Chapter 11. A curated reading list for a junior blended-finance and rural-development researcher

You will not read all of these papers properly. Nobody does. A reading list is not a thing to finish; it is a thing to know. Where to look when a referee says “cite X.” A working stock of identification strategies you can call up without searching.

This chapter has three pieces: a protocol for triaging any paper in eight minutes, fifty papers worth knowing if you are going into blended finance, rural credit, or place-based development, and a six-month schedule plus the books and weekly feeds that sit behind it.

1. How to triage a paper in 8 minutes

Set a timer. Be honest about whether the paper deserves more time.

Minute 1: title and abstract. Decide: skim (adjacent), full read (critical path), or skip (famous but irrelevant this month). Most papers should be skipped on first triage.

Minute 2: first and last paragraph of the introduction. First paragraph gives you the question. Last paragraph of the intro gives you the headline result and usually flags the identification strategy. If you cannot write down the question and the headline after 2 minutes, that is information too.

Minute 3: the headline figure or Table 1. A serious empirical paper has one figure or table where the result is visible to the eye. If you need to trust the regression to see the effect, your prior should adjust downward.

Minute 4: identification. What is the design? Write it down in one phrase. Then write the identifying assumption. For an RDD it is local continuity at the cutoff. For DiD it is parallel trends. For an IV it is exclusion plus relevance plus monotonicity. For an RCT it is random assignment plus no spillovers (or modelled spillovers). If you cannot articulate it in one sentence, either the paper has hidden it or you have not read carefully. Both are signals.

Minute 5: robustness and falsification. Skip to the robustness section. What is the placebo? What is the most threatening alternative the authors address? Direct engagement with the obvious threat (a placebo, a falsification, a sensitivity bound) is a good sign. A footnote wave is also a signal.

Minutes 6 and 7: conclusion plus references. The conclusion tells you whether the authors believe their own result or whether they are hedging. Then scan the references. Names you do not recognise go in a follow-up queue. Names you expected to see and do not are worth noting.

Minute 8: decide and log. Read end-to-end or move on. Either way, write 2 lines in your notes file before closing the tab:

- Author-Year, Title
- Question: [one sentence]
- Design: [RDD / DiD / IV / RCT / structural / other], identifying assumption
- Headline: [the number, with sign and rough magnitude]
- Concern: [one sentence on the threat you are least convinced about]

Do this for every paper and by the end of year one your notes file has around 300 entries. You become the person in the room who can say “this is basically Karlan-Zinman 2010 with worse instruments,” which is the kind of remark that gets attributed and remembered.

The protocol scales down to four minutes when you are speed-reading a literature for a draft, and up to thirty minutes when the paper is on your critical path. The numbers are not sacred. What matters is making a deliberate decision about which papers earn which budget.

2. How to keep a reading library

Pick one tool and use it consistently. The tool matters less than the discipline.

Zotero is what I recommend. Free, open source, syncs across machines, integrates with browsers and with LaTeX through Better BibTeX. Mendeley works but Elsevier owns it. A plain .bib file is portable and version-controllable, with the downside that PDF organisation is manual.

Whichever you use, tag every paper on three dimensions:

Method: RDD / DiD / RCT / IV / synth / structural / descriptive / theory
Topic: credit / agri / education / health / labour / cash-transfers / spatial / climate
Status: unread / skimmed / read / cited

Re-read the “skimmed” backlog every 6 months. Half of what you skimmed in March will look different in September after you have done more of your own work.

3. The 50 papers

A. The methodological canon (10 papers)

These are the papers you should be able to cite in your sleep. The spine of modern applied econometrics.

1. Imbens and Lemieux (2008), “Regression discontinuity designs: A guide to practice,” J. Econometrics. The RDD reference handbook (bandwidth, kernels, validity tests). Method: RDD methodology. Why: anything you write using an RDD will cite this.

2. Calonico, Cattaneo, and Titiunik (2014), “Robust nonparametric confidence intervals for regression-discontinuity designs,” Econometrica. Fixed RDD inference; rdrobust is its software twin. Method: RDD inference, bias-corrected CIs. Why: referees will ask why you did not use rdrobust.

3. Goodman-Bacon (2021), “Difference-in-differences with variation in treatment timing,” J. Econometrics. Showed that staggered two-way FE DiD is a weighted average that can put negative weights on real treatment effects. Killed a generation of bad DiDs. Method: DiD decomposition. Why: every staggered DiD you read post-2021 owes this a footnote.

4. Callaway and Sant’Anna (2021), “Difference-in-differences with multiple time periods,” J. Econometrics. A standard estimator for the staggered DiD problem. Method: DiD, group-time ATTs. Why: pair with Sun-Abraham; you will use one of the two.

5. Sun and Abraham (2021), “Estimating dynamic treatment effects in event studies,” J. Econometrics. The other staggered DiD fix; interaction-weighted event studies. Method: event study, IW estimator. Why: this is the machinery behind a defensible event-study plot.

6. Abadie, Diamond, and Hainmueller (2010), “Synthetic control methods for comparative case studies,” JASA. The California tobacco paper; founded modern synth control. Method: synthetic control. Why: one treated unit and many donors is the blended-finance pilot setting.

7. Chernozhukov et al. (2018), “Double/debiased machine learning for treatment effects and structural parameters,” Econometrics Journal. The bridge between ML and causal inference. Method: DML, orthogonalisation. Why: the standard for high-dimensional control problems.

8. Athey and Imbens (2017), “The state of applied econometrics: causality and policy evaluation,” JEP. A tour of the modern frontier. Method: survey. Why: use it to find papers worth reading carefully.

9. Bertrand, Duflo, and Mullainathan (2004), “How much should we trust differences-in-differences estimates?” QJE. Serially correlated DiD standard errors are routinely too small. Started the wild cluster bootstrap industry. Method: DiD inference. Why: if you cluster at the wrong level, this paper is why your referee will know.

10. Cunningham (2021), Causal Inference: The Mixtape. Book, but the canonical entry text. Method: textbook. Why: if you buy one causal-inference book this year, this or Angrist-Pischke.

B. The development-economics canon (10 papers)

11. Banerjee and Duflo (2009), “The Experimental Approach to Development Economics,” Annual Review of Economics. Their case for RCTs in development, written pre-Nobel. Method: methodological essay. Why: this is the worldview that won. Know the arguments.

12. Duflo (2001), “Schooling and labor market consequences of school construction in Indonesia,” AER. The classic DiD-with-staggered-rollout paper; cohort variation in INPRES exposure. Method: DiD, staggered rollout. Why: shows how to exploit policy variation in space and time.

13. Miguel and Kremer (2004), “Worms,” Econometrica. Taught development economics about within-school spillovers. Method: cluster RCT with spillover estimation. Why: any village-level RCT needs a spillover answer; this is the template.

14. Burgess and Pande (2005), “Do rural banks matter? Evidence from the Indian social banking experiment,” AER. The 1:4 Indian branch licensing rule as an IV. Method: IV via regulatory cutoff. Why: foundational for “does rural credit reduce poverty.” You will cite this.

15. Townsend (1994), “Risk and insurance in village India,” Econometrica. Full-insurance test on ICRISAT data; mostly rejected, in interesting ways. Method: panel test of consumption smoothing. Why: the theoretical benchmark behind every rural risk-sharing paper.

16. Banerjee, Karlan, and Zinman (2015), “Six randomized evaluations of microcredit,” AEJ Applied. The “microcredit does not transform lives” consensus issue. Method: meta-discussion across six RCTs. Why: the empirical consensus you must argue with to claim microcredit works.

17. Karlan and Zinman (2010), “Expanding credit access,” RFS. South African consumer-credit RCT via lender collaboration. Method: lender-side randomisation. Why: best example of using a financial-institution partner for credible identification.

18. de Mel, McKenzie, and Woodruff (2008), “Returns to capital in microenterprises,” QJE. Random cash or in-kind grants; measured returns. Method: RCT, returns to capital. Why: foundational for “is there a credit constraint.”

19. Field, Pande, Papp, and Rigol (2013), “Does the classic microfinance model discourage entrepreneurship?” AER. Grace periods change borrower behaviour and raise risk-taking. Method: RCT on loan-contract terms. Why: opens the still-active contract-design literature.

20. Cole, Giné, Tobacman, Topalova, Townsend, and Vickery (2013), “Barriers to household risk management,” AEJ Applied. Why farmers do not buy rainfall insurance even when offered cheaply. Method: RCT plus structural choice model. Why: agri-insurance take-up is your problem too.

C. Rural finance and blended finance specifically (10 papers)

The operational core for your work.

21. Karlan, Osei, Osei-Akoto, and Udry (2014), “Agricultural decisions after relaxing credit and risk constraints,” QJE. Cross-randomised cash grants and rainfall insurance to Ghanaian farmers; the binding constraint was risk, not credit. Method: 2x2 factorial RCT. Why: reorients you from “credit is the constraint” to “uninsured risk is the constraint.” Central for blended-finance design.

22. Bryan, Chowdhury, and Mobarak (2014), “Underinvestment in a profitable technology: seasonal migration in Bangladesh,” Econometrica. A small migration incentive produced large persistent income effects. Method: RCT with long-run follow-up. Why: argues for tiny, well-targeted nudges over large programmes.

23. Suri and Jack (2016), “The long-run poverty and gender impacts of mobile money,” Science. M-Pesa rollout in Kenya, geographic variation, large poverty reduction. Method: DiD with geographic rollout. Why: the digital-finance evidence base for delivery-channel design.

24. Casaburi and Macchiavello (2019), “Demand and supply of infrequent payments as a commitment device,” AER. Kenyan dairy farmers preferred lumpy monthly over weekly payments at the same present value. Method: RCT plus structural model. Why: savings-as-commitment is central to rural smallholder product design.

25. Beaman, Karlan, Thuysbaert, and Udry (2023), “Selection into credit markets: agriculture in Mali,” Econometrica. Those who select into microcredit are those for whom credit has the smallest marginal effect. Method: RCT with universal offer plus take-up modelling. Why: changes how you think about the marginal borrower.

26. Carter, Cheng, and Sarris (2016), “Where and how index insurance can boost adoption of improved agricultural technologies,” JDE. Theory plus evidence on whether index insurance unlocks fertiliser adoption. Method: theory plus reduced-form. Why: connects insurance to investment, the blended-finance story.

27. Cole, Stein, and Tobacman (2014), “Dynamics of demand for index insurance,” AER P&P. Multi-year rainfall insurance take-up: learning, payouts, persistence. Method: RCT with multi-year follow-up. Why: short-run take-up estimates miss the learning dynamics.

28. Burke, Bergquist, and Miguel (2019), “Sell low and buy high: arbitrage and local price effects in Kenyan markets,” QJE. Loans to delay maize sales generated large returns and had local-price GE effects. Method: RCT with saturation design. Why: the most elegant rural-finance RCT of the last decade.

29. Pomeranz and Vila-Belda (2019), “Take-up and targeting: experimental evidence from SNAP,” AER. Information and assistance effects on programme take-up. Method: RCT on outreach. Why: the take-up problem in blended finance has the same structural form. Borrow the framework.

30. Helfrich (2026), “The rural mobilization gap in the US New Markets Tax Credit programme,” working paper. CDE fixed-effect decomposition shows the rural gap is supply-side, not demand-side. Method: panel with CDE FEs, decomposition. Why: the most direct US analogue to what you are building for Portugal.

D. Cash transfers, conditional and unconditional (5 papers)

31. Banerjee, Hanna, Kreindler, and Olken (2017), “Debunking the stereotype of the lazy welfare recipient,” World Bank Research Observer. Meta-evidence that cash transfers do not reduce work effort. Method: meta-analysis. Why: kills a standard objection to UCTs in policy debates.

32. Egger, Haushofer, Miguel, Niehaus, and Walker (2022), “General equilibrium effects of cash transfers: Kenya,” Econometrica. Multiplier effects, no inflation. Method: RCT with village saturation variation. Why: the first credible large-scale GE evidence on cash transfers; methodologically beautiful.

33. Haushofer and Shapiro (2016), “The short-term impact of unconditional cash transfers to the poor: Kenya,” QJE. The GiveDirectly RCT; big effects on assets, food security, well-being. Method: RCT with within-village randomisation. Why: foundational for UCTs and the rise of cash benchmarking.

34. Baird, McIntosh, and Özler (2011), “Cash or condition?” QJE. (Cited with 2013 follow-ups on schooling and fertility.) CCTs vs UCTs head-to-head in Malawi. Method: RCT with conditional and unconditional arms. Why: the conditionality debate starts here.

35. Akresh, de Walque, and Kazianga (2013), “Cash transfers and child schooling,” NBER WP. Burkina Faso CCT vs UCT on enrolment. Method: RCT. Why: pair with Baird et al.; together they bracket the conditionality debate.

E. Spatial, geography, and agricultural economics (5 papers)

36. Dell (2010), “The persistent effects of Peru’s mining mita,” Econometrica. RDD across the historical mita boundary; centuries-old forced-labour districts still poorer. Method: spatial RDD. Why: foundational for spatial RDDs and persistence research; the figure-1 map is iconic.

37. Nunn and Qian (2011), “The potato’s contribution to population and urbanization,” QJE. Potato suitability as a cross-country shifter. Method: DiD with crop suitability. Why: early canonical example of agronomic suitability as exogenous variation.

38. Henderson, Storeygard, and Weil (2012), “Measuring economic growth from outer space,” AER. Night-lights as activity proxy at fine resolution. Method: remote sensing plus growth proxy. Why: when you do not have admin data, you have lights. Know the limits.

39. Donaldson and Hornbeck (2016), “Railroads and American economic growth: a market access approach,” QJE. Market-access counterfactual via structural gravity. Method: structural gravity counterfactual. Why: the bridge between reduced-form spatial work and structural quantitative trade.

40. Burke, Hsiang, and Miguel (2015), “Global non-linear effect of temperature on economic production,” Nature. Temperature and growth, non-linear, large damages. Method: panel with non-linear climate response. Why: climate damages are now central to development. Start here.

F. The modern frontier (10 papers)

41. Athey and Wager (2021), “Policy learning with observational data,” Econometrica. Optimal policy from observational data via ML and orthogonalised scores. Method: doubly-robust policy learning. Why: the future of programme-targeting research.

42. Athey, Tibshirani, and Wager (2019), “Generalized random forests,” Annals of Statistics. The HTE-estimation backbone. Method: GRF, causal forests. Why: the modern default when you want HTEs with many covariates.

43. Carleton et al. (2022), “Valuing the global mortality consequences of climate change accounting for adaptation costs and benefits,” QJE. The big climate-mortality structural paper. Method: structural plus reduced-form panel. Why: state of the art for climate damage estimation.

44. Donaldson (2018), “Railroads of the Raj,” AER. Market access, trade costs, real income in colonial India. Method: structural quantitative trade with infrastructure shock. Why: the model that revived structural quantitative spatial economics.

45. Andrews, Stock, and Sun (2019), “Weak instruments in linear IV regression: theory and practice,” Annual Review of Economics. F-statistic thresholds, AR tests. Method: weak-IV inference. Why: if your first stage is weak, this paper tells you what to do.

46. Borusyak, Hull, and Jaravel (2022), “Quasi-experimental shift-share research designs,” ReStud. The Bartik framework, modernised. Method: shift-share IV. Why: Bartik instruments are everywhere in regional and labour economics; this paper tells you when they are valid.

47. Crépon, Devoto, Duflo, and Parienté (2015), “Estimating the impact of microcredit on those who take it up: Morocco,” AEJ Applied. Modest impacts on existing businesses; no expansion. Method: RCT with universal offer. Why: one of the six microcredit RCTs. Read for the design as much as the result.

48. Lagakos (2020), “Urban-rural gaps in the developing world: does internal migration offer opportunities?” JEP. The urban-rural wage gap and what to do about it. Method: survey plus structural framework. Why: the spatial-misallocation lens on rural development.

49. Bryan and Mobarak (2014) and the seasonal-migration follow-ups. The original Econometrica paper plus AER:Insights and JDE follow-ups on long-run effects and household composition. Method: RCT, long-run. Why: a body of work, not a single paper. Know the arc.

50. The most recent J-PAL working paper on the programme type you are studying. Not a fixed paper, a discipline. Skim the J-PAL feed weekly. Method: varies. Why: keeps you current; the frontier of your sub-literature moves quarterly.

4. Books for the bookshelf

Wooldridge (2010), Econometric Analysis of Cross Section and Panel Data, MIT Press. The reference. When in doubt, look it up here.
Hansen (2022), Econometrics, Princeton. Modern, free in PDF, complements Wooldridge.
Cameron and Trivedi (2005), Microeconometrics: Methods and Applications, Cambridge. Heavy reference rather than a read-through.
Angrist and Pischke (2009), Mostly Harmless Econometrics, Princeton. Read it once. Short, opinionated, shapes how a generation of applied economists thinks.
Cunningham (2021), Causal Inference: The Mixtape, Yale. The most accessible entry book.
Banerjee and Duflo (2011), Poor Economics; (2019), Good Economics for Hard Times. How to talk about your work to non-economists.
Glennerster and Takavarasha (2013), Running Randomized Evaluations, Princeton. Read before you design any field experiment.
Karlan and Appel (2011), More Than Good Intentions. Popular history of the RCT movement, useful for context.
Imbens and Rubin (2015), Causal Inference for Statistics, Social, and Biomedical Sciences, Cambridge. The potential-outcomes reference. Heavier than Cunningham.

If you only buy four: Wooldridge, Angrist-Pischke, Cunningham, Glennerster-Takavarasha.

5. Blogs, podcasts, and weekly digests

You need a steady incoming stream. Pick 4 or 5 and stick with them.

NBER Development Economics weekly working-paper digest. Free, email-based. The most efficient way to know what is being written this month.
J-PAL working papers RSS / email. The applied-RCT frontier in development.
World Bank Development Impact blog. Goldstein, McKenzie, and others. Practical, methodological, often funny.
VoxDev. Short author-written summaries of new development papers.
CGD blog. Policy-oriented. Useful for the institutional context behind your data.
Marc Bellemare’s blog. Trade and rural economics, plus tactical advice on grad school and publishing.
Bruegel. EU-focused. Useful when your work touches European policy or Portugal-specific institutions.
Twitter or Bluesky lists of active development researchers (Karlan, Duflo, Mobarak, McKenzie, Atkin, Bryan, and the people they retweet). Curate the list, do not let the algorithm pick for you.

A weekly habit: twenty minutes on Friday morning, go through the week’s NBER and J-PAL output, triage two to four papers using the 8-minute protocol. Over a year that is one to two hundred triaged papers. By year three you have a reading library deeper than most of your peer cohort.

6. A reading schedule for the first 6 months

Assume one paper a day on weekdays, plus a longer reading session on Saturday mornings.

Month 1: methodological canon (papers 1 to 10). Read end-to-end. Take notes. Run the example code if there is any. By the end you should be able to articulate the identifying assumption behind RDD, DiD, IV, and synth control in one sentence each.

Month 2: development-economics canon (papers 11 to 20). Read for argument structure, not just for results. How does Duflo set up the INPRES design? How does Miguel-Kremer treat spillovers? Good empirical papers spend a lot of time on the threats-to-identification section.

Month 3: rural finance (papers 21 to 30). This is your home field. Read these slowly. By the end you should know the six microcredit RCTs cold, the Karlan-Osei “risk not credit” result cold, and have an opinion on whether the rural mobilisation problem is supply, demand, or product design.

Month 4: cash transfers and spatial (papers 31 to 40). Cash transfers are the benchmark for any anti-poverty intervention. Spatial because most rural finance is also about geography.

Month 5: frontier (papers 41 to 50). ML in causal inference, modern IV, shift-share, structural-quantitative spatial economics. Do not feel bad if some are hard. They are hard.

Month 6: re-read and produce. Pick the 5 papers most relevant to your project. Re-read end-to-end. Redo the 8-minute triage notes from memory and compare with first-pass notes. Then start a working paper of your own. Even a 6-page draft. Writing forces you to read the literature differently and reveals which papers you actually understood.

The schedule fails if you treat it as a checklist. It works if you treat it as a frame. Some weeks you read three papers. Some weeks you read none because you are running regressions. After 6 months you have moved from “I have heard of those papers” to “I have read those papers and have opinions on them.”

A final word

Reading lists do not make researchers. Writing does. The goal of reading fifty papers in six months is not to know fifty papers. The goal is to have read enough that when you sit down to write your own paper, you can produce the first three paragraphs of the introduction without looking anything up. If you can do that on your topic, you are ready. If you cannot, go back to the list.