As part of our interview series, we ask renowned experts in the field about the future of research in development economics, and for their advice for young researchers. For this interview, we got the opportunity to talk to Prof. Karthik Sastry.
Karthik Sastry is an Associate Professor of Economics and Public Affairs at Princeton University. His research focuses on the role of information, bounded rationality, and social contagion in the macroeconomy, and on the causes and consequences of technological change.
ETRM: Your work moves between how people’s beliefs and attention shape the economy and how agriculture and food systems cope with a warming climate. These seem like very different worlds. How did you end up working on both, and do you see them as part of one bigger question?
Karthik: My story reflects something that happens often in academia: the starting points can be closer than the ending points. Once you’re on a research trajectory and new questions open up, there’s a temptation, and a genuine desire, to dig further and understand the issues better.
When I was studying for my PhD, my general interest was macroeconomics. One topic within macroeconomics that I found especially fascinating was the role of beliefs: how people form expectations about what’s going to happen in the economy, how they are moved by rational and irrational forces, how to formalize that idea, and what the implications are for macro policy. That was the topic my main advisor was studying, and there were open research questions I was working on.
A second interest of mine was the macroeconomics of growth and technology development. At a theoretical level, these are problems that share some similarities. A lot of that work ended up being with one of my classmates, and still one of my best friends and collaborators, Jacob Moscona. When we started our first paper, I thought of it as the economics of technology and climate change, with an application to agriculture and food systems. But once we started thinking deeply and seriously about that application, other questions came up that took us further along the track of studying food systems per se. The paper I presented last week is pretty far down that track, but there’s also a common thread connecting it to the work we did before.
ETRM: You have used remote sensing data in several of your papers. For students who are just beginning, these datasets might look ready to use: you just download them and run a regression. What should students consider before working with this type of data? And do you have examples of how misunderstanding these data could bias results?
Karthik: A lot of my work has started with unfamiliar data, and at the beginning, remote sensing data was very much in that category. One sincere recommendation I’d make is to reach out to librarians. They were always the first people I contacted, and talking to them was consistently the best way to learn how to use certain data, and how not to misuse them.
On the nitty-gritty, GIS or geographic data is generally where you’ll find remote sensing data, and a few things are worth keeping in mind. First, unless you spend a lot of time looking at the data, it is hard to know which data are more or less smoothed or imputed. You can look online for a dataset on tree cover or temperature, to use examples from my talk last week, and think, “That’s the data on where all the trees are.” But it usually takes more reading to understand how the data were put together. In GIS, as in national accounts if you think about it, there is often a lot of smoothing across space and time, and that interacts a lot with the kinds of analysis we tend to do in economics.
Second, and this is less intellectual but leads to the most tragic mess-ups, are map projections. You think you have two maps that you can just stack on top of each other, but if they’re coded in slightly different ways, they won’t align the way you expect.
Third, the algorithms you use for many standard calculations matter a lot: for example, how you get the intersection of two things, or how you measure the relative prominence of two densities. The software you use matters, but helpfully, this is usually a solved problem. Coding things from scratch is generally not nearly as smart as using the standard tools that ship with GIS software. These are little things you figure out along the way, and they’re pretty small hurdles to clear.
A new wrinkle, of course, is that if you started these projects now, generative AI might be quite helpful. But in my experience working with AI and data, AI is good at doing what you tell it to do; it’s not super clever at finding the smartest way to do something. Working with remote sensing and GIS data is full of situations where you could do some crazy calculation on a dataset, but there already exists a dataset in which it’s been done. Talking to people with experience will get you to that answer much more quickly.
ETRM: In your paper on Embrapa, you show that investing in local R&D can substantially improve agricultural productivity; you estimate that Embrapa raised Brazil’s agricultural productivity by 110%. Why do you think we don’t yet see similar initiatives in other developing countries? And what constraints do you think are limiting local R&D?
Karthik: That’s a research agenda I’m passionate about continuing precisely for this reason: it feels like there are more mysteries to unravel about how you could or couldn’t replicate that success in other contexts, and what other constraints exist.
If I had to think of the best parallel to Embrapa, reaching in a slightly different direction, I might say the expansion of the USDA’s agricultural research centers and the land-grant universities in the United States. You can see aspects of that if you drive out to the more land-grant parts of Cornell. In several ways, the United States made a coordinated, large-scale effort both to do agricultural R&D and to decentralize it. Like Brazil, it was in a situation where there was potentially enough human capital to go around, but many unique and under-researched agricultural and ecological problems. To put it briefly, the U.S. grew, and still grows, a mix of effectively indigenous crops like corn and crops effectively imported from Europe like wheat, the latter in very unfamiliar conditions. So the history of U.S. crop breeding involved an enormous amount of experimentation and solving novel problems. There is a lot of fantastic research on it, but it comes from such a different time and place that, in some ways, it isn’t ideal for policy advice.
As for what other countries are doing right now, I wonder and worry, and I’m painting in broad strokes here, that a wrong turn in many countries’ agricultural investment has been to invest too much in extension services, broadly defined, relative to genuine R&D. By extension, I mean anything based on training farmers, providing information, and trying to intervene directly in the production process. I think most countries are making a good-faith effort to do both at the same time, but I think that is a potential misstep.
There are also aspects of Embrapa that I’m not sure have been replicated. First, as I just implied, Embrapa was R&D-centric from the beginning: R&D plus information provision, and quite a bit less extension by that definition. Second, Embrapa invested heavily in human capital alongside its research. Concretely, Embrapa, together with CNPq and CAPES, Brazil’s other scientific organizations, paid for many people to get PhDs in the U.S. and come back to government service. We don’t do an independent calculation of the returns to that investment, although much of its cost is in the denominator of the benefit-cost ratios we compute, but I suspect the returns are quite high. It would be independently interesting to study how effective programs like that are, in agricultural science and other areas, at paying for foreign education and bringing people back into public service. Third is scale. The investment in Embrapa is low relative to its benefits, but it is quite enormous. In the paper, we give some metrics, such as the number of Embrapa researchers or its budget as a share of agricultural GDP. The analogs are mainly the modern USDA, and in those normalized units, Embrapa’s investment is enormously more than even that of large developing countries like India and China.
ETRM: For students who want to work across fields, say macro and environment, or theory and empirics, how do you build credibility in each without being seen as a generalist?
Karthik: Part of the answer, especially as a student, is to really spend a lot of time learning about the field. My background is in macroeconomics. In my case, one answer is having amazing collaborators, but that’s not the entire answer, because to build research and do it correctly, everyone needs to understand every part.
I was interested in the macroeconomics of technological change. I had never run a regression with crop yields on the left-hand side, I didn’t know what a growing degree day was, none of that. So, I forced myself to read all of those papers, not just to say, “Here’s this method, so I can apply it and run the standard regression to test my macro model,” but to really go through them and spend time thinking about them. Go to the seminars and the conferences and build your body of knowledge. It takes time, but there’s no substitute for it.
There’s also a more on-the-ground challenge as a student: you have an advisor who works on macro, half of your project is about something else, and you have to convince them that you know what you’re talking about. Putting on my advisor cap, it is hard to teach people new things, but it’s easier than you think to tell when someone has spent a lot of time thinking and reading about a topic and has built a large body of knowledge. That shines through even if you’re not doing anything conscious to bring it to the forefront.
ETRM: You have a prolific research career. What advice would you give to students and early researchers about staying focused and productive while working on multiple research projects?
Karthik: In my experience, both as a student and now, I have enjoyed working on multiple things at once. It keeps the mind moving, and you learn a lot of new things very quickly. But to really get things done, it often requires focusing on individual projects in more concerted chunks. Depending on the project, that could be a big-push week, a big-push month, or a big-push year, but at some point, it becomes necessary. Where one starts to lose the plot is, “We have our weekly check-in,” or “Every Tuesday, I have a standing 30-minute calendar block to think a little about that project.” That’s not a bad way to broaden one’s horizons, but I have found that the work never gets done that way.
I would also advise graduate students not to put too many steps between themselves and their first results. If you have a cool idea and you know the regression you want to run, just run it. That may sound contradictory to what I said earlier about building knowledge, but you want to get quickly to the point where you can determine what to do next. If you have a good research question and your first empirical method doesn’t work, you can try the next one. It’s important to identify the choke points and fix those first. That can get you pretty far in life.
ETRM: Thanks for your time!
