A friend of mine recently forwarded me a recent publication titled The Limitations of Microcredit for Promoting Microenterprises in Bangladesh that appeared in the Jan-Mar 2012 edition of the Economic Annals. I sort of feel bad for picking on this one study but it’s somewhat representative of a few others I’ve seen where a couple of things just bothered me a brick ton.. Here’s a sampling. Apologies in advance to the authors; and I’m sure my turn will come soon enough
Claim 1: The field survey shows that about 11.7% of the microcredit borrowers are this kind of potential or growing microentrepreneur (Abstract).
Except, the survey this paper is based on was not randomly sampled (p.g. 43):
Samples were selected from urban (32.4%), semi-urban (27.2%), and rural (40.4%) areas, to ensure that microborrowers of different sized loans engaged in various categories of economic operations in rural and urban settings were adequately represented. In the absence of full knowledge of the structure and distribution of the microcredit borrower population in the country, random sampling as representative sampling is neither possible nor desirable (Molla and Alam, 2011). Moreover, in many situations random sampling is neither effective nor cost effective in serving the purpose for which sample data are collected. Purposive or judgment sampling is effectively used in such cases. Accordingly, a judgment sampling procedure was thought more effective and appropriate for this survey.
A couple of things:
- The 11.7% is not representative of the universe of Bangladeshi microcredit recipients, but of the non-random sample used in the study. Indeed, unless the distribution of microcredit borrowers is exactly 32.4%:27.2%:40.4% over urban:semi-urban:rural areas, the number is anything but 11.7%. Not a lot of MFIs want to work in urban areas, specially slums, where a large proportion of the urban poor and ultra-poor live. Ask Shakti Foundation, one of the few MFIs that serve this challenging demographic. The real number could be 5%, it could be 20% – who knows..
- I don’t buy the excuse that random sampling is not appropriate. By the authors’ own admissions, there are 15 million microcredit borrowers. (I think that’s a low ball number, but let’s go with it for now.) There are 150 million people in the country, including infants, the middle class, the elderly – demographics which are not obvious target populations of MFIs. You are almost guaranteed to hit a MFI borrower if you throw a … pillow a few times. Purposive or judgemental sampling is done when you are either targeting a very specific group and you don’t care about being representative, or there are so few of those you want to talk to that you have to search them out with deliberation. I can’t figure out how that could possibly be the case here.
- I find the suggestion that “in the absence of full knowledge of the structure and distribution of the microcredit borrower population in the country, random sampling as representative sampling is neither possible nor desirable” quite counter-intuitive. Indeed, if one does not know the underlying distribution of borrowers, a random sample would have illuminated that unknown too, contributing to the findings of this paper. Also, one has to look at the big three – Grameen, BRAC and ASA- and one can guesstimate fairly accurately what the distribution is..
It would have made much more sense to present the findings in three silos – urban, semi-urban and rural, and share all the findings within those segments. It would have been more appropriate than as an aggregate too since, conceivably, the urban implications of microcredit on microenterprises is somewhat different from rural ones.
Claim 2: A sizeable chunk of all borrowers, microentrepreneurs or not, have issues with the terms of credit, which are inadequate for entrepreneurial purposes. (p.g. 47, my summary)
The entire para is as follows:
About 20.7% of all the borrowers and 15.4% of the microenterprise borrowers believe that they do not have the scope to effectively use the entire loan amount at the start of activities. In practice about 29.2% of all the borrowers and 20% of the microenterprise borrowers did not use the entire loan amount at the start of their business operations (Table 3). On the other hand, about 27.9% of all the borrowers and 55.4% of the microenterprise operators had to top-up the loan fund with personal or other borrowed funds to start operations. On top of that about 21.4% of all the borrowers and 8.6% of the microenterprise operators invested additional funds during the year, either from personal sources or from credits obtained from other microcredit providers. About 28.3% of all the sample clients and 40% of the microenterprise clients received multiple loans (2-3 or more) from 2-3 or more microcredit institutions.
I agree with the general thrust of the message. The rigid disbursement and repayment schedules are not conducive to the fluid needs of business, and borrowers often have to borrow from other sources to make up working capital shortfalls.
But the numbers I see here actually don’t seem that bad:
- If 20% of borrowers don’t believe they can use the entire loan amount right away, then 80% believe they can, right?
- Similarly, if almost 3/4 of borrowers and 1/2 of microentrepreneurs do not have to top-up funds right at the beginning, that’s not too bad, right?
- Also, similarly, if almost 80% of borrowers and > 90% of microentrepreneurs did not have to invest additional funds, that’s not terrible either, right?
- 40% of the clients borrowing from multiple MFIs could be seen as a bad thing, but we have to be careful not to equate miltiple-borrowing with overindebtedness. PotP is chock full of examples which clearly demonstrate how sophisticated the poor are in their financial management, and Bangladesh was one of the study countries too.
I mean, it looks like microcredit is able to satisfy funding needs at various levels for 75-80% of borrowers in general and 50-90% of microentrepreneurs in particular, more or less. If we demand more, are we not holding microcredit to an unrealistically high standard, given the realities of the products and distribution channels?
Again, the bone I pick is not with the underlying message, but that the numbers put forward seem to weaken the case being made.
Claim 3: (The) preference for women as clients for credit is found to be a strong methodological limitation of the microcredit delivery system in promoting microenterprises. (p.g. 45-46)
The authors make a compelling enough case, up to a point. Men tend to run businesses in Bangladesh, and their survey shows how the female clients simply pass on the microcredit to their male counterparts. The respondents note the following as reasons for dependence on men (p.g. 45):
- inability and lack of skill of the women borrowers,
- more investment opportunities in man-relevant activities,
- male-dominated family structure where male members maintain and control family,
- social environment and custom where business activities are considered to be men’s work, and
- women are not expected or respected in the domain of men’s activities (business activities)
So .. Why don’t MFIs simply lend to men? Blind ideology, or is this something based on reality?
Google “men microfinance” and you’ll get a ton of useful discussion, interspersed by a couple of good studies on this issue. The short answer is that we may not always know why, but men tend make for crappy borrowers. There is something in the woman-borrower/man-entrepreneur dynamic that “works.” (But may not always “work” in a way that is comforting – check out Lamia Karim’s work for societal dynamics gone bad.) Man-borrower-entrepreneur models don’t tend to work.
It is not constructive to simply take out the borrower intermediary when she clearly has something big to do with things.
It is also why SME lending has been so hard.
There are bunch of other things that gave me reason for pause, including:
- The study relies too much on the Grameen model. BRAC and specially ASA do not do things like Grameen, and the results might be quite different for them. The authors may find that the “stereotyped microcredit delivery system” may have considerable variation within it.
- There is no “counterfactual” to the claim that “microcredit is not sufficiently productive to generate enough revenue for interest payments if market rate wages are paid for family labour” for a significant portion of the borrowers. What if they did not borrow? Would they earn more? Would they starve?
- It calls 25%-65% interest rates exorbitant, citing Bangladesh Bank lending rates of 4-5%, and commercial lending rates of 10-12% (p.g. 42). 65% could be considered exorbitant, but 25%? And most importantly, there are very, very real reasons why microfinance interest rates are so high. And it’s not because Grameen/BRAC/ASA are wannabe loan-sharks.
- Its citations are .. unimpressive. One study used to comment on male-female gender dynamics is from 1996 – arguably an eternity in terms of the evolution of microfinance (p.g. 45). Commercial lending rates are quoted from 1997 (p.g. 42). More than half the references are the authors’ own, and the rest are mostly links to MFI reports.
Overall, this piece has decent analysis behind it. I think it gets into trouble trying to hammer out conclusions from it that are not adequately supported by the data.
By the way, if 11.7% of the (non-random) sample are microentrepreneurs of some stripe, what about the remaining 88.3%? What are they using microcredit for? If microcredit has limitations for “promoting microenterprises in Bangladesh,” what is it overwhelmingly succeeding in doing?
Wouldn’t that be fun to know!
In my last post, we saw how SKS’ write-offs levels went up from 0.29% in FY2008 to 0.60% in FY2009, and then up again to 0.86% in FY2010, and that PAR numbers followed a similar trend. Today, we’re going to look into this in more detail, and then ask why this might be an issue for SKS.
Tracking the increase
Before we go any further, let’s cross-check the numbers we got from MixMarket. Their numbers are generally very good, and allowing for reporting period disparities, differences in interpretation of similar terms, or differences in how something is bucketed, there is broad agreement in most cases with the numbers one gets from MFIs.
This is no different for SKS, which notes the following for it’s write-off figures in its Annual Report for 2009-2010:
(Source: Schedule 19, SKS Annual Report 2009-2010, p.g. 82)
Ok, SKS agrees that write-offs levels increased between FY2009 and FY2010. What’s their story for FY2008 to FY2009?
Turn to page 203 of the Red Herring, and you’ll learn that the increase in bad debt write offs was a whooping 2,205.8%!
Yes, that’s right, bad debt write-offs were up 22 times in FY2009, compared to FY 2008. Umm… what!?
And let me take this opportunity to say again why I am really happy about getting my hands on the Red Herring document – SKS has gone out of the way to provide an abundance of information, and it’s all very informative. There are two primary reasons that account for this, albeit partly:
- Under their provisioning policy, they used to write off 50% of the loans overdue between 25 to 50 weeks (or 6 to 12 months, roughly) in FY2008. In FY2009, they wrote off 100%, i.e. the entire amount. Makes sense that this increased write-off, since this change doubled the number of loans they had to bump off the books for that category of loans, all other things remaining the same.
- The portfolio grew during this period, to the tune of 81.5%. Again, makes sense that this increased write-offs – if you give out more loans, proportionately more loans will also have to be written-off, all other things remaining the same.
Here’s what’s interesting – once you normalize for these changes, (essentially deflating the 2,205.8% number by 100% and then by 81.5% again to account for the explanations) the write-offs still go up by approximately 600% … ! Gonna have to say it again – umm… what!?
Something is pushing up SKS’s bad loan count, and pushing it beyond what seems to be the average for Indian MFIs. As we saw in the previous post, the average write-off for the market was 0.52% in 2009; SKS’ figure was 0.73% for the corresponding period.
SKS needs to identify the drivers behind bad loans. I could hazard a couple of guesses:
- Quality of staff – any any given point, three-quarters of SKS’ staff has been around for less than year. This is partly because of phenomenal growth leading to continuous hirings, and partly because of high staff turnover.
- Characteristics of new markets – borrower demographics, existing competitors etc.
- Credit policy – particularly those related to due diligence on loan disbursal and procedures for loan recovery
- Resource focus on growth, as opposed to consolidation
Without looking at more detailed numbers or talking to SKS staff though, we’d be hard pressed to know which of these, if any, are key drivers of deteriorating portfolio quality.
Additional Mitigating Factors
Irrespective of the underlying drivers, there are two effects to keep in mind that make deteriorating portfolio quality a particularly pernicious issue for MFIs enjoying rapid growth:
- Lag Effect: Loan vintages (i.e. loans given out during the same period – say a month) do not display full delinquency levels until a few months after disbursement. Given that most SKS loans have a 50-week term, a short 3 month catch-up period would mean that ¾ of the portfolio are displaying full delinquency levels, while a 6 month one would imply that only ½ of them were doing so – assuming 0% portfolio growth. This also means that if, for some reason, the overall portfolio quality deteriorates, it will not manifest itself fully for a good couple of months.
- Dilution Effect: Of course, SKS’ portfolio is not growing at 0%. It grew 99% between FY2009 and FY2010, and according to the 2009-2010 Annual Report, has a CAGR of 150% over the last 4 years. Consider what this means for the lag effect. The earlier vintages that are now displaying full delinquency levels are essentially watered down, and the low or non-existent delinquency levels of newer vintages have a higher per-rupee weight. How much watering down happens is dependent on the vintage disbursement amounts, but in general, the dilution effect depresses the delinquency levels further.
Why is this an issue? Well, SKS cannot continue to grow at a CAGR of 150% for too many more years. The Indian MFI market is getting increasingly saturated, and the industry as a whole will need to slow down. As growth slows, the delinquency and write-off levels will catch up. If SKS does not anticipate this catch-up and prepare accordingly, investor confidence will take a blow, along with portfolio quality.
Vintage-level Analysis Has Some Answers
One can get a sense of what will happen in the future by literally seeing how loans in SKS’ portfolio age. Loans given out during the same period, usually a month, are said to be in the same vintage. I have not found any public information on SKS vintages, let alone their delinquency profiles. It’s generally not very helpful to guesstimate on this because such profiles can vary greatly between MFIs with similar top-level PAR and write-off numbers, and can differ substantially for different products even within the same MFI, so I’m not going to dive into a modeling exercise.
It’s not too difficult to figure out the lag and dilution effect though, if one has access to granular portfolio data. SKS would have to take each vintage, figure out its delinquency profile, derive historic trends by superposition, and account for differences based on product characteristics, branch office location or any other salient factors. It can then combine projected performance of existing vintages with internal growth targets, and see where the PAR and write-off numbers end up at.
Sure, the past is not necessarily the best predictor of the future, but given the purported out-of-the-box nature of SKS products in particular and microfinance products in general, and the short loan lifecycles that reveal profile changes relatively quickly, this should be a pretty useful exercise.
This is the first of a series of posts that will take a look at the numbers behind MFI operations. I find it to be quite an instructive exercise to wade through MFI data in the rare instances where they are available in any level of detail beyond mere institution-level aggregation, as presented in annual reports and the like.
We’ll first look at delinquencies and write-offs, and use the Indian MFI SKS as a case study of sorts. This is partly because the Red Herring released prior to its IPO provides a wealth of information that allows for more meaningful and in-depth analysis, and in general, affords a rare look at the inner workings of an institution enjoying prodigious growth while carving out a place in a rapidly evolving market. SKS is also the largest MFI in India, followed by Spandana and SHARE, and according to 2009 MixMarket data, serves about a fifth of the Indian MFI market. What we might glean from this is therefore relevant to a large chunk of the Indian microfinance market too.
PAR and Write-off as Measures of Delinquency
Well-run MFIs fastidiously maintain high portfolio quality. Many MFIs have primarily lent to women, who traditionally have excellent repayment rates. Many adhere to the Grameen group-lending model because they rely on peer support and pressure to enforce regular repayment habits. Very low delinquency and default rates have made this sector a favourite for investors, domestic and international.
Portfolio-at-Risk (PAR) is one of the standards of measuring delinquency for microfinance loan portfolios. It is defined as the total outstanding principal balance of loans with any amount of arrears due. Thus, if a $100 loan still has $45 in outstanding principal, and the $2 of that was due last week is not paid up, the PAR amount is noted as $45. In a way, PAR is the most conservative measure of how much loss the portfolio would suffer, since this is the maximum value that the MFI would lose from this loan if nothing more was ever paid back.
PAR is usually associated with a number of days count, where PAR30 would mean the total outstanding principal balance with any amount of arrears due for over 30 days, PAR60 for 60 days, and so on. Many MFIs will typically provision against 100% of the PAR120 amount, thus assuming that none of it can be recovered.
Once a loan is written off, it is no longer on the books of the MFI.
SKS Delinquency Profile
The unweighted averages for PAR30 and write-offs were 1.84% and 0.52% respectively for Indian MFIs with more than USD 10m in gross loan portfolio. When one considers the weighted averages, the corresponding figures are 0.59% and 0.54%. (The weight applied is principal outstanding.) We are interested in looking at the weighted average, by the way, because it gives us a sense of how every dollar (or rupee) in the portfolio is doing, on average, as opposed to every loan, which would be the case for the unweighted measure. Comparing the two PAR30s tells us that larger portfolios have much lower PAR amounts on average – this is quite interesting, and we’ll come back to why a little later.
This provides some context to PAR and write-off data for SKS for the last 5 years:
(Screenshot Source: MixMarket. Note that FY2010 denotes fiscal year ended Mar 31, 2010.)
Two things of interest jump out at me from this data:
- Write-offs levels went up from 0.29% in FY2008 to 0.60% in FY2009, and then up again to 0.86% in FY2010.
- The recent PAR30 and PAR90 rates are much less than the write-off rates.
Rising write-offs are an obvious issue. The PAR numbers also follow a similar trend. We’ll take a closer look at this in my next post.
Do you think it’s normal that the PAR numbers are less than write-offs, by the way? It’s quite an interesting phenomena, and we’ll focus on it on the third post of this series.