1. Introduction

An important resurgence of Microfinance Institutions (MFIs) took place in the 70’s when initially they were constituted as non-profit NGOs¹, though the first antecedents are given since the 19th century in Europe (Germany).² The MFIs mainly focus on providing financial services to the population that does not have access to conventional bank credit for lack of real collateral. MFIs face, continuously, two main objectives: 1) achieve financial sustainability, and 2) in-crease the number of clients. Currently, there are more than 2000 Microfinance Institutions (MFIs) in the world, with a total of 130 million clients and a Gross Loan Portfolio of 108 billion (USD)³.

On the other hand, the number of MFIs that are listed on stock exchanges is still small worldwide. However, they account for 16.8 % of the total number of clients since they have a high degree of concentration -in their respective markets where they operate. It is worth mentioning that MFIs seek in the stock market resources more efficiently (quickly and at a lower cost). The main question that arises is: if this objective can be achieved without being affected by the volatility from financial crises, which may discourage other MFIs from entering the stock market.

Due to the above concerns, Wagner and Winker (2011) and Di Della (2011) have studied the impact of financial crisis (for example, the sub-prime crisis, 2008) on the MFIs, particularly in periods of high volatility. However, the contagion in the volatility in the returns of the MFIs listed on stock markets has not yet been studied. In this sense, the goal of this paper is to analyze how volatility affects them. This will allows us to better understand the effects and consequences of high levels of volatility, which could be generated via global financial crises or high-volatility clusters. Thus, the main contribution of this research is to analyze, under a DCC-M-GARCH framework, how the contagion occurs in the volatilities of the returns of MFIs that are listed on stock exchanges in different emerging economies.

This paper analyzes five MFIs from three emerging economies: India, Indonesia and Mexico.⁴ Table 1 shows the studied MFIs and their characteristics such as number of clients, size of their portfolio and country of origin. In addition, for each MFI two reference variables are used in its corresponding stock market, and a global reference index -All Countries World index (ACWI). Table 2 describes each of the variables used (14 in total), it is shown the corresponding characteristics of: activity, currency in which they operate (in their respective stock market), and the period of analysis that comprises for each one of the MFIs, with daily frequency data.

It is important to point out that the period of analysis is not the same for all MFIs. The reason is that, on the one hand, this research seeks to obtain as many observations as possible with the aim of achieving more robust results and, on the other hand, it is understandable that the periods are not homogeneous since the MFIs began to operate in the stock market on different dates; see the last two columns of Table 2.

In what follows, a descriptive statistical analysis is carried out in Table 3, According to the fifth column, all the returns are leptokurtic. Moreover, none of the returns are normally distributed according to the Jarque-Bera test. In Table 4, the unit root test was performed on each variable, using the Dickey Fuller Augmented test, under the three specifications: intercept, trend and intercept, and none. The results show that there is no empirical evidence of an explosive behavior in the analyzed variables.

In order to detect whether there are long-term memory effects in the returns of each variable, Hurst exponent is calculated. The latter is a useful indicator to examine whether returns have long-term memory -a characteristic useful to forecast future values. It is worth mentioning that Hurst’s exponent can be equal to 0.5 (without long-term memory), greater than 0.5 (long-term memory), and less than 0.5 (mean reversion). It is also important to notice that long-term memory violates the Efficient Market Hypothesis (EMH), established by Fama (1970). Moreover, this research computes an index of stock market liquidity⁵ in order to find some relationship between the effects of long memory and low stock market liquidity.

The obtained results, in relation to the Hurst exponent, show that Gentera and BRI do not present strong empirical evidence of long-term memory in their returns with Hurst exponents of 0.519 and 0.493, respectively.⁶ These results can be seen in the fourth column (in descending order) of Table 5. With respect to the obtained results in the liquidity index, see the sixth column (in ascending order) of Table 5, the MFIs that appear with less liquidity are FI and BRI with an index of 81.6 and 90.9, respectively.

Based on the previous results, there is not a pattern in the behavior between long-term memory and low market liquidity (as would be expected at first). However, we can highlight the case of FI with long memory and low liquidity, 0.56, 81.6 %, respectively. In contrast, Gentera shows an acceptable liquidity of 98 % and a Hurst exponent of 0.519. Notice also that BFIL_NSE and BFIL_BSE provide empirical evidence of long memory in its returns but with high levels of stock market liquidity -see columns fifth and seventh of Table 5.

It is important to point out that not all the MFIs analyzed have had an acceptable long-term performance (in the period of analysis) in their accumulated returns. Only Gentera and BRI had a better performance than the local reference index in their respective markets, Mexico and Indonesia, respectively.

After the descriptive exploration of the variables, this research will be structured in the following way: section 2 provides a brief description of the M-GARCH model of Dynamic Conditional Correlation (DCC); section 3 presents the empirical findings for each specification of the MFIs (benchmark variables and a global index); finally, section 4 exposes the conclusion.

2. Dynamic Conditional Correlation (DCC)-M-GARCH

Recently, Bala and Takimoto (2017) analyze the effects of the dynamic correlation, during periods of financial crisis, by using a DCC-M-GARCH econometric approach. These authors found that the dispersion of volatilities among deve-loped markets is greater than in emerging markets. Previously, Mollah et al. (2014) study 63 countries during the period of the global financial crisis. These authors find -through the use of DCC- that contagion of volatility occurs in 46 out of 63 countries that is, in 73 % of the analyzed countries. Although the previous studies allow us to see some results related to the DCC-M-GARCH methodology in assessing the effects of contagion within emerging and developed markets, there is no recent literature that focuses on volatility contagion of returns of the MFIs listed on the stock market.⁷

On the other hand, Visconti (2009) argues that MFIs are affected in diffe-rent ways, depending on the country in which they operate and the degree to which these countries are integrated into the global economy. In addition, the author points out that MFIs operating in developing countries are less affected by financial crises due to close ties and constant monitoring of their clients. Moreover, Krauss and Walter (2008) applied a panel data approach (with fixed effects) in order to examine whether MFIs are a good option to reduce the volatility of an investment portfolio since their clients are mostly micro-entrepreneurs. In this regard, it is important to consider that the main disadvantage that MFIs face is that their clients do not have real guarantees. However, they do have the advantage of being able to maintain a continuous monitoring of their clients, which allows having a better quality in the portfolio of microcredits granted.

3.Empirical result

MFIs systems

We present now the results obtained with a system of non-related variables for each MFI (with benchmark variables) for every stock market analyzed. The selection of the GARCH model and the error specification, to obtain the DCCs in each market, are chosen according to the less explosive parameters, see Table 6. The estimations are shown in Table 7. Subsequently, we present the DCCs for each MFIs and their benchmark variables, see Figures 4 and 5. Finally, in Table 7 we calculate the arithmetic and geometric maens for each estimated DCC considering the period of analysis.

Table 6.

Model GARCH family (specifications).

MFIs Systems Not considering ACWI	GARCH Model	Error Specification
Gentera, IPC, MF	GARCH (1,1)	t-student
FI, IPC, ;MF	GARCH (1,1)	t-student
BFIL_ NSE, N50, iI50	GJR-GARCH (1,1)	t-student
BFIL_BSE, BD&MFG, BRFL	GJR-GARCH (1,1)	GED
BRI, JII, TLK	GARCH (1,1)	Normal
Gentera, IPC, ACWI	GARCH (1,1)	Normal
FI; IPC, ACWI	EGARCH (1,1)	GED
BFIL_NSE, N50, ACWI	GARCH (1,1)	t-student
BFIL_BSE, BD&MFG, ACWI	GARCH (1,1)	t-student
BRI, JII ACWI	EGARCH (1,1)	t-student

^TFN8Note: criteria selection is according to the less explosive parameters.

^TFN9The results were obtained by using the R software, "fGARCH"library.

Table 7.

Econometric results (MFIs and Benchmarks).

	Coef. (1) Std. Err.		Coef. (2) Std. Err.		Coef. (3) Std. Err.		Coef. (4) Std. Err.		Coef. (5) Std. Err.
Gentera_ω	0.000009	FI_ω	0.000083	BFIL_NSE_ω	0.000241	BFIL_BSE_ω	0.000225	BRI_ω	0.000011
	(0.000014)		(0.000051)		(0.000121)*		(0.00005)*		(0.000009)
Gentera_α	0.08659	FI_α	0.348359	BFIL_NSE_α	0.158495	BFIL_BSE_α	0.130941	BRI_α	0.055043
	(0.037231)*		(0.12433)*		(0.049523)*		(0.030491)*		(0.027312)*
Gentera_β	0.89872	FI_β	0.650641	BFIL_NSE_β	0.618593	BFIL_BSE_β	0.629758	BRI_β	0.926208
	(0.027132)*		(0.136032)*		(0.136525)*		(0.056455)*		(0.016712)*
IPC_ω	0.000001	IPC_ω	0.000001	N50_ω	0.000003	BD&MFG_ω	0.000037	JII_ω	0.000006
	(0.000002)		(0.000003)		(0.000005)		(0.000004)*		(0.00001)
IPC_α	0.065897	IPC_α	0.066201	N50_α	0.000003	BD&MFG_α	0.02046	JII_α	0.09792
	(0.02605)*		(0.029512)*		(0.026756)		(0.000026)*		(0.024537)*
IPC_β	0.922272	IPC_β	0.921662	N50_β	0.928628	BD&MFG_β	0.951826	JII_β	0.873711
	(0.029107)*		(0.032855)*		(0.01574)*		(0.009623)*		(0.053657)*
MF_ω	0.000005	MF_ω	0.000005	iI50_ω	0.000021	BRFL_ω	0.000016	TLK_ω	0.000007
	(0.000006)		(0.000006)		(0.000009)*		(0.000023)		(0.000003)*
MF_α	0.115317	MF_α	0.115912	iI50_α	0.036828	BRFL_α	0.178185	TLK_α	0.044075
	(0.028479)*		(0.027632)*		(0.020836)*		(0.091392)**		(0.009175)*
MF_β	0.871226	MF_β	0.870042	iI50_β	0.822952	BRFL_β	0.756588	TLK_β	0.937472
	(0.039183)*		(0.03935)*		(0.054078)*		(0.165688)*		(0.013027)*
DCC_I	0.022961	DCC_I	0.02871	DCC_I	0.012854	DCC_I	0.018947	DCC_I	0.006419
	(0.009062)*		(0.009961)*		(0.00433)*		(0.011151)**		(0.005797)
DCC_II	0.928073	DCC_II	0.937239	DCC_II	0.97705	DCC_II	0.914855	DCC_II	0.952879
	(01.031519)*		(0.02274)*		(0.007167)*		(0.02149)*		(0.028827)v

^TFN10Parameters are significant at: 5 % p-value (*) and 10 % p-value (**), respectively.

^TFN11Source: own elaboration. The results were obtained with the use of the R software, using the libraries: “rugarch” and “rmgarch”.

[Figure ID: f1] Figure 1.

Dynamic Conditional Correlations (MFIs: Gentera, FI, and BFIL_N SE).

  —Source: own elaboration with the use of the R software..

[Figure ID: f2] Figure 2.

Dynamic Conditional Correlations (MFIs: BF IL_BSE, BRI)

  —Source: own elaboration with the use of the R software..

[Figure ID: f3] Figure 3.

Dynamic Conditional Correlations with ACWI (MFIs: Gentera, FI, and BFIL_N SE).

  —Source: own elaboration with the use of the R software..

[Figure ID: f4] Figure 4.

Dynamic Conditional Correlations with ACWI (MFIs: BFIL_BSE; BRI).

  —Source: own elaboration with the use of the R software..

Table 7 shows the estimated coefficients for each MFI and its benchmark variables (based on the criteria in Table 6). The obtained results show that the specification in each equation successfully captures the estimated variances with persistence in both series

<mml:msubsup> <mml:mrow> <mml:mi>ε</mml:mi> </mml:mrow> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>t</mml:mi> <mml:mo>-</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mrow> <mml:mn>2</mml:mn> </mml:mrow> </mml:msubsup> <mml:mi> </mml:mi> <mml:mi>a</mml:mi> <mml:mi>n</mml:mi> <mml:mi>d</mml:mi> <mml:mi> </mml:mi> <mml:msubsup> <mml:mrow> <mml:mi>h</mml:mi> </mml:mrow> <mml:mrow> <mml:mi>i</mml:mi> <mml:mi>t</mml:mi> <mml:mo>-</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> <mml:mrow> <mml:mn>2</mml:mn> </mml:mrow> </mml:msubsup>

respectively. However, in the case of the parameters

<mml:msub> <mml:mrow> <mml:mi>ω</mml:mi> </mml:mrow> <mml:mrow> <mml:mi>i</mml:mi> </mml:mrow> </mml:msub>

, some specifications of the GARCH models capture that persistence in a smaller proportion; see columns (1)-(5) in Table 6. It can also be observed (in the same table) that for the estimated

<mml:msub> <mml:mrow> <mml:mi>ω</mml:mi> </mml:mrow> <mml:mrow> <mml:mi>i</mml:mi> </mml:mrow> </mml:msub>

in the columns (1) and (2) it is not possible to obtain a long-term variance.

It can be also observed in Figures 1 and 2 that it is not possible to appreciate a pattern (or contagion) common in the DCCs in each system of the MFIs. In other words, the particular specifications for each model of the GARCH family, according to the criteria in Table 5, do not capture a common contagion in the volatilities for each MFI. Contagion in volatilities can only be seen in some peaks with high volatility clusters, but not within the entire analysis period. Furthermore, we can see, in Table 8, that the arithmetic and geometric means do not show common patterns in the behavior of the DCCs in the analyzed stock markets.

Table 8.

Means of Dynamic Conditional Correlations (DCCs).

Arithmetic Mean		Arithmetic Mean		Arithmetic Mean		Arithmetic Mean		Arithmetic Mean
DCC_Gentera_IPC	0.43	DDC_FI_IPC	0.14	DCC_BFIL_NSE_N50	0.36	DCC_BFIL_BSE_BD&MFG	0.28	DCC_BRI_JII	0.61
DCC_Gentera_MF	0.12	DCC_FI_MF	0.09	DCC_BFIL_NSE_iI50	0.41	DCC_BFIL_BSE_BRFL	0.12	DCC_BRI_TLK	0.35
DCC_IPC_MF	0.22	DCC_IPC_MF	0.22	DCC_N50_iI50	0.82	DCC_BD&MFG_BRFL	0.13	DCC_JII_TLK	0.47
Geometric Mean		Geometric Mean		Geometric Mean		Geometric Mean		Geometric Mean
DCC_Gentera_IPC	0.43	DCC_FI_IPC	0.15	DCC_BFIL_NSE_N50	0.35	DCC_BFIL_BSE_BD&MFG	0.27	DCC_BRI_JII	0.61
DCC_Gentera_MF	0.11	DCC_FI_MF	0.12	DCC_BFIL_NSE_iL50	0.41	DCC_BFIL_BSE_BRFL	0.12	DCC_BRI_TLK	0.35
DCC_IPC_MF	0.21	DCC_IPC_MF	0.20	DCC_N50_iI50	0.82	DCC_BD&MFG_BRFL	0.12	DCC_JII_TLK	0.47

^TFN12Source: own elaboration with the use of the R software.

4. Conclusions

This research has shown that there is not a pattern between long-term memory and liquidity in the studied MFIs. According to the analysis carried out on the DCC-M-GARCH approach, the effects of contagion (in MFIs returns) only occur in periods of high volatility when considering local benchmark variables. Moreover, when considering the global index All Countries World Index (ACWI), the results confirm the empirical evidence.

As a recommendation arising from the empirical findings, the MFIs that obtain resources via the stock market should operate with efficient methodologies in the selection of clients, which will impact in their level of liquidity in the stock market. It is also recommended for investors, both institutional and individual, consider MFIs in their investment portfolios in stability periods given that contagion only occurs in periods of high volatility.