Research

Working Papers

"Social Network Structure and the Radius of Risk Pooling." Putman, D.S.; Working Paper. Under Review.

Abstract: While risk sharing transfers are mediated by bilateral networks, the network neighborhood may not serve as the ultimate measure of the radius of risk pooling. In two empirical illustrations, I explore how social network structure drives risk sharing participation using community detection and dyadic regression. In both cases, I find evidence consistent with a radius of risk sharing that extends beyond the network neighborhood. First, using data from a behavioral experiment in Colombia, I find that (detected) community co-membership and distance-2 connections (i.e., being a friend of a friend) help explain co-membership in experimental risk pooling groups. Second, using detailed social network and transfer data from Tanzania, I find that while community co-membership matters for making intra-village transfers, it is crowded out by distance-2 and 3 connections. In the discussion, I explain the nuances in which network structures are relevant for risk sharing by appealing to the role of group formation, asymmetric information, and the role of sampled networks. Additionally, to address concerns around validity due to sampled networks in the Colombia data, I simulate the effect of network sampling on dyadic regression results using the Tanzania data, which form a village census.  These insights are particularly important in informal contexts where groups are loosely defined, rarely self-identified, and often illegible to outsiders.
Note: this paper replaces "Social Network Structure and the Scope of Risk Pooling" and "The Scope of Risk Pooling."
First draft (as "Scope"): June 30th, 2020

"Sharing Covariate Risk in Networks: Theory and Evidence from Ghana." Putman, D.S.; Working Paper. Preparing for Submission.

Abstract: When risk preferences are heterogeneous, pooling covariate risk can lead to welfare improvements by shifting correlated shocks from more risk averse to less risk averse agents in exchange for a premium. However, the ability to pool covariate risk in this way depends crucially on whether agents prefer to share risk with others who have similar risk preferences. Do agents assortatively match on risk preferences? To investigate this question, I build a theoretical model of covariate risk pooling with heterogeneous risk aversion. I use rich data from four villages in southern Ghana to construct a bilateral risk sharing network and community detection algorithms to detect risk pooling communities, which bound the scope of risk pooling. Using econometric models of network formation, I estimate that individuals prefer to assortatively match in risk sharing networks. But, in detected communities the magnitude of assortative matching falls considerably. I compare the allocation of agents in communities to three benchmarks, including an optimal and worst-case scenario. In terms of assortative matching, I find that the observed networks deviate only slightly from optimal networks for this form of risk pooling.
An older version of this paper appeared at PacDev 2020 as "Risk Preferences and Risk Pooling in Networks: Theory and Evidence from Community Detection in Ghana"
First draft: November 30th, 2019. 
Additional Materials, NoBeC Early Career Research Short Talk

"Probing the Limits of Mobile Phone Metadata for Poverty Prediction and Impact Evaluation." Barriga-Cabanillas, O., J. Blumenstock, T.J. Lybbert, D.S. Putman; Working paper [please email for latest draft]. Preparing for submission. 

Abstract: Recent work demonstrates that machine learning can predict household wealth and target resources to the poor using the call detail records (CDRs) generated by individual mobile phone subscribers. We use survey data from an emergency cash transfer program in Haiti and CDRs to explore whether similar methods can be tapped for impact evaluation. CDR-based wealth predictions using nationally-representative data are on par with others in the literature, but expenditure predictions are much noisier. Focusing on the target  populations of the cash transfer program further reduces prediction performance. While a conventional survey-based impact evaluation shows positive impacts of the program on food security and dietary diversity, the CDR-based approach fails to detect comparable impacts. In postmortem discussion, we assess reasons for this failure. We conclude with a discussion of the implications and limitations for predicting welfare outcomes using big data in poor countries.

"Strategies for Reducing Non-Institutional Fraud and Building Trust in a Digital Market Platform" Byrne, S., D.S. Putman, M. King, C. Jang; Working Paper [please email for latest draft].  Preparing for Submission.

Abstract: The high prevalence of digital financial fraud stresses businesses’ ability to distinguish between real communications from digital financial service (DFS) providers and fraudulent impersonations. Besides the financial and psychological costs to businesses, fraud may erode trust in, and usage of DFS. We test two strategies for preventing non-institutional fraud: a series of anti-fraud learning interventions and a technical solution to authenticate inbound communications from a digital platform. Using a pre-registered behavioural laboratory experiment in Nigeria, we find evidence that timely educational interventions increased trust in DFS, its likely future usage, and improved knowledge about fraud four weeks post intervention. However, when we task micro business owners with evaluating the authenticity of a series of fictionalised scenarios, we do not find evidence of improvement in fraud detection, either overall, or when considering only genuine or fraudulent scenarios. Surprisingly, we find increased self-confidence in fraud detection ability, highlighting the risk of overconfidence.
Project Page, AEA Registry
Abstract: Which connections do people value most in times of crisis? We examine a period of social unrest in Haiti to understand how social networks respond to social unrest. We construct communication networks using mobile phone metadata from a major mobile network operator in Haiti and a detailed geo-referenced timeline of severe unrest. These episodes are geographically isolated, persist for a matter of days, and vary in their degree of coordination and predictability. We use the more spontaneous of these events to estimate how calling behavior responds day-to-day. Estimating treatment effects using a difference-in-differences estimator robust to variation in treatment timing with heterogeneous treatment effects, we find that daily contacts decrease, but total talk time remains constant. These results are consistent with predictions from a theoretical model of social network response to social unrest. However, the number of daily contacts with network neighbors who are strong ties and have low centrality does not fall during the period of social unrest, in contrast to other subgroups. This finding suggests a pattern of checking in on close friends, family or other associates in lieu of information search.
First draft: September 24th, 2021
MeasureDev 2022 Presentation
Abstract: I analyze a novel dataset of digital credit transactions in the Kenyan market collected as part of the Digital Credit Market Inquiry undertaken by the Competition Authority of Kenya. This information request yields detailed transaction data on over five million consumers, allowing exploration of consumer outcomes including the size, nature, and evolution of the market; the price, size, and tenure of loans; the timing and type of fees applied; and late repayment and default. Moreover, a unique de-identification approach allows me to match consumers across providers without observing direct identifiers, allowing for the study of multiple borrowing behavior among these consumers. Additionally, I disaggregate these outcomes by gender, age, and provider type. In reflection, I identify five stylized facts about digital credit in Kenya, along with six working hypotheses, and set the stage for future use of administrative data-based market monitoring tools.
First draft: March, 30th 2022 
Abstract: What is the optimal rate of passing in professional football? We derive a simple model of play efficiency where efficiency is highest when the play comes as a surprise and falls when a more expected play is called. That is, playcalling efficiency is determined by a usage curve. We propose using fumbles lost as an instrumental variable to estimate the effect of passing, expected passing, and their interaction on the per play efficiency on early downs. More specifically, we argue that conditional on total cumulative fumbles by each team, cumulative fumbles lost serves as a valid instrument in each of these cases. Using fifteen years of play-by-play data, we use this strategy to estimate linear usage curves and are able to recover the optimal passing rate on early downs. On first and second down, coaches have their quarterbacks drop back to pass on about 52% of the time, whereas the optimal rate is around 71%. These results suggest that coaches deviate considerably from the optimal pass rate, overestimating the element of surprise.
First draft: November 26th, 2022

Selected In Progress

"Robust Inference for Audit Studies" Putman, D.S.; In Progress

"Many Economists" Huntington-Klein, N., C. Pörtner, and many other co-authors; In Progress.

"Poverty, Social Capital, and the Formation of Subjective Expectations" Putman, D.S., C. Bicchieri, K. Vallier, M. Movahed, S. Lahiri, A. Shpenev; In Progress.

Policy Reports

"Translating Digital Credit Transaction Data into Consumer Protection Supervision." Putman, D.S.; Innovations for Poverty Action (2022)

"Report on the Competition Authority of Kenya Digital Credit Market Inquiry." Putman, D.S., R.K. Mazer, W. Blackmon; Innovations for Poverty Action (2021).  

Abstract: Management of marine recreational fishing for Pacific halibut (Hippoglossus stenolepis) off Alaska has changed considerably in recent years due to concerns over stock declines and allocation issues. Since 2007, increasingly restrictive limits have been placed on Pacific halibut fishing of charter boat anglers, and a limited entry program was established in 2011 to curb the growth of the charter sector. In 2014, the Alaska Halibut Catch Sharing Plan (CSP) was implemented. It formalized the process for both (a) determining allocation of halibut between the commercial and recreational charter sectors and (b) initiating changes to harvest restrictions on charter fishing. One provision in the CSP allows Alaska saltwater sport fishing charter businesses that hold charter halibut permits (CHP) to lease pounds of commercial individual fishing quota (IFQ), which get converted into guided angler fish (GAF). These GAF can be used by charter businesses to offer their clients harvesting opportunities that are less restrictive in terms of the number and size of fish they catch and keep on a charter fishing trip.
This report describes and summarizes the results from a survey of CHP holders (charter businesses) conducted during 2015 that collected information on CHP holders’ attitudes and preferences toward Pacific halibut management in Alaska and preferences and behavior related to the GAF lease market, including values they place on GAF/leased IFQ under different sets of user or transactional restrictions. The mail survey was administered during 2015 to all CHP holders (565 charter businesses) and involved multiple mailings and a telephone contact. The survey response rate was 48% (271 completed surveys).
The survey results suggest that CHP holders generally had a negative view of the CSP and the GAF leasing program, with the majority believing that the GAF leasing program negatively impacts their business. Only a small percentage of respondents had participated in the program during 2014. Among those who had not leased GAF, the costs to lease GAF and generally opposing the GAF leasing program were cited by the most CHP holders as the primary reason for not participating in the program. About 84% of respondents did not plan to lease GAF in 2015 either. The majority of respondents also felt that relaxing restrictions on how GAF could be used (lease terms and transferability) were not likely to be helpful to their business. Respondents were also asked about their knowledge of, and attitudes toward, the Catch Accountability Through Compensated Halibut (CATCH) Proposal, which aims to create a recreational quota entity that can buy and sell commercial halibut IFQ. About 32% were not at all familiar with the CATCH Proposal, and over three-quarters of respondents indicated that they were not supportive of funding the proposal through a fee based on the number of endorsements held by CHP holders or a charter halibut tax per fish based on charter logbook records. Instead, the favored funding mechanism in terms of support was a charter halibut stamp, which would be purchased directly by charter anglers (70% were at least a little supportive). Respondents were split on whose responsibility (angler clients, charter businesses, or both) it was to fund the CATCH proposal, but the majority indicated that they did not feel the cost should be borne solely by charter businesses (about 68%).
There were several differences between responses from CHP holders in International Pacific Halibut Commission (IPHC) regulatory Areas 2C (Southeast Alaska) and 3A (Southcentral Alaska). Specifically, Area 3A respondents viewed the CSP, the GAF leasing program, and how the current CSP would affect their businesses more negatively than those in Area 2C. They also differed in terms of their support for the CATCH Proposal, with Area 3A respondents being less supportive on average than Area 2C respondents. Area 2C and 3A respondents also seemed to feel differently about how supportive they would be of alternative programs, such as a GAF ownership program (that would allow individual charter businesses to buy and sell commercial fishing quota as GAF) and GAF leasing programs that were more flexible than the current program. In general, Area 2C respondents were a little more supportive than Area 3A respondents of these alternative programs. However, Area 2C and 3A respondents were similar in their statements about whose responsibility they felt it was to pay for the CATCH Proposal (in terms of charter anglers, charter businesses, or both) and their beliefs about how effective it would be if implemented.

Blog Posts

"New Consumer Protection and Competition Policies Are Needed for Kenya’s Digital Credit Market." Blackmon, W., R. Mazer, D. Putman, and N. Mwarania.; Innovations for Poverty Action Blog (2021).