Week of September 17, 2018

1. MicroDigitalFinance: A few weeks ago I wrote that small-dollar short-term loans have always been the bane of the banking industry. We're getting a new test of that. US Bank is launching an alternative to payday loans: loans are between $100 and $1000 and repaid over three months. Interest rates are well below payday lending rates, but still around 70% APR--interestingly on US Bank's page about the loan they very clearly say: "Simple Loan is a a high-cost loan and other options may be available." All of that is good news. But the loans are only available to people with a credit rating (even if it's bad), who have had bank accounts with US Bank for 6 months and direct deposit for 3 months. It will be fascinating to watch take-up, repayment rates, and outcomes--those are where banks have always struggled in this market. Here's Pew's Nick Bourke's take on the US Bank move and the potential for others, with some more regulatory action, to follow suit.
I occasionally remark on insurance being the most amazing invention of all time. It's astounding that it works at all, even in the most developed, trusting and well-regulated markets (see this attempt by one of the US's oldest life insurance providers to collapse the market); it's not surprising that it's a struggle to make it work elsewhere, in the places where households face more risk and would most benefit from access to insurance. So I'm always interested in new work on insurance innovation. Here's a new paper on a lab-in-the-field insurance experiment in Burkina Faso. The basic insight is that many potential purchasers struggle with the certain cost of an insurance premium versus the uncertain payoff. It turns out that framing the premium around an uncertain rebate if there is no payout--which makes both premium and benefit uncertain--increases take-up, especially among those that value certainty most. Yes, you probably need to read that sentence again (and then click on the link to see that even that obtuse sentence is marginally clearer than the abstract). If we want to delve into the details of insurance contract construction, there's also a new paper that delves into how liquidity constraints--a huge factor that hasn't generally gotten enough attention--affect the perceived value of insurance contracts, and how to adjust the contracts accordingly.
And finally, William Faulkner's dictum that "The past is never dead. It's not even past." applies to fintech. A new paper finds that common law countries in sub-Saharan Africa have greater penetration of Internet, telecom and electricity infrastructure, and thus much greater adoption of mobile money and FinTech. That's consistent with history of banking literature that finds common law countries do better on financial system development, financial inclusion and SME lending. 
For the record, I've clarified in my own mind the difference between the MicroDigitalFinance and Household Finance categories. The former provides perspective on providers, the latter on consumers. I reserve the right to break that typology as necessary or when it suits me.  

2. Household Finance: I suppose another way to distinguish between the two categories is that MicroDigitalFinance features bad news only most of the time, while Household Finance is just all bad news. At least that's the way it feels when I come across depressing studies like this: Extending the term of auto loans (e.g. from 60 months to 72 months as has become increasingly common during this low-quality credit boom) leads to consumers taking loans at a) higher interest rates, and b) paying more for the vehicle. Liquidity constraints mean consumers pay much more attention to the monthly payment and get screwed.
It's not just auto loans where liquidity constraints lead to people making sub-optimal choices (yes, I'm thinking a lot about managing liquidity lately). For instance, when people move from traditional health insurance to high-deductible plans they suddenly reduce spending on health care--but not in the ways you want. People don't learn to price shop, even after two years, and they don't reduce spending only on optional or low-value services. And here's the JP Morgan Chase Institute study that shows how much liquidity constraints or their removal affect health care spending using a different approach.
Now if you are a loyal faiV reader, I know you're not thinking, "We need financial literacy training!" But just in case, here's some more bad news: "peer-to-peer communication transmits financial decision-making skills most effectively when peers are equally uninformed, rather than when an informed decision maker teaches an uninformed peer." Or this: "provision of effective financial education to one member of a pair...does not lead to additional improvements in the quality of the untreated partner's decisions." 
If you're thinking, "That hasn't ruined my Friday yet, Tim, give me more," don't worry. How about "Twenty-four million homeowners think it's acceptable to tap into home equity to cover everyday payments." Granted, that's from one of those ridiculous bankrate.com surveys that should be taken with several kilos of salt, but still. 

3. Our Algorithmic Overlords: Here's a quick story about an egregiously bad algorithm the State of Idaho was using to determine how much assistance Medicaid recipients should receive. You can probably already guess--bad data, bad software, bad implementation. But it took a lot of work, and a lawsuit, to figure that out. 
Stories like that emphasize that before handing over decisions to our algorithmic overlords we should want those algorithms to be understandable and fair. Here's a new paper from Jon Kleinberg and Sendhil Mullainathan developing a model that shows you have to pick between simple and equitable. You can't have both.
And here's the "Anatomy of an AI System" that in some ways is a visual proof of the Kleinberg and Mullainathan paper. It's also one of the coolest visualizations I've seen in a while--both in scope and because it isn't reductionist about AI. It takes into account all of the surrounding processes as well. You won't regret clicking on this, unless you have something else really important to do.

4. Global Development: So many things to include this week. Let's start with the biggest: Asher, Novosad and Rafkin have assembled an incredible dataset on incomes in India that allows them to measure intergenerational mobility in a country of more than a billion people, down to the level of 5600 rural districts and 2300 cities and towns. One key finding: increasing mobility among scheduled castes is offset by decreasing mobility among Muslims.
At a necessarily smaller scale, but still big in terms of scope and time, Casey, Glennerster, Miguel and Voors have a long-term follow up on the results of a large scale experiment on Community Driven Development in Sierra Leone, finding that CDD doesn't break down traditional autocratic governance mechanisms enough to allow full exploitation of human capital, which as I understand it was part of the motivation for CDD, and there are easier and cheaper ways to to do so. Of note, they also look at the "prior beliefs of experts on likely impacts"--which, given the "Everything Is Obvious" responses research like this often generates, is pretty cool. Here's Rachel's Twitter thread summary.
Another of the arguments I've heard both for and against CDD-style programs is side-stepping difficult targeting questions--just let the community decide who needs help. Rema Hanna and Ben Olken have a new paper on targeting, specifically on the relative welfare gains of universal basic income versus means-testing. They find means-testing wins using data from Indonesia and Peru, despite some issues; and they discuss adding community-targeting to means-testing.
Meanwhile, here's a piece by Josh Blumenstock that tries to deflate some of the excitement around using high-tech means of targeting, like satellite maps, social networks and call records. In summary, data without theory is useless, and so is data + theory without anthro/soc (or at least anthro/soc informed economics).

5. Methods and Evidence-Based Policy : That's a good lead-in to methods. Let's start with some quick hits. Brian Wansink, whose scandals I've covered in this item in the past, has resigned from Cornell. Noah Smith has a column on the replication crisis in Economics though it's about a very different kind of replication crisis than the one Wansink faced. Now that I type that, it occurs to me that it was in fact easy to replicate Wansink--just making up numbers that matched his would apparently be both a literal and conceptual replication. And here's a new paper on improving diff-in-diff methods to account for effects changing over time.
The idea of evidence-based policy sort of requires that there is evidence of something working. But y'know, nothing does. Encouraging women to get mammograms? Those most likely to respond are those least likely to need one, and because of false positives, the net welfare effect is negative. The health effect of better trade and transport links in the United States in the early 19th century? So negative that it made it people shorter (I mean, as a whole, not specific people). What else? Oh, those gains we all know of like improved water and sanitation, and food safety standards during the early 20th century...no effect on total or infant mortality. That last one reminds me of an old LantRant about assessing whether development interventions matter based on whether they were important in the history (or present) of developed countries. Shall we scratch food safety and urban sanitation off that list? 
I suppose we can hope that these results won't replicate, like the examples that Noah Smith cites. But on the other hand, it's already too late. Once a result is published, no one (or at least no doctors) changes their mind, or changes their behavior.
Wow, this has been bleak. So here's one hopeful note on something that did work. Women's suffrage caused large gains (via demand for more spending on education) in educational attainment of poorer/disadvantaged children, and long-term earnings gains. So go out this weekend and help a woman register to vote (and then go back and make sure she has everything she needs to follow through and vote on election day).

 I would have had the Anatomy of an AI visualization here, but it's way too big, and  Justin Sandefur  created this really great example of how simple choices in the visual representation of data can radically change the way we interpret it. The two charts are of the same data, on the left from the World Bank and on the right from The Economist. Via  Justin Sandefur .

I would have had the Anatomy of an AI visualization here, but it's way too big, and Justin Sandefur created this really great example of how simple choices in the visual representation of data can radically change the way we interpret it. The two charts are of the same data, on the left from the World Bank and on the right from The Economist. Via Justin Sandefur.

Return to the Weekly faiV