District Beta: Assessing risk in Microfinance

Introduction

Microfinance sector focuses on providing last mile financial services to the under banked segment of the population. The growth of microfinance sector in India has been tremendous, registering a staggering 51% year-on-year as of FY 2019 Q2 with total gross loan portfolio of INR 1.4 trillion. Given the sheer size, reach across under-penetrated geographies and the impact microfinance institutions create; RBI had carved out a separate deposit taking franchise under which 10 different microfinance institutions across geographies were awarded small finance bank (SFBs) licenses; prior to demonetization. From its inception, apart from sporadic localized events/risks; the sector has witnessed two large shock events:

1. Andhra (2010) crisis
2. Demonetization (2016)

Demonetization affected microfinance industry across the country and was not specific to any geography. The event also brought differential performance of geographies/districts with collection efficiencies ranging as low as 10% to an average of 70% which historically have exhibited ~99% collection efficiency. This contrast in performance could be attributed to multiple reasons including availability of cash during demonetization to the communities present in the geography, livelihood profile, support extended by operating microfinance institutions in geography etc. The performance of the corresponding microfinance institutions was largely a function of geographies it operates in during stress event.

Given such trends during a stress event; there is a need for investors to understand and possibly prune exposures (wherever applicable) with the help of deep understanding of these districts and using a universal scoring-based model utilizing either a factor based/factor less approach. In this article, we have tried implementing a factor less approach namely Beta (β)

Performance of the sector

As mentioned above, microfinance sector has historically performed at near nil delinquency levels for large periods of time. However, it has been perceived as an inherently risky sector by rating agencies and other financial bodies given its volatility in performance during stress events, lack of assessment of credit worthiness of the borrower and the underlying profile of the borrower and its vulnerability to socio-political events etc. Hence there is a lot of merit in understanding the factors that affect the performance of MFI sector during stress events and geographical profiling. Factors like LTV, IRR, FOIR play a significant role in sectors like CV, 2w while these don’t affect MFI sector. Therefore, other factors that could affect MFI performances are

• Communities and their livelihood
• Socio Political factors like riots, likelihood of socio-political events, historical crime rates
• Historical likelihood of geographies being prone to natural disasters
• Indebtedness

• Multiple Lending
• Penetration
• No. of MFI institutions

All of the factors mentioned above can be clubbed/quantified at a district level, making it an important variable to track . The cultural and economic landscape of India is strikingly diverse resulting in large variation across districts and states. This highlights the importance of need in monitoring the microfinance loan performance at a district/state level over a period

As highlighted in the first section; although there are different ways to evaluate the performance at a geography level, in this article we focus on factor-less method i.e., not identifying reasons/factors for performance and using only the performance variable and its trends over a time period to evaluate a district

Performance measure

Since we are evaluating districts using only the performance variable, we will have to define the variable considered to ascertain the value of performance measure. Portfolio at Risk (PAR) is a standard measure to quantify performance of district and PAR 0% helps to evaluate more conservatively

As we are adopting a factor-less approach to understand district level performance, historic PAR 0% and volatility of PAR 0% are very important. The variable arising out of volatility of the district historic PAR performance is Beta on which the entire article is about.

Beta?

Beta traditionally is a measure of stock’s volatility in comparison to the market. Beta is the ratio of the co-variance of stock’s and market’s returns to variance of the market’s returns over a specific time period. One can look at equity indexes like SENSEX /NIFTY which comprises stocks of 30/50 well-established and financially sound companies.

Where,

β = Beta

rj = Market capitalization of a particular stock

ri = Market capitalization of the Sensex

What does value signify?

0 < β < 1 Stock is less volatile but correlated with the index

β = 1 Stock is correlated with the index || Systematic Risk

β > 1 Stock is highly volatile

β = -ve Stock is inversely correlated to the index benchmark

District Beta: Why?

While performance of a district can be quantified by PAR 0, there is a need to quantify volatility of the district performance over a period. This will give us a sense of how this performance measure (PAR 0) has been moving historically especially during stress events particularly in this sector. We can draw a parallel from beta calculations used in equity market and customize it according to our use case.

Applications of Beta include

• Clustering districts with similar traits
• Pool selection criteria and loss modelling for off-balance sheet transactions
• Benchmarking and scoring the performance of off-balance sheet transactions over portfolio
• Time series analysis of Beta to ascertain recovery patterns for the geographies

District Beta: How?

The idea is to draw an analogy from Beta in equity market to MFI district performance. Since we are going to compare a district’s performance with the market, we might tend to use the whole market data as an index. However, it is meaningful only when we choose an index (created from a selected set of districts — hereafter called “MFI Index”) and compare districts to this MFI Index. Since district beta is analogous to stock market beta, we decided to shortlist 30–35 districts from the set of 550 MFI operating districts which will be basically like the 30 companies in the SENSEX. Index creation can be done in different ways based on one’s desired outcome and requirement.

• Top 30 districts (based on POS)
• Beta 30 districts (in terms of PAR)
• Weak 30 districts
• Ever stressed

In this case, we have adopted an approach where the districts are shortlisted based on factors like sizable district POS, state POS and geographical diversification both at a state and country level to ensure that the index is a good representation of the market as such

Figure 1: Representative 35 Districts highlighted in the map

PAR 0% data of the selected MFI Index districts across 12 quarters were taken

From the above table, weighted average PAR 0 % (based on POS) across districts for each quarter is calculated to obtain the MFI index. This MFI Index will be used as the benchmark while calculating beta for all districts.

Calculation OF Beta — Illustration

Let us assume, MFI Index calculated by using the aggregated PAR 0 values over 12 quarters is

Let’s say that district “A” has its historical PAR 0 values as

Is “MFI INDEX” a good representation of MFI market?

To understand if the MFI index is good representation of the entire MFI market, the market performance along with the MFI index has been plotted. Note that market here is aggregated PAR 0 of all 550 MFI operating districts.

Figure 2: Market PAR (vs) MFI Index

β of the market is computed to be 0.821. Also, the POS of MFI Index is approximately 20% of the Market. Since (Market) is close to 1, the MFI index created can be considered as highly correlated to the market, which in turn shows that it is a good representation.

Figure 3: Beta values of all districts. Y axis represents Beta

What does Beta tell us about the district?

0 < β <1 District is less volatile but correlated with the MFI Index

β = 1 District is correlated with the MFI Index || Systematic Risk

β > 1 District is highly volatile

β= -ve District is inversely correlated to the MFI index benchmark

Sample districts: Historic PAR 0

Microfinance sector has been cash driven sector and while there has been consistent effort towards cashless operations across various institutions; significant lending/collections still happens using cash. Demonetisation had a severe impact on MFI industry and resulted in INR 7000 crores of write-offs on account of sustained delinquencies. MFI Index’s PAR also increases to all time maximum during Demonetisation (Nov-16). To understand different districts performed before and after the stress event, beta is a very good measure.

Figure 4: PAR 0 (Vs) Months. Amravati, Tiruvannamalai, Purulia are districts

In the above graph,

β(Amravati) = 4.84 — Highly Volatile

β(Tiruvannamalai) = 0.00001 — Didn’t get affected by

β(Purulia) = -1.64 — Inversely correlated to the index

Time series Beta analysis:

Say Beta of a district “A” is high during demonetization (as of Dec-16). Is district A always a high beta district historically? This question basically answers how well the district recovered from a stress event.

In microfinance sector, till demonetization most of the districts had very low PAR 0. During demonetization, the PAR 0 of few districts shot up much more than the index. This results in high Beta. While Beta helps us understand the volatility of a district with respect to index, time series beta will help us understand the recovery of district from stress events.

Historical beta of all the districts were also calculated. Districts were classified into three broad categories based on Time series Beta.

• Good Districts: Beta should be low historically even during a stress event
• Recovering Districts: Beta can shoot up during a stress event, but beta keeps reducing to attain a saturated/stabilized state at low PAR
• Bad Districts: Beta shoots up and keeps increasing or stay high

Figure 5: Amravati -Bad District whose Beta hasn’t stabilized after it got affected by demonetisation

Figure 6: Agra — Recovering District whose Beta has reduced by significant amount and got stabilized

Figure 7: Tiruvannamalai — Good District whose beta has remained low even during demonetization

Scoring Entities:

While there are many Microfinance institutions working in India, its very important to understand the portfolio related characteristics. We can incorporate Beta in this analysis to understand the portfolio volatility.

Though in order to achieve lowest weighted average beta portfolio, it’s intuitive to allocate 100% of the portfolio in the district which has the lowest beta. But this can lead to very risky portfolio due to low geographical diversification. It’s always better to find out an optimal portfolio which has very beta as well as high geographical diversification. In this article, we are using a measure called GHHI. GHHI is Generalized Herfindahl-Hirschman Index which quantifies the portfolio diversification, 1 being highly diversified and 10 being highly concentrated. We aren’t going in detail about this since it’s not our area of focus.

To understand the portfolio characteristics, District wise Portfolio cuts of all the entities were taken. Beta for each district can be obtained by the methodology that was explained in the previous section. Weighted Average Beta based on POS was calculated for every entity and its corresponding GHHI (Index that describes geographical diversification) was calculated.

Figure 8: GHHI vs Weighted Avg. Beta for different MFIs

The portfolios of entities which fall under top left are under a risky zone since they are volatile as well as geographically concentrated.

Portfolio Expansion:

A model was developed to intelligently allot concentration limits for districts where the entity wants to expand its portfolio. This model takes input of new disbursement amount and set of potential districts where it wants to grow its portfolio. Using its existing portfolio, this model randomly creates different sets of portfolio concentration district wise. It also calculates GHHI and Weighted Average Beta for every simulation i.e., for every portfolio.

Let’s say, an entity wants to grow its portfolio by 100 crores in 20 new districts. This model randomly allocate concentration in which the portfolio should be distributed in these 20 districts.

For every portfolio (for every simulation), Weighted Average Beta and GHHI were calculated and it has been plotted below.

GHHI vs. Weighted Avg. Beta

Each point in the above graph represents one portfolio distribution. The desired portfolio should have low beta (low volatility) and low GHHI (high diversification). The entity can choose to disburse the loans in the portfolios highlighted in the graph so as to achieve an optimal Beta-GHHI combination.

The concept of measuring volatility of performance using Beta can be used extensively in achieving optimal portfolios by microfinance entities or structuring desks as per above illustrations. Though there may not be a high correlation to final delinquency numbers during regular course, it can possibly act as significant buffer during shock events. Using & monitoring the volatility of pool/portfolio using Beta will definitely add value in crisis management/monitoring mechanisms of an organization.

Authored by Anand TS, Rahul PR & Robin Tyagi