NBER WORKING PAPER SERIES

FROM MICRO TO MACRO DEVELOPMENT

Francisco J. Buera
Joseph P. Kaboski
Robert M. Townsend

Working Paper 28423

http://www.nber.org/papers/w28423

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
February 2021

We thank David Lagakos, Jeremy Majerovitz, Jacob Moscona, and Steven Durlauf and the anonymous
reviewers for helpful early feedback. The views expressed herein are those of the authors and do not
necessarily reflect the views of the National Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been peer-
reviewed or been subject to the review by the NBER Board of Directors that accompanies official
NBER publications.

© 2021 by Francisco J. Buera, Joseph P. Kaboski, and Robert M. Townsend. All rights reserved. Short
sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided
that full credit, including © notice, is given to the source.
From Micro to Macro Development
Francisco J. Buera, Joseph P. Kaboski, and Robert M. Townsend
NBER Working Paper No. 28423
February 2021
JEL No. O1,O11,O12,O2

ABSTRACT

Macroeconomic development remains an important policy goal because of its ability to lift entire
populations out of poverty. In our review of the literature, we emphasize that the best way to
achieve this objective is to embrace a synthesis of methods and ideas, with the science of
experiments as a unifying feature. RCTs need representative data and structural modeling, and
macro models need to be designed and disciplined to the realities and data of developing country
economies. Macroeconomic models have key lessons for gathering and analyzing micro evidence
and for moving to an evaluation of macro policy. Resource constraints, heterogeneity, general
equilibrium effects, obstacles to trade, dynamics, and returns to scale can all play key roles. A
synthesis for macro development is well under way.

Francisco J. Buera Robert M. Townsend

Department of Economics Department of Economics, E52-538
Washington University in St. Louis MIT
One Brookings Drive 77 Massachusetts Avenue
St. Louis, MO 63130 Cambridge, MA 02139
and NBER and NBER
[email protected] [email protected]

Joseph P. Kaboski
Department of Economics
University of Notre Dame
3039 Nanovic Hall
Notre Dame, IN 46556
and NBER
[email protected]
1 Introduction
A chief reason why many remain poor in the world is simply the macroecon-
omy into which they were born. Thus, the macroeconomic question of what
determines the aggregate level and distribution of income remains of utmost
importance. The focus of this review is to examine how macroeconomic
theory and micro empirical research can complement one another to increase
knowledge and to improve macro development policy. By macro development
policy, we mean simply any policy targeting economic development that is
large enough to have important aggregate or economy-wide distributional
and welfare implications. In principle, by this definition, macro development
policies include important large-scale policies in a wide range of areas: tax-
ation, transfers, and public spending; trade and industrial policy; roads and
other physical infrastructure; financial products and access; education; and
health.
Over the past 15 years, there has been a surge in empirical research
based on randomized controlled trials (RCTs) and related work evaluating
development policies at the small-scale level.1 This literature has redirected
research efforts toward micro-level program evaluations while often eschewing
formal theory and macro-level policy modeling. Nevertheless, and perhaps
paradoxically, evaluations of micro-level interventions are generally viewed
as pilot studies with the policy goal of scaling successes to large numbers
of people (Muralidharan and Niehaus, 2017; Banerjee et al., 2017b). Over
the same period of time, a macro economic literature has made advances
in building and solving models incorporating rich micro-structure, that is,
with well-defined agent problems, with heterogeneity, and with contracting
and market frictions. However this line of work has tended to rely on strong
structural assumptions, e.g., assumptions on functional forms and distribu-
tions of unobservables, and on somewhat stylized calibration strategies, and
thus economists often view it as disconnected from micro empirical research.
This article is about the interplay of economic theory and data as an
experimental science and how economists can use it to formulate macro de-
velopment policy. The implicit juxtaposition in the previous paragraph may
suggest little overlap between micro development economics featuring RCTs
and structural macro models. We emphasize the opposite: a much needed
1
One measure of its impact is the award of the 2019 Nobel Prize in Economic Sciences
to the leaders in the micro-experimental development field: Abhijit Banerjee, Esther Duflo
and Michael Kremer.

2
synthesis is already underway, from bottom up to top down. RCT practi-
tioners need and are using economic theory, and macro economists scrutinize
micro underpinnings and are employing experimental evidence.
Almost all development policies studied in experiments have aggregate
effects that go beyond their direct impacts on the individuals treated. To
give just one example here, education policies affect the distribution of skills
in the workforce which affect wages and incentives for investment and so on.
Indeed, synthesis of micro with macro can be exemplified by efforts to scale
up local experiments, that is, to assess what would be the impact if a policy
were implemented at the national level.
We care about the welfare consequences of these policies, i.e., not just
their treatment effects but how to interpret what these effects mean for hu-
man well-being. We therefore need to be able to combine the modern empir-
ical toolkit for reduced-form estimation of treatment effects with tools that
let us address both aggregate treatment and the distribution of welfare gains
and losses. Modern macro provides just such a toolkit.
There is historical precedent for synthesizing structural macro and exper-
imental methods, evidenced by the development of these methods. Modern
macro was forged conceptually as an economic science within the framework
of the Cowles Foundation: contributors were instrumental in merging ad-
vances in general equilibrium theory with those in econometrics, and they
confronted identification issues in observational data. The development of
the distinction between exogenous and endogenous variables, that is, exter-
nally versus internally-determined, variables and variation is an illustrative
example. Moreover, efforts went beyond patterns and correlations to models,
and then, as we have emphasized, to policy prescriptions. (Koopmans, 1947)
described the business cycle empirics of Burns and Mitchell (1946) as “Mea-
surement Without Theory”. Koopmans emphasized that theories need to be
micro-founded on economic behavior, especially if theories are to be used for
policy, which he advocated as the key end goal. The debate was substan-
tive and also impactful on institutional lines, with the Cowles Commission
modifying their slogan from “Science is Measurement” to “Measurement and
Theory” in 1952. These arguments were later echoed in Lucas (1976)’s cri-
tique of the policy reliability of large-scale macroeconomic models. Since that
time, the macroeconomic methodological debate has overwhelmingly settled
on the idea of micro-founded behavioral models.
Field experiments in micro economics also have a long, rich history. They
date back at least to Fisher’s work on crops, which were analyzed in the

3
1920s but based on experiments started in the 1840s (Fisher, 1921). Likewise
there is early quasi-experimental work evaluating theory in agriculture of
developing countries. For example, (Schultz, 1964) used the 1918-19 influenza
epidemic as a natural experiment to test the doctrine of labor of zero value
in agriculture.
The synthesis also aligns with an appreciation on both micro and macro
sides of the lack of an ideal laboratory setting for testing theory and policy
prescriptions in macroeconomics. For example, Angrist and Pischke (2010)
were pleased to discover that Lucas (1988a) uses experimental language in
trying to assess the impact of monetary policy; in lieu of experimenting on
an entire economy, he imagined engineering a laboratory depression to test
monetary theory by one day shocking the number of ride and game tickets
granted per dollar in Kennywood amusement park. However, Lucas concedes
that the informativeness of such an experiment would be nonetheless limited,
“[w]e are not really interested in understanding and preventing depressions
in hypothetical amusement parks. We are interested in our own, vastly more
complicated society.”
A bridge is needed to relate small-scale experimental work to the macroe-
conomy. We view the interplay between experimental empirics and macro
theory as an important part of that bridge.2
To demonstrate, consider a thought experiment closer to the field of de-
velopment than Kennywood Park: a laboratory of villages in Thailand with
the introduction of microcredit as an experiment.3 In our laboratory villages,
most people earn their living through a combination of subsistence, day labor,
and/or entrepreneurial activities. While they face a variety of risks to their
income, productivity, wealth, and consumption needs, financial instruments
like credit, savings, and insurance are limited. The most typical experiment
might be to measure a treatment effect that randomly gives some villagers
access to microcredit, while keeping others as controls. Some of treated vil-
lagers might borrow to invest, others to consume in various ways, others to
2
Our call for a bridge between experimental micro work and structural macro analysis
naturally relates to Heckman (2010)’s proposed third way to do policy analysis, combining
structural and program evaluation approaches.
3
Naturally this is an example dear to our hearts and own expertise, but we consider
a wide variety of policies in the paper’s body. Especially relevant here with its explicit
general equilibrium foundations, is work on villages as entire economies, e.g., Townsend
(1993, 1995), Udry (1995) and a large and growing literature which has followed. See
recent theory based work on village network in Chandrasekhar et al. (2018).

4
re-lend to family or friends. Would this tell us what would happen if all
villagers were given access? What if, instead, all villages were given access?
What if persistent institutions were set up, with explicit subsidies funded
through the national government budget? What might the long run and dis-
tributional impacts be on credit, income, savings, wages and prices, and how
might these differ from alternative policies? These are the types of ques-
tions we are interested in answering, but clearly answering such questions
will require more than empirical evidence, even though experiments can help
greatly in identifying key mechanisms and parameters. It will also require
models that are explicit about the basic decisions and frictions that villagers
face, the relevant resource constraints and dimensions of heterogeneity, the
level of integration of the various villages both amongst themselves and with
Bangkok and the rest of the world, the dynamic considerations of various
actors, and any potential sources of increasing or decreasing returns to scale.
Such a blend of models and data needs to appropriately balance parsi-
mony and realism in order to be informative. Such work is a difficult process,
requiring judicious decisions about theory and measurement. Although not
definitive, this is the type of work that we feel will best inform macro devel-
opment policy. We feature the details of various work in this essay. Several
feature teams of researchers leveraging skills by integrating field and meth-
ods: theoretical, empirical, and computational.
From our review, we draw several lessons for researchers involving funda-
mental aspects of modern macro theory. Aggregate resource constraints on
key inputs can make the aggregate impacts of scaled policies proportionately
smaller than implied by RCT evidence. Heterogeneity means that the largest
players have disproportionate impacts on aggregates, but these are often un-
derrepresented in micro studies. Heterogeneity is also particularly important
in economies with incomplete financial markets, prevalent market frictions,
strongly nonlinear Engel curves, strongly gendered behavioral differences,
and large variation in productive technologies, e.g., from traditional to mod-
ern. General equilibrium (GE) effects can redistribute, either reinforcing or
counteracting partial equilibrium distributional impacts. Furthermore, spa-
tial frictions and financial frictions, with heterogeneity in obstacles to trade
can make the transmission of GE effects vary from across the populations
and regions. Likewise, capital accumulation responds to prices. This can
make the long run impacts substantially larger than from short run impacts,
but there are key instances that go the other way. Finally, economies of scale
can make large-scale policy proportionally more powerful than micro inter-

5
ventions, and can be a motivation for aggregate policy interventions. We
stress that each lesson is a qualitative pattern with quantitative implications
that researchers must discipline with models and data.
We proceed as follows. Section 2 underscores the needs for methodological
integration and provides concrete examples of such integration. In Section
3, we address our principle lessons from macroeconomics for development
policy at scale and discuss ways in which a micro-macro synthesis can better
inform the key mechanisms.

2 The Micro-Macro Synthesis

We first discuss the relative strengths of micro experimental and macro struc-
tural work, underscoring the need for an integration of the two approaches.
We then follow up with some concrete examples of such integration to inform
macro development policy using various methodological approaches.

2.1 Micro: RCTs and Quasi-Experimental Techniques

Need Theory
The obvious strength of experiments, whether natural or randomized, and
instruments is that they can in principle produce credible exogeneity, ideally
free from selection bias and other types of endogeneity. The controlled vari-
ation that is created is like a medical researcher injecting dye into a patient
or an experimental physicist colliding particles, each to create new data in
a researcher-controlled environment to shed new light on a phenomenon of
interest. This explains the appeal of experiments for economics, by analogy.4
Valid natural experiments, as instruments, can produce an unbiased es-
timate of the local average treatment effect (LATE), the impact on those
people induced into program participation (Imbens and Angrist, 1994). Sim-
ilarly, micro randomized experiments can produce an internally valid sta-
tistical estimate of the average treatment effect (ATE) in the trial sample,
since randomization is an instrument that induces participation across the
full sample.
4
There have been an active recent debate on the merits and limitations of RCTs. We
refer the interest reader to recent reviews by Banerjee and Duflo (2009), Deaton (2009),
Heckman and Urzua (2010), and Imbens (2010).

6
 Treatment effects can be heterogeneous, however, as is often the case in
economic applications. When the trial sample is not representative of the
most relevant population, the ATE in an experimental setting could be dif-
ferent from the ATE for the population of interest. Moreover, in a heteroge-
neous population the ATE is only one of a host of other possible informative,
relevant and distinct impact moments, e.g., the fraction of people who benefit
or the median impact. These include the average effect of treatment on the
treated (TT) or, important for predicting the impacts of an expansion, the
average effect of treatment on the untreated (TUT). Heckman and Vytlacil
(2005) introduce the unifying parameter (for a particular treatment) called
the marginal treatment effect (MTE), the infinitesimal counterpart of the
LATE, and show that other desired impact estimates are weighted averages
of the MTE. Furthermore, when take up by the treatment group is limited,
the focus is often on the average intent to treat effect (ITT) as distinct from
treatment, actual voluntary participation in the policy or program.
The importance of heterogeneous impact effects is also a central motiva-
tion of the analysis of the conditions for an instrument to identify the LATE
(Imbens and Angrist, 1994; Imbens, 2010). In the case of an instrument for
treatment, the econometrician observes a variable that induces treatment for
some individuals but does not directly affect the outcomes. In addition the
instrument has to induce the treatment in a monotone way, i.e., it cannot
simultaneously induce an otherwise treated person to abstain from the treat-
ment. This assumption is needed precisely because of the heterogeneity, i.e.,
the possibility that treatment effects differ between those induced to treat-
ment by the instrument and those induced to drop the treatment by the
instrument.
To illustrate the importance of heterogeneous treatment effects, return
to our example of an economy populated by individuals that differ in their
entrepreneurial and worker abilities. Individual choices are potentially con-
strained by their financial resources, e.g., entrepreneurs’ investment is con-
strained by their beginning-of-period wealth. Individuals choose their occu-
pation and further, choose whether to pay a fixed cost to participate in a
financial market, i.e., to save or borrow at a fixed interest rate. Imagine a
policy-maker interested in measuring the benefits of fostering entrepreneur-
ship or the effect of a policy that improves financial access. Townsend and
Urzua (2009) use a version of this simple example as a laboratory to study
the answers provided by alternative experimental and quasi-experimental
methods.

7
 Townsend and Urzua consider two experiments in the model that pro-
vide potential instruments to identify the LATE of entrepreneurship and the
LATE of financial participation on income net of subsidies.5 The first experi-
ment consists of the random assignment of a lump-sum subsidy to individuals
that become entrepreneurs. Importantly, it is assumed that the subsidy can-
not be used to finance the capital of the business. Thus, the experiment
provides an instrument that has a monotone effect on the probability that
an individual becomes an entrepreneur, but it does not have a direct effect
on the outcome of interest, income net of the subsidy. The estimate of the
LATE of entrepreneurship on income in the model data is negative, reflecting
the negative selection of entrepreneurs induced by the subsidy; they would
have been wage earners, given their talent, if it were not for the subsidy. It is
straightforward to calculate other treatment effects using the structure of the
estimated model. Revealingly, in their application, the ATE of entrepreneur-
ship is positive and the TT is twice as big.
In the second experiment, individuals in the model are randomly assigned
a lower cost of accessing a financial market where they can save and borrow
at a fixed interest rate. This intervention provides an estimate of the ITT of
financial participation, as slightly less than 20% of the agents in the model
choose to participate when offered a lower cost of participation. Equivalently,
the intervention provides an instrument to estimate the LATE of financial
participation. The estimated LATE of participation in the financial market
is positive. In this case the ATE and TT of participation in the financial mar-
ket are positive as well, although their relative magnitudes are substantially
different. Interestingly, this second intervention does not provide a valid
instrument to estimate the LATE of entrepreneurship, as it violates both
the monotonicity and independence conditions. Monotonicity is violated be-
cause, while financial participation induces some productive individuals who
are now able to take up loans to become entrepreneurs, it also induces some
less talented entrepreneurs, who are now able to obtain a higher return on
their assets, to close their unproductive businesses. Independence is violated
because participation in the financial market affects the ability to finance the
capital and, therefore, it has a direct effect on the income of entrepreneurs.
5
In this application a pure experimental strategy can not be used to identify the ATE
of entrepreneurship, as it is not realistic to randomly assign the occupational choice to
an individual. Rather, an RCT can be designed to randomly assign monetary incen-
tives that affect occupational choices. Thus, the experiment provides an instrument for
entrepreneurship.

8
 This simple example is intended to highlight the importance of prior
theoretical knowledge in the design and interpretation of experimental and
quasi-experimental empirical strategies. As Imbens (2010) points out, a priori
theory is critical to determine the validity of instrumental variables. In the
context of his argument, an atheoretical view of micro work is not valid.

2.2 Macro: Structural Analyses Need Measured Micro

Foundations
Structural models utilize theory to link behavior with outcomes and welfare.
The advantage of theory is that it provides a coherent framework through
which to understand, interpret, and assess empirical results, incorporating
them in an organized fashion. Macro theory in particular is our focus, as
it is well suited for thinking about macro development policies, i.e., policies
that can have sizable impacts on the aggregate economy, whether in levels
or distribution.
The strength of macro models is helping researchers think clearly and
quantitatively about the roles of aggregate resource constraints, general equi-
librium effects, various potential obstacles to trade, forward-looking behavior
and dynamics, scale economies and their implications for aggregates, distribu-
tions, and welfare, all through the lens of micro-founded behavior. Macroe-
conomists have much to add to the micro empirical literature in thinking
through how a complex set of changes in an economy affect the economic
outcomes of any given individual and the economy as a whole.6
There is nevertheless a need for strong microfoundations, appropriately
tested and disciplined with data. Empirically, the work of Kydland and
Prescott (1982) introduced calibration strategies that allowed the researcher
to quantitatively discipline and evaluate models by comparing moments in
simulated data to those in actual data. However, in the context of represen-
tative agent models, this analysis was done at the aggregate level. The twin
criticism of this macro agenda was its harsh anti-econometric tone and the
insufficiency of frictionless representative agent models to capture important
economic realities, e.g., Summers (1986), Hansen and Heckman (1996).
The critiques are not new, but progress continues. Cambridge-Cambridge
debates concerned whether an aggregate capital stock and aggregate (or rep-
resentative) production function exist (e.g., Robinson, 1953, 1959; Solow,
6
We elaborate on these claims in Section 3.

9
1957; Fisher, 1969). Aggregation research has been extended even quite re-
cently (Baqaee and Farhi, 2019a). In the 1970s and 1980s, macroeconomics
relied almost exclusively on representative agents, as noted, but seminal work
by Bewley (1986), Aiyagari (1994), Hopenhayn (1992), Huggett (1993), and
Krusell and Smith (1998) has allowed macroeconomists to model heteroge-
neous consumers and firms in an aggregate, general equilibrium economy.
The point: these and related innovations allow macro models to consider
richer sets of frictions and to increasingly embrace micro data to discipline
them. Thus, while these critiques still apply to some work, virtually all these
critiques have been incorporated to some extent in other contemporary macro
work.
Closer to development, Banerjee and Duflo (2005) discussed the shortcom-
ing of representative agent macro models, and the need to consider important
empirical evidence on heterogeneity and frictions, including spatial frictions
and financial frictions, even if accounting for aggregates were the focus. This
critique proved influential in both the theory and empirical work on misal-
location (e.g., Restuccia and Rogerson, 2008; Hsieh and Klenow, 2009), one
literature that is now salient in both development and in macro.
All models are abstractions, but when the assumptions built into the
model veer too far from reality on important dimensions, the predictions of
macroeconomic models and recommendations for policy can be off the mark.
This is true both of the key assumptions but also the key parameter values,
whether calibrated, estimated, or assigned. The full set of microfoundations
and parameter values needs to be well justified, and this depends on the
policy to be analyzed.
An illustrative example is recent work on agricultural misallocation. A
series of macro development papers used detailed micro data and applied
the framework of Restuccia and Rogerson (2008) and Hsieh and Klenow
(2009) to agriculture, attributing large productivity losses to the misallo-
cation across farmers of varying productivity. These exercises in Malawi
(Santaeulalia-Llopis and Restuccia, 2017), Ethiopia (Chen et al., 2019), and
China (Adamapoulos et al., 2017) attributed much of the productivity gap
to losses from resource misallocation, especially misallocation connected with
market frictions on land and other inputs. The work assumed a uniform, cal-
ibrated Cobb-Douglas production function in agriculture. Follow up work,
however, by Gollin and Udry (2019) showed that similar levels of misalloca-
tion are observed in Tanzania and Uganda, but that this apparent misallo-
cation is also observed across plots owned by the same farmer, who does not

10
face market frictions in allocating those internal resources.7 Their analysis
is also model-based, but more detailed, and it attributes much of this seem-
ing misallocation to measurement error, heterogeneity in technologies, and
ex post shocks to output.8 Thus, a stark macro model with strong predic-
tions yielded a faulty conclusion from the data because it didn’t rest on the
appropriate micro foundations.9 .

2.3 The middle ground between theory and data: how

much structure is needed?
The more dynamics, frictions, heterogeneity and realism that macro models
incorporate, however, the more difficult and challenging they are to solve
and estimate. Basic intuitions and pen and paper solutions surrender to
computational results, which can themselves be black box predictions. This
is a valid criticism.10
There exists an important middle ground in terms of solving and stat-
ing models. Sometimes, depending on the policy and class of models under
consideration, a given model does not have to be fully solved, nor does a
particular model need to be chosen among a class. The field of macro trade,
for example, has built explicit models but also relied on exact hat algebra
results and sufficient statistics. Dekle et al. (2008) specify a trade model for
the world economy in terms of changes from the current equilibrium, avoiding
the need to assemble proxies for bilateral trade costs or inferring parameters
of technology; Arkolakis et al. (2012) show that the welfare implication of
a large class of trade model depend on only the trade share and the trade
elasticity, very much in the tradition of public finance (see Chetty, 2009, for
a review). Another example is aggregating firm-level distortions. Baqaee
and Farhi (2019b) provide approximate formulas for the aggregate impacts
7
Esfahani (2018) has related work on showing high factor ratio variation among same-
owned plots in Ethiopia and Malawi.
8
The macro development work on misallocation in manufacturing allows for heterogene-
ity in technology across industries, but measurement error also seems to play an important
role in those findings, as well, see Bils et al. (2018).
9
Related, a natural use of small experiments is to test and sometimes reject conven-
tional theories incorporated in macro models. Behavioral economics has both old and new
contributions on this front, including in development, e.g. Kahneman and Tversky (1979)
and Kaur et al. (2015)
10
Sometimes the predictions of models can be illuminated by comparative static simu-
lations, robustness exercises, and the like.

11
of distortions on aggregate TFP and output in general equilibrium. The
model allows for a fairly general input-output structure with a wide class of
distortions. Related, Sraer and Thesmar (2018) propose a method to use
experimental data to analyze the aggregate effects of policies. In particu-
lar, propose moments of the distribution of revenue-to-capital as sufficient
statistics for aggregate outcomes.
Both papers have limitations of course. The frictions in Baqaee and
Farhi (2019b) must be exogenous. Sraer and Thesmar (2018)’s frictions can
be endogenous, but they must be homogeneous of degree one with respect to
the wage and aggregate demand and firm-level technologies should be Cobb-
Douglas. More importantly, they require experimental data about the long
run effects of policies. Neither paper allows for firm entry.
An example spelling out an explicit non-competitive market structure is
Joaquim et al. (2019), which develop an industrial organization model of
imperfect competition in financial services with two components, a utility-
profit contract frontier coming from mechanism design and allowing a variety
of frictions versus the industrial organization part that delineates the num-
ber of bank branches and the extent of competition. It is shown that the
contracting frontier can be identified with market share data, hence policies
that impact competition do not need to identify the underlying frictions.
They have not explored the issue of dynamics - both in contracting and com-
petition of financial service providers, which may be relevant in particular
applications
The overall lesson is that sufficient statistics can be utilized and useful
in accounting for GE effects, but they are typically sufficient only for a well-
defined class of models. Some policy welfare questions can be answered
without much structure. The point also is to use theory to figure out when
and how this can be accomplished.

2.4 From bottom up to top down, the literature is al-

ready integrated
We want to emphasize that an integration is already occurring, and so we
offer some concrete representative examples. We organize these examples
by their different methodological approaches, ordering the approaches from
micro to macro but emphasizing the synthesis. For each approach, we offer
a few salient examples to convey the flavor of the approach, rather than a

12
comprehensive review of the literature.
The first methodological approach involves integrating structural mod-
els with experimental evidence. Staying at the micro level, this structural
synthesis nevertheless helps interpret and leverage experimental evidence.
The studies of the negative income tax experiments to estimate the effect of
taxes on the labor supply response of low-income workers in the US are early
examples (see the review in Hausman, 1985, Section 7).
An early development policy example of the synthesis of structural and
RCT work is Todd and Wolpin (2006), who use the randomized implemen-
tation of the Mexican program, Progresa, which offered conditional cash
transfers and nutritional supplements to families who kept their children
in school. Simulations of counterfactual policies using the estimated model
establish that the program could have had similar effects at lower cost by
eliminating subsidies. Attanasio et al. (2011) also examines Progresa with a
structural model, but they use the experimental variation to estimate, rather
than validate, their more general model.
Kaboski and Townsend (2011) evaluate a large microfinance intervention
in Thailand, anticipated in the earlier discussion of village experiments. The
village fund credit program placed a million Thai baht (roughly $25,000) in
each village. They estimate their structural model of household investment
and saving from pre-intervention data, and then use the Thai Million Baht
village fund policy experiment to validate the model, nearly replicating the
difference-in-differences estimates in Kaboski and Townsend (2012). Hetero-
geneity is key in distinguishing households, e.g., those on the margin of mak-
ing an indivisible investment (for whom consumption actually drops); those
who are credit-constrained hand to mouth (for whom consumption increases
one-for-one), and those neither constrained nor investing (who increase con-
sumption, no longer needing as much buffer stock saving with future credit
availability now assured.)11
A second emerging methodological approach involves using structural
macro models together with small-scale experimental and quasi-experimental
evidence to evaluate the impacts of policies at an economy-wide scale. A
broad range of different policies have been studied in this growing literature,
at the heart of our envisioned synthesis.
11
In the same spirit, Lagakos et al. (2018) evaluate transportation subsidies to seasonal
migration for rural workers using experimental evidence in Bryan et al. (2014). RCT
practitioners also bring structure into evaluations, typically focusing on estimation of
preferences (for example, Kremer et al., 2011; Kreindler, 2018; Mahajan et al., 2019).

13
 Buera et al. (2020) studies the macroeconomic impacts of universal access
to micro-credit. The model is calibrated with macro and firm distributional
non-experimental moments but is shown to match at the partial equilibrium
level the micro evidence of impact from RCT and natural experimental vari-
ation, in India (Banerjee et al., 2015) and Thailand (Kaboski and Townsend,
2012), respectively. Buera et al. (2014) use a similar model and methods to
examine the impact of cash grants to the poor. Both papers are clear that
the partial equilibrium versus general equilibrium implications prices, wages,
interest rates, TFP, saving and capital can be quite distinct, as will be the
distribution of winners and losers.
These modeling efforts have a close parallel in Bergquist et al. (2019) who,
as a team, are concerned with the impact of local experiments versus scaled-
up agricultural policies. The do this within the context of general equilibrium
model of farm production and, importantly, trade, featuring spatial trade
frictions in the markets for agricultural outputs and inputs. The model is
estimated with a multitude of micro data from Uganda. As in the first
example, they find that the direction and magnitude of the GE effect is a
non-linear and non-monotone function of the scale of the intervention.12
A third methodological approach is to empirically evaluate experiments
that are actually introduced at scale, whether RCTs or quasi, natural ex-
periments. This can provide valuable concrete evidence of impact that goes
beyond what might happen through the lens of a model, to what actually
happens, ideally using models to provide an interpretation.
For example, while Kaboski and Townsend (2012) analyzed the village
fund credit program as an intervention at the village level, it was introduced
in every village in the entire country. Ehrlich and Townsend (2019) establish
that wages increase more for highly treated isolated villages, as modeled, due
to production demand for inputs and limited labor inflow. Likewise, if the
collection of nearby villages was highly treated, the wage in a given target
village goes up, but this is due to higher out migration, as modeled.13
12
Other examples include: Donovan (2020) who examines the impact of introducing
farmers to a low-risk seed, validating the model with RCT findings on risk-reducing seeds
from Emerick et al. (2016); Fujimoto et al. (2019) who examine the effects of free secondary
schooling vouchers to talented, poor households matching aggregate moments and evidence
from the secondary school scholarships experiment by Duflo et al. (2017); Greenwood et
al. (2019) who examine the impact of circumcision on AIDS prevalence using experimental
evidence and a general equilibrium search and matching model.
13
Cai et al. (2016) is an example of a study of a large-scale village banking intervention

14
 A well-known national anti-poverty program is India’s National Rural
Employment Generation Scheme (NREGS), which guaranteed 100 days of
government funded employment every year to every adult willing to work.
Muralidharan et al. (2017) evaluated in this context a electronic payment
system introduced experimentally, to improve the program, at a scale of
19 million people. They find that 90% of the increase in income caused
by the program comes from non-program earnings, driven by increases in
both market wages and, somewhat surprising, private-sector employment.
The further exploration of the positive effects on private-sector employment
provide a good candidate for the use of an explicit theoretical framework to
interpret experimental results.
A fourth approach is to utilize the natural experiments that are often
gifts to researchers stemming directly, yet accidentally, from the details of
macro policy, which is often implemented from the top down. These natural
experiments can often allowing relatively transparent difference-in-differences
analysis. This evidence is then used in structural modeling.
A prominent example is the study of the effects of transfers using tax
rebates in the US with timing randomized, that is, based on the last few
digits of social security numbers which are thus essentially random: Johnson
et al. (2006) study the effect of income tax rebates of 2001, and Parker et
al. (2013) study the effect of the stimulus payments of 2008. Kaplan and
Violante (2014) use this evidence to motivate and calibrate a life cycle model
with costly access to illiquid assets, yielding rich hand-to-mouth agents14
Other recent natural experiments in developing economies shed light on
classical questions in monetary economics. Chodorow-Reich et al. (2019)
recently analyze the 2016 Indian “demonetization”, a dramatic initiative that
made 86% of cash in circulation illegal tender overnight. Detailed data from
the Reserve Bank of India allowed them to pin-point variation in the timing
of the release of replacement notes. In one sense, this is a version of Lucas’
imagined Kennywood experiment carried out in India in real time.15
in China. The migration outcomes are studied by (Cai, 2020).
14
As in Sarto (2018), Wolf (2019) and many others, a stand must be taken on a class of
models.
15
Related, Alvarez and Argente (2019) estimate the private benefits for Uber riders of
using alternative payment methods. They rely on a combination of large field experiments
and quasi-natural experiments given by the ban of cash as a payment method in certain
regions in Mexico.

15
3 Gains from Trade: Lessons for Modeling
and Empirical Work
We now review some take away lessons for those interested in thinking about
macro policy organized around five themes influencing modern macro theory:
i) aggregate resource constraints; ii) heterogeneity; iii) general equilibrium
effects; iv) obstacles to trade; v) dynamic optimization and capital accumu-
lation; and vi) diseconomies and economies of scale. For each theme, we
introduce a key lesson using a concrete example examining the scale up of a
microintervention policy that has been evaluated using an RCT. We then ex-
tend the discussion with additional examples of macropolicy and additional
opportunities for integrating microempirics with macrotheory.16

3.1 Aggregate Resource Constraints

A starting point for thinking about macroeconomic theory is the importance
of resource constraints. Individuals and firms may be able to flexibly ad-
just scarce resources at the prevailing market prices, but in the aggregate
these resources are typically constrained. Consequently, the aggregate con-
sequences of a scaled policy can be much smaller than what is measured in
a microintervention.
A concrete example of this is Buera et al. (2020)’s study of the aggregate
effects of microcredit. In their model, a small-scale microcredit interven-
tion leads credit to flow to some microentrepreneurs with lower marginal
productivity. The impacts on those individuals with access include higher
capital and higher income but those individuals are lower average productiv-
ity entrepreneurs. Without any external capital injection, however, aggregate
capital is fixed in the short run. That is, capital is a scarce, constrained re-
source, and demand for microcredit capital must compete with other uses in
the economy. As a consequence, the interest rate increases so that individual
capital choices are consistent with the aggregate resource constraint. This
16
The five themes in which we organize the lessons have somewhat artificial boundaries.
For instance, the discussion of the (general equilibrium) effects of policies on prices will
show up naturally when discussing the importance of resource constraints. Still, we choose
to delay a careful discussion of general equilibrium effects until after highlighting the
importance of heterogeneity, as this sets the stage to an analysis of how general equilibrium
effects are central to understand the heterogeneous impacts of policies.

16
leads to much lower immediate impacts of microcredit on capital and in-
come. Moreover, realized aggregate income comes from improved allocation
of capital toward higher productivity entrepreneurs.
In the previous example, the important resource constraint is on capi-
tal, but aggregate constraints can be important for many different resources.
When considering policy more broadly, a key question is which scarce re-
sources are most important for success. In scaling micro interventions, the
relevant scarce resources – e.g., physical capital, human capital, government
capacity, firms and other organizational capital – may be quite specific to
the intervention. Administrative capacity at the level of the organization or
government may be an important constraint in scaling, for example.17 One of
the primary arguments for randomized controlled trials has been that, given
scarcity in funding and administrative capacity, randomized assignment is
a fair allocation rule. In some of the most famous RCTs (e.g., Progresa),
randomized rollout was indeed justified as an equitable way to proceed given
limited budget and administrative resources (Parker and Teruel, 2005).
Since scarce resources can be specific to the macro development policy
in question, modeling relevant constraints requires case-by-case judgement
based on local knowledge of the environment and program being scaled, eco-
nomic theory, and prior empirical results as the following examples illustrate.
In Donovan (2020), the key aggregation and scarcity in the analysis involve
not only the labor, which can be assigned to produce agricultural and non-
agricultural goods, but the output of the non agricultural sector, which can
be used for final consumption or as an intermediate input into the production
of agricultural goods. With a focus on risk aversion, scarcity of agricultural
output plays a key role at the village level because of the incomplete mar-
kets and minimum consumption requirements. Bergquist et al. (2019) uses
a similar commodity space, but they add land as a scarce factor for agri-
cultural output. In the human capital model of Fujimoto et al. (2019), the
direct costs of education need to be funded. The key dimensions of scarcity
and aggregation involve the government budget constraint – where the real
direct costs of education need to be financed by taxes – and the exogenous
distribution of inherent ability. The labor market and governments are both
national in their model, and that is the key level of aggregation.
17
When the administrative capacity is the limiting factor, the equilibrium adjustment
after scaling micro interventions may involve the rationing of these resources across com-
peting programs instead of changes in market prices. Thus, aggregate resource constraints
need not work through GE channels.

17
 There are additional possibilities for integrating microempirics with macro
models. Micro can learn from macro. For example, even micro-level empir-
ics themselves can benefit from applying resource constraints. While budget
constraints are rarely satisfied at the individual level (because of measure-
ment error, for example), applying such constraints at a more aggregate level
may be more reasonable. For example, Kaboski et al. (2019) analyze a cash
grant experiment. They apply a budget constraint at the village level to es-
timations of impacts on income, consumption, savings, investment, lending,
etc. to yield cross-equation restrictions that increase efficiency in estimation.
They require that changes in investment, consumption, and savings equal
changes in income and transfers (including the cash grant). The restriction
cannot be rejected and yields more reasonable and easily interpretable esti-
mates. A common practice in reduced form empirical work is to log outcome
variables in order to reduce the influence of outliers. This practice can often
increase significance in the estimation, but also leads to less obvious, non-
linear cross-equation restrictions. Coefficients don’t “add up” in clear ways.
Bonferroni type adjustments have become common in the micro-empirical
literature when multiple outcomes are analyzed, but utilizing cross-equation
constraints of aggregation have not been typically applied.
Macro models can also be better informed and disciplined by micro stud-
ies. Work that estimates the geographical level of impacts of programs (on
individuals, local areas, and neighboring areas, for example) can give in-
sights into the most appropriate level of aggregation. Similarly, take up
rates from experiments can tell us which sub-populations of a distribution
are more likely to be directly impacted by an intervention. That is, experi-
mental results with extensive margin elasticities can be used as part of the
specifications of macro models.
Microexperiments can also discipline key elasticities that determine the
consequences of scarce resources. For example, Lagakos et al. (2018) use the
estimated micro impacts of the migration subsidy in Bangladesh on migration
and rural wages to estimate the labor demand elasticity in rural labor markets
in the context of their Roy model framework.

3.2 Heterogeneity
Many micro experimental studies of policy interventions find that not ev-
eryone is impacted the same by treatment. Heterogeneity matters in macro
models, both because of its distributional implications but also for aggregates

18
when decisions don’t scale linearly, e.g., a rich person with double the wealth
spends differently than the totals of two poor persons. Consequently, a key
lesson for scaling interventions is the need to ask how a policy impacts the
biggest players for aggregate effects, since they often have the largest weight
in aggregates. Unfortunately, these players often don’t show up in survey
data. The largest firms are unlikely to show up in firm survey samples, but
this is especially true when experimental work focuses on microenterprises.18
The same issue holds for wealth and the impacts of policies on aggregate
savings, or income and the impacts of tax policies on aggregate revenue.
An interesting example of this is the evaluation of village funds in Kaboski
and Townsend (2011). One might think that considering large players would
be less important in village economies when poverty is the main outcome
of interest rather than national aggregates, but big investors matter in their
study. Their structural model emphasizes the very heterogeneous responses
to credit that different agents can have: some increasing consumption, while
others decreasing consumption in order to increase investment. In their es-
timated model, there is wide variation in the the size of household’s desired
indivisible investment. Large investments have disproportionate weight on
aggregate investments. In theory, there is always a chance that the avail-
ability of even a small amount of credit enables a much larger indivisible
investment, and such cases, while rare, matter substantially for aggregate
investment consequences of credit. Using simulations, the paper shows that
small sample sizes are pooorly suited for capturing the impact on aggregate
investment given the importance of large, infrequent indivisible investments.
In theory, heterogeneity is particularly important in economies with in-
complete financial markets, prevalent market frictions, strongly nonlinear
Engel curves, strongly gendered behavioral differences, large variation in
productive technologies, e.g., from traditional to modern. Each of these are
especially applicable in developing countries, and indeed there are important
micro studies emphasizing dimensions of heterogeneity in these economies.
These dimensions include wealth, productivity, ability, entrepreneurial back-
ground, education, and gender. Moreover, when we are especially concerned
about inequality, those with less social standing in society, and the poorest
of the poor dimensions like gender, education, and wealth can be especially
18
The opposite problem occurs with some firm-level data from business registers in
developing countries, which tend to report data for firms above a minimum size threshold
(Bartelsman et al., 2004).

19
important from a distributional standpoint. Many microstudies find that
such heterogeneity is critical.19
Macro models can now accommodate heterogeneity much more so than in
the past, and good reduced-form empirical work (whether RCT or otherwise)
can help direct and discipline heterogeneity in macro models. This opens up
additional opportunities to integrate the two modes of research. We elaborate
on three such opportunities.
First, RCTs can be used to help identify the important dimensions of
heterogeneity that need to be modeled and to validate or reject particular
model or distributional assumptions. For example, de Mel et al. (2008) eval-
uated randomized grants to non-employer entrepreneurs in urban areas of
Sri Lanka. They found sizable impacts on investments and monthly profits
with implied monthly return on the grants that substantially exceeded mar-
ket interest rates. Yet the researchers emphasized the strong heterogeneity
in returns to capital. those with disproportionately low levels of wealth and
those with higher education or cognitive ability drove the overall impacts.
In line with this evidence, wealth and talent are two important dimensions
of heterogeneity that have dominated the theoretical and quantitative macro
literature on financially-constrained entrepreneurs, which have been used to
evaluate the impacts of the financial system more broadly.20
Second, combinations of experimental and observational data can be used
to better discipline macro models. There are limitations associated with ex-
clusive use of samples from micro RCTs, since evidence from RCTS often
comes from non-representative populations. Consider the study by Muralid-
haran et al. (2017), which examines an experimental improvement in the
payment system used for the India’s NREGS employment guarantee pro-
gram. On the one hand, the sample is impressively large and heterogeneous,
involving 19 million people in India across 157 subdistricts. Yet, this sample
may not be the most appropriate for disciplining a macro model for cer-
tain aggregate questions: by its very nature, the population is rural, so it
may not be appropriate for thinking about urban labor supply. However,
19
A comprehensive list would be impossible, but we note a couple examples. Banerjee
et al. (2017a) show that microfinance impacts in India are concentrated among existing
“gung-ho” entrepreneurs. Duflo (2003) finds pensions received by women in South Africa
had a large impact on the anthropometric status of girls but little effect on that of boys.
20
See for instance, Giné and Townsend (2004); Jeong and Townsend (2007); Townsend
and Ueda (2006); Buera (2009); Banerjee and Moll (2010); Townsend and Ueda (2011);
Buera et al. (2011); Midrigan and Xu (2014); Moll (2014).

20
non-representative data can be extrapolated to the relevant macro popu-
lation using structural assumptions and more representative observational
data. For example, to discipline labor supply from rural-urban migration,
Lagakos et al. (2018) combine RCT data with representative household data
in their evaluation of a scaled migration subsidy in Bangladesh. Parameters
governing aggregate moments are targeted using the representative data, and
yet the RCT data gives important evidence for disciplining the productivity
parameters and disutility cost governing Roy-like selection into migration.
Third, macro models suggest the importance of attempts to sample more
representatively. Samples of micro studies are often targeted toward specific
populations (e.g., the poorest or least served, marginal populations in terms
of take up), the populations that are easiest to survey (e.g., spatially clus-
tered, those with lower opportunity cost of time, those available at the time
of survey), or those that are easiest to incorporate into randomized evalua-
tion methods (e.g., targeted expansion areas). Though targeting for poverty
reduction makes sense as a laudable goal, this neglects the great force that
macro growth and improved distribution can bring. For that we need data to
estimate the underpinnings of macro development models. One way to bal-
ance these goals in collecting data is to combine surveying frequencies with
representative national samples in order to obtain effective sampling weights
that can be used to aggregate. That is, one can combine over-sampled pop-
ulations of particular interest with partial samples from the full population.

3.3 General Equilibrium Effects

In a market economy general equilibrium (GE) effects are the price counter-
part of aggregate constraints on quantities discussed in Section 3.1.21 Never-
theless, we view them as conceptually distinct, especially when heterogeneity
is important as discussed in Secton 3.2: GE effects on relative prices or wages
can impact not only aggregates but the distribution of income. A key lesson
here is that these GE effects can be important, either reinforcing or counter-
acting the partial equilibrium distributional impacts.22
21
Acemoglu (2010) also emphasized the importance of GE and resource constraints as
a reason that theory needs development.
22
Many emerging empirical studies highlight the distributional consequences of GE ef-
fects. Cunha et al. (2018) show drops in prices of goods transferred in kind, especially
in less developed villages. Similarly, Filmer et al. (2018) show that cash-transfers led
to an increase in the prices of protein-rich foods, which affected the health outcomes of

21
 An interesting example of this is the work of Donovan (2020), a macro
study that examines the role that risk and the subsistence consumption re-
quirements of poor farmers plays in the use of productivity-enhancing in-
termediates like high yield seeds and fertilizer. In the model, farmers with
expected consumption near the subsistence requirement are reticent of the
additional risk that intermediate expenditures entail. He uses the model to
investigate the impact of the introduction of lower risk seed, which was eval-
uated in an RCT by Emerick et al. (2016). When done at scale, the adoption
of this seed leads to higher output, and a GE decrease in the price of agri-
cultural goods. This reduces the cost of subsistence in the event of crop
failure, actually reinforcing the initial risk reduction from the seed, making
the induced take up of intermediates and welfare gains larger in GE than in
PE, especially for poor farmers.
In macro development models incorporating micro experiments, GE ef-
fects have been analyzed in many different settings and policy analyses.
The experiment in Cai and Szeidl (2018) randomized loan takeup to small-
medium enterprises (SMEs), both within (some firms receive the treatment)
and across local markets (a higher fraction of firms receive the treatment
in some markets) in China. The direct effect of credit access is to increase
sales, profits, and factor payments, but the indirect effect is to reduce the
sales, profits, and factor payments of one’s competitors; a fair amount of the
direct effect of treatment comes from a business stealing effect. They use a
Melitz-type model to show how this works through prices. Hence, the GE ef-
fects in this case lead to smaller welfare impacts than the direct effects might
indicate. In another recent paper, Fujimoto et al. (2019) examine the effects
of free secondary schooling vouchers to talented, poor households in Ghana.
They calibrate a model with human capital accumulation to match aggregate
moments and evidence from the secondary school scholarships experiment by
Duflo et al. (2017). They find important general equilibrium effects, which
are key drivers of the fiscal and distributional effects of the program. The skill
premium falls in consequence, and rich, skilled lose through lower high skill
wages, while the poor, unskilled gain even though they do not attend school,
as their productivity increases by working with more skilled, complementary
workers. Examining the aggregate impact of circumcision on AIDS preva-
lence in Malawi, Greenwood et al. (2019) apply experimental evidence and a
general equilibrium search and matching model, in which heterogeneous in-
non-beneficiary children.

22
dividuals selectively adopt different sexual practices while knowing inherent
risk. They find that there are important GE changes in prices in markets
for marriage, protected, and unprotected sex, although they have relatively
small effects on behavior, since prices affect men and women’s behavior in
opposite directions.
Again, there are additional lessons and opportunities for integration.
First, the experimental practitioner must always be concerned with GE ef-
fects; an ideal RCT requires an absence of any spillover of treatment onto
non-treated groups, e.g., individuals, villages, and regions within a country.
If treatment changes the prices that the nontreated face, that can impact the
outcomes of the non-treated households thereby biasing the estimates.
Second, experiments themselves can intentionally generate exogenous price
variation in either the prices that sellers face or the prices that buyers face.
These can be used to illuminate the relevant elasticities of supply and de-
mand that are key to macro models. Randomized price variation has been
used inside some of the most influential randomized experiments, e.g., Fehr
and Goette (2007), who estimate a labor supply elasticity by varying the
wages workers received.
Third, because highly concentrated interventions can generate endoge-
nous GE effects themselves, microevidence from these interventions can be
used to discipline or validate the parameters governing the strength of GE
forces in macro models. This can be done through estimation, calibration, or
additional validation checks. For example, Akram et al. (2017) use variation
in the intensity of their migration subsidy in Bangladeshi villages, i.e., the
fraction of villagers receiving the subsidy, to show that the outflow of labor
from local villages also increased wages. On the flip side, Breza and Kinnan
(2018) use the impact of coordinated default on microfinance loans in Andhra
Pradesh, India, the balance sheet impacts it had on microfinance banks,
and the variation in pre-existing presence across other Indian provinces to
show that a contraction in microcredit led to a drop in wages. Kaboski and
Townsend (2012) similarly estimate positive impacts on wages using exoge-
nous variation in treatment from the village fund. This evidence is used as an
out-of-sample test on the reasonableness of the labor-entrepreneurship forces
in Buera et al. (2020).
Finally, while localized GE mechanisms can be measured directly using
cross-sectional or spatial variation, any impact that is truly at the aggregate
level can only be quantified through the lens of a model (unless policies are
randomized at the country level, for many countries!), and indeed models

23
are needed to interpret any local GE effects that happen through market
segmentation. These analyses incorporate theory to varying degrees.
In some cases, assessing the welfare gains from price effects require only
reduced form elasticities, while in others more explicit models and simulations
are required. For example, the work of Arkolakis et al. (2012), shows that
a reduced form elasticity is sufficient for calculating gains from a reduction
in the price of imports in a wide class of tractable and relatively standard
models. The analysis in Donaldson (2018) is a nice application of this in
evaluating the benefits of infrastructure investment in India on the gains to
agricultural trade. This follows the tradition of the public finance literature
(Harberger, 1964; Chetty, 2009). For other questions, a full scale GE model
with microfoundations may be necessary. Van Leemput (2016) looks at trade
costs within India, but she does so within a two-sector model of agriculture
and non-agriculture where agriculture is a necessity good as in Fieler (2011);
the constant elasticity assumption is violated in the aggregate and moreover
the model has heterogeneity. Van Leemput’s welfare gains cannot, therefore,
be reduced to a single summary statistic, and indeed they require quantitative
simulation.
In sum, macro development policy needs to consider price effects because
there is strong theoretical and empirical reasons to believe that price effects
are important for aggregates, efficiency, distribution, and welfare. Moreover,
incorporating and quantifying GE effects into analysis requires both models
and data.

3.4 Obstacles to Trade

Spatial and other trade frictions are an important source of endogenous het-
erogeneity and can be intimately linked with GE mechanisms. The economic
environments, available markets, and prevailing prices of those in remote ar-
eas of developing countries can be strikingly different from those in capital
cities. We have emphasized the importance of careful modeling. A key les-
son for extrapolating RCT evidence to scaled up policies for an entire nation
is that the transmission of GE effects and the impacts of policy can vary
considerably region-to-region and even person-to-person.
A concrete example of this is the paper evaluating the aggregate and
distributional consequences of a scaled subsidy to agricultural intermediates
by Bergquist et al. (2019). They develop a (static) trade model in which trade
costs vary from household to household and map it to the Ugandan economy

24
using a clever combination of calibration and estimation.23 Since markets
are only partially integrated, wages vary from market to market and effective
prices vary from household to household. They contrast a local saturation
of the intermediates subsidy intervention with a nationwide saturation. One
headline finding is that the intermediate subsidy causes larger GE effects on
wages when nationwide than when just local, since labor is less elastic in
aggregate, a point we have made in the sections on resource constraints and
general equilibrium effects. More relevant to the point in this section, they
find that some areas benefit considerably more at scale, while others benefit
considerably less at scale depending on how integrated they are and the
extent to which they specialize in particular crops. These patterns depend
not only on factor endowments but the full set of trade costs, which they
estimate.
Other examples incorporate simpler geographies to look at spatial fric-
tions. Ehrlich and Townsend (2019) model not only trade but also migration
flows using a geography of villages along a circle, while Lagakos et al. (2018)
model migration using a simple two location urban-rural model. Naturally,
there is a tradeoff between simplicity and the level of detail of predictions.
Bergquist et al. (2019) can assess household level impacts, while Ehrlich and
Townsend (2019) can assess village level impacts. Lagakos et al. (2018) can
only show urban-rural distinctions, but this simplicity allows them to incor-
porate rich Roy-model heterogeneity and migration dynamics.
There are more opportunities for integration on these fronts, in part be-
cause of advances in modeling, improvements in data availability, and inno-
vations in combining the two.
In the past, modeling such regional heterogeneity in ways that could
be taken to data was not practical. However, in recent years, building on
trade models, there have been important advances in not only proving the
existence of spatial equilibria with mobility of goods and workers but also in
developing methods to solve these models.24 The models often use Frechet
distribution assumptions on productivity to derive gravity-model patterns of
trade, but they add in labor mobility. Researchers have used these models
most directly to think about the spatial distributional effects of international
23
Like Donovan (2020), they are concerned with the productivity benefits of agricultural
intermediates and they too model nonhomothetic preferences, but they abstract from
endogenous dynamics. The these two papers illustrate important trade-offs between richer
model elements and tractability.
24
Redding and Rossi-Hansberg (2017) provide an excellent review.

25
trade policy (e.g., Monte et al., 2018; Caliendo et al., 2018, 2019), but they
have also considered the benefits of infrastructure investment (e.g., Allen and
Arkolakis, 2014) and consequences of migration policy (e.g., Desmet et al.,
2018b). All of these are important issues for developing countries.
Many of these models allow for a great deal of detail, but, similar to the
sufficient statistics approach, do not require solving the full model in order
to generate the predictions of interest. Instead, practitioners quantify the
model using “exact hat algebra”. The moniker was introduced by Dekle et al.
(2008) to describe a development on the older “hat algebra” approximation
that dates back to Jones (1965)’s contribution to general equilibrium theory.
His “algebra” was the matrix algebra from the comparative statics exercise of
log-linearizing a model and totally differentiating it. The resulting equations
were functions of log changes in factor prices and endowments as well as
elasticities and shares, which could be taken from either outside estimates
(elasticities) or read from the data (shares). The log changes were eventually
denoted by “hats” (or carats). “Exact” hat formulas occur when modeling
choice allow such equations to no longer be approximations but exact, even
for large changes. In this case, without solving the model or even specifying
its deep parameters, counterfactual predictions can be calculated using the
hat deviations from a benchmark together with knowledge of the relevant
elasticities of substitution between goods and factors. Gravity models with
trade/migration/capital flow costs typically require additional information
on bilateral flows. In that sense, the elasticities, shares, and flows together
constitute a sufficient statistics approach.
These models are especially appropriate for developing country analysis,
since a robust and important feature of both the countries and their devel-
opment policy is that regional heterogeneity matters. Development involves
urban centers, peri-urban peripheries, and rural areas, and wide spatial vari-
ation. There are important questions involving both the international and
within country flow of labor, goods, firms, and capital. Mapping models
to developing countries requires nationally representative samples, however.
Such samples exist for the largest countries (e.g., China, India, and Brazil)
and are emerging for others, e.g., the World Bank’s Enterprise Surveys for
firms and Living Standard Measurement Surveys (LSMS) for households.
The panel length of these data sets are limited but also typically not re-
quired for these spatial models, since they are static. However, one does
need to augment these using knowledge of geographic location.
The methods of Bergquist et al. (2019) for Uganda are resourceful in lever-

26
aging limited data from disparate sources, and thus are particularly insight-
ful.25 They require key elasticities, shares, and trade costs. They combine
multiple data sources. They use the detailed data of the LSMS samples to
calibrate the relevant, region-specific, share parameters. They then combine
these rich data with less detailed Ugandan census data and geocoded data on
crop suitability from the Food and Agriculture Organization (FAO) in order
to project shares from the representative LSMS sample to the full country.
For key elasticities, they use RCT evidence from a nearby country to disci-
pline the relevant elasticity of substitution governing demand and separate
RCT evidence from another nearby country to discipline farmers’ elastic-
ity of substitution across traditional and modern (input-intensive) farming
technologies. They then use global crop price variation as an instrument
to estimate farmers’ elasticity of substitution across crops. Identification
of trade costs is likewise tricky since they lack the full set of household-
household trade flows. They combine survey data on market-level flows and
prices together with use household production and consumption decisions to
infer farm gate prices, with price gaps across farms thereby identifying trade
costs. Lacking data on distance at the household level, they project price
gaps (trade costs) on observed demographic characteristics.
Another important example is the work of Townsend and coauthors, who
have applied structural models of entrepreneurship decisions under financial
frictions and mapped them to rich observational data from multiple datasets
in Thailand. Financial frictions limit trade across time and states of the
world and can also have a spatial dimension. For instance, the work of Paul-
son and Townsend (2004) and Paulson et al. (2006) shows regional hetero-
geneity in the underlying causes of financing frictions: moral hazard-driven
frictions in the more developed Central region of Thailand and a limited
commitment collateral constraint in the more rural Northeast in combina-
tion with other features. Ahlin and Townsend (2007) derive the signs of
covariances among variables for a variety of models of limited liability lend-
ing, which allow them to rule out models of limited liability that are counter
to these observations in the data, yielding a similar regional pattern. Finally,
25
Other useful approaches applying regional variation and microdata to give potential in-
sights for aggregate spatial models include: experimental methods, e.g., Egger et al. (2019);
applying national income accounting methods to villages or regions, e.g. Paweenawat and
Townsend (2012, 2019); using the spatial expansion of programs, e.g., Jack and Suri (2014),
and using region-specific data to discipline region specific market imperfections, Moll et
al. (2017).

27
Karaivanov and Townsend (2014) perform pairwise comparisons of the like-
lihoods of cross sectional and panel data on consumption and investment to
distinguish between a variety of non-nested structural models. A rural-urban
pattern emerges similar to that of Northeast vs. Central. This work leads to
aggregate and regional policy analysis. For example, Moll et al. (2017) study
migration and capital flows in a model with regional variation in financing
environments. Consistent with earlier work, they impose that urban con-
straints be moral-hazard driven, while rural constraints derive from limited
commitment. Mapping the data to unique regional flow of funds data, they
show substantial rural-urban flows of labor and capital, and they show that
policies to restrict these flows lead to substantial aggregate losses, but with
heterogeneous impact across regions. This work is an example where model
predictions together with observational data help us moves beyond reduced
form modeling of financial frictions and distinguish between underlying mech-
anisms, and then showing their importance for policy and understanding the
Thai growth.
The applications move beyond Thailand, however, as Dabla-Norris et al.
(forthcoming) develop a general equilibrium framework featuring multiple
realistic sources of financial frictions to study how different financial con-
straints interact in equilibrium. They show that financial inclusion policies
should target the most binding constraint, and this can vary across countries
in ways that may not be obvious, a priori. There are important tradeoffs
between financial inclusion, GDP, and the distribution of income when con-
ducting such policies. There are also inter-temporal tradeoffs and policy
commitment issues, as some variables move in the short term in the opposite
direction from their longer run changes.

3.5 Dynamic Optimization and Capital Accumulation

Dynamic considerations are indeed important, and dynamic models have
been well analyzed in macroeconomics. At the most basic level, savings and
investment decisions respond to changes in income and interest rates. Thus,
dynamics are likely to be important any time GE effects are important, but
dynamics and price effects can depend critically on the specifics of theory. A
key lesson for scaling up the results of micro interventions is that long run
impacts can differ substantially, in direction and magnitude, from short run
impacts when laws of motion depend on prices.
Usually, long run elasticities are bigger than short-run elasticities, hence

28
the impacts on prices will be smaller in the long run but the impacts on quan-
tities will be larger. A key example of this is the work of Fried and Lagakos
(2020), who study the aggregate impacts of reliable electrity-generating in-
frastructure on development. Micro empirical work in developing countries
focus on short-run impacts of power outages and find comparatively small
impacts. In their model, for a given set of producers’ capital and technology,
they can replicate these short run impacts since power outages only lower
productivity by leaving both existing output-producing capital and private
generator capital idle. This lowers the return to capital and to the use of
modern technologies, however, so investment falls both on the intensive mar-
gin (capital of existing establishments) and the extensive margins (new entry
and the adoption of modern technologies among encumbents). In the long-
run, both the capital stock and productivity of producers decline relative to
a world without power outages and the aggregate impacts on output and
productivity are many times larger.
In some cases, however, the long run impacts on quantities can be smaller
than short run impacts. Recall the aggregate microfinance analysis of Buera
et al. (2020) in a world with financial frictions in the form of collateral con-
straints discussed in Section 3.1. In that paper, microcredit increases interest
rates both in the short and long run, but the capital stock still falls because
the direct effect of the availability of microcredit is to reduce precautionary
savings and to transfer income from individuals with high saving rates to
those with low saving rates. Since savings has collateral value for acquiring
capital, the capital stock also falls and this accumulates over time. Wage
gains are experienced in the short run, but not the long run. Indeed, since it
is the accumulation of savings that is so pivotal, the endogenous decline in
capital can happen even when the country is modeled as a small open econ-
omy with a fixed interest rate, and especially so when microfinance capital
is subsidized, where the wage can actually fall.
Other examples of the importance of dynamic considerations abound. In
some cases, the laws of motion for dynamic models can lead to multiple
dynamically stable steady states, where the less desirable steady states are
referred to as poverty traps. The presence of such poverty traps – whether
individual poverty traps, where these exist for the individuals’ laws of motion,
or aggregate poverty traps, where they exist in the aggregate dynamics –
are of keen interest to development economists concerned with aggregate

29
and distributional outcomes.26 An earlier theoretical heterogeneous agents
literature in development emphasized the possibility for multiple equilibria
and aggregate poverty traps (e.g., Banerjee and Newman, 1993; Galor and
Zeira, 1993; Piketty, 1997), and the positive role of permanent redistribution
policies (Aghion and Bolton, 1997). In these models, the wage or interest
rate depend crucially on the distribution of income and/or wealth, and this
feedback loop between distribution and aggregates is crucial. These models
had policy prescriptions that opened up the idea that one-time redistribution
might jolt the economy from the bad to the good equilibrium, improving
aggregate surplus. However, the combination of high levels of heterogeneity,
idiosyncratic shocks that cause churning in the distribution of entrepreneurial
productivity, and forward-looking behavior can lead to a singular stationary
equilibrium in related models of entrepreneurship and investment as in Buera
et al. (2011), Buera and Shin (2013a), Buera and Shin (2013b), and Buera
et al. (2014). While these models still yield poverty traps at the individual-
level, one-time redistributions dissipate nearly completely within 10 years,
and instead policies that continually redistribute are necessary to maximize
aggregate output.27
Beyond these examples, there are again additional opportunities for in-
tegrating microempirics and macro models to better inform dynamic models
and dynamic implications for policies more broadly.
Empirically, a number of long run experimental results are emerging in
the RCT literature.28 These studies are potentially highly informative for
models. For example, focusing on the poverty trap dynamics, randomized
recipients of one-time asset grants have been tracked by several studies.
Research on cash grants in Uganda (Blattman et al., 2018) and Ethiopia
(Blattman et al., 2019) shows no long term impacts on measures like earn-
ings, consumption, employment, and health at nine years and five years,
respectively, although the initial results were quite promising (Blattman et
26
To be more precise, we refer to an individual poverty trap when there is a threshold
wealth such that individuals that start with wealth below this threshold remain poor,
while those that start with wealth above this threshold save to overcome poverty. An
aggregate poverty traps arises when an economy is in a stationary with low per-capita
income due to its initial distribution of wealth, but would transition to a stationary with
higher per-capita income is a particular redistribution of wealth is implemented.
27
Bowles et al. (2016) give a comprehensive treatment of poverty traps more broadly,
including outside of development.
28
Bouguen et al. (2019) combine a review of long term RCT evaluations together with
many practical suggestions for addressing empirical issues, such as respondent tracking.

30
al., 2014; Blattman and Dercon, 2018, respectively). In contrast, two RCTs
in South Asia find more sustained impacts from grants. de Mel et al. (2012)
returned to the entrepreneurs in their de Mel et al. (2008) grant experiment
in Sri Lanka and found sustained impacts on business survival and business
profitability for male grantees. Similarly, Bandiera et al. (2013) studied ul-
tra poor asset grants of livestock to poor women in Bangladesh. They find
sustained poverty reduction and increased labor supply among the women
after seven years. Our point is that not only is the evidence mixed, but that
variations in underlying environments may lie behind the puzzles. Again,
this evidence can help improve modeling, with assumptions appropriate to
not only the environment but the policy considered. For instance, are the
findings of the experiments consistent with the implications of quantitative
models featuring financial frictions, and if not, what are the model elements
or institutional details required to match this evidence?29
At the same time models might assist long run tracking of experiments
in interpreting the data. Absent theory, the challenges of interpretation are
formidable. Consider a “phase-in” RCT like Progresa described above, where
one can always compare those who were treated early with those who were
treated late. One cannot estimate the impact of long-term treatment, how-
ever. At best, one can estimate the long-term effects of difference in treat-
ment timings or the long-term effects of a short difference in total amount
of time the treatment was received. For example, if one were to observe a
dissipation of treatment effects over time after phase in, it could reflect (i)
the benefits of treatment falling over time, (ii) persistent benefits to treated
communities, combined with catch up of later treated communities, or (iii)
a weakening of the effective treatment through channels like migration, e.g.,
contamination of treatment and control areas/populations. The last possi-
bility is particularly problematic; the assumption of no spillovers required
for valid RCT comparison become increasingly dubious over long periods of
time, and so the definition of treatment can become much more nuanced.
An exacerbating source of spillovers is the presence of spatially-segmented
GE effects, as described above. In principle, dynamic models based on sound
theory could help both in the interpretation of the results of the few long
29
Buera et al. (2014) is a crude first attempt at confronting the implications of quantita-
tive macro development models featuring financial friction with the recent evidence from
asset grants, and related historical evidence by Bleakley and Ferrie (2013). The aforemen-
tioned work by Kaplan and Violante (2014) is a more developed example showing how
evidence from quasi-experiments can be used to identify important model elements.

31
term RCT studies, but also in the prediction of the long term consequences of
policies at the macro level. One could model the interactions, dynamics, and
spillovers directly, and simply use the differences in outcomes among treat-
ment and control to identify key parameters. The long run empirics would
then become additional identifying moments. The challenge here is to have
dynamic extensions of models with realistic spatial frictions as in Desmet et
al. (2018a).
Micro empiricists increasingly appreciate that aggregate shocks can be
important, even for understanding and extrapolating RCT evidence, because
they impact the interpretation of individual studies. For example, the results
found in agricultural studies from a particular year may depend crucially on
the weather experienced during that year. As a concrete example, Zhang
and Carter (1997) argue that Lin (1992)’s estimates of the effect of agricul-
tural reforms in post-Mao China were substantially overestimated because
he failed to consider the effects of weather on agricultural output. Even if
there is cross-sectional heterogeneity in weather patterns, this variation alone
cannot be simply extrapolated to account for the aggregate shock, since the
average or aggregate shock can affect the price of crops, labor, etc. in gen-
eral equilibrium. Indeed, Rosenzweig and Udry (2016) make the point that
aggregate shocks are important for determining the external validity of micro-
level RCTs themselves, because of these aggregate channels. Naturally, this
evidence calls for an integration of microevidence with quantitative macro
models featuring aggregate shocks, a task that can be computationally chal-
lenging when considering environments with rich heterogeneity and various
frictions.
While dynamic, general equilibrium models with heterogeneity, idiosyn-
cratic shocks, and aggregate shocks can be difficult to compute and solve,
new methods enable easier analysis. The models are difficult because invest-
ment responds to expectations of the future path of prices, but the state
that determines prices at any point in time is a high dimensional object (the
full distribution). Advances involve combinations of discretization and local
linearizations. Krusell and Smith (1998) argued that most of the impor-
tant aspects of the wealth distribution in their context could be captured
with just a single moment, the mean of the wealth distribution. But more
generally, the distribution of wealth matters. Recent work by Ahn et al.
(2018) provides more efficient computational methods to solve models with
heterogeneity and aggregate shocks that allow for larger numbers of indi-
vidual state variables. The approach is to solve (an approximation to) the

32
global stationary equilibrium without aggregate shocks using efficient finite-
difference methods, and then to add in the aggregate shocks using linear
approximations. They provide a precoded toolbox to make these methods
available for broader applications. Silva and Townsend (2019) propose a
new form of solving for the transitional dynamics, by combining perturba-
tion and finite difference methods to discretize the system of ODE/PDEs
that does not assume aggregate or idiosyncratic shocks are small. These ad-
vances have in principle enabled macro models to more easily address the
environments of developing countries with extensive heterogeneity across the
macroeconomy, highly uncertain environments, incomplete markets, and im-
portant additional frictions. These richer models can be used, for example,
to perform Monte Carlo analyses of estimates of the impacts of large scale
policies in environments with large aggregate shocks or, more generally, to
study the role of aggregate shocks in development (Aguiar and Gopinath,
2007).
Of course, assets and capital are not the only laws of motion that are of po-
tential importance. We have already noted the distribution of entrepreneurial
ability. In models with one-time set up costs for firms, the full distribution
of firms can be an important state variable, while in the “ideas” models of
Lucas (2009), Lucas and Moll (2014), Perla and Tonetti (2014), for example,
the state space can be the distribution of productivity of all individuals in an
economy. Human capital may be an important dynamic state, or locational
distribution of workers or firms in the spatial models described above, includ-
ing migration models. Culture itself can be a slow-moving state influenced
by policy, for example sexual norms in the Bangladesh contraception inter-
vention (Munshi and Myaux, 2006), and ignoring dynamic policy impacts on
culture could lead to short-sighted policy.
In sum, quantitative models with rich state spaces provide a way of ac-
counting for and understanding the multidimensional heterogeneity in treat-
ment effects that RCTs often test for and find. Moreover, rich dynamics,
anchored in theory, allow policy makers to account for changing dimensions
of this heterogeneity, both the dynamics that are responses to the interven-
tions, and those that are independent of the intervention and can therefore be
taken as exogenous to the intervention but with endogenous dynamics nev-
ertheless. Both long and short-run empirical work can be helpful in testing
and disciplining theory.

33
3.6 Economies of Scale
Returns to scale are another important consideration for macro development
policy and for extrapolating the results from micro experiments. The behav-
ior of economies under aggregate decreasing returns to scale is very similar
to the behavior under the presence of aggregate resource constraints (fixed
factors are a key reason for decreasing returns), so we do not address this
in more detail.30 However, economies of scale can also exist and can open
opportunities for development and have aggregate policy implications (e.g.,
coordination, Pigouvian subsidies). Focusing on scaling interventions, the
key lesson is that impacts at scale can be larger because of economies of
scale.
Here we turn to a concrete empirical example: mobile money, a banking
and money transfer program. It has created a strong foothold in some coun-
tries, most famously M-PESA in Kenya, where it is almost universal, but
it has far fewer users in other countries (Suri, 2017). Mobile money works
through the mobile phone system and a network of existing retail establish-
ments who act as agents. It is therefore much more useful at scale than in a
small pilot and network effects are important. Indeed, Jack and Suri (2014)
document the rapid rollout of M-PESA and the importance of the underly-
ing retail network in adoption. There are at least two sources of increasing
returns to scale at play. First, the more users, the more potential bilateral
transfers, and that relationship increases with the square of the number of
users. Second, the more users, the denser the network of agents, and the
smaller the transaction costs of using the network.31
30
Nevertheless, the channels through which smaller impacts under scale can arise are
interesting and varied. For example, information can undermine scaled policy. For exam-
ple, the experimental work of Pomeranz (2015) and Carrillo et al. (2015) who randomized
warning letters of tax enforcement to firms that increased reporting, even though they
were largely bluffs. If one were to implement a national policy based on this, it is likely
that firms and tax lawyers would soon (perhaps even ahead of time) learn of the policy
and learn to ignore the false letters. Othe surprising channels include bureaucratic and
political opposition (Bold et al., 2018), counterproductive competition (Galle, 2019), and
information.
31
Batista and Vicente (2013) use an experiment in Mozambique to assess the impacts of
mobile money, finding that it indeed increases a willingness to send remittances, but there
is no assessment of the role of scale. Mobile money is interesting, not only because it is
a scalable microintervention that can assist, for example, in the delivery of social service
payments – the large-scale NREGS experiment in India by Muralidharan et al. (2017)
is a nice example – but also because it has regulatory questions that relate to the more

34
 Returning to theory, at the micro-level of firms and households, indivis-
ibilities, learning by doing, or minimum efficient scales in production, con-
sumption, and/or delivery may cause impacts to be larger when scaled up
on the intensive margin. If there are regions of increasing returns to scale
at this micro-level, one-time increases in scale can have persistent impacts.
Beyond network effects, other potential theoretical mechanisms for external
economies of scale can stem from critical mass, industrial concentration, tech-
nology spillovers, and external learning. On the firm side, many of these are
the traditional agglomeration forces emphasized by Marshall (1890). Their
external nature gives them particular policy relevance. Within the macro
literature, external scale mechanisms played an important role in the en-
dogenous growth theory of the early 1990s (Romer, 1990; Grossman and
Helpman, 1991; Aghion and Howitt, 1992) in promoting policies of openness
and subsidies to human capital and research and development expenditures.
The question of whether there are increasing returns to scale is therefore
important in macro models, important for scaling policies, and important for
guiding efficient growth policy. How can microstudies help inform us on this
question?
Recent empirical work has used randomized short-term increases in de-
mand to evaluate impacts. Carrillo et al. (2019) use the randomization of
government procurement contracts in Ecuador to examine their impact on
firms. They find essentially no evidence of increasing returns: firms expand
by the amount of the government procurement contract and quickly contract
after it is complete. In contrast, Atkin et al. (2017) randomize export orders
to Egyptian carpet manufacturers and find evidence of improved technical
efficiency over time, which they attribute to learning-by-doing mechanisms.
Alfaro-Urena et al. (2019) use a quasi-experimental approach to look at the
impact of multinational contracts on domestic suppliers in Costa Rica. They
find firms grow 20 percent in response and TFP increases by 3-9%, depending
on whether using a model-based measurement or not.
Another way to help uncover nonconvexities is to vary the endowment
of treatment. Such is the approach in Balboni et al. (2018), who evaluate
livestock grants in rural Bangladesh. The treated households in their exper-
iment all receive the same sized grant of assets (livestock), but they vary in
traditional macroeconomic topics of money and banking. Electronic payments systems, for
example, proved important in the 2016 demonetization policy in India (Chodorow-Reich
et al., 2018).

35
their initial wealth. Focusing on asset dynamics, they show s-shaped returns
consistent with poverty trap asset dynamics, and evidence of twin peaks in
the asset distribution, consistent with multiple equilibria. Examining the
same issue in Uganda, Kaboski et al. (2019) use a choice across different
lotteries as an alternative way of combining intensive margin variation with
randomization, and tying it to risk preference. When agents are in regions
of non-convexities, they can be locally risk-loving. They offer a choice over
cash lotteries to an underbanked population in Uganda and show that a
substantial share of the population prefers a high-risk lottery even when it
has a slightly lower expected payout. This study is unique in that it uses
the macro theory to design the experiment, motivating the experiment by
the local risk-loving that accompanies non-convexities. They attribute the
non-convexity to the market for land, which has high returns.
The diffusion of technologies typically follow s-shapes, implying regions
with increasing returns and regions with decreasing returns. The aforemen-
tioned recent macro literature that focuses on idea diffusion models thinks
about the role of interaction in productivity growth (Lucas, 2009; Lucas and
Moll, 2014; Perla and Tonetti, 2014), providing microfoundations for human
capital externalities (Lucas, 1988b). Interactions between agents lead to the
transfer of ideas and hence productivity increases for the less productive
agents. The search externalities in these models allow for efficiency enhanc-
ing policy. These frameworks have been used to explicitly model at the
micro level how macro policies like trade and openness to foreign direct in-
vestment impact the distribution of productivity across producers (Alvarez et
al., 2013; Sampson, 2015; Buera and Oberfield, 2020; Perla et al., 2021), and
how knowledge is diffused within and across firms through the reallocation
of workers (Herkenhoff et al., 2018; Jarosch et al., 2019). In a recent paper
combining macro models with experimental micro evidence, Brooks et al.
(2018) use the evidence from an experiment combining novice entrepreneurs
with more experienced mentors to quantitatively discipline the productivity
diffusion process in Kenyan slums. The paper is quite innovative, although it
leaves open the question of how to extrapolate narrow experimental evidence
to the broader economy.
To summarize, economies of scale are important considerations for guiding
macro development policy and interpreting the results of small-scale studies
in light of macro policy. Nevertheless, both large scale and small-scale studies
can help inform macro theory, especially when designed in conjunction with
it.

36
4 Conclusions
We have discussed advances in empirical development and macroeconomics
which we argue are best seen as complementary.
On the one hand, beginning at the micro level, grounded on new study
designs, namely, natural and randomized experiments in conjunction with
models, we are now able to have clearer snapshots and understanding of the
realities in developing countries, and to identify causal effects of various micro
policy interventions. There are obvious limits to a purely empirical strategy,
which we discussed at length.
On the other hand, at the macro level, there have been parallel advances in
the construction of quantitative macroeconomic models that incorporate rich
heterogeneity across agents and space, and various contracting and trading
frictions. We argue that the rich quantitative heterogeneous agents models
are ideally suited to connect the dots of micro interventions and, therefore,
evaluate (and devise) macro development policies. The recent wealth of micro
evidence is central to better discipline the structure of the new generation of
policy oriented macroeconomic models.
Although as we have shown that there is already much work integrating
these approaches, this is not a review of a mature literature. Rather, we view
the paper as equal parts review and call to arms for more work in an exciting
and promising direction. The development problem is important but hard to
crack, with no magic bullets. There are certainly no magic methodological
approaches.
Successfully moving this agenda forward requires not only the integra-
tion of the seemingly methodological extremes, randomized experiments and
macro modeling, but advances from other complementary areas in economics,
from the collection and homogenization of representative country datasets to
better understanding of the identification of structural models. The needs
span across many fields in economics: field work, macro and growth mod-
eling, econometrics, spatial methods, trade, industrial organization, labor,
family economics, public economics, etc. The most important questions re-
quire people with diverse expertise and skills. We encourage macro or trade
development economists to bring models to collaborate with micro empiri-
cists who have collected RCT data to evaluate scaled policies, and we have
cited many such examples. More ambitious, however, is for such collabora-
tions to begin earlier in the process, with the macro question and conceptual
model leading the design of experiments and data collection. The ground is

37
therefore fertile for both young researchers and for new collaborations.

38
References
Acemoglu, Daron, “Theory, General Equilibrium, and Political Economy
in Development Economics,” The Journal of Economic Perspectives, 2010,
24 (3), 17–32.

Adamapoulos, Tasso, Loren Brandt, Jessica Leight, and Diego

Restuccia, “Misallocation, Selection and Productivity: A Quantitative
Analysis with Panel Data from China,” NBER Working Paper 23039, Na-
tional Bureau of Economic Research January 2017.

Aghion, Philippe and Patrick Bolton, “A Theory of Trickle-Down

Growth and Development,” Review of Economic Studies, 1997, 64 (2),
151–172.

Aghion, Phillipe and Peter Howitt, “A Model of Growth Through Cre-

ative Destruction,” Econometrica, March 1992, 60 (2), 323–351.

Aguiar, Mark and Gita Gopinath, “Emerging Market Business Cycles:

The Cycle Is the Trend,” Journal of Political Economy, 2007, 115, 69–102.

Ahlin, Christian and Robert M. Townsend, “Using Repayment Data to

Test Across Models of Joint Liability Lending,” Economic Journal, 2007,
117 (517), 11–51.

Ahn, SeHyoun, Greg Kaplan, Benjamin Moll, Thomas Winberry,

and Christian Wolf, “When Inequality Matters for Macro and Macro
Matters for Inequality,” NBER Macroeconomics Annual, 2018, 32 (1), 1 –
75.

Aiyagari, S. Rao, “Uninsured Idiosyncratic Risk and Aggregate Saving,”

The Quarterly Journal of Economics, 1994, 109 (3), 659–684.

Akram, Agha Ali, Shyamal Chowdhury, and Ahmed Mushfiq Mo-

barak, “Effects of Emigration on Rural Labor Markets,” NBER Working
Papers 23929, National Bureau of Economic Research, Inc October 2017.

Alfaro-Urena, Alfonso, Isabela Manelici, and Jose P. Vasquez, “The

Effects of Joining Multinational Supply Chains: New Evidence from Firm-
to-Firm Linkages,” NBER Working Paper April 2019.

39
Allen, Treb and Costas Arkolakis, “Trade and the Topography of the
Spatial Economy,” The Quarterly Journal of Economics, 2014, 129 (3),
1085–1140.

Alvarez, Fernando E. and David Argente, “Consumer Surplus of Alter-

nativePayment Methods: Paying Uber with Cash,” BFI Working Papers
2019-96, Becker Friedman Institute September 2019.

, Francisco J. Buera, and Jr. Robert E. Lucas, “Idea Flows, Eco-

nomic Growth, and Trade,” NBER Working Papers 19667, National Bu-
reau of Economic Research, Inc November 2013.

Angrist, Joshua D. and Jörn-Steffen Pischke, “The Credibility Revo-

lution in Empirical Economics: How Better Research Design Is Taking the
Con out of Econometrics,” Journal of Economic Perspectives, June 2010,
24 (2), 3–30.

Arkolakis, Costas, Arnaud Costinot, and Andres Rodriguez-Clare,

“New Trade Models, Same Old Gains?,” American Economic Review,
February 2012, 102 (1), 94–130.

Atkin, David, Amit K. Khandelwal, and Adam Osman, “Export-

ing and Firm Performance: Evidence from a Randomized Experiment,”
Quarterly Journal of Economics, May 2017, 132 (2), 551–615.

Attanasio, Orazio P., Costas Meghir, and Ana Santiago, “Education

Choices in Mexico: Using a Structural Model and a Randomized Exper-
iment to Evaluate PROGRESA,” The Review of Economic Studies, 08
2011, 79 (1), 37–66.

Balboni, Clare, Oriana Bandiera, Maitreesh Ghatak, and Anton

Heil, “Why Do People Stay Poor?,” 2018. Working Paper.

Bandiera, Oriana, Robin Burgess, Narayan Das, Selim Gulesci,

Imran Rasul, and Munshi Sulaiman, “Can Basic Entrepreneurship
Transform the Economic Lives of the Poor?,” IZA Discussion Papers 7386,
Institute for the Study of Labor (IZA) May 2013.

Banerjee, Abhijit, Emily Breza, Esther Duflo, and Cynthia Kin-

nman, “Do Credit Constraints Limit Entrepreneurship? Heterogeneity

40
 in the Returns to Microfinance,” Working Papers 17-104, Global Poverty
Research Lab 2017.
, Esther Duflo, Rachel Glennerster, and Cynthia Kinnan, “The
Miracle of Microfinance? Evidence from a Randomized Evaluation,”
American Economic Journal: Applied Economics, 2015, 7 (1), 22–53.
, Rukmini Banerji, James Berry, Esther Duflo, Harini Kannan,
Shobhini Mukerji, Marc Shotland, and Michael Walton, “From
Proof of Concept to Scalable Policies: Challenges and Solutions, with an
Application,” Journal of Economic Perspectives, November 2017, 31 (4),
73–102.
Banerjee, Abhijit V. and Andrew F. Newman, “Occupational Choice
and the Process of Development,” Journal of Political Economy, 1993, 101
(2), 274–298.
and Benjamin Moll, “Why Does Misallocation Persist?,” American
Economic Journal: Macroeconomics, January 2010, 2 (1), 189–206.
and Esther Duflo, “Growth Theory through the Lens of Development
Economics,” in Philippe Aghion and Steven N. Durlauf, eds., Handbook of
Economic Growth, Vol. 1A, Amsterdam: Elsevier, 2005, pp. 473–552.
and , “The Experimental Approach to Development Economics,” An-
nual Review of Economics, April 2009, 1 (1), 151–178.
Baqaee, David Rezza and Emmanuel Farhi, “The Microeconomic foun-
dations of Aggregate Production Functions,” Journal of the European Eco-
nomic Association, October 2019, 17 (5).
and , “Productivity and Misallocation in General Equilibrium*,” The
Quarterly Journal of Economics, 09 2019, 135 (1), 105–163.
Bartelsman, Eric, John Haltiwanger, and Stefano Scarpetta, “Mi-
croeconomic evidence of creative destruction in industrial and developing
countries,” Policy Research Working Paper Series 3464, The World Bank
December 2004.
Batista, Catia and Pedro C. Vicente, “The Price of Political Opposi-
tion: Evidence from Venezuela’s Maisanta,” Working Paper Series, NO-
VAFRICA 2013.

41
Bergquist, Lauren, Benjamin Faber, Thibault Fally, Matthias Hoel-
zlein, Edward Miguel, and Andres Rodriguez-Clare, “Scaling Agri-
cultural Policy Interventions: Theory and Evidence from Uganda,” Unpub-
lished manuscript, University of California at berkeley 2019.

Bewley, Truman, “Stationary Monetary Equilibrium with a Continuum

of Independently Fluctuating Consumers,” in W. Hildenbrand, A. Mas-
Colell, and G. Debreu, eds., Contributions to Mathematical Economics in
Honor of Gérard Debreu, North Holland, 1986.

Bils, Mark, Peter J. Klenow, and Cian Ruane, “Misallocation or Mis-

measurement?,” Unpublished manuscript, Stanford University 2018.

Blattman, Christopher and Stefan Dercon, “The Impacts of Industrial

and Entrepreneurial Work on Income and Health: Experimental Evidence
from Ethiopia,” American Economic Journal: Applied Economics, July
2018, 10 (3), 1–38.

, Nathan Fiala, and Sebastian Martinez, “Generating Skilled

Self-Employment in Developing Countries: Experimental Evidence from
Uganda,” The Quarterly Journal of Economics, 2014, 129 (2), 697–752.

, , and , “The Long Term Impacts of Grants on Poverty: 9-year Ev-

idence From Uganda’s Youth Opportunities Program,” Technical Report,
National Bureau of Economic Research 2018.

, Stefan Dercon, and Simon Franklin, “Impacts of Industrial and

Entrepreneurial Jobs on Youth: 5-year Experimental Evidence on Factory
Job Offers and Cash Grants in Ethiopia,” NBER Working Papers 25788,
National Bureau of Economic Research, Inc April 2019.

Bleakley, Hoyt and Joseph P. Ferrie, “Up from Poverty? The 1832
Cherokee Land Lottery and the Long-run Distribution of Wealth,” NBER
Working Papers 19175, National Bureau of Economic Research, Inc June
2013.

Bold, Tessa, Mwangi Kimenyi, Germano Mwabu, Alice Ng’ang’a,

and Justin Sandefur, “Experimental evidence on scaling up education
reforms in Kenya,” Journal of Public Economics, December 2018, 168,
1–20.

42
Bouguen, Adrien, Yue Huang, Michael Kremer, and Edward
Miguel, “Using Randomized Controlled Trials to Estimate Long-Run Im-
pacts in Development Economics,” Annual Review of Economics, 2019, 11
(1), 523–561.

Bowles, Samuel, Steven N. Durlauf, and Karla Hoff, Poverty Traps.,

Vol. 8, Princeton: Princeton University Press, 2016.

Breza, Emily and Cynthia Kinnan, “Measuring the Equilibrium Impacts

of Credit: Evidence from the Indian Microfinance Crisis,” NBER Working
Papers, National Bureau of Economic Research 2018.

Brooks, Wyatt, Kevin Donovan, and Terence R. Johnson, “Identifi-

cation and Quantitative Implications of an Equilibrium Diffusion Model,”
Technical Report, University of Notre Dame 2018.

Bryan, Gharad, Shyamal Chowdhury, and Ahmed Mushfiq Mo-

barak, “Underinvestment in a Profitable Technology: The Case of Sea-
sonal Migration in Bangladesh,” Econometrica, 2014, 82 (5), 1671–1748.

Buera, Francisco J., “A Dynamic Model of Entrepreneurship with Bor-

rowing Constraints: Theory and Evidence,” Annals of Finance, 2009, 5,
443–464.

and Ezra Oberfield, “The Global Diffusion of Ideas,” Econometrica,

2020, 88 (1), 83–114.

and Yongseok Shin, “Financial Frictions and the Persistence of History:

A Quantitative Exploration,” Journal of Political Economy, 2013, 121 (2),
221 – 272.

and , “Financial Frictions and the Persistence of History: A Quantita-

tive Exploration,” Journal of Political Economy, 2013, 121 (2), 221–272.

, Joseph P. Kaboski, and Yongseok Shin, “Finance and Development:

A Tale of Two Sectors,” American Economic Review, August 2011, 101
(5), 1964–2002.

, , and , “Macro-Perspective on Asset Grants Programs: Occupa-

tional and Wealth Mobility,” American Economic Review, May 2014, 104
(5), 159–64.

43
 , , and , “The Macroeconomics of Microfinance,” The Review of
Economic Studies, 08 2020.

Burns, Arthur F. and Wesley C. Mitchell, Measuring Business Cycles,

Cambridge, MA: National Bureau of Economic Research, 1946.

Cai, Jing and Adam Szeidl, “Interfirm Relationships and Business Perfor-
mance,” The Quarterly Journal of Economics, 2018, 133 (3), 1229–1282.

Cai, Shu, Albert Park, and Sangui Wang, “Microfinance Can Raise
Incomes: Evidence from a Randomized Control Trial in China,” Technical
Report, HKUST 2016.

Caliendo, Lorenzo, Fernando Parro, Esteban Rossi-Hansberg, and

Pierre-Daniel Sarte, “The Impact of Regional and Sectoral Productivity
Changes on the U.S. Economy,” Review of Economic Studies, 2018, 85 (4),
2042–2096.

, Maximiliano Dvorkin, and Fernando Parro, “Trade and Labor Mar-

ket Dynamics: General Equilibrium Analysis of the China Trade Shock,”
Econometrica, 2019, 87 (3), 741–835.

Carrillo, Paul, Dave Donaldson, Dina Pomeranz, and Monica Sing-

hal, “The Bigger the Better? Using Lotteries to Identify the Allocative
Efficiency Effects of Firm Size,” Technical Report 2019.

, Dina Pomeranz, and Monica Singhal, “Dodging the Taxman: Firm

Misreporting and Limits to Tax Enforcementa,” American Economic Jour-
nal: Applied Economics, 2015, 9 (2), 144–164.

Chandrasekhar, Arun G., Cynthia Kinnan, and Horacio Larreguy,

“Social Networks as Contract Enforcement: Evidence from a Lab Exper-
iment in the Field,” American Economic Journal: Applied Economics,
October 2018, 10 (4), 43–78.

Chen, Chaoran, Diego Restuccia, and Raul Santaeulalia-Llopis,

“The Effects of Land Markets on Resource Allocation and Agricultural
Productivity,” Working Papers tecipa-649, University of Toronto, Depart-
ment of Economics September 2019.

44
Chetty, Raj, “Sufficient Statistics for Welfare Analysis: A Bridge Between
Structural and Reduced-Form Methods,” Annual Review of Economics,
2009, 1 (1), 451–488.

Chodorow-Reich, Gabriel, Gita Gopinath, Prachi Mishra, and Ab-

hinav Narayanan, “Cash and the Economy: Evidence from India’s De-
monetization,” NBER Working Papers 25370, National Bureau of Eco-
nomic Research, Inc December 2018.

, , , and , “Cash and the Economy: Evidence from India’s De-

monetization*,” The Quarterly Journal of Economics, 09 2019, 135 (1),
57–103.

Cunha, Jesse M, Giacomo De Giorgi, and Seema Jayachandran,

“The Price Effects of Cash Versus In-Kind Transfers,” The Review of Eco-
nomic Studies, 2018.

Dabla-Norris, Era, Yan Ji, Robert M. Townsend, and D. Filiz Un-

sal, “Distinguishing Constraints on Financial Inclusion and Their Impact
on GDP, TFP, and the Distribution of Income,” Journal of Monetary Eco-
nomics, forthcoming, 48 (2).

de Mel, Suresh, David McKenzie, and Christopher Woodruff, “Re-

turns to Capital in Microenterprises: Evidence from a Field Experiment,”
The Quarterly Journal of Economics, 2008, 123 (4), 1329–1372.

, , and , “One-Time Transfers of Cash or Capital Have Long-Lasting

Effects on Microenterprises in Sri Lanka,” Science, 2012, 335 (6071), 962–
966.

Deaton, Angus, “Instruments of development: Randomization in the trop-

ics, and the search for the elusive keys to economic development,” Pro-
ceedings of the British Academy, 2008 Lectures, 2009, 162, 123–160.

Dekle, Robert, Jonathan Eaton, and Samuel Kortum, “Global Re-

balancing with Gravity: Measuring the Burden of Adjustment,” IMF Staff
Papers, July 2008, 55 (3), 511–540.

Desmet, Klaus, Dávid Krisztián Nagy, and Esteban Rossi-

Hansberg, “The Geography of Development,” Journal of Political Econ-
omy, 2018, 126 (3), 903–983.

45
 , Robert E Kopp, Scott A Kulp, Dávid Krisztián Nagy,
Michael Oppenheimer, Esteban Rossi-Hansberg, and Ben-
jamin H Strauss, “Evaluating the Economic Cost of Coastal Flood-
ing,” Working Paper 24918, National Bureau of Economic Research August
2018.

Donaldson, Dave, “Railroads of the Raj: Estimating the Impact of Trans-

portation Infrastructure,” American Economic Review, April 2018, 108
(4–5), 899–934.

Donovan, Kevin, “The Equilibrium Impact of Agricultural Risk on Inter-

mediate Inputs and Aggregate Productivity,” 12 2020.

Duflo, Esther, “Grandmothers and Granddaughters: Old-Age Pensions and

Intrahousehold Allocation in South Africa,” The World Bank Economic
Review, 2003, 17 (1), 1–25.

, Pascaline Dupas, and Michael Kremer, “The Impact of Free Sec-

ondary Education: Experimental Evidence from Ghana,” manuscript, MIT
2017.

Egger, Dennis, Johannes Haushofer, Edward Miguel, Paul Niehaus,

and Michael W Walker, “General Equilibrium Effects of Cash Trans-
fers: Experimental Evidence from Kenya,” Working Paper 26600, National
Bureau of Economic Research December 2019.

Ehrlich, Dan and Robert Townsend, “Spatial Spillovers and Labor Mar-
ket Dynamics: Village Financial Interventions in Thailand,” manuscript,
MIT 2019.

Emerick, Kyle, Alain de Janvry, Elisabeth Sadoulet, and Man-

zoor H. Dar, “Technological Innovations, Downside Risk, and the Mod-
ernization of Agriculture,” American Economic Review, June 2016, 106
(6), 1537–61.

Esfahani, Saraa, “Agricultural Misallocation: Mismeasurement, Misspeci-

fication, or Market Frictions?” - Manuscript sub- mitted for publication,”
Technical Report, University of Notre Dame, Department of Economics
2018.

46
Fehr, Ernst and Lorenz Goette, “Do Workers Work More if Wages Are
High? Evidence from a Randomized Field Experiment,” American Eco-
nomic Review, March 2007, 97 (1), 298–317.

Fieler, Ana Cecı́lia, “Nonhomotheticity and Bilateral Trade: Evidence and

a Quantitative Explanation,” Econometrica, July 2011, 79 (4), 1069–1101.

Filmer, Deon P., Jed Friedman, Eeshani Kandpal, and Junko On-
ishi, “Cash Transfers, Food Prices, and Nutrition Impacts on Nonbenefi-
ciary Children,” Policy Research Working Paper Series 8377, The World
Bank March 2018.

Fisher, Franklin M., “The Existence of Aggregate Production Functions,”

Econometrica, 1969, 37 (4), 553–577.

Fisher, Ronald A., “Studies in crop variation. I. An examination of the

yield of dressed grain from Broadbalk.,” Journal of Agricultural Science,
1921, 11 (72), 107–135.

Fried, Stephie and David Lagakos, “Electricity and Firm Productiv-

ity: A General-Equilibrium Approach,” Working Paper 27081, National
Bureau of Economic Research May 2020.

Fujimoto, Junichi, David Lagakos, and Mitchell Vanvuren, “Aggre-

gate and Distributional Eects of ‘Free’ Secondary Schooling in the Devel-
oping World,” manuscript, University of California at San Diego 2019.

Galle, Simon, “Competition, Financial Constraints and

Misallocation:Plant-Level Evidence from Indian Manufacturing,” Techni-
cal Report, University of California, Berkeley 2019.

Galor, Oded and Joseph Zeira, “Income Distribution and Macroeco-

nomics,” Review of Economic Studies, 1993, 60 (1), 35–52.

Giné, Xavier and Robert M. Townsend, “Evaluation of Financial Lib-

eralization: A General Equilibrium Model with Constrained Occupation
Choice,” Journal of Development Economics, 2004, 74 (2), 269–307.

Gollin, Douglas and Christopher Udry, “Heterogeneity, Measurement

Error, and Misallocation: Evidence from African Agriculture,” Working
Paper 25440, National Bureau of Economic Research 2019.

47
Greenwood, Jeremy, Philipp Kircher, Cezar Santos, and Michèle
Tertilt, “An Equilibrium Model of the African HIV/AIDS Epidemic,”
Econometrica, July 2019, 87 (4), 1081–1113.
Grossman, Gene M. and Elhanan Helpman, “Quality Ladders in the
Theory of Growth,” Review of Economic Studies, 1991, 58 (1), 43–61.
Hansen, Lars P. and James J. Heckman, “The Empirical Foundations
of Calibration,” Journal of Economic Perspectives, 1996, 10 (1), 87–104.
Harberger, Arnold C., “The Measurement of Waste,” The American Eco-
nomic Review, 1964, 54 (3), 58–76.
Hausman, Jerry A., “Taxes and labor supply,” in A. J. Auerbach and
M. Feldstein, eds., Handbook of Public Economics, Vol. 1, Elsevier, 1985.
Heckman, James and Sergio Urzua, “Comparing IV with structural
models: What simple IV can and cannot identify,” Journal of Economet-
rics, 2010, 156 (1), 27–37.
Heckman, James J., “Building Bridges between Structural and Program
Evaluation Approaches to Evaluating Policy,” Journal of Economic Liter-
ature, June 2010, 48 (2), 356–98.
and Edward Vytlacil, “Structural Equations, Treatment Effects, and
Econometric Policy Evaluation,” Econometrica, May 2005, 73 (3), 669–
738.
Herkenhoff, Kyle, Jeremy Lise, Guido Menzio, and Gordon M.
Phillips, “Production and Learning in Teams,” NBER Working Papers
25179, National Bureau of Economic Research, Inc October 2018.
Hopenhayn, Hugo A, “Entry, Exit, and Firm Dynamics in Long Run
Equilibrium,” Econometrica, September 1992, 60 (5), 1127–1150.
Hsieh, Chang-Tai and Peter Klenow, “Misallocation and Manufacturing
TFP in China and India,” Quarterly Journal of Economics, November
2009, 124 (4), 1403–1448.
Huggett, Mark, “The Risk-Free Rate in Heterogeneous-Agent Incomplete-
Insurance Economies,” Journal of Economic Dynamics and Control, 1993,
17 (5-6), 953–969.

48
Imbens, Guido W., “Better LATE Than Nothing: Some Comments on
Deaton (2009) and Heckman and Urzua (2009),” Journal of Economic
Literature, June 2010, 48 (2), 399–423.

and Joshua D. Angrist, “Identification and Estimation of Local Aver-

age Treatment Effects,” Econometrica, 1994, 62 (2), 467–475.

Jack, William and Tavneet Suri, “Risk Sharing and Transactions Costs:
Evidence from Kenya’s Mobile Money Revolution,” American Economic
Review, January 2014, 104 (1), 183–223.

Jarosch, Gregor, Ezra Oberfield, and Esteban Rossi-Hansberg,

“Learning from Coworkers,” NBER Working Papers 25418, National Bu-
reau of Economic Research, Inc January 2019.

Jeong, Hyeok and Robert M. Townsend, “Sources of TFP Growth:

Occupational Choice and Financial Deepening,” Economic Theory, 2007,
32 (1), 197–221.

Joaquim, Gustavo, Robert Townsend, and Victor Zhorin, “Optimal

Contracting and Spatial Competition among Financial Service Providers,”
2019. MIT.

Johnson, David S., Jonathan A. Parker, and Nicholas S. Souleles,

“Household Expenditure and the Income Tax Rebates of 2001,” American
Economic Review, December 2006, 96 (5), 1589–1610.

Jones, Ronald W., “The Structure of Simple General Equilibrium Models,”

Journal of Political Economy, December 1965, 73 (6), 557–572.

Kaboski, Joseph P. and Robert M. Townsend, “A Structural Evalua-

tion of a Large-Scale Quasi-Experimental Microfinance Initiative,” Econo-
metrica, 2011, 79 (5), 1357–1406.

and , “The Impact of Credit on Village Economies,” American Eco-

nomic Journal: Applied Economics, April 2012, 4 (2), 98–133.

, Molly Lipscomb, Midrigan Midrigan, and Carolyn Pelnik, “How

Important are Indivisibilities for Savings and Development? Experimental
Evidence from Uganda,” manuscript, University of Notre Dame 2019.

49
Kahneman, Daniel and Amos Tversky, “Prospect Theory: An Analysis
of Decision under Risk,” Econometrica, 1979, 47 (2), 263–291.

Kaplan, Greg and Giovanni L. Violante, “A Model of the Consumption

Response to Fiscal Stimulus Payments,” Econometrica, July 2014, 82 (4),
1199–1239.

Karaivanov, Alexander and Robert M. Townsend, “Dynamic Finan-

cial Constraints: Distinguishing Mechanism Design From Exogenously In-
complete Regimes,” Econometrica, 2014, 82 (3), 887–959.

Kaur, Supreet, Michael Kremer, and Sendhil Mullainathan, “Self-

Control at Work,” Journal of Political Economy, 2015, 123 (6), 1227 –
1277.

Koopmans, Tjalling C., “Measurement Without Theory,” The Review of

Economics and Statistics, 1947, 29 (3), 161–172.

Kreindler, Gabriel E., “The welfare effect of road congestion pricing,”

Working Paper, International Growth Center 2018.

Kremer, Michael, Jessica Leino, Edward Miguel, and Alix Peterson

Zwane, “Spring Cleaning: Rural Water Impacts, Valuation, and Property
Rights Institutions,” Quarterly Journal of Economics, 2011, 126 (1), 145–
205.

Krusell, Per and Anthony A. Smith Jr., “Income and Wealth Hetero-
geneity in the Macroeconomy,” Journal of Political Economy, 1998, 106
(5), 867–896.

Kydland, Finn E. and Edward C. Prescott, “Time to Build and Ag-

gregate Fluctuations,” Econometrica, 1982, 50 (6), 1345–1370.

Lagakos, David, Mushfiq Mobarak, and Michael Waugh, “The Wel-

fare Effects of Encouraging Rural-Urban Migration,” Working Papers
2018–002, Human Capital and Economic Opportunity Working Group
January 2018.

Leemput, Eva Van, “A Passage to India: Quantifying Internal and Ex-

ternal Barriers to Trade,” 2016. International Finance Discussion Papers
1185.

50
Lin, Justin Yifu, “Rural Reforms and Agricultural Growth in China,” The
American Economic Review, 1992, 82 (1), 34–51.

Lucas, Robert and Benjamin Moll, “Knowledge Growth and the Allo-
cation of Time,” Journal of Political Economy, 2014, 122 (1), 1 – 51.

Lucas, Robert E., “What Economists Do,” Unpublished Manuscript 1988.

, “Ideas and Growth,” Economica, 2009, 76 (301), 1–19.

Lucas, Robert Jr, “Econometric policy evaluation: A critique,” Carnegie-

Rochester Conference Series on Public Policy, January 1976, 1 (1), 19–46.

Lucas, Robert Jr., “On the mechanics of economic development,” Journal

of Monetary Economics, July 1988, 22 (1), 3–42.

Mahajan, Aprajit, Christian Michel, and Alessandro Tarozzi, “Iden-

tification of Time-Inconsistent Models: The Case of Insecticide Treated
Nets,” Manuscript, University of California at Berkeley 2019.

Marshall, Alfred, Principles of Economics, London: MacMillan, 1890.

Midrigan, Virgiliu and Daniel Yi Xu, “Finance and Misallocation:

Evidence from Plant-Level Data,” American Economic Review, February
2014, 104 (2), 422–58.

Moll, Benjamin, “Productivity Losses from Financial Frictions: Can Self-

financing Undo Capital Misallocation?,” American Economic Review, Oc-
tober 2014, 104 (10), 3186–3221.

, Robert M. Townsend, and Victor Zhorin, “Economic development,

flow of funds, and the equilibrium interaction of financial frictions,” Pro-
ceedings of the National Academy of Sciences, 2017, 114 (24), 6176–6184.

Monte, Ferdinando, Stephen J Redding, and Esteban Rossi-

Hansberg, “Commuting, Migration and Local Employment Elasticities,”
2018. American Economic Review, forthcoming.

Munshi, Kaivan and Jacques Myaux, “Social Norms and Fertility Tran-
sition,” Journal of Development Economics, 2006, 80 (1), 1–38.

51
Muralidharan, Karthik and Paul Niehaus, “Experimentation at Scale,”
The Journal of Economic Perspectives, 2017, 31 (4), 103–124.
, , and Sandip Sukhtankar, “General Equilibrium Effects of (Improv-
ing) Public Employment Programs: Experimental Evidence from India,”
NBER Working Papers 23838, National Bureau of Economic Research, Inc
September 2017.
Parker, Jonathan A., Nicholas S. Souleles, David S. Johnson, and
Robert McClelland, “Consumer Spending and the Economic Stimulus
Payments of 2008,” American Economic Review, October 2013, 103 (6),
2530–53.
Parker, Susan W. and Graciela M. Teruel, “Randomization and Social
Program Evaluation: The Case of Progresa,” The ANNALS of the Amer-
ican Academy of Political and Social Science, 2005, 599 (1), 199–219.
Paulson, Anna L. and Robert M. Townsend, “Entrepreneurship and
Financial Constraints in Thailand,” Journal of Corporate Finance, 2004,
10 (2), 229–262.
, , and Alexander Karaivanov, “Distinguishing Limited Liability
from Moral Hazard in a Model of Entrepreneurship,” Journal of Political
Economy, February 2006, 114 (1), 100–144.
Paweenawat, Archawa and Robert M. Townsend, “Village Economic
Accounts: Real and Financial Intertwined,” American Economic Review,
May 2012, 102 (3), 441–46.
and , “The Impact of Regional Isolationism: Disentangling Real and
Financial Factors,” PIER Discussion Papers 109, Puey Ungphakorn Insti-
tute for Economic Research May 2019.
Perla, Jesse and Christopher Tonetti, “Equilibrium Imitation and
Growth,” Journal of Political Economy, 2014, 122 (1), 52–76.
, , and Michael E. Waugh, “Equilibrium Technology Diffusion, Trade,
and Growth,” American Economic Review, January 2021, 111 (1), 73–128.
Piketty, Thomas, “The Dynamics of the Wealth Distribution and the In-
terest Rate with Credit Rationing,” Review of Economic Studies, 1997, 64
(2), 173–189.

52
Pomeranz, Dina, “No Taxation without Information: Deterrence and Self-
Enforcement in the Value Added Tax,” American Economic Review, 2015,
105 (8), 2539–2569.

Redding, Stephen J. and Esteban Rossi-Hansberg, “Quantitative Spa-

tial Economics,” Annual Review of Economics, 2017, 9 (1), 21–58.

Restuccia, Diego and Richard Rogerson, “Policy Distortions and Ag-

gregate Productivity with Heterogeneous Establishments,” Review of Eco-
nomic Dynamics, 2008, 11 (4), 707–720.

Robinson, Joan, “The Production Function and the Theory of Capital,”

Review of Economic Studies, 1953, 21 (2), 81–106.

, “Accumulation and the Production Function,” The Economic Journal,

1959, 69 (275), 433–442.

Romer, Paul M., “Endogenous Technological Change,” Journal of Political

Economy, 1990, 98 (5), S71–S102.

Rosenzweig, Mark and Christopher Udry, “External Validity in a

Stochastic World,” NBER Working Papers 22449, National Bureau of Eco-
nomic Research, Inc 2016.

Sampson, Thomas, “ Dynamic Selection: An Idea Flows Theory of Entry,

Trade, and Growth *,” The Quarterly Journal of Economics, 09 2015, 131
(1), 315–380.

Santaeulalia-Llopis, Raul and Diego Restuccia, “Misallocation and

Productivity,” NBER Working Paper 23128, National Bureau of Economic
Research February 2017.

Sarto, Andres, “Recovering Macro Elasticities from Regional Data,” Un-

published manuscript, Princeton University 2018.

Schultz, T.W., Transforming Traditional Agriculture, New Haven: Yale

University Press, 1964.

Silva, Dejanir and Robert Townsend, “Risk-taking over the Life Cy-
cle: Aggregate and Distributive Implications of Entrepreneurial Risk,”
manuscript, University of Illinois at Urbana-Champaign 2019.

53
Solow, Robert M., “Technical Change and the Aggregate Production
Function,” The Review of Economics and Statistics, 1957, 39 (3), 312–
320.
Sraer, David and David Thesmar, “A Sufficient Statistics Approach
for Aggregating Firm-Level Experiments,” NBER Working Papers 24208,
National Bureau of Economic Research, Inc January 2018.
Summers, Lawrence H., “Some skeptical observations on real business
cycle theory,” Quarterly Review, 1986, 10 (Fall), 23–27.
Suri, Tavneet, “Mobile Money,” Annual Review of Economics, 2017, 9,
497–520.
Todd, Petra E. and Kenneth I. Wolpin, “Assessing the Impact of a
School Subsidy Program in Mexico: Using a Social Experiment to Validate
a Dynamic Behavioral Model of Child Schooling and Fertility,” American
Economic Review, December 2006, 96 (5), 1384–1417.
Townsend, Robert M., The Medieval Village Economy: A Study of the
Pareto Mapping in General Equilibrium Models., Princeton: Princeton
University Press, 1993.
, “Financial Systems in Northern Thai Villages,” Quarterly Journal of
Economics, November 1995, 110 (4), 1011–1046.
and Kenichi Ueda, “Financial Deepening, Inequality, and Growth: A
Model-Based Quantitative Evaluation,” Review of Economic Studies, 2006,
73 (1), 251–293.
and , “Welfare Gains From Financial Liberalization,” International
Economic Review, 2011, forthcoming, p.
and Sergio S. Urzua, “Measuring the Impact of Financial Interme-
diation: Linking Contract Theory to Econometric Policy Evaluation,”
Macroeconomic Dynamics, 2009, 13 (S2), 268–316.
Udry, Christopher, “Risk and Saving in Northern Nigeria,” American Eco-
nomic Review, December 1995, 85 (5), 1287–1300.
Wolf, Christian, “The Missing Intercept: A Demand Equivalence Ap-
proach,” Unpublished manuscript, Princeton University 2019.

54
Zhang, Bin and Colin A. Carter, “Reforms, the Weather, and Produc-
tivity Growth in China’s Grain Sector,” American Journal of Agricultural
Economics, 1997, 79 (4), 1266–1277.

55