Editor's Note: You might think you've seen this announcement before, but it's different. The US Financial Diaries project has an upcoming free webinar on May 16th. If you're in the New York City area, join us on May 23rd at New America NYC.
1. American Inequality: The exceptionalism of the United States in promoting home ownership asthe signifier of middle class status and/or upward mobility, and a generally accepted keystone of building wealth has persisted despite the Great Recession/housing crisis. But that doesn't mean that things haven't changed--the availability of housing that costs less than 30% of a household's income has dramatically decreased. Matt Desmond, author of Evicted, writes in the New York Times magazine that the American emphasis on home ownership has become one of the primary engines of inequality. Non-profits--or at least how we measure and fund them--are another (unintended) engine of inequality. In New York state, non-profits pay wages just above retail and food service (and 80 percent of these workers are women, and 50 percent people of color).
2. Our Algorithmic Overlords: The goal of machine-learning and using algorithms to analyze data is to yield better decisions, at least better than human beings would make given biases and the challenges of causal inference. A(nother) new book looking into the way this works is Everybody Lies. I haven't read it yet, but I'm looking forward to it. In the meantime, there's an excerpt in the Science of Us, taking a look at one of those areas that humans always struggle to make good decisions: who is credit-worthy. The substitution of bias against minorities (or at least people different from the loan officer) and the poor for careful judgment is well documented and wide-spread. Netzer, Lemaire and Herzenstein turn the machine loose on data from Prosper, an online platform for peer-to-peer lending, and find that the words that borrowers use are predictive of repayment behavior. You should read the whole excerpt because it does focus on the unintended consequences of using machine learning and big data. I, of course, immediately wonder how quickly borrowers and lenders will adapt to the findings.
Meanwhile, here's a Quora forum with Jennifer Doleac on the American criminal justice system, which dwells a lot on how machine learning is affecting decisions in another area humans have a lot of trouble with: who's guilty and who is a threat for recidivism. And of course, on the unintended consequences of our efforts to punish people. And here's a speculation that Donald Trump is a dynamic neural network/machine-learning algorithm with narrow goals. Here's an alternate version of the same argument, which in addition to being even more frightening, provides additional insight into the potential unintended consequences of data analysis without theory (of Mind).
3. Digital Finance: The item on Prosper and algorithms determining credit-worthiness based on language used by borrowers is about digital finance of course. But in the domain of more traditional ways of thinking about digital finance, here's a story about M-Pawa in Tanzania, interesting for it's integration of savings, lending and education. The bottom line: more savings, larger loans, better repayment. In other news, M-Pesa is supporting proposed regulations for cross-platform transfers in Kenya. And MicroSave has some ideas on how to enable digital finance among the illiterate, since traditional approaches to inclusion through digital have the unintended consequence of excluding the illiterate.
More specifically on the "unintended consequences" theme, though having relatively little to do with digital finance, here's some new research on how global de-risking in banking has cut the number of correspondent banking relationships (what makes cross-border payments even somewhat efficient) have declined by 25% since 2009, pushing whole regions out of the regulated banking sector.
Week of May 1, 2017
1. Households Matter!: If you've followed research on microfinance at all, you've probably come across work by de Mel, McKenzie and Woodruff about giving cash grants to microenterprises (in Sri Lanka and Ghana), finding that the returns to investment in women's firms is much lower (and close to 0) than in men's enterprises. It's a bit of puzzle for several reasons (e.g. why do women borrow if their returns are so low, and why don't men borrow more if their returns are so high?) and there have been various explanations tried out (you can see one of mine in this paper). Bernhardt, Field, Pande and Rigol (paper here, overview from Markus Goldstein here) have a new one that seems pretty compelling based on reanalyzing data from several experiments, including the cash grant experiments. It's an explanation that points back to Gary Becker and Robert Townsend ideas (household's maximizing returns across the household assuming money is fully fungible) about how households work, and away from Viviana Zelizer's (money is often not, in fact, fungible and different income streams in the household are treated differently) or in some ways against Yunus's idea of focusing on women. Bernhardt et al. see that in general when it appears that when women's enterprises show little or no return to capital it's often because the household has another microenterprise that the capital is invested in instead--and those enterprises (where data is available) show gains from the capital injection into the household. When women own the only microenterprise in the household, they see returns (and are often in similar industries) as men.
This is a big deal and it emphasizes how far we still have to go in understanding household finance. This doesn't say that Zelizer's insights are wrong--they are clearly right in lots of cases--but we don't have a solid grasp on when we should think of households as a single utility-maximizing unit and when we should disaggregate.
2. Pre-K Matters? (and other scale-ups): One of the things that households--or if you read some of the charity marketing that has dominated the last decade or so, only women--invest in is their children's education. Unfortunately, it seems that they often under-invest in education and so a lot of effort is invested in getting children into and keeping them in school. In the United States, the current frontier is about universal Pre-K since most every child is enrolled through the beginning of secondary school. The idea is that children from poorer households start school already well behind their wealthier peers, those gaps persist and if we close them early, well the gaps will stay closed. There are some studies that suggest that's true and Jim Heckman in particular among economists has been a big advocate of significantly increasing investment in early childhood education programs. But there are other studies that suggest it's not. I called the arguments on this "Pre-K" wars in my book because a lot of the argument has been over experimental design and methodological issues in the studies.
Russ Whitehurst at Brookings has a new post on the Pre-K wars that I learned a lot from, including new data from Tennessee that shows the returns from pre-K there were negative and the randomization in the famous Abecedarian study was violated in ways that are impossible to correct for. The bottom line for Whitehurst is that while small-scale, intensive interventions with very high-skill staff can make a big difference, programs at scale don't have any solid evidence they work. Which sounds a lot like some of the things we're seeing from scale up of successful programs in other areas of development.
Week of April 24, 2017
1. Sweatshops: You've probably seen the New York Times piece by Chris Blattman and Stefan Dercon about their experiment with Ethiopian factory employment--finding that while many people wanted the jobs initially, they quickly learned that they didn't want them after all. The jobs are dangerous and unpleasant, and people who didn't get the jobs did just as well if not better via self-employment. Meanwhile, Lee et. al. look at urban-to-rural remittances from Bangladesh factory workers and find large positive effects for the folks back home, while the factory workers were less likely to be poor, but also less healthy. Morduch (one of the et als) also notes the workers felt pressure to work more despite poor conditions in order to send money home. It's an interesting compare/contrast.
I'm of several minds about this. First, the Blattman/Dercon piece notes that much of the problem in the Ethiopian factories is that they were poorly run, not that the owners were deliberately trying to exploit workers. If you're a reader of the faiV you know I'm somewhat obsessed with the "technology of management" and how to spread it (and that there's a good bit of evidence that its a big problem, Google Nick Bloom for more). Second, there's the perennial issue of external validity: what do these experiments tell us about sweatshops more generally in other places and times. Here's an overview by Heath and Mobarak (HT Asif Dowla) on the impact of factory labor on Bangladeshi women; and here are some emerging financial diaries of garment workers in several different countries. Third, factory jobs have almost always been terrible, despite the romanticization of those jobs in developed countries of late--and they still are even in places like the United States. So what to make of the fact that they do seem instrumental in the process of countries and households becoming wealthier? And what of my strong prior that most people in developing countries are "frustrated employees and not frustrated entrepreneurs"?
2. Our Algorithmic Overlords: Continuing last week's theme on Seeing Like A State and algorithms, the Royal Society has a new report suggesting easier access to public data sets so that machine learning can help improve policy. You'll be shocked, shocked, to learn that Google DeepMind, Amazon and Uber leaders were all part of drafting the report. The New Inquiry has used data to create a predictive algorithm and heat map for people and places likely to commit white-collar crime. Here's the methodology behind it, which you should definitely at least glance through through to see Figure 4 on page 4. On a related note, here's a story about racial and gender biases being "learned" by machine learning programs.
The white collar crime piece came via Matt Levine, and it's worth scrolling down to his item on Facebook for this gem: "What if human history isn't a video game at all?" Hopefully that will soon be a standard response to FinTech triumphalism: "What if people's financial lives aren't a video game at all?" It all brings to mind this piece from several years ago: The Reductive Seduction of Other People's Problems. You should definitely read it. It's about social entrepreneurs from developed countries traveling to developing countries but it does easily apply to algorithms, fintech and seeing like a state. Hat tip to Lee Crawfurd and Justin Sandefur for reminding me about it.
Week of April 17, 2017
1. FinTech Like a State: Aadhaar, the Indian government's unique identifier system, is now ubiquitous with 99% of citizens over 18 having an ID. That makes it a powerful platform for delivering both government programs and digital financial services. But it also raises a lot of concerns about what the government might do--or what others could do if they gain access to or corrupt the system--when it can track and/or regulate citizen behavior at a detailed level. That certainly plays into the longer-term ramifications of Indian demonetization, especially since it appears that it has driven many more people to digital transactions. CGD held an event this week with Annie Lowery interviewing Arvind Subramanian about Aadhaar, demonetization and universal basic income. I haven't gotten all the way through it yet, so I don't know whether my pre-submitted question was asked, "Which governments should be trusted with the power to deny people the ability to transact legally?"
And for some reason I feel like this piece, nominally about why Silicon Valley keeps getting biotechnology wrong, is really about FinTech.
2. Financial Literacy Like A State (University): "Shut Up About Financial Literacy" says Sara Goldrick-Rab contemplating how higher education institutions blame a lack of financial literacy for the problems students have paying for college. Here's Helaine Olen documenting the head of Penn State University's FinLit program saying: "The real problem is not the rising cost of education, it is in the... lack of financial literacy..." Goldrick-Rab cites a new paper from Sandy Darity and Darrick Hamilton (and here's a Chronicle of Higher Education write-up) making the case that the financial literacy movement as a whole tends to blame the victim rather than acknowledging that many of the choices that look like "low financial literacy" are in fact choices born of poverty and the racial wealth gap. That's a key element of Scott's Seeing Like A State: The drive to solve problems at scale often leads to simplified measurement systems that obscure important distinctions, or miss reality altogether, and ultimately reinforce the problems they are meant to address or create worse ones.
3. Financial Services Regulation: You pretty much have to do financial services regulation like a state. In the United States one of the main financial regulators is the Office of the Comptroller of the Currency (OCC). This week we learned that the OCC had received more than 700 whistleblower complaints about Wells Fargo's practice of opening accounts without customer knowledge or consent, but did nothing. Well not quite nothing. Matt Levine points to part of the OCC's report where it admits it focused too much on process and not enough on outcomes: "You spend so much time making sure that there are processes to stop bad things that you forget to actually stop the bad things." [You have to scroll past the amazing JuiceTech story] That's certainly another part of seeing like a state. And it's a particular concern when you get isomorphic mimicry, in Lant Pritchett's application, of financial services regulation.
On the bright side, I worry a bit less about the progress of our algorithmic overlords when apparently none of the deep learning programs noticed that videos about Wells Fargo like this or this (and many, many, many others) have been on YouTube since at least 2010. But then there's also this about how United's algorithms led to it's disastrous decision-making.
Week of April 10, 2017
1. Social Investment Dissent: Last week I had an item about "social investment wars"--unfortunately Felix Salmon's critical take ("How Not to Invest $1 Billion") on the Ford Foundation's announcement came out just a bit too late to be included. It does pair nicely with a video of Xav Briggs of the Ford Foundation talking about the decision and the future of impact investing.
In the item last week I criticized the sector for not acknowledging trade-offs, principal-agent problems and the like. To be fair, there are people in the sector talking about these issues. Here's a piece from Omidyar Network staff in SSIR about a "returns continuum" rather than "no tradeoffs." Here's a piece from Ceniarth staff concurring. And there are two recent pieces from the CFI blog on responsible exits from social investments: first, pointing out that who a social investor sells to should be part of the impact calculation, and second making an important point about the "missing middle" in social investment (though they don't use that term).
The missing middle they are pointing out is investors who are willing to buy on the secondary market but maintain social goals. This echoes a long-standing problem in foundation philanthropy: most large foundations want to be first movers and believe that there are "followers" who will come after them to support organizations or programs after the initial grants. It seems in both cases, the followers just don't meaningfully exist.
2. Financial Literacy: April is financial literacy month in the United States at least. I continue to use financial literacy as my barometer for the evidence-based policy movement: if evidence isn't making an impact here, why should we expect to have an influence elsewhere? But on to the links. Here's perhaps the dumbest idea currently circulating--making financial literacy a requirement for high school graduation. Here's Graham Wright de-mythifying financial education in the developing world. And on a brighter note, here is IPA's review of what's been learned from impact evaluations of financial literacy programs around the world (it's not just "they don't work!").
3. The Technology of Management: Having written a couple of books about Toyota, this is a particular fascination of mine--and of course I therefore think other people should be paying more attention to it. Management matters a lot to firm performance (explaining about 20% of firm-to-firm productivity gaps), which in turn matters a lot to wages and job creation/growth. Here's Nick Bloom in Harvard Business Review on rising firm inequality. Here's Bloom et al. on why the technology of management diverges (or alternatively, doesn't converge as much as expected given the returns).
Read MoreWeek of April 3, 2017
1. Cash vs Chickens Wars: Within development circles, the most widely read point/counterpoint began with Chris Blattman's piece in Vox, written almost as a letter to Bill Gates. Blattman takes issue with Gates' idea to provide livestock, specifically chickens, to poor households and instead proposes a test of the benefit of just giving cash. To be clear Blattman isn't saying that cash is better, but that we don't know--and we do know that giving chickens is much more expensive (and everyone who's been involved in aid knows at least one story about how "the chickens all died")--so we should run a test and compare. Lant Pritchett responds on CGD's blog, saying in all his years in development, never once has the question of "chickens versus cash" arisen as a pressing question. One reason is that Pritchett believes the goal of development shouldn't be marginal improvements for the poorest but generating the kind of growth that has seen hundreds of millions escape poverty in China, Vietnam, Indonesia and other countries. Of course, Blattman responds and does a good job keeping the focus on what I would call the competing theories of change proposed by Chris and Lant. In fact, I have called it that, and if you're interested in a deeper dive into the issues in this debate, I know a good book you should read (or at least check out Marc Bellemare's and Jeff Bloem's review of it).
2. Mortality Wars: Those in the US policy community, on the other hand, have probably been too occupied following the "mortality wars" to even know there's a battle between cash and chickens happening next door. Here's the quick background: Anne Case and Angus Deaton have a new paper about mortality rates in the US--I would say more about their results but, of course, this wouldn't be a war if there wasn't vehement disagreement over what their results actually are. As with an earlier paper, Jonathan Auerbach and Andrew Gelman take issue particularly around the composition of Case's and Deaton's aggregate results, and provides charts decomposing mortality rates by race, gender and state. There are a lot of other critiques, including about the data visualization in Case's and Deaton's paper, but you can save yourself a lot of time by just reading Noah Smith's excellent post about the data and the debate which brings the attention squarely to where it should be: that mortality rates for white Americans stopped following the trajectory of other developed countries and a massive gap has opened up between the US and others.
Then there's a secondary discussion of why this is happening and what it all means so here's some supplementary reading on that, courtesy of Jeff Guo at the Washington Post: An interview with Case and Deaton; "if white Americans are in crisis, what have black Americans been living through?"; and it's more than opioids. But if there's one related thing you aren't likely to read, but should, it's this article from Bloomberg on automobile manufacturing in the South.
This also seems like the best place to insert my favorite new aphorism: "Being a statistician means never having to say you are certain." (via Tim Harford)
Week of February 27, 2017
1. Mobile Money: The GSMA published it's annual mobile money "state of the industry" report, except that this time it's a review of the 2006-2016. Here's a summary (I know which one I would click on). As you'd expect, the GSMA heavily touts some impressive statistics on growth and usage. And I suppose you can't be surprised at the sometimes more than implied leaps from outputs to outcomes. But the more I look at things like this, the more I'm reminded of Lant Pritchett's book The Rebirth of Education: Schooling Ain't Learning and the history of using school enrollment as a very bad proxy for the outcome that everyone actually cares about, learning. Or to use a closer to the finance industry analogy, was there anyone tracking the spread of ATMs and debit cards and getting excited about how much it was going to help the poor?
2. The New Redlining: Fisman, Paravisini, and Vig have a new paper (and a summary) in AER on the effect of loan officer "cultural proximity" with borrowers in India. Loan officers who share religion, ethnicity and other traits with a borrower provide larger loans on better terms, and borrowers have higher repayment rates, meaning the loans are more profitable for the bank. The proposed mechanism is reduced "information frictions" in the lending process. It's a more subtle form of redlining--a systematic way that banks denied credit to minority communities in the United States. Fisman et. al. suggest hiring and promoting minority loan officers as the obvious way to combat the discrimination they document (that's a version of "immigrant" banks that you can still find in places like New York and San Francisco). It's also part of the reason that algorithmic approaches to credit, like this effort to use exam scores as a proxy for student lending in Kenya--remain appealing: you can simply skip past the bias inherent in human-to-human interactions! If only. The long battle against algorithmic redlining is only just beginning and will be much harder to win as long as we succumb to the fiction that algorithms fix bias. I wonder which socioeconomic class the people doing better on exams come from?
Week of February 20, 2017
1. Ken Arrow: Ken Arrow died this week, at age 95. Arrow is the youngest economist to win a Nobel (51), and probably could have won more than once so wide-ranging was his work and influence. He won the Nobel for his work on general equilibrium, but he made foundational contributions to health economics, insurance, risk analysis, and more. Still, he was most famous for his Impossibility Theorem, showing that no majority voting system can be free of arbitrary outcomes. It was also apparently impossible to discuss a subject he wasn't well read in. Here is Tim Harford's short obituary. Here is the Monkey Cage Blog's appreciation ("Arrow proved the existence of a solution to the problem of economics and the the non-existence of a solution to the problem of politics."). And here is a three part interview with Arrow from 2009.
2. A Certain Kind of Aid: Speaking of impossible, it's impossible that the combination of subject and price of this new book isn't trolling, isn't it? To be fair, aid does go in cycles, and this was the explicit strategy during colonialism. The item name is a reference to this, if you were wondering. (Hat tip: Justin Sandefur)
3. Pick Your Crisis: Is the next US financial crisis going to come from widespread default on auto loans? Americans now owe $1.16 trillion on car loans, an average of $6000+ per licensed driver. Who is loaning all that money? The car manufacturers; 3/4s of lending to subprime borrowers is underwritten by the manufacturers. Or will the next crisis be the result of the large numbers of Americans who aren't saving for retirement? New data from the US Census Bureau based on tax records finds only 41% of American workers eligible to for a workplace retirement account are using them (another reason why the idea, noted in last week's faiV to make withdrawals from retirement accounts even harder may not help very many people). Or perhaps the next crisis will be based on uncertainty. The Trump administration seems to already be mucking with government statistics. In other words, you should probably lower your expectations of new data insights coming from the Federal government.
Read MoreWeek of February 13, 2017
1. F*ck Nuance: I know what you're thinking, but that's not what this item is about. It's actually about Kieran Healy's forthcoming paper in Sociological Theory called, well, F*ck Nuance. He argues that the rising demand for "nuance" in sociological theories inhibits clear thinking and useful research. It reminds me of what I've heard a lot of economists say about the demand for complicated formal models in economics papers. It's not what Healy intended, but here's a story about a FinTech start-up ditching FICO scores while offering "the fastest [credit] on the market," which certainly doesn't bother with any nuance like whether the product is good for customers.
2. F*ck Impact: So that's not what Jishnu Das's blog post is actually titled, but it might as well have been. Das (quite ironically, as David McKenzie noted) blogs about how researchers being held accountable for having impact beyond academia, for instance by writing blog posts, is a drag. It's worth reading because there are some valuable nuggets especially about the "poorly specified model" of impact in use and the breakdown of trust between funders and researchers. If you were interested in hearing the thoughts of some development economists who care a lot about having an impact, you could do worse than checking this out. On a different note, the subdued reaction to the post convinces me that the development blogosphere really is dead.
3. Commitment Savings: In the WSJ, Bernartzi and Beshears argue that evidence from commitment savings evaluations suggest that restrictions around retirement accounts should get even more severe, particularly citing the original Ashraf, Karlan, Yin work in the Philippines. It's true that retirement accounts in the US are very leaky, but the cause isn't just temptation or present bias as Benartzi and Beshears imply. Volatility of incomes and expenses seems to play a large role. Here's a video of Dean Karlan discussing the possibility that less restrictive accounts may work better.
Read MoreWeek of February 6, 2017
1. The World's Largest Financial Inclusion Experiment(s): That's a descriptor that applies to basically any Indian or Chinese national policy, but India is definitely where the interesting financial inclusion experiments via sweeping policy changes action is right now. Whether we'll learn from these experiments remains to be seen. Here's an attempt to learn something from the Indian government's JDY scheme to make bank accounts widely available--finding that usage is growing, and the primary actions are person-to-person transfers (sounds familiar doesn't it?). And here's a story from the FT on how insurance companies are attempting to use the shock of demonetisation and the increasing use of bank accounts (via JDY) to increase penetration of microinsurance.
2. Economists' Social Skills: A big concern in developed economies is the "skill mismatch": the skills that people have today (and by implication that our education system is focused on imparting) are not the skills that will be required for jobs in the near future (if not the present). The two items I see most frequently in such discussions is "coding/programming" and "social skills." I've never understood the former--computers are obviously better at coding than people are already. Social skills though, those do seem important. The surprising thing in this piece on skills and future jobs is the chart about mid-way through which says that economists require strong social skills, even more so than physicians, nurses' aides, and police(!). Perhaps that mismatch between practicing economists and social skills explains why the chart also notes that employment share for economists is stagnant.
