Investigating SME’s credit constraints: a trinary approach

Abstract

This study aims to provide an understanding of factors that might affect three-level credit constraints of small and medium-sized enterprises (SMEs) in Vietnam, using data from the Vietnam’s Manufacturing SME Survey database. We apply a random-effects generalised ordered probit model to accommodate two types of heterogeneity: unobserved individual heterogeneity and the heterogeneous effect of independent variables across three constraint categories. Results show that several firm and owner characteristics are in play. Several of these are indeed heterogeneous, such as varying across constraint categories, a new finding in the literature. A key contribution of this study is the use of a trinary rather than the usual binary approach to measure and more fully understand determinants of credit constraints in the case of small and medium-sized enterprises in Vietnam.

Keywords:

• Credit constraints
• generalised ordered probit
• SMEs
• Vietnam

JEL CLASSIFICATION:

• C23
• C25
• O16
• O53

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 The ability of a firm to access a mix of formal and/or informal credit might be influenced by the level of credit constraint it faces (Archer, Sharma, and Su 2020). A firm’s appetite for and ability to access either or both types of credit may influence their constraints, depending on the definition of constraint used. Our study focuses on constraints in the formal credit markets with information from survey tools asking firms if they had applied for formal loans from banks and/or other formal financial institutions. Further details are presented in Section 3.

2 There is basis to suspect a complementarity or additionality relationship between these two types of constraints (e.g., Archer, Sharma, and Su 2020; Nguyen and Luu 2013). If the impact of one constraint becomes more serious in the presence of another one and reduces otherwise, then a more efficient approach to development policy would be to tackle the constraints one at a time rather than simultaneously. Other possibilities exist: the two constraints might mutually re-inforce each other or the effect of one might be passing through the other if we combine two constraints altogether. Thus, it is important to separate the partially constrained firms from the fully constrained ones.

3 Nguyen et al. (2022), in a study that also examines determinants of credit constraints in Vietnam, analyse firm’s credit demand in a two-step process, in which the first step shows that credit demand relies on loan application, debt holding, and selectivity, while the second step presents results of unadjusted and adjusted determinants of credit constraints. Our study is different regarding the nature of the dependent variable – a hierarchical three-level credit constraint, and the method applied to address the potential econometric issues of heterogeneity arisen from the variable nature and the ordinal panel data.

4 Literature has also determined multi-level credit constraints based on responses of managerial respondents (for example, Nguyen and Luu 2013; Hoque, Sultana, and Thalil 2016), or as a continuous variable that is measured by a ratio of credit access to collateral (Aghion, Fally, and Scarpetta 2007), or by cost of external credit (Gorodnichenko and Schnitzer 2013). For example, Kira (2013) measures credit constraints as a multi-level variable through manager’s perceptions on how severe the constraint is. A study by Nguyen and Luu (2013) proposes a multi-level definition with four unordered categories of financing access: formal, informal, both formal and informal, and none. The authors highlight that several firm characteristics do matter for firm’s financing access.

5 Although Ha Tay has been merged into Ha Noi since 2008, in this study we treat Ha Tay as a separate province to ensure the consistency of the sample over years.

6 A binary approach proposed by Tran and Santarelli (2014) takes into account three cases to identify if a firm is capital constrained when it applied for loans and being rejected, and/or applied for loans and was in need of more loans, and/or did not apply for loans due to strict requirements including collateral, indebtedness, and interest rates.

7 When $\mathrm{\Phi }(.)$ is replaced by the cumulative density function of the logistic distribution, the model becomes generalised logit/probit.

8 Constraints for parallel lines imposed for five variables with large P-values in parentheses: revenue (0.591), networking (0.555), gender of owner (0.819), age of owner (0.154), and the overall PCI (0.603), implying no violation of the parallel-lines assumption.

9 Constraints for parallel lines are not imposed for eight variables with small P-values in parentheses: entry age (0.000), firm type (0.002), firm size (0.000), leverage ratio (0.000), investment (0.000), registration (0.008), change of owner (0.041), and education of owner (0.006), implying a rejection of no violation of the parallel-lines assumption.

10 Our result based on the usual probit is more in line with the findings of Tran and Santarelli (2013) and Rizov (2004).

11 In addition, data in Table 1 indicates that the two subgroups of those firms with "no demand for loans" and those with “demand for loans but not applying for loans due to other reasons” make up around 66 percent of the data sample. Thus, we look at these subgroups more closely by providing an empirical analysis reported in Table A1 in the Appendix. Results in Table A1 show the similarities as those reported in Table 8 with a binary credit-constraint variable, except variable registration turns from statistical significance into non-significance.

12 Crisis is a dummy variable, with value of 1 if year is 2009 and 0 otherwise. Inflation is a variable that denotes the inflation rate over year of the study period.

13 Concerning firm’s location, the Vietnam’s SME Surveys cover three urban cities including Ha Noi (the capital), Hai Phong, and Ho Chi Minh City, and seven rural provinces namely Phu Tho, Ha Tay, Nghe An, Quang Nam, Khanh Hoa, Lam Dong, and Long An.

14 A balanced sample is generated by combining the biennial datasets from 2005 to 2013 and filtered the observations that are repeated for all of five surveying years.

15 Non-constraint (N.C.), partial constraint (P.C.), and full constraint (F.C.).

16 Since the average partial effects obtained from the ordered probit model are similar to those obtained from the random-effects ordered probit model, we only report the latter for the sake of brevity.

17 The sub-indices are calculated averagely in the period between 2006 and 2013.