Model 1: Diversification and Specialisation in Microfinance

To study the impact of lending strategy on liquidity creation, I first need to know whether the level of diversification of lending activities on its own has an impact on liquidity creation. For this purpose, I measure the degree of specialisation across MFIs, by computing a normalized Herfindahl-Hirschman Index (HHI) of lending methodologies. The HHI Index was traditionally developed to measure the level of market concentration, by taking the square of each market share, and adding the numbers for the final sum. In this study, I adapt the same approach, to calculate the concentration of a particular type of loan (individual, group and village), as a share of the overall loan pool of each MFI.

I thus calculate the following:

Equation 5.1: Loan Type as Share of Total Loans

𝑆𝑆

Where 𝑆𝑆_𝑖𝑡 comprises the sum of the squared lending methodologies for MFI i in year t. IND refers to individual loans, VB is village bank loans, GR is group loans and finally TLEND is total loans. I then find:

Equation 5.2: Herfindahl-Hirschman Index (HHI) for Specialisation

𝐻𝐻𝐼_𝑖𝑡 =𝑆𝑆𝑖𝑡−

1 3 1 −1₃

where MFIs that only provide one type of lending will obtain an HHI score of 1, while MFIs which engage equally in all lending strategies will gain scores closer to 0. Higher HHI values will thus indicate the MFI’s degree of product specialisation.

135 Estimations for Specialisation

To investigate the relationship between the degree of specialisation and levels of liquidity created by the MFI, I estimate the following fixed effects model:

Equation 5.3: Liquidity Creation and Specialisation, DV I-III

𝐿𝐼𝑄𝑖𝑡 = 𝛼 + 𝛽𝑘𝐻𝐻𝐼𝑘𝑖𝑡−1+ 𝛿1𝐶𝐴𝑃𝐼𝑇𝐴𝐿𝑖𝑡−1+ 𝛿2𝑖𝑡 − 1 + 𝛿3𝐴𝐺𝐸𝑖𝑡+ 𝛿4𝑅𝐼𝑆𝐾𝑖𝑡−1+ 𝛿5𝑁𝐺𝑂𝑖𝑡 +

𝛿6𝑂𝑆𝑆𝑖𝑡−1+ 𝛿7𝑅𝐸𝐺𝑖𝑡+ 𝑓𝑗+ 𝑇𝑖+ 𝜀𝑖𝑡

where 𝐻𝐻𝐼𝑘𝑖𝑡−1 captures the effect of specialisation on levels of liquidity creation in MFI i for period t. As in the previous chapter, LIQ represents one of the following measures of liquidity creation of MFI i in period t: pure, absolute liquidity creation (LC), LC scaled by total assets (LC_GTA), liquidity to equity (LC_E) and liquidity creation to total loans (LC_GLP). In all 3 models, the pure LC values have been deflated using the World Bank Consumer Price Index (CPI).

Table 5.1: Variables and Definitions (Model 1)

Variable Definition

Liquidity creation (log) Log of the pure liquidity creation value, deflated using the CPI Index.

LC/GTA Liquidity creation scaled by size.

LC/E Liquidity creation scaled by equity.

LC/GLP Liquidity creation scaled by total loans.

HHI Specialisation HHI Index indicating level of lending specialization. The higher the value, the higher the level of specialization.

Capital Measured as the ratio of equity to total assets.

Regulation (dummy) Dummy indicating whether the individual MFI is regulated.

Size (log) Measured as a log of the MFI’s total assets.

Age Age of MFI, measured a number of years since year of establishment.

NGO (dummy) Dummy variable taking the value 1, if the MFI is registered as an NGO.

Bank risk Bank risk, measured as the ratio of portfolio at risk, 30 days.

Operational self-sufficiency Ratio of financial revenue divided by financial and operating expense, and impairment loss.

136

The HHI model includes a set of variables, which estimate the impact of MFI characteristics, as well as control for more general traits.59 These are the age of the MFI, calculated as year y minus the year of establishment (AGE), as well as the size of the MFI, measured as the one-period-lagged log of total assets (SIZE). I also control for the impact of capital (CAPITAL), measured as the one-period-lagged proportion of equity to assets (Diamond and Rajan, 2001; Berger and Bouwman, 2009; Horvath and Seidler, 2013).

The level of bank risk present in the MFI is measured as the one-period-lagged ratio of portfolio at risk > 30 days to the MFI’s gross loan portfolio (RISK). Finally, two dummy variables are included, indicating whether the MFI is employing a non-profit organisation (NGO), and whether the MFI is regulated by a state banking supervisory agency (REGULATION).60

Aside from regulation, macro-economic variables are not included in the main regressions presented here. The addition of the variables, while insignificant themselves, reduced the adjusted r-squared from 20.3 to 18.1. The relationship between the remaining variables remains qualitatively the same.

All right-hand side variables in the estimation are lagged one period, to account for potential endogeneity problems and robust standard errors are clustered at MFI level. This is done to address possible serial correlation of residuals between the observations, and the dispersion of coefficient estimates across clusters61.

As in the previous chapter, I measure liquidity creation using a cross-country approach, but recognise that there is likely to be regional differences in the creation of liquidity. For this reason, I include a fixed effect to capture the unobserved geographical traits over time.

59 _{I do not include deposit insurance or capital requirements in the main regressions for two reasons. First,} including them reduces the sample with over 20%. Given the reduced data on loan strategy from our main dataset, the loss of observations has significant impact on reliability. Second, while results remain qualitatively the same, the inclusion of the two significantly reduces the explanatory power (adjusted r-squared) of the model.

60

Deposit insurance is not included in the results presented here. Including the dummy variable provided by the Barth index is positive but insignificant, and lowers the r-squared value significantly. The relationship between variables remains the same.

61

In addition to the vif test, correlation matrices were created for all regressions in the dissertation, with upper limits set to 0.4. As can be seen in Appendix E, no correlation among the independent variables breaches this limit.

137 62 _𝑓

𝑗 is therefore a vector of regional dummies, using the six MIX regional category for division: Sub-Saharan Africa (SSA), Middle East and North Africa (MENA), Eastern Europe and Central Asia (EECA), East Asia and Pacific (EAP), South Asia (SA) and Latin America (LAC).

Finally, I include a time trend (𝑇𝑖) to capture any banking-related methodological changes over the time. ε is a stochastic error term. 𝛼 is a constant term. 𝛽 and 𝛿 are the unknown coefficients to be estimated. The subscripts i and t refer to MFI and time, respectively.