Week of July 26, 2019

1. MicroDigitalFinance: The nominal intention of the faiV is to keep you aware of what's happening in various domains, especially microfinance. But there is a more systematic approach to documenting trends in the industry, e-MFPs survey of people in the industry on what they perceive to be the most important trends and developments. Here's their report from last year's survey. This year's survey is now open--so click here (French; Spanish) and go produce some data on trends in the industry.
There's some new experimental evidence on the impact of mobile money from Christina Weiser, Miriam Bruhn and co-authors, who managed to work with Airtel to randomize the expansion of mobile money agents in Northern Uganda. The findings, to my eye, are broadly similar to Jack and Suri's work in Kenya (keep that in mind, we'll be coming back to it later), though without the direct impact on income poverty.
And here's a report from Karandaaz Pakistan on the regulatory and policy bottlenecks limiting the spread of digital financial services there. The basic issue is a lack of clear policy and regulation, rather than existing policies that prevent action--which raises a question of why the lack of clear policy and regulation was a boon to digital financial services development in so many places, but a hindrance in others.

2. Digital Security: One of the areas the report on Pakistan highlights is lack of clarity on data privacy and protection, but mostly from the compliance side. One of the things I've been thinking a lot about lately is the other side of digital security and the huge burden we are rapidly putting on individuals and firms to protect themselves from bad actors.
I'll admit this is somewhat driven by personal anecdote--I've spent a good bit of time over the last few weeks helping my in-laws recover after falling for one of the "Microsoft" security alert scams. These are older folks, obviously, but both are highly educated, experienced professional people--and they found it completely plausible that Microsoft had a customer service department that was monitoring their computers and helping protect them. Which is not a crazy thing to think, unless you've spent most of your life living in an era where digital service providers have effectively declaimed all responsibility for the damage using their products could do.
But apply this more broadly to people and institutions. As digital financial services spread, we are asking essentially the entire world to become immediately savvy about what a plausible claim is in the digital world. Google and Firefox and other browser providers who have policies and authorized "stores" for browser extensions actually enforce those policies right? Ha, ha, no, of course not, why would you think that? If you get a call from travel agent to book your accommodation at the conference you have been invited to give a keynote at, that's safe right? I mean, how would some scammer know that you're the keynote speaker and the right dates, etc. No of course that's a scam too.
But individuals aside, institutions should have the expertise to protect themselves. Unless it's say, local governments who keep being compromised by ransomware. Or you know, institutions that don't deal with anything particularly crucial, like say, elections.
But the old adage is that "banks" will be the most attacked targets because that is where the money is. Here's an interview with the former CEO of Thomson Reuters and now founder of a digital security company that touches on some important points, especially for financial services providers that aren't behemoths. Here's a blog post about how you can't "hire enough people to fix your cybersecurity problems." The nominal solution is to hire data scientists, which while great for the job market prospects of future economics PhD cohorts, isn't much of an answer for most organizations. How on earth are microfinance institutions going to be able to secure their digital infrastructure?
The bottom line on all of this for me is simply this: we pushed digital financial services as a way of pushing down transaction costs but it seems increasingly likely that if we add up the costs of keeping up with technology and ensuring digital security, we actually radically increased total cost.

3. Our Algorithmic Overlords: Digital security is even more of a concern because of our algorithmic overlords--I keep picturing the movie Brazil, which a minor change to data triggers an unstoppable set of processes, ruining a man's life. It was conceived in an era where those processes were totalitarian/bureaucratic; it's all the more plausible today when those processes are automatic.
What does the age of algorithmic control mean for how governments make decisions. Here's video of a session on that topic from the Institute for Government's recent conference featuring Sendhil Mullainathan and Rachel Glennerster. Rachel makes the point that decision rules hidden by machine learning and algorithms are a big problem for good governance.
I've argued before that decision rules tend to be much more hidden when the decisions are carried out by people, who actually have motivation to hide their real reasoning. Sendhil (et al.) makes a similar point in the video and in this paper which I've featured in the faiV before but is now published in the Journal of Legal Studies.
Consider a few specific instances. Amazon has secret deals with local police forces to turn them into salespeople for the company's Ring video surveillance doorbells; in return the cops get expedited access to video from the installed Ring devices. This is frighteningly terrible, but the core issue is not the technology, it's the people who made the decision to enter into such partnerships and agreed to keep them secret from the public. Or how about Palantir's secret manual for local police forces to use the company's mass digital surveillance technologies. Again, terrible, but mostly because of the people who can use this data to retroactively come up with justification for the surveillance, and the people who agreed to give the government data to Palantir that enables this type of mass surveillance.
Another factor is how people react to the algorithms: who uses them, and how do they react to the information or decision the algorithm generates? We're getting more research on both of these questions and they tend to point to people problems. Last week I included, though it was somewhat buried, a paper on howKentucky judges react to algorithmic risk scores in determining bail. The short version: the judges follow the algorithm for white defendants but overrule it for black defendants, imposing harsher bail conditions. Here's a new paper from Kate Bundorf, Maria Polyakova and Mig Tai-Seale on using algorithms to help people choose optimal insurance plans. The key headline for me: the people most likely to gain from using the tools were least likely to use them.

4. Subsistence Retail: Although I've done a recent faiV on the "research production" function, there are still lots of parts of the system that I don't understand. For instance, the most recent set of NBER papers featured four papers on agricultural extension services. You may be wondering why that fits into the category of subsistence retail. Here's how: the problems of low agricultural productivity, the problems that ag extension aims to solve, are essentially the same problems of low microenterprise productivity. As urbanization occurs globally, the numbers of people engaged in subsistence retail is increasing and the number engaged in subsistence agriculture is decreasing. If we're going to figure out interventions to help subsistence retailers, the best place to learn is from decades of work on subsistence agriculture. So it's helpful to know about a) spillovers from training delivered via video, b) how well groups of undifferentiated producers can effectively coordinate for improved profits, c) the success of an intensive program that targets multiple constraints at once, and d) how information interacts with social networks. And although it's not about ag extension, here's another relevant paper that explores how much small-scale farmers are simply wrong about their plot sizes and how that can mislead conclusions in other work.
Here's an interesting overview of one of the few programs that I've heard of that are directly targeting subsistence retailers: "The Informal Retailer Platform aims to deliver sustainable, large-scale social and financial impact by building the business capacity and income potential of low-income shopkeepers in the global South via education and access to services."
I mentioned that you should keep the Jack and Suri paper on mobile money in Kenya finding increased economic activity, and a decrease in extreme poverty in mind. This is where it comes up again. There's a new paper from Milford Bateman, Maren Duvendack and Nicholas Loubere critiquing that paper and it's findings. Here's the layman's version. I've had my own problems with the way the Jack and Suri paper was framed, in line with Pritchett's larger point about "kinky" development: there's too much emphasis on moving people from slightly below to slightly above an arbitrary line. But I don't find the new critiques particularly compelling--e.g. whether the majority of the benefits of mPesa flowed to wealthy investors in the form of profits is an interesting point but it doesn't tell us anything about benefits to users. My interpretation of the Jack and Suri result is from the perspective of subsistence retail and consumption smoothing: mobile money provides a big boost to liquidity in poor communities where literal absence of physical currency is a real constraint. That in turn enables people to jump into and out of subsistence retail in a way that helps them smooth income and consumption. That again is consistent with the new paper from Uganda cited above. And it's consistent with mobile money having some positive impact but not a large one on individual households. Here I think we're going to learn a lot from a few papers Emily Breza has in the works that shed light on the decisions that Indian village households make in switching between casual labor and subsistence retail.

5. Methods: Two very important items that I've left for the end, largely because they speak for themselves. First, Guido Imbens has a new paper on the Rubin Causal Model and Directed Acyclic Graphs and their relevance and usefulness in empirical economics. It's one of the few times when I've been actually tempted to use the popcorn-eating animated gif. I'm hoping to spend most of my upcoming flight to the West Coast reading it.
And if you pay attention in this space you know about the controversy over a new paper that appeared to incent young people to participate in protests in Hong Kong. Berk Ozler has a lengthy post on the controversy and a better approach to ethics in experimental economics that is a must read, including the comments where Berk has a long response to some of the critics of the critiques.

L-IFT , a diaries research firm, is a proud sponsor of the faiV. See our video on  diaries research with 800 women microentrepreneurs in Myanmar .

L-IFT, a diaries research firm, is a proud sponsor of the faiV. See our video on diaries research with 800 women microentrepreneurs in Myanmar.

A map of the current location for authors of World Development articles created by editor-in-chief  Arun Agrawal . I find it surprisingly diverse, especially considering this doesn't count where the author is originally from, just where they are currently are. I'm not sure how to think about the difference between those two metrics.  Source .

A map of the current location for authors of World Development articles created by editor-in-chief Arun Agrawal. I find it surprisingly diverse, especially considering this doesn't count where the author is originally from, just where they are currently are. I'm not sure how to think about the difference between those two metrics. Source.

Return to the Weekly faiV