{X}_{i,t-k}+{\epsilon }_{it}$$

where CG represents commercial banks’ credit growth rate, X indicates a vector of independent variables include non-performing loans (NPLs), equity to total assets ratio (ETA), operating expenses to total assets ratio (OTA), liquid assets to total assets ratio (LTA), and sensitivity to market risk (SMR), which measure the ratio of individual bank assets to total bank assets in the industry in each particular year. ${\alpha }_i$ is the unit effect, ${\beta }_{ik}^{\prime }$ represents parameters of regressors, and ${\epsilon }_{it}$ is a residual term. Let units $i=1,\dots ,N$ and time $t=1,\dots ,T$. More specifically, i represents an individual bank and t indicates the time series period. In addition, for the measurement of the short- and long-run relationship between credit growth and the regressors, especially estimating the of a speed of adjustment, an error correction model (ECM) is specified, which has the following general form:

$${∆CG}_{it}={\psi }_i\left({CG}_{i,t-1}+{\mathsf{\theta }}_{\mathit{i}}^{\prime }{X}_{it}\right)+\sum _{k=0}^{q-1}{\mathsf{\delta }}_{ik}^{\prime }{\mathsf{\Delta }\mathit{X}}_{i,t-k}+{\alpha }_i+{\epsilon }_{it}$$

where ${\mathsf{\theta }}_i$ defines long-run coefficients and ${\psi }_i$ is the error-correction speed, which is expected to be negative and statistically significant. In order to estimate common long-run coefficients and heterogeneous short-run coefficients (${\psi }_i,{\mathsf{\delta }}_{\mathit{i}\mathit{k}}$), this study employed the pooled mean group (PMG) and dynamic fixed effect (DFE) techniques. Prior to the estimation of short- and long-run models, it was necessary to implement a co-integration test to evaluate whether banks’ credit growth and the regressors were co-integrated or had a long-run relationship.

It is worth highlighting that in the estimation of the models’ parameters, balance panel data were applied; hence, any banks that had missing data were omitted from the study. The period of this study covered from 2010 to 2023, accounting for 13 years. As of December 2024, there are 59 commercial banks (local and international banks) that operate in Cambodia; however, for the period considered in this study, 23 commercial banks had a complete dataset. The dataset comprised cross-sectional observations for these 23 banks over 13 years, and therefore the total sample size was 322 observations, with all data being extracted from the National Bank of Cambodia website.

4. Empirical Results

Data analysis begins with descriptive statistics, including interpretations of the mean, standard deviation (SD), minimum, and maximum values for banks’ credit growth, non-performing loans, equity to total assets ratio, operating expenses to total assets ratio, liquid assets to total assets ratio, and sensitivity to market risk. Next, a panel co-integration test is implemented to assess the long-run relationship between credit growth and banks’ characteristics using the Pedroni panel co-integration procedure. Finally, this section presents and interprets the empirical results obtained from the pooled mean group (PMG) and dynamic fixed effect (DFE) estimators.

As indicated in Table 1, the sample includes 322 bank–year observations and all variables are presented as percentages. Credit growth (CG) was at 47.0% on average, but showed a high dispersion (SD = 287.1%) and ranged widely between −55.1% and 4736.1%. This finding suggests that the distribution is very skewed, with the much more moderate or even declining portfolio growth of many banks being overwhelmed by a few experiencing explosive growth of their portfolios. Such right-tail events could correspond to mergers, base-effect arithmetic from very small starting portfolios, or data irregularities, meaning there is a need for robustness checks.

Table 1. Summary statistics.

Variable	Observation	Mean	Median	Standard Deviation	Minimum	Maximum
CG	322	47.04	17.71	287.08	−55.11	4736.08
NPL	322	2.95	1.87	3.46	0.00	19.40
ETA	322	25.31	18.86	42.01	0.00	742.10
OTA	322	2.22	1.87	2.61	0.34	44.84
LTA	322	38.24	35.83	14.63	12.25	97.81
SMR	322	3.58	1.87	4.52	0.18	19.70

The average level of asset quality, proxied by the NPL ratio, was low (2.95%), but varied significantly (0–19.4%). Instances of double-digit NPL ratios point to bouts of stress, which could limit lending ability. The amount of bank capital relative to assets (ETA, expressed as a percentage) averaged 25.3%, but values were highly spread out (SD = 42.0%), with an implausible maximum ETA = 742.1%, indicating outliers due to small denominators, restructurings, or data entry problems, suggesting a need for close auditing before inferences are drawn from these data.

Mean operating costs (OTA) were relatively low, averaging 2.22%, with a right-tail maximum of 44.84%, once more suggesting special events or mismeasurement in some observations. The extent of LTA buffers appeared substantial and legitimate: 38.24% was a large enough mean (with a range between 12.25 and 97.81%) to suggest diverse banking sector funding and risk management characteristics. Market risk sensitivity (SMR) had average values of 3.58% and a long right tail (19.70%), reflecting a generally reduced but sometimes high exposure to the market.

When taken together, the data uncover a sector characterized by good overall liquidity and moderate NPLs but considerable differences in growth, capitalization, and costs. For empirical research, non-normality and outliers are expected; researchers may employ robust estimators, cluster standard errors (where applicable), and report medians in addition to means after carefully authenticating ETA, OTA, and CG outliers.

The Pedroni panel co-integration result, presented in Table 2, provided very strong evidence that a long-run common equilibrium exists among the variables in the 23 banks monitored over 13 years. The specification permitted panel-specific co-integrating vectors and AR parameters, which included unit (bank) means, disregarded deterministic trends, and employed a Bartlett kernel with Newey–West bandwidths of two and one augmented lag. The three within-dimension statistics were also all highly significant under the null hypothesis of no co-integration: Modified Phillips–Perron t > 0 (5.226; p = 0.0000), Phillips–Perron t < 0 (9.528, p = 0.000), and Augmented Dickey–Fuller t < 1 in absolute terms (7.585, p = 0.000). These values uniformly reject the null hypothesis because of the right-tailed modified PP statistic and left-tailed PP/ADF statistics.

Table 2. Pedroni test for co-integration.

Ho: No Co-Integration	Number of Panels:	23
Ha: All panels are co-integrated	Number of periods:	13
Co-integrating vector: Panel-specific
Panel means: Included	Kernel:	Bartlett
Time trend: Not included	Lags:	2.00 (Newey–West)
AR parameter: Panel-specific	Augmented lags:	1
Test	Statistic	p-value
Modified Phillips–Perron t	5.2259	0.0000
Phillips–Perron t	−9.5277	0.0000
Augmented Dickey–Fuller t	−7.5847	0.0000

In substantive terms, this implies that deviations from the long-run relationship were mean-reverting: shocks which moved banks away from the equilibrium would tend to dissipate rather than exacerbate. This confirms the modeling of both long-run coefficients and short-run adjustment, for example using panel ARDL/ECM with PMG/DFE methods. In the second ECM, the error-correction (EC) term is expected to be negative and significant, measuring how fast credit expansion adjusts back to the equilibrium following shocks.

Inferentially, the pooled mean group (PMG) estimates indicated in Table 3 provide strong evidence of a significant long-run relationship between CG and asset quality, capitalization, costs, liquidity, and market risk across 23 banks, with over 299 data points. Over the long term, a higher NPL ratio was accompanied by slower credit growth (−0.832, p < 0.001), in line with the theory that problem loans destroy balance-sheet capacity and risk tolerance. In contrast, capital robustness (ETA) positively underpinned credit growth (0.518, p < 0.001), indicating that banks with larger capital cushions demonstrated superior rates of growth. Operating costs (OTA) displayed a long-run positive coefficient (2.49, p = 0.003), likely based on the fact that banks in periods of strong growth have higher non-interest expenses (branching, staff, and IT expenses) with more rapid loan growth, rather than high expenses causing growth. LTA was not significant (p = 0.57), meaning that there is no long-run trade-off between liquid-asset buffers and growth when we control for the other influences. SMR was positively associated with credit growth (2.66, p < 0.001), suggesting that banks engaging in market-facing activities—and being willing to take on some risk—also grew, lending more in the long run.

Table 3. Pooled mean group regression.

Model	CG	Coefficient	Standard Error	z	p > z	[95% Conf. Interval]
Long Run (LR)	NPL	−0.8319	0.2037	−4.08	0.0000	−1.231	−0.433
	ETA	0.5178	0.0688	7.52	0.0000	0.383	0.653
	OTA	2.4922	0.8337	2.99	0.0030	0.858	4.126
	LTA	0.0343	0.0601	0.57	0.5680	−0.084	0.152
	SMR	2.6632	0.5080	5.24	0.0000	1.668	3.659
Model	CG	Coefficient	Standard Error	z	p > z	[95% Conf. Interval]
Short Run (SR)	EC	−0.9692	0.0530	−18.29	0.0000	−1.073	−0.865
	ΔNPL	−1.0116	1.2176	−0.83	0.4060	−3.398	1.375
	ΔETA	−1.5067	0.5295	−2.85	0.0040	−2.545	−0.469
	ΔOTA	−9.7585	5.0539	−1.93	0.0530	−19.664	0.147
	ΔLTA	−1.3879	0.2316	−5.99	0.0000	−1.842	−0.934
	ΔSMR	37.9371	10.8576	3.49	0.0000	16.657	59.218
	_CONS	−3.3533	2.6788	−1.25	0.2110	−8.604	1.897

The story, however, is different for short-run dynamics. The error correction term is large, negative, and highly significant (−0.969, p < 0.001), corresponding to very fast adjustment: approximately 97% of any deviation from the long-run equilibrium corrects in one year. Short-run dynamics reflect cyclical prudence: ΔETA is negative (−1.51, p = 0.004) and ΔLTA is strongly negative (−1.39, and p < 0.001), indicating that within the year, movements in capital or toward liquidity were associated with a deceleration of credit growth, consistent with temporary balance-sheet repair or precautionary liquidity hoarding. ΔNPL was insignificant (p = 0.413), indicating that year-on-year jumps in problem loans did not instantaneously transmit into growth once the error correction channel was incorporated. ΔOTA was marginally negative (p = 0.053), suggesting that abrupt increases in cost dampen short-term expansion. ΔSMR was both substantial and positive (37.94, p < 0.001), suggesting surges in risk-taking with short-run credit growth.

Economically, the findings suggest a story of long-run discipline and short-run trade-offs: durable growth co-moves strongly with sound asset quality and capitalization, while banks moderate lending during shorter spells of reduced capitalization and liquidity. This error correction concept ensures co-integrated specification and provides a justification for estimating long-run coefficients.

As referring to Table 4, the DFE error correction estimates suggest evidence of co-integration and a stable long-run relationship between banks’ credit growth and banks’ balance-sheet fundamentals with rapid short run adjustment. In the longer run, banks with higher NPL ratios experienced weaker credit growth (−1.707, p < 0.001), in line with provisioning requirements and subdued risk-taking. Capitalization (ETA) and liquidity (LTA), on the other hand, were found to be statistically insignificant, indicating no common long-run impact given the other controls. Meanwhile, expense ratio intensity (OTA) was positively associated with long-term growth (4.833, p = 0.036). This is likely an expression of expansionary strategies—branches, staffing, and technology—that boost non-interest costs hand-in-hand with durable loan growth rather than inefficiency driving growth. Market risk sensitivity (SMR) was also non-predictive in the long run.

Table 4. Dynamic fixed effect regression.

Model	CG	Coefficient	Standard Error	z	p > z	[95% Conf. Interval]
Long Run (LR)	NPL	−1.7068	0.4835	−3.53	0.0000	−2.654	−0.759
	ETA	−0.1685	0.1762	−0.96	0.3390	−0.514	0.177
	OTA	4.8328	2.3047	2.1	0.0360	0.316	9.350
	LTA	0.0988	0.1543	0.64	0.5220	−0.204	0.401
	SMR	−0.5593	0.8287	−0.67	0.5000	−2.183	1.065
Model	CG	Coefficient	Standard Error	z	p > z	[95% Conf. Interval]
Short Run (SR)	EC	−1.0121	0.0044	−231.5	0.0000	−1.021	−1.004
	ΔNPL	−0.4634	0.5740	−0.81	0.4190	−1.588	0.662
	ΔETA	0.4014	0.1302	3.08	0.0020	0.146	0.657
	ΔOTA	−6.9226	2.1296	−3.25	0.0010	−11.096	−2.749
	ΔLTA	−0.7343	0.1711	−4.29	0.0000	−1.070	−0.399
	ΔSMR	7.9122	2.0533	3.85	0.0000	3.888	11.937
	_CONS	18.6177	9.5701	1.95	0.0520	−0.139	37.375

Short-run dynamics present cyclical trade-offs. The error-correction term is large, negative, and highly significant (−1.012, p < 0.001), thus indicating a more than full adjustment within one period—that is, deviations from the long-run target not only close rapidly but also mildly over-shoot before settling. In the short run, ΔETA is positive (0.401, p = 0.002), which indicates that an increase in equity per asset basis is matched by faster credit growth, much as one would expect if new capital allows for lending. On the contrary, ΔLTA is negative (−0.734, p < 0), which means liquidity hoarding in the short term crowds out credit lending. ΔOTA is negative (−6.923, p = 0.001), indicating that cost spikes suddenly truncate credit growth in the short term regardless of whether higher steady-state costs continue with further expansion. ΔSMR is positive (7.912, p < 0.001) for risk-on episodes that lead to temporary expansions in lending, while ΔNPL is insignificant (p = 0.419), implying that changes in problem loans in one period do not instantaneously translate into shifts in growth once the error correction channel has been controlled for.

The DFE results are consistent with a narrative of long-run discipline carrying short-term frictions. Continued expansion while maintaining sound asset quality and continued strategic investments in costs is supported because there were some short-term increases in liquidity buffers, while there was pressure on some operating costs of banks as they adjusted. Moreover, the evidence of fast and strong error correction supports the co-integration framework, justifying the long-run coefficients reported along with short-run dynamics.

5. Conclusions and Policy Implication

Based on evidence of panel co-integration, the results from both the PMG and DFE estimators provide consistent evidence of how balance-sheet fundamentals influence credits growth in banks. Poorer asset quality due to a higher rate of NPL was found to be negatively associated with credit growth across specifications. The long-run coefficient for NPLs was large and negative under PMG, while the DFE error correction form also produced a statistically significant negative relationship, suggesting that problem loans need to be subtracted from lending capacity as provisioning requirements increase, capital buffers are thinned out, and risk-taking decreases. Capitalization (ETA) helps to sustain long-run growth in the PMG estimates, though the DFE specification did not find evidence of a common long-run effect once homogeneous slopes were enforced. This is not to imply that there was no capitalization–growth nexus—instead, perhaps it was heterogeneous across banks and better captured when unit-specific short-run dynamics were accommodated by PMG. Operating expense intensity (OTA) had a positive long-run relationship with growth in both model estimations (statistically strong for PMG and significant for DFE), which is another finding most consistent with expansion-related expenses (for processes, people, and equipment) accompanying durable loan growth rather than costs being causal of lending. Only the long-run effects on liquidity (LTA) and market-risk sensitivity (SMR) were ambiguous or weak: LTA was significant in PMG and DFE, while SMR was positive in PMG but not significant in DFE, once again reflecting cross-sectional heterogeneity indicating that DFE’s common-slope restriction might have been dampened.

There was a negative and very significant error correction term in both cases, suggesting rapid convergence back to the long-run path. Both partial adjustment and speed of adjustment were rapid: under the PMG specification, most of the deviation closed within one period, while the DFE estimate indicated more than full adjustment with slight overshooting. Short-run coefficients illustrated balance-sheet trade-offs during adjustment: increases in liquidity ratios were associated with slower same-period credit growth (consistent with a time-series of precautionary liquidity hoarding), while temporary upticks in operating costs muted short-term lending even if higher steady-state costs could be consistent with expansion. Short-term movements in NPLs were largely negligible after correction for errors, suggesting that the NPL channel worked mainly through the long-run equilibrium instead of short-period shocks. The aggregate impact of capital changes reversed over time: under DFE, financial institutions increased equity and lending growth in the short run, while freshly raised capital eased constraints.

In conclusion, our findings imply that regulators should strengthen their actions to improve the asset quality of banks and ensure sustainable capitalization to facilitate stable credit growth. As the strong negative relationship between NPLs and lending persists, more stringent initiatives related to loan classification and timely provisioning or other sound policies regarding an understaffed effective credit-risk management system should be organized and run by competent supervisory authorities. As credibly shown above, NPLs tend to increase lending beyond its equilibrium level. As a result, the authorities must persuade banks to build up capital cushions in an upswing to cover potential losses they may incur. In particular, procyclical spending is a concern, although supervision policies may need to differentiate between operations that enhance productive and sustainable credit growth (e.g., technology and risk spending) and those that do not. It is therefore important to develop and refine liquidity and market risk management systems to minimize banks’ procyclical multiplier behavior and ensure continued lending capacity during periods of financial market stress.

The difference between PMG and DFE regarding some long-run coefficients was enlightening. PMG, which estimated joint short-run bank effects on the spread and allowed bank-specific short-run linear adjustment to be incorporated, found a significant capitalization effect and a significant linkage between SMR and growth; DFE, which constrained both short- and long-run slopes (except intercepts), together with pooling the data over the long run, did not. This behavior pattern indicated that long-run relationships were also quite common for all banks, while short-run responses and eventually the transformation from market-risk activity into lending differed in a meaningful way among institutions. In these contexts, PMG’s flexibility was better suited to homogeneity restriction than DFE.

Taken as a whole, the results are consistent with a story of long-run discipline and short-run frictions. In the long run, it is a sustainable quality of assets and capitalization which maintain credit expansion, while then non-performing loans hold back growth. In the short term, balance-sheet rebalancing—increasing liquidity buffers or absorbing cost shocks—depresses lending as banks adjust back to their equilibrium. In terms of policy, this evidence suggests that maintaining surveillance over asset quality and capital planning would help to protect the long-run credit supply, but not to the extent that shocks inducing liquidity creation binges or cost adjustments could stop even sound systems from lending briefly.

In conclusion, PMG’s and DFE’s central message is the same—NPLs matter and adjustment is swift—but PMG’s greater variability reveals a further long-run role for capitalization and moderate risk-taking which is not evident in DFE.