# 2019-100

Impacts of Extreme Climate Events on

Technical Efficiency in Vietnamese Agriculture

Yoro DIALLO*, Sébastien MARCHAND †, Etienne ESPAGNE ‡

March 2019

Please cite this paper as: DIALLO, Y., S. MARCHAND and E. ESPAGNE (2019), “Impacts of Extreme
Climate Events on Technical Efficiency in Vietnamese Agriculture”, AFD
Research Papers Series, No. 2019-100, March.

Contact at AFD: Etienne ESPAGNE ([email protected])

* Université de Clermont Auvergne, CNRS, CERDI.

† Université de Clermont Auvergne, CNRS, CERDI.
‡ AFD, CERDI.
Agence Française de Développement / French Development Agency

Papiers de Recherche de l’AFD

Les Papiers de Recherche de l’AFD ont pour but de diffuser rapidement les résultats de travaux en
cours. Ils s’adressent principalement aux chercheurs, aux étudiants et au monde académique. Ils
couvrent l’ensemble des sujets de travail de l’AFD : analyse économique, théorie économique, analyse
des politiques publiques, sciences de l’ingénieur, sociologie, géographie et anthropologie. Une
publication dans les Papiers de Recherche de l’AFD n’en exclut aucune autre.

L’Agence Française de Développement (AFD), institution financière publique qui met en œuvre la
politique définie par le gouvernement français, agit pour combattre la pauvreté et favoriser le
développement durable. Présente sur quatre continents à travers un réseau de 72 bureaux, l’AFD
finance et accompagne des projets qui améliorent les conditions de vie des populations, soutiennent la
croissance économique et protègent la planète. En 2014, l’AFD a consacré 8,1 milliards d’euros au
financement de projets dans les pays en développement et en faveur des Outre-mer.

Les opinions exprimées dans ce papier sont celles de son (ses) auteur(s) et ne reflètent pas
nécessairement celles de l’AFD. Ce document est publié sous l’entière responsabilité de son (ses)
auteur(s).

Les Papiers de Recherche sont téléchargeables sur : https://www.afd.fr/fr/ressources

AFD Research Papers

AFD Research Papers are intended to rapidly disseminate findings of ongoing work and mainly target
researchers, students and the wider academic community. They cover the full range of AFD work,
including: economic analysis, economic theory, policy analysis, engineering sciences, sociology,
geography and anthropology. AFD Research Papers and other publications are not mutually exclusive.

Agence Française de Développement (AFD), a public financial institution that implements the policy
defined by the French Government, works to combat poverty and promote sustainable development.
AFD operates on four continents via a network of 72 offices and finances and supports projects that
improve living conditions for populations, boost economic growth and protect the planet. In 2014, AFD
earmarked EUR 8.1bn to finance projects in developing countries and for overseas France.

The opinions expressed in this paper are those of the author(s) and do not necessarily reflect the
position of AFD. It is therefore published under the sole responsibility of its author(s).

AFD Research Papers can be downloaded from: https://www.afd.fr/en/ressources

AFD, 5 rue Roland Barthes

75598 Paris Cedex 12, France

 [email protected]

ISSN 2492 - 2846

Impacts of Extreme Climate Events on Technical Efficiency in Vietnamese Agriculture

Yoro Diallo, Université de Clermont Auvergne, CNRS, CERDI

Sébastien Marchand, Université de Clermont Auvergne, CNRS, CERDI

Etienne Espagne, AFD, CERDI

Abstract

The aim of this study is to examine farm household-level impacts of weather extreme events on
Vietnamese rice technical efficiency. Vietnam is considered among the most vulnerable countries
to climate change, and the Vietnamese economy is highly dependent on rice production that is
strongly affected by climate change. A stochastic frontier analysis is applied with census panel
data and weather data from 2010 to 2014 to estimate these impacts while controlling for both
adaptation strategy and household characteristics. Also, this study combines these estimated
marginal effects with future climate scenarios (Representative Concentration Pathways 4.5 and
8.5) to project the potential impact of hot temperatures in 2050 on rice technical efficiency. We
find that weather shocks measured by the occurrence of floods, typhoons and droughts
negatively affect technical efficiency. Also, additional days with a temperature above 31°C
dampen technical efficiency and the negative effect is increasing with temperature. For instance,
a one day increase in the bin [33°C-34°C] ([35°C and more[) lessen technical efficiency between
6.84 (2.82) and 8.05 (3.42) percentage points during the dry (wet) season.

Keywords: Weather shocks, Technical efficiency, Rice farming, Vietnam

JEL Classification: D24, O12, Q12, Q54

Original version: English

Accepted: March 2019

 Impacts of Extreme Climate Events on Technical
Efficiency in Vietnamese Agriculture
Yoro Diallo*, Sébastien Marchand†and Etienne Espagne‡

March 12, 2019

Abstract

The aim of this study is to examine farm household-level impacts of weather ex-
treme events on Vietnamese rice technical efficiency. Vietnam is considered among
the most vulnerable countries to climate change, and the Vietnamese economy is
highly dependent on rice production that is strongly affected by climate change.
A stochastic frontier analysis is applied with census panel data and weather data
from 2010 to 2014 to estimate these impacts while controlling for both adaptation
strategy and household characteristics. Also, this study combines these estimated
marginal effects with future climate scenarios (Representative Concentration Path-
ways 4.5 and 8.5) to project the potential impact of hot temperatures in 2050 on
rice technical efficiency. We find that weather shocks measured by the occurrence of
floods, typhoons and droughts negatively affect technical efficiency. Also, additional
days with a temperature above 31°C dampen technical efficiency and the negative
effect is increasing with temperature. For instance, a one day increase in the bin
[33°C-34°C] ([35°C and more[) lessen technical efficiency between 6.84 (2.82) and
8.05 (3.42) percentage points during the dry (wet) season.
Keywords: Weather shocks, Technical efficiency, Rice farming, Vietnam
JEL Classification: D24, O12, Q12, Q54.

*
Corresponding author: [email protected]; Université Clermont Auvergne, CNRS, CERDI
†
[email protected]; Université Clermont Auvergne, CNRS, CERDI
‡
[email protected]; AFD, CERDI. Acknowledgements: This work is part of the GEMMES Vietnam
project, launched by the Agence Française de Développement in order to study the socio-economic impacts
of climate change and adaptation strategies to climate change in Vietnam. It belongs to the ECO package.
We are thankful to Nicolas De Laubier Longuet Marx, Olivier Santoni, Ngo Duc Thanh and the GEMMES
team for helpful comments.

1
1 Introduction
Given the increasing awareness about climate change and the growing concern about
its downside consequences, the question of a quantitative assessment of the economic
consequences of climate change is of great importance. This issue is particularly topical
in countries heavily exposed to the risks of weather variability and climate change like
Vietnam which is among the countries most vulnerable to climate change according to the
Climate Change Knowledge Portal (CCKP) of the World Bank1 . Since the country lies
in the tropical cyclone belt, it is heavily exposed to climatic-related risks like droughts,
floods, tropical storms (typhoons), rising sea level and saltwater intrusion (Bank, 2010).
This vulnerability is also increased by the topography of the country. Vietnam is a
long narrow country consisting of an extensive coastline of more than 3,000 km long
subjected to accelerated erosion and rising sea level. It contains two major river deltas
(the Mekong delta in the South and the Red River delta in the North) highly exposed to
floods and rising sea level (Dasgupta et al., 2007), which concentrate a high proportion
of the country’s population and economic assets such as rice farming, and mountainous
areas on its eastern and northeastern borders.
In Vietnam, rice farming has played a central role in economic development since
1980 and the beginning of market and land reforms2 . Paddy rice is by far the main crop
produced in Vietnam and employs two thirds of total rural labor force. However, it is
also one of the most climate-change affected sectors due to its direct exposure to, and
dependence on, weather and other natural conditions (Bank, 2010). The ongoing climate
change and its related effects have and will have significant impacts on rice production
and farmer livelihoods. From census data between 2010 and 2014, and weather data
on temperature and precipitation, this study examines farm household-level impacts of
weather shocks, defined as extreme events such as extreme temperatures, floods, droughts
and typhoons, on agricultural productivity in the Vietnamese rice farming3 .
This study contributes to the growing literature that uses farm-level panel data (here
the Vietnam Household Living Standard Survey (VHLSS)) coupled with finely-scaled cli-
mate data to estimate the weather change impacts, here on Vietnamese rice farming (Yu
et al., 2010; Trinh, 2017). The use of a panel structure allows to control for time-invariant
omitted variables correlated with weather extreme events that may confound the climatic
1
Source: the CCKP website.
2
These last years, while the contribution of Vietnamese farming to national GDP has become less
important (from about 40% in 1990 to about 18% in 2016), rural areas still generate employment and
income for a significant part of the population (Bank, 2010). In 2016, 66% of the population live in rural
areas where 43% of the country’s active workforce is employed.
3
This study investigates weather impacts rather than climate impacts (Auffhammer et al., 2013).
More precisely, the former is defined as the conditions of the atmosphere over a short time horizon while
the latter is the variability of the conditions of the atmosphere over a relatively long period. Thus, the
interpretation of the coefficients associated with climatic variables have to be interpreted as weather
shocks in the short run and climate change in the long run.

2
effect in pure cross-sectional studies (Blanc and Schlenker, 2017). Also, we study two
weather effects on rice farming from a Stochastic Frontier Analysis (SFA). We first esti-
mate the effect of weather trend, defined as the mean daily temperature and the mean
daily precipitation, on rice farming output. Then, we assess the effect of weather extreme
events, measured by the occurrence of floods, typhoons, droughts and extreme tempera-
tures, on rice farming productivity defined as technical efficiency4 . As a result, we find
that weather shocks measured by the occurrence of floods, typhoons and droughts nega-
tively affect technical efficiency. Also, daily temperatures above 31°C dampen technical
efficiency in the dry season, an effect which is increasing with temperature. For instance,
a one day increase in the bin [33°C-34°C] lowers technical efficiency between 6.84 and 8.05
percentage points. Simulation results show dramatic drops in technical efficiency after
2040. In the case of the RCP8.5 scenario, technical efficiency collapses from 40%, while
it stabilizes in the RCP4.5 scenario around 10% below the reference period5 .
The remaining of the paper is organized as follows. Section 2 presents the literature
related to the climate-agriculture nexus. Section 3 details the rice sector and climate
conditions in Vietnam. Sections 4, 5 and 6 present respectively the empirical methodology,
data and descriptive statistics, and econometric results. Section 7 gives the results from
simulations and Section 8 concludes.

2 Literature reviews
This section reviews the theoretical and empirical studies that estimate the economic
impact of climate change on agriculture. The literature can be divided between the long-
run climate effect approach using the Ricardian hedonic model with cross-sectional data
(see Mendelsohn and Massetti (2017) for a discussion of main advantages and weaknesses
of this approach), and the weather-shock approach using Ricardian hedonic model with
panel data (see Blanc and Schlenker (2017) for a discussion of main advantages and
weaknesses of this approach).
The first approach consists in examining how the long-run climate (the distribution of
weather over 30 years) affects the net revenue or land value of farms across space using
the Ricardian method (also called the hedonic approach). The principle of this method is
to estimate the impact of climate on agricultural productivity by regressing net revenue
or farmland value (use as a proxy for the expected present value of future net revenue)
on climate in different spatial areas. The model assumes that competitive farmers are
profit-maximizing agents. Farmers choose an optimal combination of inputs and output
4
To our knowledge, only Key and Sneeringer (2014) use a SFA methodology to study heat stress on
technical efficiency on dairy production in United States.
5
We only used one CORDEX-SEA model for climate projections in this version of the paper, which
limits the level of confidence we can have for these projections. We will use all the existing simulations
as soon as they are available, so that we can discuss the uncertainty issue about future climates.

3
to maximize net agricultural income, subject to the exogenous variable such as climatic
conditions that are beyond the farmer’s control. Put differently, if climate is different, the
farmer has to adapt his production and choose a different output (crop switching) and
different inputs (new pesticides for instance). This is probably the main advantage of the
Ricardian approach that allows to capture long-run adaptation to climate. So the goal is
to regress net revenue on different arrays of climates to estimate the impact of climate.
According to Mendelsohn and Massetti (2017), this approach has been used in 41 studies
over 46 countries. The first attempt is Mendelsohn et al. (1994) who estimate the impact
of temperatures on land prices in 3,000 counties in the United States. They found from
simulation based to the econometrically estimated impacts of temperature that global
warming may have economic benefits for the U.S. agriculture.
This initial approach has been then improved in different ways to take into account
many empirical issues. One of them concerns the measure of climate. Most studies used
seasonal climate variables but the type of variable changes from one study to another.
Some studies include mean seasonal temperature and rainfall (Mendelsohn et al., 1994;
Schlenker et al., 2005) while other use the degree days over the growing season that are
the sum of temperatures above a floor (Schlenker et al., 2006; Deschênes and Greenstone,
2007)6 .
Another important empirical issue is related to the cross-sectional nature of the method.
In fact, many existing studies estimate a Ricardian model using data for a single year or
two. However, a main disadvantage of cross-sectional data is potential omitted variables
that might bias the results since average climate over a long period is not random across
space. For instance, Dell et al. (2009) find that poorer countries tend to be hotter. But
this relationship can be considered as spurious correlation if there are some omitted vari-
ables correlated with climate that can explain income (institutions for instance). The
model has to control for these potential omitted variables. Two solutions have been de-
veloped in the literature to avoid omitted variable bias. The first solution is to account for
all factors that are both correlated with climate and the impacted farmland values. One
first example is irrigation that is correlated with temperature. For instance, Schlenker
et al. (2005) show that access to subsidized irrigation water is both capitalized into farm-
land values and correlated with hotter temperatures. This means that the impacts of
irrigation has to be control while estimating the impact of temperature on land value.
If not, the regression estimates not only the direct effect of temperature, but also the
beneficial effect of access to irrigation water (which is positively correlated with higher
temperatures). To resolve this issue, Schlenker et al. (2005) separate irrigated and rain-
fed farms and estimate models for each sample. Another solution is the one implemented
6
See Massetti et al. (2016) for a discussion of these two approaches and the pitfalls of the degree days
approach with the Ricardian method. Note that this issue concerns also the weather approach discussed
infra.

4
by Kurukulasuriya et al. (2011). The authors first estimate the probability of making the
irrigation choice and then estimate the conditional Ricardian model given the choice of
making irrigation. However, this solution can never completely eliminate the possibility
of omitted variables. In fact, there might always be an additional control variable (e.g.
soil quality) that is correlated with climate (e.g. temperature) but unfortunately not
correlated with the other control variables (e.g. irrigation) included in the specification.
The second solution may address this concern and consists in using panel data into
the Ricardian model (i.e., estimate long-run climate impact) (Deschênes and Greenstone,
2007). Panel data allow for the use of fixed effects, which control for any time-invariant
confounding variation. However, in a model with fixed effects, it is impossible to estimate
the effect of the long-run climate averages because climate has no temporal variation.
However, while Deschênes and Greenstone (2007) show that the Ricardian results are
not robust when estimated as a series of repeated cross sections, Schlenker et al. (2006);
Massetti and Mendelsohn (2011) provide evidences that the Ricardian model is stable
when estimated with panel methods. Massetti and Mendelsohn (2011) for instance provide
two robust methods to estimate Ricardian functions with panel data: (1) a two-stage
model based on Hsiao (2014) where agricultural outcome is regressed on time varying
variables using the covariance method with fixed effects and then, in the second stage,
the time-mean residuals from stage 1 are regressed on non-varying time variables such as
climate variables (also used by Trinh (2017)); (2) a single stage “pooled” panel model.
While the Hsiao model is less vulnerable to the omitted variable bias than the pooled
panel model, it is less efficient than the pooled panel model estimated in one step. The
main result of Massetti and Mendelsohn (2011) is that the overall effect of climate change
is likely to be beneficial to U.S. farms over the next century.
The second main approach is the weather-shock approach using Ricardian hedonic
model with panel data (Schlenker and Roberts, 2009; Schlenker et al., 2013; Deryugina
and Hsiang, 2017). The starting point of this approach is to take advantage of fine-scaled
weather data in both time and space to detect for instance nonlinearity through the large
degree of freedoms that give panel data. For instance, Schlenker and Roberts (2009) find
a non-linear relationship between temperature and U.S. crop production. Beyond the
respective thresholds of 29°, 30° and 32°, the temperature generates major damage on
wheat, soybean and cotton yields respectively. Also, this approach allows to avoid the
omitted variable bias by controlling for fixed effects. Another advantage of this approach
is to account for short-term adaptation. Although panel analysis allows for spatial and
temporal heterogeneity, it is not free of limits (Blanc and Schlenker, 2017). One of them
is the consideration of spatial autocorrelation in crop yields and climatic variables which
is necessary in order to limit the estimation bias. Chen et al. (2016) take this criticism
into account in their analysis of the link between climate change and agricultural sector
in China. They find that Chinese agricultural productivity is affected by the trend in

5
climate and the existence of a non-linear and U-inversed shape between crop yields and
climate variability.
Our study uses the weather-shock approach with panel data. However, instead of
using a Ricardian hedonic model, we follow Key and Sneeringer (2014) and estimate the
relationship between weather and rice farming productivity defined as technical efficiency
using a stochastic production frontier model.

3 Rice production and climate condition in Vietnam

3.1 Rice production in Vietnam
Since the beginning of the Vietnam’s Đổi Mới (renovation) process launched in 1986, Viet-
nam has witnessed unprecedented transitions from planned and collectivized agriculture
to market and household-based farming.
The market reform periods of Vietnamese rice farming began with the output contracts
period (1981–87) which launched the move to de-collectivize agriculture (Kompas et al.,
2002, 2012). It was the first attempt towards private property rights. Farmers were
allowed to organize production activities privately but the most part of rice production
had still to be sold in state markets at low state prices. However, private domestic
markets emerged for some portion of output sold (approximately 20%). This period was
thus characterized by a “dual price” system (a low state price and a competitive market
price) with strict state controls.
From 1988 on, the period of trade and land liberalization began with the aim to
establish effective private property rights over both land (initially 10–15 year leases) and
capital equipment while restrictions on farm size and prohibitions against the removal of
land from rice production were maintained. In 1990 the central government abolished the
dual price system and rice was authorized to be sold on competitive domestic markets.
However, while those reforms were intended to incite farmers to invest, in practice, farmers
were reluctant to undertake long-term investments because the land-use rights were not
seen as secure as they were not tradable. Consequently, the government passed a new
Land Law in 1993. This law extended the lease period to twenty years for land used to
produce rice (increased to 30 years in 1998 revisions) and allowed farmers to transfer,
trade, rent, mortgage and inherit their land-use rights (Scott, 2008). Also, from 1993,
farmers could now sell rice freely in international markets.
From the mi-1990s, land and market reforms implemented from 1981 allowed the
decentralization of production decisions at the farmer’s level and guaranteed that all farm
income (after tax) was retained by the farmer. Individual efforts were rewarded in order
to push farmers to invest and produce more. More recently, beyond these market and
land reforms, government implemented a rice policy helping to increase yield through the

6
development of rice varieties, large investment in irrigation (roughly 85% of rice area are
applied with active irrigation drainage system), the support in case of emergency cases,
the ease of credit access, input support (reducing valued-added tax for key inputs as
fertilizers), etc.
As a consequence, Vietnam has become the fifth rice producer in the world with a
total production of 42.76 millions of tons per year and a yield of 5.55 tons per hectare in
2017, a lot more than annual 12.4 millions of tons produced and the yield of 2.19 tons per
hectare in 1980, and a leading world exporter (about 7 millions tons)7 .
Regarding the geographical distribution of rice production, rice area covers roughly
7,8 millions hectares (23% of total land area and 82% of arable land) owned by 9 millions
of households (accounting for more than 70% of rural households) so that the average
farm size is below one hectare. Rice area is located mostly in the Mekong River Delta
(about 55% of total rice production (23 millions of tons produced in 2017) and 90% of
rice exports) followed by the Red River Delta in the northeast (about 15% of total rice
production) and the north-central coast8 .
Despite the increase of the yield in rice production these last decades, some important
pitfalls remain. For instance, rural inputs and land markets or access to agricultural ex-
tension services and farm credit remain still far less developed in some provinces, trapping
farmers into poverty (Kompas et al., 2012). Also, the expansion of rice production for the
last thirty years was reached by focusing on quantity increases. The abuse of chemical
inputs (Berg and Tam, 2012) produce important environmental damages in terms of soil
fertility or depletion of fishery resources for instance. Besides, past international successes
of Vietnamese rice production was based mainly on high production of low quality rice
sold at very low price on international markets, a strategy that the recent increase in in-
put prices (fertilizer, fuel, and labor) could well jeopardize (Demont and Rutsaert, 2017).
Vietnamese rice farming now has to deal with significant issues both at national and inter-
national levels. At the national level, Vietnamese farming has to deal both with poverty
alleviation of rice households (by encouraging crop diversification on rice), food security
(feeding both Vietnamese with good quality rice products) and environmental preserva-
tion (by promotion organic rice farming, soil preservation, etc.) (Tran and Nguyen, 2015).
These national challenges have also implications at the international level. The compet-
itiveness of Vietnamese farming depends on the performance of farmers and companies
to deliver rice products with reliability regarding the quality (i.e. switching to high value
rice to follow change in world demand), safety and sustainability of the products supplied
(Demont and Rutsaert, 2017). Beyond these national and international issues, the ongo-
ing climate change is also an imperious issue that Vietnamese have to face in order to
preserve their rice production and the livelihood of millions of farmers.
7
Data come from FAOSTAT.
8
Data come from GSO, the general statistics office of Vietnam.

7
3.2 Climate condition
Seasonal variability of temperature and precipitation

Due to the diversity in topography, it is likely that the impacts of climate change will be
different depending both on the place and the months of the year. The curves in figure 1
transcribe the seasonal variations of the temperatures and precipitations according to the
months of the year. High temperatures are observed from May to October (the average
is 27.12°C) and lower temperatures from November to April (the average temperature
is 22.73°C). In addition, a greater instability of temperatures appears in the middle of
the year (an average amplitude of 2°C). Similarly precipitations are higher during the
period from May to October (an average rainfall of 238.20 mm) and relatively low from
November to April (an average rainfall of 68.93 mm). These observations allow us to
distinguish two major climatic seasons in Vietnam: a dry season (November to April)
and a wet season (May to October).
As in Hsiang (2010) and Trinh (2017), we use these seasonal temperature and precip-
itation variables to measure the impact of seasonal variability to test the dependence of
technical efficiency on the periodic occurrence of weather shocks. However, we are aware
that these time intervals can vary from one region to another throughout the country.

Figure 1: Seasonal variation of precipitation and temperature (1950-2015)

Seasonal variation of precipitation

500
400
(mean) preM
200 300
100
0

1 2 3 4 5 6 7 8 9 10 11 12

Month

Seasonal variation of temperature

30 25
(mean) tmpM
20 15

1 2 3 4 5 6 7 8 9 10 11 12

Month

Source: authors from CHIRPS and MODIS data

8
Temperature and precipitation trends by Vietnamese regions (1950-2015)

Figures 2 and 3 represent the average trends in temperatures and precipitations over
the period 1950-2015 at sub-national levels. There is a strong spatial heterogeneity in
the variability of climatic conditions. The Mekong region in the south has experienced
a more pronounced global warming which is manifested by a mean annual temperature
trend increase of 0.02°C corresponding to an increase of 1.3°C over the period 1950-2015.
However, the temperature in the Red Delta region in the north-east remains pretty stable.
On the other hand, there is an average decrease in precipitations in the south, unlike
in the north and center where there has been a relative increase in monthly precipitations.
In addition, Figures A1 and A2 in Appendix respectively show the evolution of the level
of temperatures and precipitations according to the month of the year. In these figures,
we are able to perceive the occurrence of short-term climatic shocks by month and by
year.

9
Figure 2: Average annual temperature trend increase over the period 1950-2015

Source: authors from MODIS data

10
 Figure 3: Average annual rainfall trend increase over the period 1950-2015

Source: authors from CHIRPS data

4 Empirical methodology
The link between agricultural productivity in rice farming and weather (trends and shocks)
is analyzed through a two-step approach. Agricultural productivity is first defined in terms
of technical efficiency calculated from a stochastic production frontier model in which
weather trend is also used to explain agricultural production. In the second step, the
estimated technical efficiency is explained by weather extreme events. Before presenting
the econometric model, we present the conceptual framework on which the econometric
analysis relies.

11
4.1 Conceptual framework
4.1.1 Definition of efficiency

Farrell (1957) defines agricultural productivity as productive efficiency which is the ability
of producers to efficiently use the available resources, called inputs hereafter, in order to
produce maximum output at the minimum cost. It differs from effectiveness that refers
to the degree of achievement of a desired goal. In addition, productive efficiency is the
combination of allocative efficiency (AE hereafter) and technical efficiency (TE hereafter).
AE is based on the optimal combination of inputs given their market prices, production
technology and the market prices of the output. It necessarily leads to the maximization
of profit or even the minimization of production costs. TE refers to the performance of
the producer to avoid waste of inputs to produce. This waste can be avoided in two
ways: either by reducing the quantity of inputs for the same level of production (the
input-oriented measure of TE), or by increasing the production for the same given level of
inputs (the output-oriented measure of TE). While AE is estimated from a profit function
or a cost function, TE is estimated from a production function. In this study, we work
on TE because we do not have price informations on inputs and output.

4.1.2 Estimation of technical efficiency

Technical efficiency is estimated under three auxiliary hypotheses regarding the choice
of the estimation method, the choice of the production function and the choice of the
functional form of TE over time.
Firstly, the estimation of TE relies on either the non-parametric method or the para-
metric method. The principle of the non-parametric method also called data envelopment
analysis (DEA) is to impose no restriction on the distribution of inefficiency, no behavioral
assumptions (goal of profit maximization) unlike the parametric method which is based
on the methods and techniques of econometric estimation. However, DEA imposes to con-
sider that all shocks to the value of output have to be considered as technical inefficiency
whereas some factors (ex. climatic conditions) are not related to producer behavior and
can directly affect the production frontier. This explains why parametric method is often
preferred in the literature, by using stochastic production functions called stochastic fron-
tier analysis (SFA)(Aigner et al., 1977; Meeusen and van Den Broeck, 1977). This method
allows the error term to have two components: a negative component that measures ineffi-
ciency and an idiosyncratic error that represents all other idiosyncratic shocks. However,
imposing on the inefficiency component to be negative requires strong assumptions about
its distribution law. The most used distributions are the half-sided normal law, the ex-
ponential law and the normal truncated law (Stevenson, 1980). The use of the half-sided
normal law and the exponential law assumes that the majority of the observation units

12
are efficient relative to the truncated normal law9 .
Secondly, the form of the production function has to be chosen in a SFA technique. In
microeconomics, the production function expresses the relationship between outputs and
inputs. Its functional representation has to respect certain properties10 , taking into ac-
count the presence or not of economies scale and the nature of the substitutability between
inputs. The production function is often modelled using a Cobb-Douglas form (Cobb and
Douglas, 1928) or a transcendental logarithmic (“translog”) specification (Christensen
et al., 1971) in the literature. The Cobb-Douglas form is often preferred because it gives
convex and smoothed isoquantes. However, it is based on strong assumptions such as the
constancy of the elasticities and the hypothesis that all the elasticities of substitutions are
supposed to be equal to -1. More flexible forms of production such as the translog form
have emerged by not imposing restrictions on the production technology, especially with
regard to the substitution between inputs. In our analysis, we estimate the production
function by considering the translog form11 .
Thirdly, the estimation of TE in panel model implies to model the functional form
of TE over time. The first models are those of Pitt and Lee (1981) and Schmidt and
Sickles (1984) where inefficiency is supposed not to vary over time. This type of model
is comparable to a fixed effect in panel model. However, these models are based on very
strong assumptions. On the one hand, the model is valid under the assumption that
the inefficiency is uncorrelated with the inputs used to estimate the production function.
On the other hand, inefficiency has not to vary over time. Thus, other models emerged
to allow temporal variation of TE. However, the problem that has arisen concerns the
functional form of the temporal variation of inefficiency. Cornwell et al. (1990) proposes
the CSS model in which inefficiency varies with time according to a quadratic form.
While the temporal variation of TE is not necessarily quadratic, this hypothesis is very
restrictive. Battese and Coelli (1992) and Kumbhakar et al. (2000) develop a model in
which the temporal variation of the inefficiency term takes an exponential form. Lee and
Schmidt (1993) provide more flexibility in the form of temporal variation in inefficiency.
Their time-varying fixed-effects model does not impose restrictions on the functional form
of inefficiency. In other words, inefficiency is supposed to vary over time without imposing
a particular functional form on this variation. This model is particularly advantageous
for studies with a fairly large number of observation units and a relatively short time
dimension. This advantage also makes possible to circumvent the concern of incident
parameters (Chamberlain, 1979) potentially present with panel models12 .
9
The mode of the semi-normal law and the exponential law is equal to 0.
10
The production frontier requires monotonicity (first derivatives, i.e., elasticities between 0 and 1 with
respect to all inputs) and concavity (negative second derivatives). These assumptions should be checked
a posteriori by using the estimated parameters for each data point.
11
And the Wald test applied to interactive terms confirm the using of this model.
12
Other models such as Greene (2005) make it possible to dissociate the individual fixed or random
effect from TE. However, the large number of parameters to be estimated in these models is still subject

13
 In this study, we implement the stochastic frontier analysis by using both the Cobb-
Douglas and the translog production functions following the literature as well as the model
developed by Lee and Schmidt (1993) given that the time dimension of our base is quite
short (three years), while the number of farms is large (2,592 households).

4.2 Econometric strategy

We now apply the conceptual framework explained above to an econometric model in
agricultural production to firstly estimate TE and secondly to estimate the effects of
weather shocks on TE.

4.2.1 First step: Estimation of Technical Efficiency

Consider a farmer i at time t who uses x inputs (defined later) to produce rice defined by
y. The production function can be written as follows:

yi,t = f (xi,t ), (1)

where f is a function that defines the production technology. The rational producer
aims at maximizing his total production of rice while minimizing the total use of his
inputs. On the frontier, the farmer produces the maximum output for a given set of
inputs or uses the minimum set of inputs to produce a given level of output. Thus, the
definition of the production frontier and the estimation of technical efficiency depend on
the type of orientation: input-oriented or output-oriented. We use the output-oriented
measure of technical efficiency (more output with the same set of inputs) that gives the
technical efficiency of a farmer i as follows:

T Ei,t (x, y) = [maxϕ : ϕy ≤ f (xi,t )]−1 , (2)

where ϕ is the maximum output expansion with the set of inputs xi,t .
The output-oriented measure of technical efficiency defined by Eq. 2 is estimated
under three auxiliary hypotheses.
Firstly, Eq.1 is applied to an econometric model as follows:

yi,t = f (xi,t , β).e−Ui,t (3)

where yi is a scalar of output, xi is a vector of inputs used by farmers i=1,…,N, f (xi ; β)

is the production frontier and β is a vector of technology parameters to be estimated. Ui
are non-negative unobservables random variables associated with technical inefficiency
that follow an arbitrary half-sided distribution law.
to the incidental parameter concern.

14
 Secondly, we use a stochastic frontier analysis in which we assume that the difference
between the observed production and maximum production is not entirely attributed to
TE and can also be explained by idiosyncratic shocks such as weather. Eq.3 becomes:

yi,t = f (xi,t , β).e−Ui,t .eVi,t , (4)

where Vi,t represent random shocks which are assumed to be independent and iden-
tically distributed random errors with a normal distribution of zero mean and unknown
variance. Under that hypothesis, a farmer beneath the frontier is not totally inefficient
because inefficiencies can also be the result of random shocks (such as climatic shocks).
Since T Ei,t is an output-oriented measure of technical efficiency, a measure of T Ei,t is:

obs
yi,t f (xi,t , β).e−Ui,t .eVi,t
T Ei,t = max
= . (5)
yi,t f (xi,t , β).eVi,t

Thirdly, the production function is modeled by a translog specification. The general

form of the translog is as follows:

∑
4 ∑
4 ∑
4
ln(yi,t ) = β0 + βj ln(Xij,t ) + 0.5 βjk ln(Xij,t )ln(Xik,t ) − Ui,t + Vi,t , (6)
j=1 j=1 k=1

where i = 1, N are the farmer unit observations; j, k = 1, ..., 4 are the four applied inputs
explained later; ln(yi,t ) is the logarithm of the production of rice of farmer i at time t;
ln(Xij ) is the logarithm of the jth input applied of the ith individual; and βj ,βjk are
parameters to be estimated.
The final empirical model estimated in the translog case is twofold. It does first not
take into account the weather variables as follows:

ln(Ricei,t ) = β0 + β1 ln(f amlabori,t ) + β2ln(hirlabori,t ) + β3ln(capitali,t )

+β4ln(runningcostsi,t ) + β5ln(f amlabori,t )2 (7)
+... + β9ln(f amlabori,t )ln(hirlabori,t ) + ... + αt − Ui,t + Vi,t ,

Rice is the output defined as the total rice production over the past 12 months.
f armlabor and hirlabor define respectively family labor (in hours) and hired labor (in
wages). capital is the total value of investment in machinery and runningcosts is the value
of running costs (e.g. fertilizers and irrigation). Both output and inputs are normalized
by farm land area devoted to rice farming. More information can be found in Table A1 in
the Appendix. t refers to the year of the last three surveys used in this study (2010-2012-
2014)13 . Each household i has been surveyed two or three times. αt measures temporal
13
The year 2016 will be added in a future version.

15
fixed effects which represent the unobserved characteristics common to each region and
which vary over time and which affect agricultural yields (e.g. inflation, macroeconomic
policy, price shock of commodities ...). In addition, this variable takes into account the
possibility of neutral technical progress in the sense of Hicks.
Then, the empirical model integrates both irrigation and the weather variables as
follows:

ln(Ricei,t ) = β0 + β1 ln(f amlabori,t ) + ... + β15 irrigi,t + β16 climm,t + αt − Ui,t + Vi,t ,
(8)

where irrig is a dummy variable (1 = irrigation) and clim represents both the av-
erage daily temperature over the production period and the total precipitation over the
production period in the municipality m14 .
Fourthly, the functional form of TE over time is defined following Lee and Schmidt
(1993) as follow:

Ui,t = δt ∗ γi ⩾ 0, (9)

where δt encompasses the parameters that capture the variability of technical ineffi-
ciency. In this model both the components of δt and γi are deterministic. Although Lee
and Schmidt (1993) estimated this model without any distributional assumptions on γi .
This specification makes the temporal variability of inefficiency quite flexible. γi is the
measure of the technical efficiency of producer i.
Finally, efficiency scores are computed from the estimation of Ui,t in Eqs. 8 and 9 as
follows:

T Ei,t = e−Ui,t (10)

The maximum likelihood estimator is used to estimate the technical efficiency under
a half-sided normal law.

4.2.2 The second step: the effects of weather shocks on TE

Once TE is computed from the first stage, it is used as dependent variable in the second
stage as follows:

T Ei,t = α0 + α1 .Zi,t + α2 .Wi,t + ϵi,t , (11)

Equation 11 is estimated with the fixed effects model. W includes control variables
14
The production period is defined as the twelve last months before the household is surveyed. Climatic
variables are available at municipality level so that all household living in the same municipality share
the same climatic variables.

16
(household size and gender, education level and age of the household head) and Z repre-
sents weather shocks15
Weather shocks are considered as short-term extreme climatic events measured by the
occurrence of extreme temperatures and natural disasters (flood, typhoon and drought).
We define a weather shock in terms of temperature in two ways. Firstly, we follow
Schlenker and Roberts (2009) who use data on daily precipitations and temperatures to
calculate the Growing Degree Days (GDD) index. This measure consists in calculating
the optimal daily temperature and the optimal daily precipitation required for the growth
of each crop. Thus, climate variability is captured by a deviation of temperature or
precipitation from these optimal thresholds16 . GDD can be computed as follows:

∑
N
GDDbase,opt = DDi , (12)
i=1


0 if Ti < Tlow or Ti > Tup
DDi = (13)
T − T base if Tlow ≤ Ti ≤ Tup
i

Where i represents day, and Ti is the average of the minimal (Tmin ) and maximal
(Tmax ) temperature during this time-span. Tlow and Tup are respectively the lower and
upper thresholds of a given temperature range. DD represents the degree day of each day
during the growing stage. N is the number of days within a growing season. However, this
way to compute GDD has some limitations. Indeed, through these equations, we note that
below the minimum threshold or beyond the maximum threshold, the temperature makes
no contribution to the development of the plant. Thus, we do not capture the negative
effect of extreme temperatures on the plant’s development process. To tackle this issue,
our strategy is close to that of Schlenker and Roberts (2009) and Chen et al. (2016). Here,
the GDD is calculated in terms of days where the temperature is in an interval considered
optimal for the growth of the plant. The days when the temperature is outside this
range are considered harmful to the plant. It will be called Killing Growing Degree Days
(KGDD). We follow Sánchez et al. (2014) to define the optimal temperature thresholds
for rice in Vietnam. The authors make a meta-analysis on the different temperature
thresholds (Tmin , Topt and Tmax ) that rice needs according to the phase of the cycle of its
growth. Thus, we have identified the temperature levels of 10°C and 30°C respectively
as the minimum and maximum temperature levels necessary for the development of rice
culture throughout its growing cycle. In our climatic base, the average of the numbers of
days where temperature is below 10°C during the growing season for rice is equal to one
15
More information in Table A1 in Appendix.
16
There also exist several works (McMaster and Wilhelm (1997), Lobell et al. (2011) and Butler and
Huybers (2013)) which propose a different way to compute GDD

17
day. Its range of temperature is not considered. Then, two measures of weather shocks
are considered: KGDD heat dry and KGDD heat wet which are respectively the number
of days when the temperature is above 30°C during either dry season or wet season.
In our regression, we decompose the KGDD by 1°C bin interval ([30-31[, [31-32[, [32-
33[, [33-34[, [34-35[ and [35-plus[) for both dry and wet seasons17 . Then, for each interval,
we compute the following variable:

∑
N
IT[a−b[ = DDi , (14)
i=1


0 if T < a or T ≥ b
i i
DDi = (15)
1 if a ≤ T < b
i

Secondly, we define other weather shocks by using the occurrence of floods, typhoons
and droughts over the production period.

5 Data and descriptive statistics

The data used in this study are derived from both socio-economic and climate data.

5.1 Socio-economic data

The socio-economic data come from the Vietnam Household Living Standard Survey
(VHLSS) provided by the GSO (General Statistics Office of Vietnam). The main objective
of VHLSS is to collect data at the household and commune level to define and evaluate
national policies or programs that include assessing the state of poverty and inequality of
individuals. The survey questionnaire is administered at two levels.
On the one hand, a questionnaire is administered at the household level. It collects
data on agricultural production (outputs and inputs), income (farming and off-farming)
and socio-demographic characteristics of individuals within a household (gender, age, level
of education, ...). In this study, variables in monetary values (i.e. output and some inputs)
are calculated based on the 2010 consumer price index. Table 1 gives descriptive statistics
of variables used in this study. Inputs and output variables are normalized by the area
allocated to rice production. The average rice production is 2,420 VND per squared
meter with a very strong heterogeneity (minimum = 190 VND/m2 – households called
“small producer” ; maximum = 25,540 VND/m2 – households called “large producers”).
Regarding socio-demographic variables, we note from Table 1 that women are very poorly
represented in rice farming (only 16% of all household heads are women). Also, only 1%
Variables [34 − 35]Dry and [35 − plus[Dry do not exist because there are no days where average daily
17

temperature is above 34°C during the dry season.

18
of the household heads reached the university level, compared to 27% at no level, 26% at
the primary level and 46% at the secondary level. Also the average age of the household
head is 48 years with a high degree of dispersion (a standard deviation of about 13 years).
The average household size is about four persons with a standard deviation of 1.54.
On the other hand, there is a questionnaire at the municipal level. It is administered
to the local authorities of each municipality. It collects information on infrastructure
(schools, roads, markets...) and economic conditions (work opportunities, agricultural
production...) of each municipality. Through this questionnaire, we get information on
the occurrence of extreme events by category (typhoons, floods, cyclones ...).
All of these questionnaires collect data from 9,000 representative households each year.
This allows us to build our database from the last three VHLSS surveys (2010-2012-
2014)18 . In our analysis, we retain only households that produced rice and are followed
at least twice over the three years of surveys. In total, there are 2,592 households and
5,894 observations in the database.

5.2 Climate data

The climate data used in this study are daily temperatures and precipitations. These
data come from the Climate Prediction Center (CPC) database developed by the Na-
tional Oceanic and Atmospheric Administration (NOAA). It provides historical data on
maximum and minimum temperature and precipitation levels for a grid of 0.5 degree by
0.5 degree of latitude and longitude. The daily average temperature (precipitation) can
be generated from these Tmax and Tmin . Thus, at each geographic coordinate (longitude
and latitude), a mean precipitation value is associated with an average temperature value
per day, month and year.
However the geographical coordinates of this base did not correspond exactly to those
which we had for the municipalities of Vietnam. To overcome this problem, we use
the STATA geonear command. For each given municipality, we compute the average
precipitation and daily temperature of the four nearest localities weighted by the inverse
of the squared distance.
The Figure 5 gives the distribution of minimum, maximum and average daily temper-
atures over the period 2010-2014. There is a strong dispersion of the daily temperature
levels in Vietnam. The minimum daily temperature is between -1.4°C and 32°C while the
maximum temperature is up to 41°C. The averages of the minimum and maximum daily
temperatures are respectively 22.3°C and 28.8°C. Tavg is the average of the daily minimum
and maximum temperature levels. It is worth noting that extreme temperatures (above
30°C) is not negligible in Vietnam.
18
There are data for surveys before 2010. However, these data are not usable because the sampling
method and questionnaire content changed in 2004 and 2010.

19
 Figure 4: Distribution of tmin, tmax and tavg

.15
.1
Density
.05
0

0 10 20 30 40

Tmin
Tmax
Tavg

Source: authors from CPC data

Finally, from Table 1, we note an average cumulative precipitation of 1,475.24 mm

and an average daily temperature of 24.82°C over the period (2010-2014).

Table 1: Descriptive statistics for variables used for econometric analysis

Variables Obs Mean Std. Dev. Min Max

Rice yield (1,000 VND/m2 ) 5,894 2.42 .75 .19 25.54
2
Capital (1,000 VND/m ) 5,894 .22 .16 0 2.25
Hired labor (1,000 VND/m2 ) 5,894 .09 .14 0 1.26
Family labor (number of hours) 5,894 .32 .35 0 4.92
2
Running costs (1,000 VND/m ) 5,894 .75 .32 0 10.02
Temperature (°C) 5,894 24.82 1.92 19.13 28.83
Precipitation (mm) 5,894 8.35 2.40 1.23 20.77
IT_30_31 (number of days) 5,894 20.69 13.45 0 61
IT_31_32 (number of days) 5,894 9.57 8.36 0 37
IT_32_33 (number of days) 5,894 3.92 5.06 0 24
IT_33_34 (number of days) 5,894 1.26 2.58 0 23
IT_34_35 (number of days) 5,894 .19 .87 0 7
IT_35_plus (number of days) 5,894 .01 .08 0 3
Age (years) 5,894 48 13 16 99
Gender (2=female) 5,894 1.15 .35 1 2
Education (1= no education to 9= univ. level) 5,461 1.49 1.23 0 9
Household size (number of persons) 5,894 4.21 1.54 1 15

20
6 Econometric results
6.1 Estimation of the SFA model
The first step is to estimate TE scores from a translog production function within a SFA.
Table 2 presents the estimation results19 .
In the first column, we use only inputs (hired and family labor, capital and running
costs) as well as their quadratic and interactive terms as explanatory variables. However,
the results of this estimation are potentially subject to the problems of omitted variables.
First, the production technology may be different depending on whether irrigation is used
or not. In column 2, we thus include a dummy variable to control for irrigation practices.
Moreover, temperature and precipitation levels have direct effect on agricultural yields.
To limit this bias, we include the temperature and precipitation levels in column 3 by
hypothesizing that temperature and precipitation levels impact agricultural yields while
weather shocks influence technical efficiency (second step). To test the consistency of
this model, we apply the Wald test to the coefficients of the climatic variables. The test
concludes that the inclusion of climatic variables in the first step is more relevant than
their exclusion. Thus, we will continue with this model to estimate the technical efficiency
scores and proceed to estimate the second stage equation.
Also, we check the theoretical consistency of our estimated efficiency model by verifying
that the marginal productivity of inputs is positive. If this theoretical criterion is met,
then the obtained efficiency estimates can be considered as consistent with microeconomics
theory. As the parameter estimates of the translog production function reported in Table
2 do not allow for direct interpretation of the magnitude and significance of any inputs, we
compute the output elasticities for all inputs at the sample mean, minimum, maximum
and median, and report them in Table A2 in Appendix. We find that rice farming in
Vietnam depends more strongly on running costs (0.64), Hired labor (0.34) and capital
(0.29) at the sample mean. These results capture the important role of mechanization
and intensification in rice farming in Vietnam. However, the marginal productivity of
family labor appears very low (0,13) at the sample mean. This result seems to be relevant
within the context of Vietnamese agriculture where surplus labor may exist. The over-use
of labor inputs implies that the marginal productivity of labor must be very low, even
negative in some cases.
Regarding the effect of climatic variables, our results are consistent with those found
in the literature. Indeed, we find that the impact of temperature and precipitation on
agricultural production is non-linear.
19
Note that all variables are expressed in logarithm. We transform the variable X into ln(1 + X) to
account for the null values in variables. The interaction terms are reported in Table A4 of Appendix.
Note that the Wald test in column 1 of Table 2 suggests that quadratic and interactive terms of the
translog production have to be included. This test confirms the relevance of the translog production
function compared to the Cobb-Douglass production function.

21
 Table 2: Estimation of production frontier

Variables (1) (2) (3)

Hired labor 1.606*** 1.605*** 0.559
(0.483) (0.483) (0.518)
Family labor 1.277*** 1.280*** 0.264
(0.215) (0.215) (0.269)
Running costs 2.205*** 2.189*** 0.863***
(0.224) (0.227) (0.317)
Capital 0.223 0.217 0.186
(0.432) (0.432) (0.432)
Irrigation 0.0164 0.0168
(0.0351) (0.0350)
Temperature 0.065***
(0.017)
Temperature squared -0.00163***
(0.0005)
Precipitation 4.08e-05
(0.0001)
Precipitation squared -1.22e-08
(3.55e-08)
Interactions factors x x x
Observations 5,894 5,894 5,894
Number of HH 2,592 2,592 2,592
Wald test 126.69 - 39.53
Estimation method: Maximum likelihood estimator
with time-variant TE. The dependent variable is the rice
yield per square meter. *** statistical significance at
1%, ** statistical significance at 5%, * statistical signif-
icance at 10%.

22
6.2 Impact of extreme weather events on TE
Table 3 summarizes the distribution of technical efficiency (TE) scores obtained from the
column 3 of Table 2 and the formula of Jondrow et al. (1982)20 . TE scores range from
0.29 to 1 with an average of 0.67. There are 55% of households with efficiency scores
below this value. The results show that on average, Vietnamese rice farmers could save
about one third (1-0.67) of their inputs.

Table 3: Distribution of efficiency score

Efficiency score Nbr Percent Cum.

0-0.5 436 7.40 7.40
0.5-0.6 1,592 27.01 34.41
0.6-0.7 1,711 29.03 63.44
0.7-0.8 1,094 18.56 82.00
0.8-0.9 969 16.44 98.44
0.9-1 92 1.56 100.00
Average 0.67
Min 0.29
Max 1

Form these TE scores, we assess the impact of extreme weather events (extreme tem-
peratures, typhoons, droughts and floods) on TE with both a fixed effects model (Table
4) and a Tobit model (Table 5).
As a result, we find that the occurrence of temperature shocks and extreme events
relative to what is expected prevents agents to efficiently use their potential technological
resources. Thus, this expectation bias creates inefficiency in the decision making of their
agricultural activities.
In the first column of Table 4, we assess only the effect of extreme temperatures on TE
according to the dry and wet seasons. We find that the effect of extreme temperatures on
TE is differential according to the seasons. During the dry season, extreme temperatures
above 31°C lessen TE and the effect is increasing with temperature. Indeed, an increase
of one day corresponds to a reduction in TE of 0.49 percentage points in the bin [31°C-
32°c], 4.34 percentage points in the bin [32°C-33°C] and 7.94 percentage points in the bin
[33°C-34°C]. During the wet season, only the bin [30-31[ has a significant and negative
effect on TE but this effect is relatively small. The insignificant effects for wet season
above 31°C can be explained by the mechanisms of adaptation. Farmers are used to very
high frequencies during this season and they adapt to that. Thus, the level of temperature
must be very extreme to have a detrimental effect on TE. For instance, even if the effect
20
In Jondrow et al. (1982), technical efficiency is calculated as the mean of individual efficiency condi-
tional to the global error terms which encompasses idiosyncratic error term and efficiency term.

23
 Table 4: Impact of weather shocks on TE: fixed effects model

Variables (1) (2) (3) (4)

IT_30_31_Dry 0.376*** 0.424*** 0.410*** 0.239*
(0.144) (0.142) (0.140) (0.132)
IT_31_32_Dry -0.491** -0.573*** -0.571*** -0.145
(0.212) (0.207) (0.204) (0.457)
IT_32_33_Dry -4.342*** -4.528*** -4.355*** -2.928***
(1.093) (1.077) (1.008) (.939)
IT_33_34_Dry -7.940*** -8.053** -6.849** -7.001**
(3.050) (3.148) (2.782) (3.474)
IT_30_31_Wet -0.386*** -0.388*** -0.383*** -0.386***
(0.0582) (0.0579) (0.0578) (0.058)
IT_31_32_Wet 0.165* 0.141 0.152* -0.106
(0.0881) (0.0881) (0.0859) (0.082)
IT_32_33_Wet 0.134 0.189** 0.236*** 0.251***
(0.0890) (0.0897) (0.0870) (0.083)
IT_33_34_Wet 0.112 0.167 0.263** 0.126
(0.138) (0.135) (0.134) (0.134)
IT_34_35_Wet 0.430* 0.342 0.492** 0.745***
(0.248) (0.237) (0.238) (0.238)
IT_plus_35_Wet -3.000 -2.808 -3.425* -2.823*
(1.850) (1.740) (1.770) (1.728)
Flood -4.060** -3.540** -3.154**
(1.629) (1.629) (1.655)
Typhoon -7.748*** -7.367*** -7.676***
(1.389) (1.444) (1.419)
Drought -2.614** -2.522** -3.663***
(1.132) (1.105) (1.124)
Age 0.652*** 0.573***
(0.0974) (0.092)
Educ 1.254*** 1.151***
(0.408) (0.366)
HH size -0.659** -0.676**
(0.283) (0.269)
Gender -2.901 -2.767
(2.154) (2.049)
Constant 70.97*** 71.64*** 44.30*** 82.41***
(0.949) (0.955) (5.844) (21.307)

Observations 5,894 5,894 5,461 5,461

Number of HH 2,592 2,592 2,457 2,457
R-squared 0.060 0.084 0.130 0.296
Estimation method: within fixed effects estimator. The dependent
variable is the score of technical efficiency estimated from col. 3 of
Table 2. In col. 4, daily precipitation are controlled for. Robust
standard errors in parentheses. *** statistical significance at 1%,
** statistical significance at 5%, * statistical significance at 10%.

24
is not significant, it is worth noting that above 35°C, a one day increase above this degree
decreases efficiency by 3 percentage points.
In column 2, the occurrence of natural disasters such as floods, typhoons and droughts
is introduced. These events are found to be significantly detrimental for TE. More pre-
cisely, the efficiency diminishes respectively by 4.06, 7.74 and 2.61 percentage points after
the occurrence of a flood, a typhoon and a drought, respectively. The previous results
found in column 1 for the extreme temperatures remain the same.
In column 3, we test the robustness of the effect of climate variables to the inclusion of
several control variables related to the household (gender, age, education and household
size)21 . The effects of extreme temperatures in the dry season and natural disasters remain
unchanged. However, a one day increase in the bin [35°C[ during the wet season becomes
significant (a reduction of 3.42 percentage points of TE). Regarding household variables,
age and education are found to positively affect TE. These results suggest that older
and more educated rice farmers are more efficient than others. In addition, it appears
that men are more efficient than women. This result should be interpreted with caution
because women represent only 18% of our sample. Also, women have less access to
credit or insurance systems for lack of collateral, while the literature shows the important
role of these factors on efficiency (Helfand and Levine (2004), Fontan (2008)). Finally,
household negatively size affects TE. This result can reflect a problem of misallocation of
inputs mainly in term of family labor.
In column 4, average daily precipitations are introduced (Zhang et al., 2014). Previous
results concerning extreme temperatures and natural disasters remain the same except for
the coefficient of the bin [31°C, 32°C[ during the dry season that becomes non significant.
However, since TE scores are truncated to 1, the Tobit estimator is used to estimate
the impact of weather shocks on TE. Results are presented in Table 5. It is worth
noting that the effects of climate variables remain the same. During the dry season, the
more temperature increases above 31°C, the lower TE is. More precisely, over the four
estimations (col. 1 to col. 4), the effects range between -0.69 and -0.64 percentage point
for the bin [31 32]), -5.02 and -4.80 percentage point for the bin [32 33], -7.13 and - 5.93
percentage point for the bin [33 34]. However, we now find that the impact of one more
day in the bin [35 and more] during the wet season significantly reduces TE from -3.56
to -4.20 percentage points for the four estimations. In addition, floods, typhoons and
droughts have still detrimental effects on TE. However, the magnitude of the coefficients
change. While typhoons are found to have the highest detrimental impact in Table 4,
droughts have now the highest negative effect. More precisely, over the four estimations
(col. 1 to col. 4), the effects range between -1.95 and -1.83 percentage points for flood,
21
It is useful to note that there are several standard household variables that we do not take into
account (for instance, the access to credit, land tenure, ...). However, the purpose of this analysis is not
to list exhaustively the determinants of TE but to investigate whether climate shocks affect TE. Since
climate shocks are exogenous to household characteristics, we limit the problem of omitted variables.

25
 Table 5: Impact of weather shocks on TE: Tobit model

Variables (1) (2) (3) (4)

IT_30_31_Dry 0.168*** 0.174*** 0.165*** 0.0934**
(0.0440) (0.0437) (0.0450) (0.0460)
IT_31_32_Dry -0.652*** -0.691*** -0.678*** -0.536***
(0.126) (0.125) (0.128) (0.127)
IT_32_33_Dry -5.023*** -4.913*** -4.892*** -4.930***
(0.551) (0.548) (0.545) (0.538)
IT_33_34_Dry -5.931 -7.133** -6.939* -6.270*
(3.612) (3.586) (3.557) (3.509)
IT_30_31_Wet -0.0133 -0.0312 -0.0666*** -0.0212
(0.0207) (0.0207) (0.0215) (0.0216)
IT_31_32_Wet 0.144*** 0.139*** 0.136*** 0.113***
(0.0402) (0.0399) (0.0404) (0.0400)
IT_32_33_Wet 0.0271 0.0782 0.0655 0.106*
(0.0631) (0.0628) (0.0632) (0.0627)
IT_33_34_Wet 0.169 0.199* 0.152 0.0683
(0.103) (0.103) (0.103) (0.102)
IT_34_35_Wet -0.0518 0.0208 0.171 0.192
(0.250) (0.248) (0.251) (0.251)
IT_35_+_Wet -3.563* -4.054** -3.961* -3.867*
(2.079) (2.064) (2.054) (2.026)
Flood -1.873** -1.950** -1.787**
(0.757) (0.776) (0.768)
Typhon -4.214*** -4.121*** -4.069***
(0.687) (0.712) (0.703)
Drought -4.742*** -4.725*** -4.912***
(0.685) (0.708) (0.702)
Age 0.0585*** 0.0581***
(0.0133) (0.0132)
Educ 0.397*** 0.387***
(0.134) (0.132)
HH size -0.220** -0.230**
(0.109) (0.108)
Gender -0.469 -0.354
(0.480) (0.474)
Constant 65.48*** 66.20*** 65.19*** 36.43***
(0.292) (0.299) (1.037) (8.789)

Observations 5,894 5,894 5,461 5,461

Number of HH 2,592 2,592 2,457 2,457
Estimation method: Tobit estimator. The dependent variable is
the score of technical efficiency estimated from col. 3 of Table 2. In
col. 4, daily precipitation are controlled for. Robust standard errors
in parentheses. *** statistical significance at 1%, ** statistical
significance at 5%, * statistical significance at 10%.

26
-4.21 and -3.97 percentage points for typhoon, and -5.03 and - 4.73 percentage points for
drought.

6.3 Heterogeneity effects

Farm size and weather shocks

In this section, we investigate whether the effect of climate shocks on TE is different

according to the area devoted to rice farming. The sample is thus split into two categories.
One the one hand, there are the small farms defined as farms with a rice area less than
the median area of the total sample (0.40 hectare). These farms have an average size
of 0.23 hectares while larger farms (i.e., farms with an area above 0.40 hectare) have an
average size of 1.27 hectares.
For each of these two categories, we redo the estimation of column 4 in Table 4. Results
are presented in Table A4 in Appendix.
It is noted that climate shocks tend to be more harmful for small farms. In general,
the occurrence of extreme temperatures is rather detrimental for small farmers in the dry
season. For the wet season, there are no significant effects for large farms. Also, it is
observed that the occurrence of typhoons and droughts negatively affects the efficiency of
smallholders whereas only the occurrence of typhoons is harmful for large farms.

Liquidity constraint and weather shocks

The literature shows that the relaxation of liquidity constraints plays an important role in
improving agricultural productivity. Good farm management requires access to resources
(Carter and Wiebe, 1990) both ex ante and ex post. Firstly, access to resources will enable
the farmer to buy the inputs necessary for his production (hired labour, investment, and
access to land ...). Secondly, resource use may be needed after production by allowing
farmers to smooth their income when a shock hits their production. Thus, access to
resources allows the farmer to adopt better technology for her production and to smooth
her farm income. More specifically, access to new resources can enable poor farmers to
optimize the use of inputs that conditions final production.
In order to measure the liquidity constraint faced by each farmer during the growing
rice production, we sum three different sources of income obtained off the rice farming.
More precisely, we sum total value of remittances received by households (both internal
and external), total non-farm income of households and total government aid received by
households after the occurrence of disasters. We analyze the impact of this variable on
efficiency and test the conditional effect of extreme climate events on efficiency through
it.
Table A5 in Appendix presents the results of liquidity constraint relaxation effects
on efficiency. We find a non linear effect of liquidity on TE. More precisely, there is a

27
minimum of liquidity available to households which can be used to improve their TE.
From column 2, the threshold is 270.43 (1000 VND) from which total off-farm income
positively affects TE. This threshold is not trivial because 34% of the rice farmers in the
sample have an off-farm income below this threshold. In other words, these farmers can
face a liquidity constraint.
In Table A6, we interact climate shocks variables with the liquidity variable to test the
conditional effect of climate shocks on technical efficiency through liquidity availability.
The interactive terms allow to test whether farmers with more off-farm income can be
more resilient to weather shocks. Our results do not confirm this assumption. Two
explanations are possible for these results.
Firstly, the utility of these amounts as resilience factors to climate shocks may be a
function of the phase of the crop’s life cycle (land preparation, planting, cultivation and
harvesting) during which the shocks occur. Liquidity can play an important role during
the first phase by allowing an optimal adjustment of the inputs needed for production (Ex:
purchases of seeds, fertilizers, pesticides, hired labor, access to capital ...). However, the
effect of climate shocks on efficiency is not only due to a lack of adaptation to these shocks
but also to forecast and expectation errors in these shocks that affect the optimality of
farmers’ production decisions. Hence, liquidity may not be a mitigating factor of the effect
of shocks on efficiency even if it is true that these resources can be considered as a resilience
factor to climate shocks by allowing individuals to smooth their consumption (Arouri
et al., 2015). Secondly, as we have pointed out above, the amount of these resources is
not large enough to deal with climate shocks more precisely to natural disasters whose
magnitude of effects on efficiency is very high. Thus these additional resources are used
to smooth household consumption rather than to invest in the agricultural sector.

7 Simulation
We can use the previous estimations of the impact of weather shocks on technical efficiency
in rice production to derive the potential impacts of future global warming on that sector.
This is done under several important and strong hypotheses that will be detailed below.
The idea of this kind of estimation is not to predict with certainty the future impact
of global warming in terms of technical efficiency losses, but rather to have a picture of
possible futures depending on the future climate change in Viet Nam22 .
The future climate projections are obtained from the Regional Climate Model version
4.3 (RegCM) (Giorgi et al., 2012). RegCM is a hydrostatic, limited-area model with
a sigma vertical coordinate. In this study, the model was implemented with 18 vertical
22
This is all the more true that we use in this version the outcomes of only one regional climate model.
A future version will include all available simulations from the CORDEX-SEA program, which are not
yet available.

28
-levels with the top level set at 5 mb and with a horizontal resolution of 25 km. The phys-
ical options used for the RegCM4.3 experiment in this study are the radiative transfer
scheme of the NCAR Community Climate Model (CCM3) (Kiehl et al., 1996), the sub-
grid explicit moisture (SUBEX) scheme for large-scale precipitation (Pal et al., 2007), the
planetary boundary layer scheme of (Holtslag and Moeng, 1991), the MIT-Emanuel con-
vective scheme (Nilsson and Emanuel, 1999), the BATS1e ocean flux scheme (Dickinson
et al., 1993). This setting is based on the sensitivity experiments conducted previously by
the Coordinated Regional Climate Downscaling Experiment -Southeast Asia (CORDEX-
SEA) community (Cruz et al., 2017; Juneng et al., 2016; Ngo-Duc et al., 2017). Boundary
and initial conditions of RegCM are provided by the outputs of the CNRM5 GCM model
(Voldoire et al., 2013).
In order to compute the estimated yearly impacts of climate change on rice technical
efficiency, we compute for each year and for each temperature bin the difference between
the new conditions induced by warming (as a moving average on twenty years every year),
on the two Representative Concentration Pathways (RCPs) 8.5 and 4.5, and a reference
average of the years 1986 − 2005. These differences are then multiplied by the coefficients
estimated in tables 4 and 5. We thus get the relative impact on technical efficiency of
the change in climate in each pixel of the Viet Nam map for the different estimation
strategies. We will just show here the results in the case of the fixed effects model, when
taking floods, typhoons and droughts into account, but looking only at the effects of
temperature. This corresponds to the temperature coefficients in the second column of
table 4.
The results appear on figure 6 and 7. They show that in all cases, losses in technical
efficiency reach the highest levels in the Red River Delta and in Northern mountains.
Rice producers in these regions see their technical efficiency shrink sharply in 2050. After
2050, the full effect of the RCP8.5 appears, and the two scenarios really diverge, as shown
in figure 6. Technical efficiency losses diffuse through the Mekong delta in both scenarios.
Looking at the differentiated effect of the dry and the wet seasons (See figure 7) in the
RCP 8.5 scenario is particularly striking. It appears that the negative effect of the dry
season is much more pronounced than the one of the wet season. However, the effect of
the wet season is much more concentrated in specific geographical areas, such as the Red
River delta, and later on the coastal areas and the Mekong delta. On the contrary, the
effect of the dry season seems much more homogeneous around the country, the Mekong
delta emerging as a threatened area only at the end of the century
It must be recalled that only temperature increases have been taken into account in
these projections, all other factors (economic or climate) remaining constant. So the good
news on the Mekong region should not be a matter of optimism if we recall that the area
is prone to other kinds of climate threats such as storm surges, typhoons, and sea-level
rise in the longer run.

29
 Figure 5: Aggregate damage on technical efficiency (% points lost)

On aggregate, we can also calculate the average loss in technical efficiency for Viet
Nam as a whole, by simply aggregating with equal weights the losses evaluated for each
cell. This exercize shows that RCP4.5 and RCP8.5 scenarios start diverging soon after
2040. Technical efficiency losses reach 40 percentage points before the end of the century
in the case of the RCP8.5 scenario, while the RCP4.5 scenario seems to stabilize technical
efficiency losses around 8 percentage points losses. Here again, we must recall the very
simplified assumptions made around these projections. In particular, no technical progress
or adaptation strategy is taken into account here, which could make the situation better.
On the other side, no macroeconomic retrofitting of climate damages to other sectors are
taken into account, which could make matters worse.

30
Figure 6: Technical efficiency losses in 2030, 2050 and 2090 compared to the reference
period 1986 − 2005, for dry and wet seasons combined, RCP4.5 and RCP8.5.
(a) RCP4.5 - 2030 (b) RCP8.5 - 2030

(c) RCP4.5 - 2050 (d) RCP8.5 - 2050

(e) RCP4.5 - 2090 (f) RCP8.5 - 2090

31
Figure 7: Technical efficiency losses in 2030, 2050 and 2090 compared to the reference
period 1986 − 2005, for dry and wet seasons separated, RCP8.5 only.
(a) Dry season - 2030 (b) Wet season - 2030

(c) Dry season - 2050 (d) Wet season - 2050

(e) Dry season - 2090 (f) Wet season - 2090

32
8 Conclusion
In this study, we investigate the impact of extreme weather events on rice farming TE
using a SFA model. These weather events are defined as extreme temperatures in both
dry and wed season, and the occurrence of typhoons, floods and droughts during the rice
production period.
We first find that extreme temperatures in the dry season are detrimental for TE.
More precisely, temperatures above 31°C dampen TE and the effect is increasing with
temperature. An increase of one day corresponds to a reduction in TE between 0.49 and
0.57 percentage points in the bin [31°C-32°c]), 2.92 and 4.52 percentage points in the bin
[32°C-33°C]), and 6.84 and 8.05 percentage points in the bin [33°C-34°C]. Secondly, during
the wet season, only the bin [30°C-31°C[ and [35°C[ have a significant and negative effect
on TE. For instance, a one day increase in the bin [35°C[ dampens TE between 2.82 and
3.42 percentage points. Thirdly, we find that floods, typhoons and droughts reduce TE.
The magnitude is the highest for typhoons that lessen TE from 7.37 to 7.75 percentage
points.
Small farms are more vulnerable to climate shocks than larger farms. In addition,
farmers’ liquidity has a non-linear effect on their efficiency. In other words, households
with more liquidity are technically more efficient than others. However, the negative effect
of climate extremes on efficiency is not conditioned to the liquidity owned by households.
Hence, liquidity may not be a mitigating factor of the effect of climate shocks on efficiency
even if it is true that these resources can be considered as a factor of resilience to climate
shocks by allowing individuals to smooth their consumption.
From these results, some economic policy recommendations can be suggested. First,
the establishment of weather forecasting systems in less favored areas can be advocated.
A meteorological system that provides real-time data will reduce biases in individuals’
expectations. Indeed, it is difficult for households, specifically poor households, to au-
tomatically adjust to exogenous shocks in the short term. Secondly, policies aimed at
helping people affected by extreme temperatures and natural disasters should be discrim-
inating by favoring households with small farms. Regarding natural disasters, our results
confirm those of Arouri et al. (2015). In other words, natural disasters have negative
effects on the welfare of households. The implementation of irrigation and drainage sys-
tems would mitigate the negative effects of drought and flood on the farmers’ efficiency,
although the impact of climate change on the water cycle should be taken into account
as well.

33
References
Aigner, D., Lovell, C. K., and Schmidt, P. (1977). Formulation and estimation of stochas-
tic frontier production function models. Journal of econometrics, 6(1):21–37.

Arouri, M., Nguyen, C., and Youssef, A. B. (2015). Natural disasters, household welfare,
and resilience: evidence from rural vietnam. World development, 70:59–77.

Auffhammer, M., Hsiang, S. M., Schlenker, W., and Sobel, A. (2013). Using weather
data and climate model output in economic analyses of climate change. Review of
Environmental Economics and Policy, 7(2):181–198.

Bank, W. (2010). Vietnam - economics of adaptation to climate change (english). Wash-

ington, DC: World Banks.

Battese, G. E. and Coelli, T. J. (1992). Frontier production functions, technical efficiency

and panel data: with application to paddy farmers in india. Journal of productivity
analysis, 3(1-2):153–169.

Berg, H. and Tam, N. (2012). Use of pesticides and attitude to pest management strategies
among rice and rice-fish farmers in the mekong delta, vietnam. International Journal
of Pest Management, 58(2):153–164.

Blanc, E. and Schlenker, W. (2017). The use of panel models in assessments of climate
impacts on agriculture. Review of Environmental Economics and Policy, 11(2):258–279.

Butler, E. E. and Huybers, P. (2013). Adaptation of us maize to temperature variations.

Nature Climate Change, 3(1):68.

Carter, M. R. and Wiebe, K. D. (1990). Access to capital and its impact on agrar-
ian structure and productivity in kenya. American journal of agricultural economics,
72(5):1146–1150.

Chamberlain, G. (1979). Analysis of covariance with qualitative data.

Chen, S., Chen, X., and Xu, J. (2016). Impacts of climate change on agriculture: evidence
from china. Journal of Environmental Economics and Management, 76:105–124.

Christensen, L. R., Jorgenson, D. W., and Lau, L. J. (1971). Conjugate duality and the
transcendental logarithmic function.

Cobb, C. W. and Douglas, P. H. (1928). A theory of production. The American Economic

Review, 18(1):139–165.

34
Cornwell, C., Schmidt, P., and Sickles, R. C. (1990). Production frontiers with cross-
sectional and time-series variation in efficiency levels. Journal of econometrics, 46(1-
2):185–200.

Cruz, F., Narisma, G., Dado, J., Singhruck, P., Tangang, F., Linarka, U., Wati, T.,
Juneng, L., Phan-Van, T., Ngo-Duc, T., et al. (2017). Sensitivity of temperature to
physical parameterization schemes of regcm4 over the cordex-southeast asia region.
International Journal of Climatology, 37(15):5139–5153.

Dasgupta, S., Laplante, B., Meisner, C., Wheeler, D., and Yan, J. (2007). The impact of
sea level rise on developing countries: a comparative analysis. The World Bank.

Dell, M., Jones, B. F., and Olken, B. A. (2009). Temperature and income: reconciling
new cross-sectional and panel estimates. American Economic Review, 99(2):198–204.

Demont, M. and Rutsaert, P. (2017). Restructuring the vietnamese rice sector: towards
increasing sustainability. Sustainability, 9(2):325.

Deryugina, T. and Hsiang, S. (2017). The marginal product of climate. NBER Working
Paper.

Deschênes, O. and Greenstone, M. (2007). The economic impacts of climate change:

evidence from agricultural output and random fluctuations in weather. The American
Economic Review, 97(1):354–385.

Dickinson, E., Henderson-Sellers, A., and Kennedy, J. (1993). Biosphere-atmosphere

transfer scheme (bats) version 1e as coupled to the ncar community climate model.

Farrell, M. J. (1957). The measurement of productive efficiency. Journal of the Royal

Statistical Society. Series A (General), 120(3):253–290.

Fontan, C. (2008). Production et efficience technique des riziculteurs de guinée. une

estimation paramétrique stochastique. Économie rurale. Agricultures, alimentations,
territoires, (308):19–35.

Giorgi, F., Coppola, E., Solmon, F., Mariotti, L., Sylla, M., Bi, X., Elguindi, N., Diro,
G., Nair, V., Giuliani, G., et al. (2012). Regcm4: model description and preliminary
tests over multiple cordex domains. Climate Research, 52:7–29.

Greene, W. (2005). Fixed and random effects in stochastic frontier models. Journal of
productivity analysis, 23(1):7–32.

Helfand, S. M. and Levine, E. S. (2004). Farm size and the determinants of productive
efficiency in the brazilian center-west. Agricultural economics, 31(2-3):241–249.

35
Holtslag, A. and Moeng, C.-H. (1991). Eddy diffusivity and countergradient transport
in the convective atmospheric boundary layer. Journal of the Atmospheric Sciences,
48(14):1690–1698.

Hsiang, S. M. (2010). Temperatures and cyclones strongly associated with economic

production in the caribbean and central america. Proceedings of the National Academy
of sciences, 107(35):15367–15372.

Hsiao, C. (2014). Analysis of panel data. Cambridge university press.

Jondrow, J., Lovell, C. K., Materov, I. S., and Schmidt, P. (1982). On the estimation of
technical inefficiency in the stochastic frontier production function model. Journal of
econometrics, 19(2-3):233–238.

Juneng, L., Tangang, F., Chung, J. X., Ngai, S. T., Tay, T. W., Narisma, G., Cruz,
F., Phan-Van, T., Ngo-Duc, T., Santisirisomboon, J., et al. (2016). Sensitivity of
southeast asia rainfall simulations to cumulus and air-sea flux parameterizations in
regcm4. Climate Research, 69(1):59–77.

Key, N. and Sneeringer, S. (2014). Potential effects of climate change on the productivity
of us dairies. American Journal of Agricultural Economics, 96(4):1136–1156.

Kiehl, T., Hack, J., Bonan, B., Boville, A., Briegleb, P., Williamson, L., and Rasch, J.
(1996). Description of the ncar community climate model (ccm3).

Kompas, T., Che, T. N., Nguyen, H. T. M., and Nguyen, H. Q. (2012). Productivity, net
returns, and efficiency: land and market reform in vietnamese rice production. Land
Economics, 88(3):478–495.

Kompas, T. et al. (2002). Market reform, productivity and efficiency in vietnamese rice
production.

Kumbhakar, S. C., Denny, M., and Fuss, M. (2000). Estimation and decomposition of pro-
ductivity change when production is not efficient: a paneldata approach. Econometric
Reviews, 19(4):312–320.

Kurukulasuriya, P., Kala, N., and Mendelsohn, R. (2011). Adaptation and climate change
impacts: a structural ricardian model of irrigation and farm income in africa. Climate
Change Economics, 2(02):149–174.

Lee, Y. H. and Schmidt, P. (1993). A production frontier model with flexible temporal
variation in technical efficiency. The measurement of productive efficiency: Techniques
and applications, pages 237–255.

36
Lobell, D. B., Bänziger, M., Magorokosho, C., and Vivek, B. (2011). Nonlinear heat
effects on african maize as evidenced by historical yield trials. Nature climate change,
1(1):42.

Massetti, E. and Mendelsohn, R. (2011). Estimating ricardian models with panel data.
Climate Change Economics, 2(04):301–319.

Massetti, E., Mendelsohn, R., and Chonabayashi, S. (2016). How well do degree days
over the growing season capture the effect of climate on farmland values? Energy
Economics, 60:144–150.

McMaster, G. S. and Wilhelm, W. (1997). Growing degree-days: one equation, two

interpretations. Agricultural and forest meteorology, 87(4):291–300.

Meeusen, W. and van Den Broeck, J. (1977). Efficiency estimation from cobb-douglas
production functions with composed error. International economic review, pages 435–
444.

Mendelsohn, R., Nordhaus, W. D., and Shaw, D. (1994). The impact of global warming
on agriculture: a ricardian analysis. The American economic review, pages 753–771.

Mendelsohn, R. O. and Massetti, E. (2017). The use of cross-sectional analysis to mea-

sure climate impacts on agriculture: theory and evidence. Review of Environmental
Economics and Policy, 11(2):280–298.

Ngo-Duc, T., Tangang, F. T., Santisirisomboon, J., Cruz, F., Trinh-Tuan, L., Nguyen-
Xuan, T., Phan-Van, T., Juneng, L., Narisma, G., Singhruck, P., et al. (2017). Per-
formance evaluation of regcm4 in simulating extreme rainfall and temperature indices
over the cordex-southeast asia region. International Journal of Climatology, 37(3):1634–
1647.

Nilsson, J. and Emanuel, K. (1999). Equilibrium atmospheres of a two-column radiative-

convective model. Quarterly Journal of the Royal Meteorological Society, 125(558):2239–
2264.

Pal, J. S., Giorgi, F., Bi, X., Elguindi, N., Solmon, F., Gao, X., Rauscher, S. A., Francisco,
R., Zakey, A., Winter, J., et al. (2007). Regional climate modeling for the developing
world: the ictp regcm3 and regcnet. Bulletin of the American Meteorological Society,
88(9):1395–1410.

Pitt, M. M. and Lee, L.-F. (1981). The measurement and sources of technical inefficiency
in the indonesian weaving industry. Journal of development economics, 9(1):43–64.

Sánchez, B., Rasmussen, A., and Porter, J. R. (2014). Temperatures and the growth and
development of maize and rice: a review. Global change biology, 20(2):408–417.

37
Schlenker, W., Hanemann, W. M., and Fisher, A. C. (2005). Will us agriculture re-
ally benefit from global warming? accounting for irrigation in the hedonic approach.
American Economic Review, 95(1):395–406.

Schlenker, W., Hanemann, W. M., and Fisher, A. C. (2006). The impact of global warming
on us agriculture: an econometric analysis of optimal growing conditions. Review of
Economics and statistics, 88(1):113–125.

Schlenker, W. and Roberts, M. J. (2009). Nonlinear temperature effects indicate severe

damages to us crop yields under climate change. Proceedings of the National Academy
of sciences, 106(37):15594–15598.

Schlenker, W., Roberts, M. J., and Lobell, D. B. (2013). Us maize adaptability. Nature
Climate Change, 3(8):690.

Schmidt, P. and Sickles, R. C. (1984). Production frontiers and panel data. Journal of
Business & Economic Statistics, 2(4):367–374.

Scott, S. (2008). Agrarian transformations in vietnam: Land reform, markets and poverty.
In The political economy of rural livelihoods in transition economies, pages 187–211.
Routledge.

Stevenson, R. E. (1980). Likelihood functions for generalized stochastic frontier estima-

tion. Journal of econometrics, 13(1):57–66.

Tran, C. T. and Nguyen, L. H. (2015). Restructuring crop production in vietnam. FFTC

Agricultural Policy Platform (FFTC-AP).

Trinh, T. A. (2017). The impact of climate change on agriculture: Findings from house-
holds in vietnam. Environmental and Resource Economics.

Voldoire, A., Sanchez-Gomez, E., y Mélia, D. S., Decharme, B., Cassou, C., Sénési, S.,
Valcke, S., Beau, I., Alias, A., Chevallier, M., et al. (2013). The cnrm-cm5. 1 global
climate model: description and basic evaluation. Climate Dynamics, 40(9-10):2091–
2121.

Yu, B., Zhu, T., Breisinger, C., Hai, N. M., et al. (2010). Impacts of climate change
on agriculture and policy options for adaptation. International Food Policy Research
Institute (IFPRI).

Zhang, Z., Wang, P., Chen, Y., Song, X., Wei, X., and Shi, P. (2014). Global warming
over 1960–2009 did increase heat stress and reduce cold stress in the major rice-planting
areas across china. European journal of agronomy, 59:49–56.

38
A Appendix

Table A1: List of variables

Variables Definition and description

Rice income Total rice production during past 12 years: Thousand VND per squared meter.
Capital Total value of investment in machinery: Thousand VND per squared meter.
Hired labor Payment of hired labor for rice production: Thousand VND per squared meter.
Family labor Number of hours for family labours. Hours per squared meter.
Running costs Other costs (fertilizer, seeds, irrigation ...): Thousand VND per squared meter.
Irrigation = 1 if farm is irrigated.

Temperature Daily temperature average over the production period: °C.

Precipitation Daily precipitation over the production period: mm.
IT_[a b] Number of days when average daily temperature is between a and b.
Flood =1 if there is flood during rice the production period.
Typhoon =1 if there is Typhoon during rice the production period.
Drought =1 if there is drought during rice the production period.

Gender Gender of household head (1=male, 2=Female).

Age Age of household head.
Educ Education level of household head (0 (no qualification) to 9 (university level)).
Household size Number of persons living together in one house.

Table A2: Inputs elasticities

Inputs variables Mean Min Max Median

Hired labor 0.34 0.56 -0.85 0.37
Family labor 0.13 0.26 -0.60 0.14
Running costs 0.64 0.86 -0.20 0.65
Capital 0.29 0.19 0.60 0.33
Total 1.40 1.87 -1.04 1.49
Calculation method: coefficient estimates from the
results of column 3 in Table 2. Elasticities calculated
at sample mean, sample median, minimum and max-
imum of inputs.

39
 Table A3: Estimation of stochastic production frontier model.

Variables (1) (2) (3)

Hired labor 1.606*** 1.605*** 0.559
(0.483) (0.483) (0.518)
Family labor 1.277*** 1.280*** 0.264
(0.215) (0.215) (0.269)
Running costs 2.205*** 2.189*** 0.863***
(0.224) (0.227) (0.317)
Capital 0.223 0.217 0.186
(0.432) (0.432) (0.432)
Capital*Capital 1.733 1.761 0.871
(1.341) (1.342) (1.350)
Capital*Hired labor -1.810 -1.811 -0.528
(1.692) (1.692) (1.698)
Capital*Family labor -1.793** -1.786** -0.660
(0.890) (0.890) (0.906)
Capital*Running costs 0.513 0.482 0.159
(1.148) (1.150) (1.151)
Hired labor*Hired labor -0.653 -0.657 -0.389
(1.273) (1.272) (1.277)
Hired labor*Family labor -1.985* -1.981* -0.493
(1.189) (1.189) (1.212)
Hired labor*Running costs -2.598* -2.605* -0.288
(1.365) (1.365) (1.417)
Family labor*Family labor -0.472 -0.471 -0.0417
(0.303) (0.303) (0.313)
Family labor*Running costs -2.178*** -2.184*** -0.154
(0.670) (0.670) (0.749)
Running costs*Running costs -1.770*** -1.746*** -0.378
(0.561) (0.563) (0.606)
Irrigation 0.0164 0.0168
(0.0351) (0.0350)
Temperature 0.0654***
(0.0171)
Temperature squared -0.00163***
(0.000544)
Precipitation 4.08e-05
(0.000119)
Precipitation squared -1.22e-08
(3.55e-08)
Interactions factors x x x
Observations 5,894 5,894 5,894
Number of HH 2,592 2,592 2,592
Wald test 126.69 39.53
Estimation method: Maximum likelihood estimator with time-
variant TE. The dependent variable is the rice yield per square
meter. *** statistical significance at 1%, ** statistical significance
at 5%, * statistical significance at 10%.
40
Table A4: Farm size, weather shocks and TE

(1) (2)
Small farm Large farm
VARIABLES Efficiency Efficiency

IT_30_31_Dry -1.517** 0.554***

(0.635) (0.119)
IT_31_32_Dry -1.790*** 0.323
(0.654) (0.214)
IT_32_33_Dry -2.009 -3.094*
(1.285) (1.810)
IT_33_34_Dry -4.309 -2.854
(4.184) (4.056)
IT_30_31_Wet -0.785*** -0.291***
(0.0643) (0.0607)
IT_31_32_Wet 0.113 -0.490***
(0.101) (0.113)
IT_32_33_Wet 0.172 0.0198
(0.122) (0.108)
IT_33_34_Wet -0.652*** 0.0816
(0.179) (0.190)
IT_34_35_Wet 0.759** -0.345
(0.352) (0.341)
IT_35_+_Wet -4.634 -3.603
(3.970) (3.732)
Age 0.466*** 0.633***
(0.122) (0.131)
Educ 1.026*** 0.988*
(0.358) (0.516)
HH_size -0.671* -0.546
(0.378) (0.363)
Gender -1.980 0.302
(2.174) (2.956)
Flood 0.941 -0.768
(1.754) (2.069)
Typhoon -4.328* -6.543***
(2.330) (1.530)
Drought -2.750** 0.346
(1.287) (1.482)
Constant 104.8*** 99.32***
(22.79) (19.38)

Observations 2,787 2,674

R-squared 0.370 0.399
Number of hid 1,477 1,400
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

41
Table A5: Liquidity constraint, weather shocks and TE

(1) (2) (3)

VARIABLES Efficiency Efficiency Efficiency

IT_30_31_Dry 0.410*** 0.412*** 0.411***

(0.140) (0.141) (0.140)
IT_31_32_Dry -0.571*** -0.571*** -0.582***
(0.204) (0.205) (0.204)
IT_32_33_Dry -4.355*** -4.336*** -4.376***
(1.008) (1.016) (1.022)
IT_33_34_Dry -6.849** -7.022** -6.963**
(2.782) (2.787) (2.822)
IT_30_31_Wet -0.383*** -0.382*** -0.376***
(0.0578) (0.0578) (0.0577)
IT_31_32_Wet 0.152* 0.152* 0.151*
(0.0859) (0.0858) (0.0857)
IT_32_33_Wet 0.236*** 0.228*** 0.238***
(0.0870) (0.0874) (0.0872)
IT_33_34_Wet 0.263** 0.264** 0.265**
(0.134) (0.134) (0.135)
IT_34_35_Wet 0.492** 0.501** 0.498**
(0.238) (0.239) (0.239)
IT_plus_35_Wet -3.425* -3.365* -3.509*
(1.770) (1.767) (1.797)
Age 0.652*** 0.653*** 0.649***
(0.0974) (0.0976) (0.0972)
Educ 1.254*** 1.259*** 1.261***
(0.408) (0.406) (0.404)
HH size -0.659** -0.679** -0.672**
(0.283) (0.283) (0.283)
Gender -2.901 -2.938 -2.907
(2.154) (2.153) (2.138)

42
Table A5 continued

(1) (2) (3)

VARIABLES Efficiency Efficiency Efficiency

Flood -3.540** -3.589** -3.538**

(1.629) (1.631) (1.644)
Typhon -7.367*** -7.311*** -7.267***
(1.444) (1.442) (1.441)
Drought -2.522** -2.503** -2.415**
(1.105) (1.107) (1.120)
ln (liquidity) -0.0995 -0.653***
(0.0728) (0.240)
ln (liquidity) squared 0.0582**
(0.0233)
Constant 44.30*** 44.89*** 44.96***
(5.844) (5.846) (5.835)

Observations 5,461 5,461 5,461

R-squared 0.130 0.131 0.133
Number of hid 2,457 2,457 2,457
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

43
Table A6: Liquidity constraint, weather shocks and TE - interactions terms

(1)
VARIABLES Efficience

Liquidity -0.256
(0.260)
Liquidity squared 0.0699***
(0.0270)
IT_30_31_Dry 0.374**
(0.149)
IT_31_32_Dry -0.383*
(0.233)
IT_32_33_Dry -4.492***
(1.452)
IT_33_34_Dry -8.935*
(4.997)
IT_30_31_Wet -0.161**
(0.0760)
IT_31_32_Wet -0.0206
(0.105)
IT_32_33_Wet 0.245*
(0.129)
IT_33_34_Wet 0.330*
(0.198)
IT_34_35_Wet 0.0832
(0.464)
IT_plus_35_Wet -0.826
(2.218)
Age 0.646***
(0.0899)
Educ 1.224***
(0.447)
HH size -0.678**
(0.298)
Gender -3.292*
(1.947)
Flood -3.415
(2.268)
Typhon -3.699
(2.280)
Drought -1.853
(1.855)

44
Table A6 continued

(1)
VARIABLES Efficience

Liquidity*IT_30_31_Dry 0.00381
(0.0151)
Liquidity*IT_31_32_Dry -0.0331
(0.0316)
Liquidity*IT_32_33_Dry 0.0246
(0.165)
Liquidity*IT_33_34_Dry 0.221
(0.397)
Liquidity*IT_30_31_Wet -0.0371***
(0.00850)
Liquidity*IT_31_32_Wet 0.0282**
(0.0134)
Liquidity*IT_32_33_Wet -0.00281
(0.0173)
Liquidity*IT_33_34_Wet -0.00891
(0.0274)
Liquidity*IT_34_35_Wet 0.0616
(0.0504)
Liquidity*IT_35_+_Wet -0.461
(0.470)
Liquidity*Flood 0.0435
(0.264)
Liquidity*Typhon -0.739***
(0.242)
Liquidity*Drought -0.0916
(0.227)
Constant 43.02***
(4.849)

Observations 5,461
Number of hid 2,457
R-squared 0.145
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

45
 Temperature in °C Temperature en °C Temperature in °C Temperature in °C
2424.52525.52626.5 27 27.5 28 28.5 24 25 26 27 28 18 19 20 21 22

1950
1950
1950
1950

1970
1970
1970
1970

year
year
year
year

July
April

October
January

1990
1990
1990
1990

2015
2015
2015
2015
Temperature in °C Temperature in °C Temperature in °C Temperature in °C
212223242526 26.52727.52828.5 26.52727.52828.529 192021222324

1950
1950
1950
1950

1970
1970
1970
1970

46
year
year
year
year

May

August

1990
1990
1990
1990
February

November

2015
2015
2015
2015

Temperature in °C Temperature in °C Temperature en °C Temperature in °C

19 20 21 22 23 26 26.5 27 27.5 2727.52828.529 21 22 23 24 25

1950
1950
1950
1950

1970
1970
1970
1970

year
year
year
year

June
March

1990
1990
1990
1990

December
September
Figure A1: Evolution of temperature level by month (1950-2015)

2015
2015
2015
2015
 Precipitation en mm Precipitation en mm Precipitation en mm Precipitation en mm
100200300400500 100200300400500 20406080100 120 0 50 100 150

1950
1950
1950
1950

1970
1970
1970
1970

year
year
year
year

July
April

October
January

1990
1990
1990
1990

2015
2015
2015
2015
Precipitation en mm Precipitation en mm Precipitation en mm Precipitation en mm
0 100200300400 150200250300350
400 100150200250300 0 20 40 60 80

1950
1950
1950
1950

1970
1970
1970
1970

47
year
year
year
year

May

August

1990
1990
1990
1990
February

November

2015
2015
2015
2015

Precipitation en mm Precipitation en mm Precipitation en mm Precipitation en mm

0 100 200 300
150200250300350
400 100200300400500 0 50 100 150

1950
1950
1950
1950

1970
1970
1970
1970

year
year
year
year

June
March

1990
1990
1990
1990

September
Figure A2: Evolution of precipitation level by month (1950-2015)

2015
2015
2015
2015

Precipitation en Décembre