IMPACT OF AGRICULTURAL CREDIT ON AGRICULTURAL

PRODUCTION: EVIDENCE FROM BANGLADESH

Abul Kalam Azad1

Abstract
This article aims to portray the impact of agricultural credit on agricultural
production in Bangladesh using the Johansen co-integration method. However,
the Johansen co-integration test requires the concerned variables to be in the same
order of integration. Augmented Dickey-Fuller (ADF), Phillips Perron (PP), and
Kwiatkowski, Phillips, Schmidt and Shin (KPSS) tests have been performed to
check whether the variables contain a unit root. ADF, PP, and KPSS tests suggest
that all variables- the natural logarithm of agricultural production, agricultural
credit, fertiliser use, and agricultural employment- follow the integration of
order one. The test results show that credit disbursed in agriculture and fertiliser
usage significantly increase agricultural production in the long run. However, our
study suggests that agricultural employment has a negative long-run effect on
agriculture employment. Regarding post-estimation, we did not find any serial
correlation in the Vector Error Correction Model (VECM) model, and residuals
of the VECM model are also normally distributed. Our findings suggest that credit
disbursed in the agricultural sector facilities needs to be augmented to increase
and sustain agricultural output.

Keywords: Agricultural Credit, Agricultural Output, VECM Model, Bangladesh

Introduction
Bangladesh is primarily dependent on agriculture to feed its around 170 million
population. Agriculture is also a primary source of employment generation for
its increasingly growing population. Nearly 50 per cent of the country’s active
labour force directly or indirectly depends on agriculture for livelihood (Ministry
of Finance, 2020). It is evident that the agricultural sector directly determines
the overall soundness of the economy. However, a growing population and rapid
expansion of towns and cities are continuously generating pressures on cultivable
lands of the country. Moreover, arable land has declined by around ten percentage

1
Abul Kalam Azad, Assistant Professor, Department of Economics, University of Dhaka,
Dhaka-1000, Bangladesh. Email: [email protected]

Social Science Review [The Dhaka University Studies, Part-D], Vol. 40, No.1, June 2023
DOI: https://doi.org/10.3329/ssr.v40i1.69078
110 Abul Kalam Azad

points since independence1. Therefore, the declining trend of arable land threatens
the country’s food security.

Due to the declining trend of arable land, agricultural productivity needs to

be increased and sustained to safeguard the country’s food security. Given the
decreased pattern of arable land, higher production requires an improved efficiency
of the production function. One major way to foster the efficiency of the production
function is by injecting agricultural credit. With access to credit, the farmers can
invest the credit in improved seeds, high-yielding varieties (HYV) technology,
fertilisers, and other inputs that can escalate their production. Agricultural credit
also ensures the on-time supply of production factors. The timely usage of seeds,
fertilisers, irrigation, etc., can significantly affect the farm operation (Saboor,
Hussain, & Munir, 2009).

Sustainability in the agricultural sector also requires agricultural diversification.

Although diversification ensures profitability and sustainability in farming, low
diversification is found in Bangladesh due to credit constraints (Azad, 2021). In
addition to the sustainability of agriculture, the country’s continuous economic
growth and development may also be hampered. However, the smooth growth of
agriculture needs unceasing credit injection in this sector. Only the government
can support such investment in the agriculture sector. Realising the significance of
agricultural credit for sustained economic growth, the government of Bangladesh
introduced institutional credit disbursement through its central bank in the 1980s
with a small amount. Besides, the government has been increasing the amount
of credit each year. Therefore, it is also essential to investigate the effectiveness
of agricultural credit on agricultural production and economic growth from a
macro view. Moreover, the agriculture sector is the prime source of food security,
economic development and poverty reduction in the country.

Hence, investigating the causal relationship between agricultural credit disbursement

and agricultural production from the aggregate view is crucial. However, the
researchers in this field mainly examined the effectiveness of credit on production
at the farm level. As a result, a research gap remains in this field of knowledge.
The existing research gap motivated us to study the causality of agricultural credit
and output from the macroeconomic perspective in the economy of Bangladesh
using an appropriate econometric model. Therefore, the prime objective of our
paper is to find out the impact of agricultural credit on the agricultural output of
Bangladesh.
Impact of Agricultural Credit on Agricultural Production: Evidence from Bangladesh 111

In this paper, the first section depicts the study’s background, rationale and research
objective. The second section seeks to give an overall review of the prevailing
literature in the field of agricultural credit. The third section explains data,
variables and methodology, whereas the fourth section covers the estimations and
interpretation of the obtained results. Finally, the fifth section draws concluding
remarks and policy recommendations.

Literature Review
Credit in the agricultural sector plays a key role in making available agricultural
inputs and technological improvements in the production process. Improving
technical efficiency is also essential for smooth production with cost minimisation
(Iqbal et al., 2003). On the other hand, lacking access to credit prevents the farmers
from utilising the full potential of other factors of production. Using state-level
panel data from India, Narayanan (2016) found that other agricultural production
inputs were highly responsive to the increased formal agricultural credit. Saleem
and Jan (2011) found that credit facilities promoted crop production technology
like the green revolution in Pakistan. In addition to the green revolution technology,
increased agricultural credit also facilitated the purchase of other modernised
inputs in Pakistan. However, debate prevails on the efficiency of credit if the
farmers face constraints on other inputs like technical barriers. Taylor, Drummond,
and Gomes (1986) revealed that only credit inflow could not resolve the technical
obstacles of traditional agricultural farmers. Hossain et al. (2019) also suggested
that credit access without relaxing other constraints may not guarantee the profits
of the marginal and tenant farmers.

Despite the debate regarding efficiency, the agricultural credit facility is considered
the most critical and significant factor for small and poor farmers to escalate their
production in Bangladesh (Malek et al., 2022). Lack of access to credit is also
responsible for lagging behind the full utilisation of factors of production. Islam
(2020) showed that a smooth flow of credit is essential for the small and poor farmers
in Bangladesh, ensuring healthy and timely production. Iqbal et al. (2003) found
that institutional credit with labour and irrigation facilities significantly intensified
Pakistan’s agricultural GDP growth. Gershon et al. (2020) showed that the farmers
with agricultural credit enjoyed three times higher production than those without
it in Nigeria. Khandker & Koolwal (2016) found a significant positive effect of
official and microcredit on agricultural output for small landowner households.
112 Abul Kalam Azad

Institutional agricultural credit also has a substantial impact in non-farm production

and income because it overcomes credit barriers. Mitra, Khan, and Nielsen (2019)
found significantly higher output for the credit recipient farmers in Bangladesh.
Azad & Wadood (2017) also depicted that household assets positively affect fisheries
production in Bangladesh. Bidisha et. al (2015) showed that credit-recipient farms
enjoyed better fortune than non-recipient farms. Using panel data, Khandker and
Koolwal (2016) explained that borrowing credit for agricultural purposes has a
strong and positive effect on non-farm income. Therefore, credit support helps
the sustainability of agriculture by expanding its horizon. They also showed that
access to agricultural credit could also increase household consumption.

Although there are many studies regarding agricultural credit and production
worldwide, only a limited number of studies in this field have been carried out in
Bangladesh. Most studies in this field focused on a micro view (Miah et al., 2006;
Rahman, 2011; Bidisha et al., 2015; Mitra et al., 2019). Although some studies
focused on long-run and macro analysis (Rahman, 2011; Alauddin & Biswas,
2014; Khandker & Koolwal, 2016), most studies lacked greater observation or
causal analysis. Besides, empirical evidence using apt methodology is needed to
formulate an inclusive and substantial agricultural credit policy for the country. By
exploring the causal relationship between credit and output, we tried to find the
long-run effect of agricultural credit on agricultural production.

Data, Variables, and Methodology

Data and Variables: The study employs data from secondary sources to inspect
the long-run relationship between agricultural credit disbursement and agricultural
production. We have considered the Cobb-Douglas production function in the
agriculture sector to find the long-run relationship. Agricultural credit, fertilisers,
and employment in agriculture have been considered the independent variables,
whereas total agricultural production has been treated as the dependent variable.
Assuming the Cobb-Douglas function for agricultural production, our long-run
relationship can be modelled as the following relationship
𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 = 𝑓𝑓(𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴, 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹, 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴) (1.1)
Where, AgrProdn stands for total agricultural production of Bangladesh in a million
(Bangladesh Taka) BDT in constant prices using 2005-06 as the base year, AgrCrdt
means total agricultural disbursed credit in a million BDT, Fert implies total used
fertiliser in the country in Metric Ton (M. T.)2, and AgrEmpl means the employment
in agriculture as a percentage of aggregate employment in the country. Data of
Impact of Agricultural Credit on Agricultural Production: Evidence from Bangladesh 113

agricultural output have been collected from Bangladesh Bank, and agricultural
credit data have been obtained from the several yearbooks of Bangladesh Economic
Review, data on fertiliser usage have been accessed from the different statistical
yearbooks of the Bangladesh Bureau of Statistics (BBS), while data of agricultural
employment have been taken from the ILO modelled estimated data3. The
aforementioned time series contain yearly data from 1983 to 20194.

Stationarity Test, Specification of Lag Order, and Testing for Co-integration:

Conventional time series econometrics estimation of regression with non-
stationary variables results in a spurious relationship between the explained and
explanatory variables (Granger & Newbold, 1974), though the regression contains
a high value of and statistically significant value (Enders, 2008). Nevertheless,
the linear combination of non-stationary variables may reflect a meaningful long-
run relationship if all series have identical order of integration. If any meaningful
equilibrium relationship exists, the aberration from the long-run steadiness must
be transitory, resulting in a stationary error term (Enders, 2008). From equation
(1.1), we can rewrite the econometric specification as follows
𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡 = 𝜃𝜃1 + 𝜃𝜃2 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡 + 𝜃𝜃3 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝑡𝑡 + 𝜃𝜃4 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡 + 𝜖𝜖𝑡𝑡 (1.2)
If the error term of equation (1.2) is stationary series, the deviation will be
temporary. Therefore, the linear combination will be stationary, resulting in a
meaningful long-run relationship. Solving for the error term gives us
𝜖𝜖𝑡𝑡 = 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡 − 𝜃𝜃1 − 𝜃𝜃2 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡 − 𝜃𝜃3 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝑡𝑡 − 𝜃𝜃4 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡 (1.3)

As long as is stationary to hold a long-run association, the right hand of the above
equation must be stationary. That means that the linear grouping of with the
integration of order one is also stationary. Hence, their linear combination will
yield a long-run steadiness relationship.

Stationary Test and Specification of Lag Order: the Augmented Dickey-Fuller

(ADF), the Phillips Perron (PP), and the Kwiatkowski, Phillips, Schmidt and Shin
(KPSS) tests of the unit root have been performed to test stationary status of the
variables. The general form of a Data generating Process (DGP) with its first
difference that follows AR (1) process with drift and the deterministic trend can be
modelled by the following regression model:
𝑧𝑧𝑡𝑡 = 𝛾𝛾 + 𝜇𝜇𝜇𝜇 + 𝛿𝛿1 𝑧𝑧𝑡𝑡−1 + 𝜔𝜔𝑡𝑡 (2.1𝑎𝑎)

∆𝑧𝑧𝑡𝑡 = 𝛾𝛾 + 𝜇𝜇𝜇𝜇 + 𝜌𝜌𝑧𝑧𝑡𝑡−1 + 𝜔𝜔𝑡𝑡 (2.1𝑏𝑏)

114 Abul Kalam Azad

Similarly, AR(p) process can be modelled with lag order p

𝑝𝑝

𝑧𝑧𝑡𝑡 = 𝛾𝛾 + 𝜇𝜇𝜇𝜇 + � 𝛿𝛿𝑖𝑖 𝑧𝑧𝑡𝑡−𝑖𝑖 + 𝜔𝜔𝑡𝑡 (2.2)

𝑖𝑖=1

where, t stands for deterministic trend. After subtracting from both sides, the
above equation 2.2 can be written by following the representation of Kirchgässner
& Wolters (2007)
𝑘𝑘

∆𝑧𝑧𝑡𝑡 = 𝛾𝛾 + 𝜇𝜇𝜇𝜇 + 𝜌𝜌𝑧𝑧𝑡𝑡−1 + � 𝜑𝜑𝑚𝑚 ∆𝑧𝑧𝑡𝑡−𝑚𝑚 + 𝜔𝜔𝑡𝑡 (2.3)

𝑚𝑚 =1

the Augmented Dickey-Fuller (ADF) test estimates the above equation under the
null hypothesis of non-stationary variables after Dickey & Fulller (1979), using
the parametric estimation to incorporate serial correlation. Although Phillip-
Perron (PP) test is estimated under the same null hypothesis, it, unlike the ADF
test, adopts the non-parametric method to examine the heteroskedastic disturbance
term (Phillips & Perron, 1988). The Phillip-Perron (PP) test estimates equation
2.1b to investigate the existence of unit root in a series.

Unlike the ADF and PP tests of unit root, the Kwiatkowski, Phillips, Schmidt and
Shin (KPSS) unit root test examine the existence of unit root under the stationary
null hypothesis (Kwiatkowski et al., 1992). Under the hypothesis, DGP can be
modelled by following the exemplification of L¨utkepohl and Kr¨atzig (2004).

𝑦𝑦𝑡𝑡 = 𝛼𝛼𝛼𝛼 + 𝑥𝑥𝑡𝑡 + 𝑣𝑣𝑡𝑡 , 𝑤𝑤ℎ𝑒𝑒𝑒𝑒𝑒𝑒 𝑥𝑥𝑡𝑡 = 𝑥𝑥𝑡𝑡−1 + 𝜀𝜀𝑡𝑡 , 𝜀𝜀𝑡𝑡 ~(0, 𝜎𝜎𝜀𝜀2 ) 𝑎𝑎𝑎𝑎𝑎𝑎 𝑣𝑣𝑡𝑡 𝑖𝑖𝑖𝑖 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 (2.4)

Given the above DGP, D. Kwiatkowski et al., (1992) suggested the KPSS test as
follows:
∑𝑇𝑇1 𝑆𝑆̂ 2
𝐾𝐾𝐾𝐾𝐾𝐾𝐾𝐾 =
𝜏𝜏̂ 2
Where , 𝑆𝑆̂𝑡𝑡 = ∑1 𝑣𝑣�𝑡𝑡 and 𝜏𝜏̂ 2 is the estimation of the long-run variance of 𝑣𝑣𝑡𝑡 . This test
𝑇𝑇

estimate is compared with the critical value at the desired significance level under
with and without liner trend hypothesis.

However, the order of the lag selection is crucial in the autoregressive process.
The lag length of A.R. (p) process can be examined through lag selection criteria
by using the most common methods: AIC, HQIC and SBIC after (Akaike, 1998;
Hannan & Quinn, 1979). AIC, HQIC, and SBIC information criteria can be
Impact of Agricultural Credit on Agricultural Production: Evidence from Bangladesh 115

summarised as follows (L¨utkepohl & Kr¨atzig, 2004)

2
𝐴𝐴𝐴𝐴𝐴𝐴 (𝑛𝑛) = 𝑙𝑙𝑙𝑙𝑙𝑙𝜎𝜎�𝑒𝑒2 (𝑛𝑛) + 𝑛𝑛
𝑁𝑁
2 log log 𝑁𝑁
𝐻𝐻𝐻𝐻𝐻𝐻𝐻𝐻 (𝑛𝑛) = 𝑙𝑙𝑙𝑙𝑙𝑙𝜎𝜎�𝑒𝑒2 (𝑛𝑛) + 𝑛𝑛
𝑇𝑇
log 𝑇𝑇
𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 (𝑛𝑛) = 𝑙𝑙𝑙𝑙𝑙𝑙𝜎𝜎�𝑒𝑒2 (𝑛𝑛) + 𝑛𝑛
𝑇𝑇

where, 𝜎𝜎�𝑒𝑒2 (𝑛𝑛) is the estimate of disterbance variance of OLS residuals from the
autoregressive lag order n, while the capital case letter N indicates the sample size.
The order of lag that minimises the information criteria is selected as the optimal
lag order of the A.R. process.

Testing for Co-integration Rank and Co-integration: To estimate the

cointegrating equation, we have first estimated the appropriate order of lag using
the multivariate generalisation of AIC, HQIC, and SBIC information criteria to
examine the traditional VAR (p) process. Once the appropriate order of p has been
obtained, VAR (p) model in standard form has been estimated in the following
fashion (Enders, 2008),
𝒛𝒛𝒕𝒕 = 𝛱𝛱0 + 𝛱𝛱1 𝒛𝒛𝒕𝒕−𝟏𝟏 + ⋯ + 𝛱𝛱𝑝𝑝 𝒛𝒛𝒕𝒕−𝒑𝒑 + 𝝎𝝎𝒕𝒕 (𝟑𝟑)

where,𝒛𝒛𝒕𝒕 is an (n*1) vector of n (𝑧𝑧1𝑡𝑡 , 𝑧𝑧2𝑡𝑡, … , 𝑧𝑧𝑛𝑛𝑛𝑛 ) variables,𝛱𝛱0 is an (n*1) vector of

drift term,𝛱𝛱𝑖𝑖 is (n*n) coefficient matrices and 𝜔𝜔𝑡𝑡 is the (n*1) vector of disturbance
terms. After estimating the VAR (p) model, the Johansen co-integration test requires
identifying the appropriate rank order for identifying the cointegrating vector and
equation. To identify the appropriate order of rank, trace and max statistics have
been employed as follows:
𝑘𝑘

𝜏𝜏𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 (𝑟𝑟) = −𝑇𝑇 � ln(1 − 𝜏𝜏̂ 𝑖𝑖 ) (4.1)

𝑖𝑖=𝑟𝑟+1

τmax (r, r + 1) = −T ln((1 − τ�r+1 ) (4.2)

where,𝜏𝜏̂ 𝑖𝑖 is the calculated value of eigenvalues, and the number of observations is

given by T. Trace test is run under the null hypothesis that the order of rank of the
cointegrating vector is maximum , while is assumed as the cointegrating vector
under the null hypothesis against the alternative for the max statistic. Once the
rank or cointegrating vector is obtained, the Johansen cointegrating test can be
applied to S. Johansen (1988). From equation (3), the Vector Error Correction
116 Abul Kalam Azad

Model (VECM) can be reparamaterized in the following way

𝑝𝑝−1
∆𝑧𝑧𝑡𝑡 = 𝛱𝛱0 + 𝛽𝛽𝒛𝒛𝑡𝑡−1 + � 𝜃𝜃𝑖𝑖 𝑧𝑧𝑡𝑡−1 + 𝜔𝜔𝑡𝑡 (𝟓𝟓)
𝑖𝑖=1

where, 𝛱𝛱0 is an (n*1) vector of drift, 𝛽𝛽 is the cointegrating vector that shows the
number of linear combinations among the variables of concern. A null matrix of
𝛽𝛽 indicates no cointegrating relationship, while the existence of stationary linear
combination results in some nonzero parameters of 𝛽𝛽 matrices. Therefore, the
trace and max statistics of equations 4.1 and 4.2 will determine the r number of
ranks, given that all series in 𝒛𝒛𝒕𝒕 are I(1) (Engle & Granger, 1987). However, if the
series are not cointegrated, the number of ranks in 𝛽𝛽 in the above VECM equation
will be zero even though all the variables are I(1).

Post Estimation: In order to identify the existence of no serial correlation, we have

estimated the Lagrange-Multiplier test of autocorrelation under the assumption
that there is absence of autocorrelation, i.e.,𝐻𝐻0 : 𝐸𝐸(𝜔𝜔𝑡𝑡 , 𝜔𝜔𝑡𝑡−𝑖𝑖 ) = 0, 𝑖𝑖 = 1, 2, ….. . On
the other hand, Jarque-Bera (J.B.) test for normality has been applied to investigate
the normality assumption of the VECM model under the null hypothesis that the
disterbance terms are normally distributed.

Results
Summary Statistics: The summarised information of the relevant variables
employed in the study has been presented in Table 1 below. Table 1 displays that
the mean value of total agricultural production in Bangladesh during 1983-2019 in
constant prices is around 820 billion BDT using the base year 2005-06. Agricultural
credit disbursement is considered a crucial catalyst for agriculture sector output.
Considering the indispensability of agricultural credit support, the government of
Bangladesh is working on expanding the disbursement of credit in the country.
The country has supported the agriculture sector with a mean credit disbursement
of around 65 billion BDT with more than 15% average credit growth during the
period mentioned above.
Table 1. Summary Statistics
Variables Description of the variables Mean Std. Dev. Minimum Maximum
AgriProdn Total Agricultural Production 816.88 307.20 462.45 1451.37
including forestry and Fish-
eries (Billion BDT)
AgrCrdt Agricultural credit (Billion 63.592 70.04 5.96 236.16
BDT)
Impact of Agricultural Credit on Agricultural Production: Evidence from Bangladesh 117

Fert Total fertilisers use (in thou- 3058.57 1278.88 968.40 5422.00
sand Metric Ton (M.T.)
AgrEmpl Agricultural employment 56.00 9.750 38.30 69.51
(percentage of total employ-
ment)
Source: Authors’ calculation
The country’s agricultural sector has used more than 3000 thousand metric tons,
including all types of fertilisers, on average during the time used. In contrast,
more than 50% of the country’s total employment was employed in agriculture
in the given period. Despite the decline in employment, almost half of the total
employment comes from the agricultural sector in Bangladesh.

Unit Root Test Results: Unit root tests of the series are carried out to study the
stationarity of the variables, as regression of stationary variables is crucial to shun
the spurious regression. Therefore, the classical econometric texts emphasise the
stationary property of series to estimate the regression of the given variables.
According to the stationary property, a 𝒛𝒛𝒕𝒕 sequence is stationary if its disturbance
term has zero mean and constant variance (Enders, 2008). The test of stationarity
is more formally carried out through ADF, PP and KPSS tests. These tests allow
different specifications –for instance, only drift, drift and trend, or without drift
and trend – to estimate the unit root test. While ADF and PP tests allow all three
different specifications in their estimation, the KPSS test allows only drift and
trend stationary specifications. On the other hand, the former two tests perform
the tests under the assumption that the series is non-stationary, while the latter is
carried out under the stationary null hypothesis. Therefore, the acceptance of the
null hypothesis represents the series as the I(0) series, whereas the rejection of
the null regards the series as higher-order integrated. On the other hand, rejection
of the null implies that the series contains a unit root and failure to reject the null
indicates the opposite one under the Kwiatkowski–Phillips–Schmidt–Shin (KPSS)
tests.

Table 2 illustrates the summarised the unit root test results using ADF, PP, and
KPSS tests of unit root. In contrast, Table A1, A2, and A3 in the appendix section
depict the details of test results, including test statistics and probability (p-value)
of the above three tests of the unit root, respectively. Since the test estimates are
sensitive to the number of lag orders, the test results of all specifications have been
presented by minimising the information criteria in line with the direction of AIC,
HQIC, and SBIC in these tables.
118 Abul Kalam Azad

Table 2. Summary of unit root test

Augmented Dick- Phillips-Perron (PP) Kwiatkowski–Phil-

ey-Fuller (ADF) test test lips–Schmidt–Shin
(KPSS) tests
N D DT N D DT DT D
LnAgriProdn S NS NS S NS NS NS NS
D(LnAgrProdn) NS S S S S S NS NS
LnAgrCrdt S NS S S NS NS NS NS
D (LnAgrCrdt) S S S S S S S NS
LnFert S NS NS S NS NS NS NS
D(LnFert) S S S NS S S S NS
AgrEmpl NS NS S S NS NS NS NS
D(AgrEmpl) S NS NS S NS NS S S
Notes: N, D, and DT stands for ‘no Drift and Trend’, ‘Drift only’, ‘both Drift and Trend’,
respectively, whereas NS and S stand for ‘Non-Stationary’ and ‘Stationary’, respectively

The summarised test results offered in Table 2 and the detailed test results
presented in Tables A1, A2, and A3 illustrate that the hypothesis cannot be rejected
in the level form of LnAgriProdn, LnAgrCrdt, and LnFert under ADF and PP
test. In contrast, null can be rejected in the first difference form of LnAgriProdn,
LnAgrCrdt, and LnFert in all specifications except some variations in ‘no drift and
trend’ specification. This implies that ADF and PP test suggest that LnAgriProdn,
LnAgrCrdt and LnFert series are I(1) series. On the other hand, although the KPSS
test can identify LnAgrCrdt and LnFert series as I(1) series in one specification,
it cannot regard LnAgriProdn as I(1) in both of the specifications. However, since
most of the test specifications under ADF, PP and KPSS testify the series as I(1),
we can regard these series as I(1) series. Although the AgrEmpl series is not I(1) in
the specification of ‘drift’ and ‘drift and trend’, it is I(1) in the specification of ‘no
drift and trend’ specification under both ADF and PP tests.

On the other hand, both specifications of the KPSS test classify the AgrEmpl
series as I(1) series. Besides, the Dickey-Fuller test has extremely limited power
to distinguish stationary series under a small number of observations (Enders,
2008). Since our sample is 37 observations, the KPSS test may reflect better
identification resulting AgrEmpl series is I(1) series. Considering all the scenarios,
we can conclude that all series considered in our model are I(1).
Impact of Agricultural Credit on Agricultural Production: Evidence from Bangladesh 119

Test results of Co-integration Rank and Co-integration test: The co-integration

test by Johansen requires that all variables should have identical order of
integration. Unit root tests also confirm that all the concerned variables in our
model have integration of order one. After assessing the same order of integration,
vector autoregressive (VAR) model estimation should be carried out after selecting
the optimal lag order of the VAR model. Moreover, VAR model estimation needs
to be run using the level form data (Enders, 2008). The results of the multivariate
generalisation of AIC, HQIC, and SBIC information criteria estimation have been
scheduled in Table A4 in the appendix section. The test results of AIC and HQIC
show that information criteria are minimised at the length of lag order 4, while
the results of SBIC suggest one lag as the optimal order of lag. Since most of
the information criteria suggest that the optimal lag order should be 4, we have
selected four as the appropriate order of lags for our VAR model. In addition to
identifying the same order of integration and VAR model estimation, the Johansen
methodology requires the estimation of the appropriate rank of matrix to assess the
number of linear combinations among the variables. The Johansen co-integration
rank test results have been pictured in Table 3.
Table 3: The Johansen Co-integration rank test

Max- Eigen- Trace Test of cointegrating Max Test of cointegrating

imum values vector vector
Rank
Trace sta- 5% crit- 1% crit- Max statis- 5% 1%
tistics ical value ical value tics critical critical
value value
0 - 85.556 47.21 54.46 55.572 27.07 32.24
1 0.814 29.985*** 29.68 35.65 20.732*** 20.97 25.52
2 0.466 9.253** 15.41 20.04 8.964 14.07 18.63
3 0.238 0.289 3.76 6.65 0.289 3.76 6.65
4 0.009 - - - - - -
Notes: *** and ** stands for statistical significance at 1% and 5% respectively * p<0.10,
** p<0.05, *** p<0.010.

Table 3 depicts that the optimal rank order is one under both Trace and Max test of
co-integration at the 1% level of significance. However, the Trace test also shows
vector autoregressive model can also contain two cointegrating vectors at 5% level
of significance. Since both the Trace test and Max test of co-integration confirm 1
cointegrating vector at 1% significance level, we have selected 1 as the maximum
number of rank and hence the cointegrating vector.
120 Abul Kalam Azad

Given that we have the appropriate order of cointegrating vector r=1, and the optimal
number of multivariate lag lengths, we can estimate the vector autoregressive
model (VECM) modelled by equation five after S. Johansen (1988). The test results
of the long-run equilibrium relationship of the Johansen co-integration test have
been presented in Table 4 below. The results have been derived after imposing the
Johansen normalisation restriction.
Table 4: Estimated Results from the Johansen Co-integrating Equations

Variables Coefficients Standard Error z-statistic

LnAgriProdn 1 - -
LnAgrCrdt -0.1401496 0.016*** -8.72
LnFert -0.2887401 0.031*** -9.36
AgrEmpl 0.0101444 0.001*** 8.39
Constant -10.44596 - -
Notes: *** stands for statistical significance at 1% statistical level * p<0.10, ** p<0.05,
*** p<0.010.

Table 4 shows that all long-run coefficients are significant at a 1% significance level.
Therefore, all of the coefficients are highly statistically significant. According to
the estimated results, agricultural credit and fertiliser use have a long-run positive
effect on agricultural production in Bangladesh. Since the coefficients are taken
in the natural logarithm form, we can interpret the obtained results in terms of
elasticity. Assuming all other variables- fertiliser use and employment - as constant
in the agricultural sector, a 1% increase of agricultural credit disbursement
increases the agricultural production on average by 0.14%. The statistically
significant coefficient implies that agricultural credit disbursement significantly
impacts agricultural production in Bangladesh. On the other hand, holding
agricultural credit and employment constant, 0.28% of agricultural production is
escalated by a 1% increase in fertiliser use. Therefore, it is evident that fertilisers
also have a significant role in agricultural production. However, although one can
expect that employment in the agricultural sector can increase agricultural output,
the obtained results have revealed the exact opposite result. The opposite result
may come because of several reasons. Firstly, Bangladesh has already experienced
disguised unemployment in the agriculture sector that may negatively impact
output (Jabbar, 1988). Secondly, the share of employment in the agricultural sector
has been continuously being decreased since independence (Ministry of Finance,
2020).
Impact of Agricultural Credit on Agricultural Production: Evidence from Bangladesh 121

Post estimation plays a vital role in shaping how the model is good and how
the model fits the data well. We have estimated the Lagrange-Multiplier test of
autocorrelation and the Jarque-Bera test of normality to check the model’s fitness.
Table A5 of the appendix section presents the result of the Lagrange-Multiplier
test of autocorrelation, while Table A6 of the appendix shows the test result of the
Jarque-Bera normality test. Lagrange-Multiplier test displays that there is no serial
correlation in the model. Besides, Table A6 also shows that residuals of the VECM
model follow the normal distribution. The cointegrating equation and the estimated
residuals have been graphed in figure A2 in the appendix section. Besides, stability
test also revealed that the VECM model is highly stable as all of the roots of the
VECM model are within the unit value. Overall, all post estimations tests predict
that the chosen VECM model fits the data well.

Conclusion and Discussion

Credit facility in the context of Bangladesh is essential for poor farmers. Access to
credit services helps the farmers enhance their agricultural production and escalate
the productivity of other inputs. Our obtained findings confirmed that agricultural
credit and fertiliser usage significantly increases the agricultural output in the long
run (Iqbal et al., 2003; Saboor et al., 2009). These findings may significantly impact
the agricultural policy formulation of a capital-scarce country like Bangladesh.
However, a surge in agricultural employment cannot increase the outcome of this
ancient means of livelihood. Even we found a contrary result for employment
input. It may imply that the agriculture sector of Bangladesh has already employed
more than the optimum level of labour input and the marginal productivity of
labour in agriculture is negative. Therefore, the farmers should employ more non-
labour inputs to enhance the effectiveness of their production function.

The sustained growth of agricultural production has showed a key role in alleviating
the perils of hunger and reducing poverty since independence in Bangladesh. The
ancient sector is also considered one of the main catalysts of economic growth.
However, technological innovation, intensive use of mechanisation, credit facility,
etc., have contributed to sustained agricultural growth. Among the factors,
availability and accessibility of credit in the agricultural sector is indispensable
to facilitate agricultural expansion in a country like Bangladesh, where credit
constraint is a fundamental problem for poor farmers. We found that agricultural
credit has played a significant role in achieving continuous and sustained
agricultural growth through a comprehensive investigation. The obtained findings
have revealed that an additional 1% increase in agricultural credit disbursement
122 Abul Kalam Azad

significantly increases the agricultural output by around 0.14%, holding all other
inputs constant. In addition, fertiliser use also significantly increases the agricultural
output. According to the results, a 0.28 % additional agricultural output can be
achieved through a one % increase in fertiliser. It indicates that agricultural credit
support needs to be expanded among the farmers, especially poor and small farmers,
as agriculture credit has a noticeable impact on agricultural output. Though the
government of Bangladesh is expanding the agricultural credit program among
the farmers, these are not sufficient compared to its enormous demand. Therefore,
based on the findings, we suggest that the government should expand and continue
credit support to the agricultural sector at an affordable cost. Ensuring agricultural
credit support can turn the agriculture sector of Bangladesh into more sustainable
by fostering sustained agricultural production.

Notes
1
Arable land (% of total land) was about 70% in 1972, while it is now around 60% of
total land (data have been extracted from World bank Development indicators at 16th
January 2022.
2
Natural logarithm has been taken for variables, whereas AgrEmpl variable has been
kept as level for the analysis.
3
ILO modelled estimate provides data only from 1991. Data of earlier years have been
collected from different statistical yearbooks of BBS, as the study has covered the
yearly data from 1983. However, we did not find the employment data for 1983, 1986
and 1987 in the available statistical publications. We employed the linear estimation
technique of missing data to generate the data of missing years. We calculate the
𝑚𝑚 −𝑚𝑚 0
missing values using the formula, 𝑚𝑚 = 𝑥𝑥 1 −𝑥𝑥 0 (𝑥𝑥 − 𝑥𝑥0 ) + 𝑚𝑚0 , where we have calculated
m at x by using the closest point of data between (𝑚𝑚1 , 𝑥𝑥1 ) and (𝑚𝑚0 , 𝑥𝑥0 ).
4
Agricultural production, agricultural credit and usage of fertilisers data were in fiscal
year, while employment data were in calendar year format. Nevertheless, we have
used calendar year data and treated fiscal year, for instance, 2982-83 as 1983 calendar
year format.

References
Akaike, H. (1998). Information Theory and an Extension of the Maximum Likelihood Principle. In
E. Parzen, K. Tanabe & G. Kitagawa (Eds.), Selected Papers of Hirotugu Akaike (pp. 199-
213). Springer. doi.org/10.1007/978-1-4612-1694-0_15
Alauddin, M., & Biswas, J. (2014). Agricultural credit in Bangladesh: Present trend, problems and
recommendations. Journal of Economics and Sustainable Development, 5(27), 11-22. Available
at: citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.663.3720&rep=rep1&type=pdf
(Accessed on April 23, 2021).
Impact of Agricultural Credit on Agricultural Production: Evidence from Bangladesh 123

Azad, A. K. (2021). Determinants of Crop Diversification in Bangladesh: An Econometric

Analysis. Asia-Pacific Journal of Rural Development, 31(2), 195–217. doi.
org/10.1177/24551333211069752
Azad, A. K., & Wadood, S. N. (2017). Climate change and Bangladesh fisheries and aquaculture:
Evidences from the Household Income and Expenditure Survey (HIES) 2010. Social Science
Review, 34(1), 193–210. Available at: mpra.ub.uni-muenchen.de/81552/ (Accessed on April
23, 2021).
Bidisha, S. H., Khan A., Khondker B. H., & Imran K. (2015). Returns to Agricultural Microcredit:
Quasi-experimental Evidence from Bangladesh. The Bangladesh Development Studies, 38(4),
31-46. Available at: www.jstor.org/stable/pdf/26538717.pdf (Accessed on April 23, 2021).
Dickey, D. A., & Fuller, W. A. (1979). Distribution of the estimators for autoregressive time series
with a unit root. Journal of the American Statistical Association, 74(366a), 427-431. doi.org/
10.1080/01621459.1979.10482531 (Accessed on April 23, 2021).
Enders, W. (2008). Applied conometric time series (2nd edition). John Wiley & Sons. Inc.
Engle, R. F., & Granger, C. W. (1987). Co-integration and error correction: representation, estimation,
and testing. Econometrica, 55(2), 251-276. Available at: www.jstor.org/stable/1913236?seq=1
(Accessed on April 23, 2021).
Gershon, O., Matthew, O., Osuagwu, E., Osabohien, R., Ekhator-Mobayode, U.E., & Osabuohien,
E. (2020). Household access to agricultural credit and agricultural production in Nigeria:
A propensity score matching model. South African Journal of Economic and Management
Sciences, 23(1), 1-11. Available at: www.researchgate.net/publication/338820870_
Household_access_to_agricultural_credit_and_agricultural_production_in_Nigeria_A_
propensity_score_matching_model (Accessed on April 23, 2021).
Granger, C., & Newbold, P. (1974). Spurious regressions in econometrics. Journal of Econometrics, 2
(1974), 111-120. Available at: www.climateaudit.info/pdf/others/granger.1974.pdf (Accessed
on August 12, 2021).
Hannan, E. J., & Quinn, B. G. (1979). The determination of the order of an autoregression. Journal
of the Royal Statistical Society, 41(2), 190–195. doi.org/10.1111/j.2517-6161.1979.tb01072.x
Hossain, M., Malek, M. A., Hossain, M. A., Reza, M. H., & Ahmed, M. S. (2019). Agricultural
microcredit for tenant farmers: Evidence from a field experiment in Bangladesh. American
Journal of Agricultural Economics, 101(3), 692-709. doi.org/10.1093/ajae/aay070
Iqbal, M., Ahmad, M., Abbas, K., & Mustafa, K. (2003). The impact of institutional credit on
agricultural production in Pakistan [with comments]. The Pakistan Development Review,
42(4), 469-485. Available at: www.jstor.org/stable/41260420?seq=1 (Accessed on August 12,
2021).
Islam, M. M. (2020). Agricultural credit and agricultural productivity in Bangladesh: An econometric
approach. International Journal of Food and Agricultural Economics, 8(3), 247-255.
Available at: ageconsearch.umn.edu/record/305327/ (Accessed on August 12, 2021).
Jabbar, M. A. (1988). Some aspects of unemployment, migraton and income in the Bangladesh rural
labour market. Bangladesh Journal of Agricultural Economics, 11(2), 35-48. Available at:
ageconsearch.umn.edu/record/208734 (Accessed on August 12, 2021).
Johansen, S. (1988). Statistical analysis of co-integration vectors. Journal of Economic Dynamics
and Control, 12(2-3), 231-254. doi.org/10.1016/0165-1889(88)90041-3
124 Abul Kalam Azad

Khandker, S. R., & Koolwal, G. B. (2016). How has microcredit supported agriculture? Evidence
using panel data from Bangladesh. Agricultural Economics, 47(2), 157-168. doi.org/10.1111/
agec.12185
Kirchgässner, G., & Wolters, J. (2007). Introduction to modern time series analysis. Springer, Science
& Business Media.
Kwiatkowski, D., Phillips, P. C., Schm, P., & Shin, Y. (1992). Testing the Null Hypothesis of
Stationarity Against the Alternative of a Unit Root. Journal of Econometrics, 54(1992), 159-
178. doi.org/10.1016/0304-4076(92)90104-Y
L¨utkepohl, H., & Kr¨atzig, M. (2004). Applied time series econometrics. Cambridge University
Press.
Malek, A. M., Kikkawa, A., Azad., A. K. & Sawada, Y. (2022). Rural Development in Bangladesh
over Three Decades: Findings from Mahabub Hossain Panel Data and the Way Forward
(Forthcoming as ADBI Working Paper).
Miah, M. K., Alam, A. A., & Rahman, A. H. M. A. (2006). Impact of agricultural credit on MV Boro
Rice cultivation in Bangladesh. Journal of Agriculture & Rural Development, 4(1), 161-168.
doi.org/10.3329/jard.v4i1.784
Ministry of Finance. (2020). Bangladesh Economic Review. Ministry of Finance, Governement
of Bangladesh (GoB). Avalable at: mof.portal.gov.bd/site/page/28ba57f5-59ff-4426-970a-
bf014242179e/Bangladesh-Economic-Review-2021 (Accessed on August 12, 2021).
Mitra, S., Khan, M.A., & Nielsen, R. (2019). Credit constraints and aquaculture productivity.
Aquaculture Economics & Management, 23(4), 410-427. doi.org/10.1080/13657305.2019.1
641571
Narayanan, S. (2016). The productivity of agricultural credit in India. Agricultural Economics, 47(4),
399-409. doi.org/10.1111/agec.12239
Phillips, P. C., & Perron, P. (1988). Testing for a unit root in time series regression. Biometrika, 75(2),
335-346. doi.org/10.1093/biomet/75.2.335
Rahman, M.W. (2011). Policies and performances of agricultural/rural credit in Bangladesh: What
is the influence on agricultural production? African Journal of Agricultural Research, 6(31),
6440-6452. Available at: www.academicjournals.org/AJAR (Accessed on August 12, 2021).
Saboor, A., Hussain, M., & Munir, M. (2009). Impact of micro credit in alleviating poverty: An
Insight from rural Rawalpindi, Pakistan. Pak. J. Life Soc. Sci, 7(1), 90-97. Available at: www.
pjlss.edu.pk/pdf_files/2009_1/90-97.pdf (Accessed on August 12, 2021).
Saleem, M. A., & Jan, F. A. (2011). The impact of agricultural credit on agricultural productivity in
Dera Ismail Khan (District) Khyber Pakhtunkhwa Pakistan. European Journal of Business
and Management, 3(2), 38-44. Available at: www.iiste.org/Journals/index.php/EJBM/article/
view/159 (Accessed on August 12, 2021).
Taylor, T.G., Drummond, H. E., & Gomes, A. T. (1986). Agricultural credit programs and production
efficiency: an analysis of traditional farming in south-eastern Minas Gerais, Brazil. American
Journal of Agricultural Economics, 68(1), 110-119. doi.org/10.2307/1241655
Impact of Agricultural Credit on Agricultural Production: Evidence from Bangladesh 125

Appendices
Table A1: Augmented Dickey-Fuller Unit Root Test Results

H0=The series has unit root

Variables
Model specification
Test Statistics Test Statistics Test Statistics (Constant
(None) (Constant) & Trend)
LnAgriProdn 4.332*** 1.245 -2.520
D(LnAgrProdn) -1.519 -4.494*** -4.934***
LnAgrCrdt 2.738** 0.535 -3.558**
D (LnAgrCrdt) -3.811*** -5.033*** -5.440***
LnFert 2.209** -2.479 -2.845
D(LnFert) -1.945* -3.053** -3.491**
AgrEmpl -0.815 -0.390 -3.760**
D(AgrEmpl) -2.194** -2.382 -2.970
Note: * p<0.10, ** p<0.05, *** p<0.010.

Table A2: Phillips-Perron Unit Root Test Results

Variables H0=The series has unit root

Model specification
Test Statistics Test Statistics Test Statistics (Constant
(None) (Constant) & Trend)
LnAgriProdn 8.236*** 1.539 -2.419
D(LnAgrProdn) -2.108*** -5.088*** -5.432***
LnAgrCrdt 3.154*** 0.155 -2.522
D (LnAgrCrdt) -5.471 -6.535*** -6.727***
LnFert 3.372*** -2.429 -2.666
D(LnFert) -5.999 -6.903*** -7.181***
AgrEmpl -1.422*** 0.398 -1.703
D(AgrEmpl) -2.257** -2.415 -2.818
Note: Note: ** p<0.05, *** p<0.010
126 Abul Kalam Azad

Table A3: Kwiatkowski, Phillips, Schmidt and Shin (KPSS) Unit Root Test Results

Variables H0=The series has unit root

Model specification
Test Statistics (Constant) Test Statistics (Constant &
Trend)
LnAgriProdn 0.492** 0.143*
D(LnAgrProdn) 0.367* 0.156**
LnAgrCrdt 0.474** 0.127*
D (LnAgrCrdt) 0.201 0.139*
LnFert 0.499** 0.134**
D(LnFert) 0.286 0.136*
AgrEmpl 0.396* 0.131*
D(AgrEmpl) 0.274 0.116
Note: * p<0.10, ** p<0.05. Without trend, the critical values are 0.739, 0.463 and 0.347
for 1%, 5% and 10% level of significance respectively, whereas with trend the critical
values are 0.216, 0.146 and 0.119 for 1%, 5% and 10% level of significance, respectively.

Table A4: Selection-order criteria of VAR (p) model where, LnAgriProdn, LnAgrCrdt,
LnFert, LnAgrEmpl are the endogenous variables

Lags LR AIC HQIC SBIC

0 - 3.98958 4.05061 4.17097
1 315.72 -4.60786 -4.3027 -3.70089*
2 48.168 -5.09782 -4.54851 -3.46526
3 35.765 -5.2119 -4.41846 -2.85376
4 51.882* -5.81439* -4.77682* -2.73068
Notes: * indicates optimal order of lag

Table A5: Lagrange-multiplier autocorr*elation test

Lag order (K) Ho= no autocorrelation at lag order (K)

Chi-square statistic Probability
1 16.1422 0.44308
2 10.1184 0.86037
3 6.8684 0.97571
4 12.6129 0.70082
Source: Authors’ calculation
Impact of Agricultural Credit on Agricultural Production: Evidence from Bangladesh 127

Table A6: Jarque-Bera test of normality

Equation Chi-square statistic Probability

D_lnagrigdpm 0.435 0.804
D_lnagricreditm 1.763 0.414
D_agriemppercenttot 1.415 0.4493
D_lnfertthousmt 7.887 0.019
ALL 11.500 0.175
Source: Authors’ calculation

Figure A1: Time series line graph of LnAgriProdn, LnAgrCrdt, LnFert, & AgrEmpl with
level and first order, respectively.
14.5

.1
Natural logarithm of gdpagrmillion, D
Natural logarithm of gdpagrmillion
14

.05
13.5

0
13

1980 1990 2000 2010 2020

1980 1990 2000 2010 2020 Year
Year
13

.5
Natural logarithm of agricreditmillion, D
Natural logarithm of agricreditmillion
11 12

0
10

-.5
9

1980 1990 2000 2010 2020 1980 1990 2000 2010 2020
Year Year
.2
8.5

Natural logarithm of fertithousandmt, D

Natural logarithm of fertithousandmt

.1
8

-.1 0
7.5

-.2
7

-.3

1980 1990 2000 2010 2020

Year 1980 1990 2000 2010 2020
Year
70

4
agricultural employment as % of GDP, D
agricultural employment as % of GDP

2
60

0
50

-2
40

-4

1980 1990 2000 2010 2020 1980 1990 2000 2010 2020
Year Year

Source: Prepared by authors

128 Abul Kalam Azad

Figure A2: Time Series line graph of predicted cointegrating equation and residuals,
respectively

.03
.05

.02
Predicted cointegrated equation
0

.01
-.05

Residuals
0
-.1

-.01
-.15

-.02
-.2

1980 1990 2000 2010 2020 2030 1980 1990 2000 2010 2020 2030
Year Year

Source: Prepared by authors