Published on Development Impact

Learning from market-level RCTs: Why consumers might want SMEs to get more loans

This page in:

In many of the policy discussions I have about private sector development policies, there is frequently pushback about efforts to help SMEs. This is often expressed along the lines of “why should we devote resources and policy effort to helping relatively well-off, private firm owners instead of poorer individuals?”. This often leads to an emphasis on job creation as a motivation for these policies. But then a second concern is whether all we do by helping some firms is help them steal business from the firms we do not help, reducing or eliminating any benefits to society. However, even if no new jobs are created, consumers could still benefit if the combination of competition and less binding financial constraints causes firms to improve the quality of the products they offer, lower prices, and/or increase the variety of goods they have to offer.

Measuring all these effects is difficult, since it requires doing interventions and analysis at the level of markets, not just firms. A nice new working paper by Jing Cai and Adam Szeidl does exactly this, and provides new evidence on some of these indirect effects of financing firms.

Setting and Intervention

Their setting consists of retail and service firms operating in government-defined geographic clusters similar to malls or bazaars in Jiangxi, China. Markets specialize in broad product categories, such as building materials, furniture, textiles, food, or hardware. The average market has about 82 firms operating, in about 5 narrower sub-industries  - so in the market for building materials, there will be 17 or so firms selling paint, another group selling bricks, another selling stone or bricks, etc. The firms are much larger than the microenterprises used in a lot of development experiments: the average firm has 9 employees and annual sales of around US$525,000.

The authors conduct an experiment in 78 of these markets. 31 of these are pure control markets, where no intervention takes place. The remaining markets receive the intervention, which consists of a large partner bank offering a new loan product. The loan product is a 2-year loan, that differs from the existing products on the market by not requiring collateral, and having a lower interest rate (0.7% per month). The amount firms can borrow depends on their net assets, but the average loan size is US$47,000. The authors conduct a 2-stage randomization: first markets are allocated to treatment (47) or control (31). Then within the treated markets, they randomly select 80% of firms to be offered the treatment in 37 of the markets, and 50% to be offered treatment in 10 of the markets.

This type of two-stage design allows the direct effects of being treated to be calculated by comparing treated firms in treated markets to firms in the control markets, and the indirect effects to be calculated by comparing untreated firms in treated markets to firms in control markets. I use a similar design in a business training intervention in Kenya I’ve blogged about previously. The authors then conduct surveys of a sample of 3,173 firms 2 and 3 years after the intervention to measure impacts, and also survey consumers. Endline attrition is around 10%, with a further 13% of firms shut down.

Some of the challenges in measuring market-level impacts of new loans

Four challenges this sort of work faces are getting sufficient loan take-up, defining a “market”, measuring business outcomes, and having enough power. I note here how the authors deal with these issues.

1.       Getting enough take-up: One of the big challenges that has faced many microfinance interventions is that of take-up. If the sample is a random sample of households or firms, and not screened on those that have already applied for credit, then it is often the case that take-up rates in the treatment group are low, dramatically reducing statistical power. Moreover, in some cases the availability of reasonably close substitutes has meant that while firms or households borrow more from the new lender, this crowds out borrowing they would have otherwise done from other lenders, and so the net increase in borrowing is further limited. Here the partner bank had a loan officer visit each treated firm every month for a year to provide information about the product, and to help them fill in the application form if the firm wanted to borrow. This handholding and long window with which to take-up the loan results in a 32 percentage point (p.p) difference in the likelihood of taking up the new loan product (3 percent of the control group do), and this seems to all be additional lending, with no crowd-out of other loans. Moreover, the untreated firms in the treated markets are also 11-15 p.p. more likely to take up these new loans than pure control firms, which the authors attribute to information diffusion.

2.       Defining a market: it is often really hard to know what constitutes a market. It is easier in rural areas, where geographically separated villages may mean that geography dictates that there is only one marketplace to shop within a reasonable travel time. But competition is perhaps more interesting, but a lot harder to understand, in cities. These government-defined geographic clusters seem like reasonable definitions of markets here, but it would be nice to see more discussion of e.g. how close one building materials market is to another, and whether consumers shop across markets as well as across stores within a market.

3.       Measuring outcomes: measuring firm outcomes is difficult, since firms are often reluctant to share a lot of financial details. The authors had a member of the market office and a loan officer from the bank make the introduction to firms to build trust, and then are able to also use accounting records on sales as well as self-reports.

4.       Power to measure outcomes at the market level: this is really hard, since you need lots of markets, and since skewed outcomes can mean a few firms may make up a lot of the total market revenue. I don’t think there is any benefit to the 50% vs 80% saturation here given this issue, and will come back to this when discussing results.


This figure below shows the distribution of log sales growth at endline. The authors measure that firms assigned to treatment experience sales growth averaging 10 percent relative to the pure controls. That is, finance helps treated firms to grow. However, we see the distribution of the untreated firms in the treated markets lies a bit to the left of the pure controls – there is a negative spillover on sales, which suggests business stealing. The authors show this negative effect is stronger when it is nearby competitors in the same sector that get treated, and there is instead a positive spillover from having non-competitors treated, which the authors suggest is driven by their customers also shopping for other products while they are there.

Figure 1: Treated firms’ sales grow, but steal some business from untreated firms in their markets

Distributions of sales in different groups

The authors estimate that if all of a firms’ competitors were treated, then this would result in a 9 percent reduction in sales. The coefficients on this spillover are smaller if you are also treated, and since not all firms are treated, my read of this evidence is that the positive effects outweigh the negative impacts in these particular markets. The authors conduct a market level regression, and find a statistically insignificant 5.8 percent increase in revenue at the market level. This is where the power issue comes in – it is hard enough to detect an increase of 5 or 10 percent in sales at the individual level, but doing so at the level of a market is very difficult  - the confidence interval incorporates anything from a small fall in revenue at the market level to an increase of 13 percent or so.

On the jobs front, the authors find employment grows 7.5 percent in treated firms, but is offset by a fall in employment in untreated firms in treated markets, which would be 6.6 percent for a firm with all their competitors treated. So the overall impact on employment is likely to be positive, but modest and statistically insignificant (Jing tells me it is a statistically insignificant 5 percent increase).

Why might the impact not be all business stealing? The authors look at how firms spend the money, both by directly asking them and through regressions. The main things firms use the loans for are renovations and increasing the scale of production, buying new inventories and inputs, and introducing new products. This allows firms to be more productive and lower prices for consumers, as well as offer them a wider range of products. Both lower prices and more variety should lead to consumers buying more overall, not just substituting buying from one store to buying from another. The authors survey consumers and find a positive impact on overall consumer satisfaction, and estimate that these gains to consumers result in a social return on capital of around 60% per year. They note that you therefore end up with a very different view of the overall return on this policy if you just look at impacts on firms, versus also incorporating these benefits to consumers.

This idea that consumers can benefit from financing firms is also seen in a different context in the recent AER paper of Andrabi et al., who give grants to private schools in Pakistan. They find that when all schools (firms) in a market get financing, then firms compete on quality, leading to socially desirable outcomes such as higher test scores. Taken together, it suggests another reason policymakers might want to help firms grow, but also emphasizes the desirability of trying to ensure many firms can benefit from such policies, rather than just a select few.



David McKenzie

Lead Economist, Development Research Group, World Bank

Join the Conversation

The content of this field is kept private and will not be shown publicly
Remaining characters: 1000