How to Figure Out if the Standard Microcredit Model Relies on Growth

In a post on the FAI blog, Timothy Ogden asked whether the standard microcredit model depends on growth. Since growth has been a central focus following several recent crises, this is a very worthwhile question. Here, we will look a little bit at what the data says, but, more importantly, show how practitioners can answer this themselves, since it is now easier to do than ever.

The question is an empirical one at heart and deserves a look at the data for an answer. Ogden specifies two scenarios to watch for:

• “Are there any examples of an MFI shrinking, both in terms of number of borrowers and total amount lent, without experiencing repayment problems?”
• “Similarly, is there any country where the growth of microfinance has slowed substantially and or shrunk without experiencing repayment problems?”

A data-driven approach is particularly relevant here. These types of scenarios - low-growth, low-risk - are precisely those that look unremarkable and which heuristic recall is likely to lead us to miss. The scenarios are also fortunately well-specified in terms of data to look for. First, we will show how to pull together the data to answer this, and then take a quick pass at an answer.

For the first scenario, we want to look at borrowers and the total amount lent (gross loan portfolio). We want to link this with the occurrence of ‘repayment problems.’ While there are a variety of risk indicators available, if we the provision for loan impairment ratio, that should give us most of the information needed - this covers both write-offs and increases in the provisions for delinquent loans. To pull data on these indicators, we go to the new data analysis tool on MIX Market. Here, you can create custom analysis for a specific set of indicators and institutions to build this kind of a data set.

We start by selecting the indicators in question:   

Next, we need to filter the dataset. How many years of data do we want to look at? Which regions or countries? Right now, we are interested in ‘the whole world’ (since we want to minimize bias in how we look for examples). Let’s choose a few recent years though to keep the size of the dataset manageable. (It is easy to expand the time horizon, if you wish.)           

If we are talking about microcredit only, maybe we want to exclude institutions that can mobilize savings. To do this, you can filter on the ‘Non FI’ option under ‘Financial intermediation’ - that leaves only institutions with no savings on their books.

           
Now, we have a dataset that provides over 6500 records across 1850+ institutions over the last 5 years (at the time this analysis was created). The analysis is available here for reference.        

However, since we are looking at growth trends for multiple years, we might want to include only MFIs that provided data for each year. That means we are dropping new institutions that have started operations since 2005, and any institutions that closed operations or stopped providing data during that period. While this trims the sample somewhat, we should not be overly concerned for this analysis because we are only looking for examples that confirm the scenarios outlined, not counterexamples (yet). To trim the dataset in this manner, we can use the ‘Balanced’ option at the lower-right of the screen:


That gives us a smaller data set, spanning 570 institutions - a lot of attrition, but still a large sample. Changing the time horizon to make it shorter or longer would increase or decrease the sample size. Clicking the ‘download’ option exports the analysis to a file that we can open in Excel or other tools. We can even add a title or share the URL with someone else. The link to this analysis is here. (Fn: European users might need a little help working with the CSV files - some guidance is here: http://www.mixmarket.org/How-Do-I/working-with-csv-files)

The second question looks at the same variables - growth, risk, outreach - but at a country level. To get this kind of view, we have to group the data by country.  The options to group the data are in the right-hand side menu - we can just choose ‘country’ to get this level of aggregation:

That opens up another set of choices. How do we want to describe the data now that it is grouped by country? For the outreach variables, it is pretty clear that we want to look at the sum - the total number of borrowers and the total loan portfolio for the sector. For the risk variables, we can choose to either use the median or the weighted average. Both of these are valid choices, but they each tell us something slightly different. The median tells us about the ‘middle’ institution in the country, and it is less influenced by outliers (such as large or risky institutions). The weighted average, instead, tells us about the level of risk in the aggregate - risk at larger institutions is weighted more than risk at small institutions. This seems like it may be a better proxy for the client perspective, so we choose weighted average in this case.           

In this case, we might want to make some further adjustments to the data set. In a comment to the blog post, Daniel Rozas cites the examples of Bank Dagang Bali and Croatia - both cases where MFIs closed or ceased operations. If we pick the wrong time horizon for our balanced panel data set, we will exclude those institutions. So we turn off the ‘Balanced’ option for the country-level analysis. We might also want to include institutions that take deposits - if credit and savings are not directly linked, this may not be a factor in the repayment decisions of clients. So we do not use the financial intermediation filter in this case (although, again, this - or other filters - can be easily added back in by any user).      

That gives us a second data set, grouped by country, for the same time period. The link to the analysis is here. If we trim the data set further, we can look at graphs to eyeball the growth and risk levels over time. For instance, the MENA region has few countries, so we can see the trends there fairly clearly - the crisis in Morocco, coupled with steady growth in Egypt:  

By this point, you will likely have noticed that the new data analysis tool does not display growth rates directly. We have to take the data offline to compute growth rates for the variables we are interested in and to do some more analysis.           

For a first pass at this analysis, we will use Tableau since it is free and publicly available. However, the data files can be directly fed into Excel or a statistics package like R or Stata or almost any other tool that works well with data though. First, we have to include the compound annual growth rates in the file.

With those in place, one way to investigate the scenarios is to group the data by high / low growth rates and high / low risk. If we bucket MFIs and countries in this manner, we can see if any are members of both the low-growth / low-risk (high-repayment) scenarios outlined in the initial post.

Again, there are multiple ways to cut the data, but regardless of the details of how we proceed, we can see that there are some MFIs and some countries that meet both criterion. Answering the questions as to why or under what conditions these scenarios occur is a different, more complicated, question, but identifying them is a good first step.   

If we focus on MFIs, we can see that most MFIs fall in ‘moderate’ growth scenarios (10 - 50 percent), but there is a subset (shown in the table below) in the low growth / low risk bucket. We can then start to look at the attributes of those MFIs - do they all share common traits (region, legal status, lending methodology)?

Or if we dig into the MFI data one-by-one, we can look at institutional dynamics over time. For instance, BancoSol has weathered a few declines in outreach with moderate growth.

In the end, there will be many ways to answer the questions posed in the FAI post. You can use different time horizons, different variables for growth or risk, different types of institutions and so on. A first pass indicate that there are probably some cases worth exploring. The key point here is that empirical questions can (and should) be answered using data, and there are now tools available that should make getting to those answers easier and more accessible than ever.