1. Arbitrary and Biased: I feel like "arbitrary and biased" should have been the tagline for the faiV but it'll have to do as just the name this week's edition (I won't make the obvious joke). The reference here specifically is an update to my post at CGAP on impact evaluations and systematic reviews of financial inclusion interventions. Duvendack and Mader, authors of a systematic review of reviews that I've mentioned in the faiV and in that post, responded. And then I responded to them. The short version, if you don't want to click on all those links or do a lot of scrolling, is that we disagree substantially (though in good faith!) and particularly on the issues of arbitrariness and bias. My perspective on these issues have been substantially influenced by Deaton's and Pritchett's critiques of RCTs, which feels a bit ironic. Systematic reviews are useful, but they are no less arbitrary nor less biased than other attempts to synthesize the literature--they're just arbitrary and biased in different ways, albeit generally more transparent ways (though what we know about how disclosure affects people's trust leaves a question about the benefits of that disclosure).
Reveling in the arbitrarily biased essential nature of the research enterprise, here are a couple of papers that raise different questions about how the literature on microcredit may be biased. Bedecarrats, Guerin, Morvant-Roux and Roubaudreplicate the Al-Amana microcredit impact study and find errors and issues with the data and code--though exactly how much it matters to the big picture conclusion isn't clear. Meanwhile Dahal and Fiala review the microcredit RCTs focusing on whether they have sufficient power to detect likely magnitude of effects (and find that they aren't) and find significant and meaningful effects on profits when the data is pooled. I need to read both these papers more closely, but they are interesting enough that I didn't want to wait before including them in the faiV.
2. Evidence-Based Policy/Methods: Speaking of arbitrarily biased research, the 5% statistical significance threshold is perhaps the most influential arbitrarily biased feature of modern academic research. Some people are trying to change that--well more than 800 who signed onto a letter in Nature protesting the cutoff. Before you come to a conclusion on whether that letter will make a difference, I must note, as many on Twitter did, that it's not a statistically significant portion of scientists who have signed on.
Another arbitrary bias, according to Nick Lea, deputy chief economist at DfID, is the need to run regressions in economics papers. David Evans, now ensconced at CGD, responds with a defense of regressions and some ideas on how development economics can be better.
Here's a reminder that "purely evidence-based policy doesn't exist" though I'm not sure how many people thought it did. And here's a reminder from Straight Talk on Evidence that short-term impact often fades out, something evidence-based policy really needs to take into account.
And finally, here's an interesting piece from mathemetician Aubrey Clayton adjudicating a long-running dispute between Nate Silver and Nassim Taleb over probabilities, finding that Taleb "overplays his hand."
3. Household Finance: The mythology of Spanish colonialism in the Americas centered heavily on cities of gold (anybody remember this?). Here's a story about the reverse--Dominicans searching Spain (and Switzerland) for lost troves of gold. It's all a scam of course, of the sort immediately recognizable by anyone who has spent time in Latin America. It's a fascinating read because of how the story delves into the psychology that has led so many Dominicans to believe (and continue to believe) an ancestor secreted billions of dollars of gold in Spanish and Swiss banks that they stood to inherit--to the point that they quit jobs and made all sorts of other bad financial decisions. When there is little hope, believing that slow, steady abstemious frugality will matter may seem as much magical thinking as hidden inheritances. Here's a piece from Morgan Housel on how much our (macro)financial experiences affect our later decision making.
The mismatch between advice, reality and experience means that most financial advice on offer today is useless for people living on low incomes--and the piece doesn't even address the problems of volatility. Here's Helaine Olen taking Suze Orman to task on the magical thinking that buying coffee makes a meaningful difference in household budgets.
Keep that in mind when you read this announcement from Walmart that customers have "moved $2 billion through" their prize-linked savings program. Don't get me wrong--that's great. But do notice that "moved...through" sounds a lot like high-frequency savings and isn't defined, while the claims remain that people are learning saving habits and becoming more financially secure.
On a related note, here's a recent Planet Money story on Purdue University's new income-share college loan program (which grew out of an idea from a Colombian economist and was piloted in Colombia). And here's a story I stumbled across on the lingering death of a much earlier program at Yale that was a miserable failure.
4. Our Algorithmic Overlords/Digital Finance: Lucy Bernholz has a problem with "AI for good" and other such constructions. The big issue for civil society is not how to use the technology but to figure out how to manage, counteract, regulate, or build on the technology that is already in use. She suggests this essay on AI and the administrative state, which I haven't read yet, but I always trust Lucy's recommendations so I'm passing it on to you.
I think Lucy and I would have a similar perspective on this article, and I'm going to let you guess what that is via this quotation: "We believe that the lack of access to financial services is fundamentally a technology problem." It's a near perfect illustration of Matt Levine's dictum that the fate of FinTech is to relearn all the lessons of modern finance, painfully and in public. Now I'll take JUMO (the source of that quotation) at face value and believe that they are using AI and machine learning to find ways to include rather than exclude. But here's a different fundamental problem: it's always more profitable to fire bad customers than gain new ones.
Of course there are many more issues when it comes to applying AI and machine learning to financial services. Here Aaron Klein does a terrific job of walking through some of them. They of course don't apply to JUMO, yet, because Aaron is looking at how the application of AI and machine learning interacts with US anti-discrimination law. But it's an illustration of some actual fundamental problems of access to financial services and the potential benefits from much deeper engagement between regulators and practitioners in the US and developing and middle-income countries.
5. Cash Transfers and UBI: We'll end with some quick hits on some new stuff on cash transfers and UBI. Here's a write-up of a survey of poor Indian households on their preferences when it comes to cash transfers versus spending on public health, roads and jobs: cash transfers come in last, public health comes first. Perhaps one reason why is a trust gap--here's a story from Kenya about the lack of transparency and limited reliability of public nutrition cash transfer programs.
Another way of determining preferences related to cash transfers is the revealed preferences of what people do with the money. That's what Almas, Haushofer and Shapiro do with the GiveDirectly cash transfers to assess whether there is a nutrition poverty trap (there isn't).
From Natalie Michelle and Joshua Tait a typology of superheroes as neoconservatives. I include this in the hope that someone will produce a version of this for superheroes as neoliberals. Source
