Evaluation of Abrams curve in agricultural sector using the NARDL approach

Seyed Yaghoub Zeraatkish; Mohammad Javad Rajabalipour; Reza Moghaddasi

doi:10.1515/opag-2022-0278

Article Open Access

Evaluation of Abrams curve in agricultural sector using the NARDL approach

Seyed Yaghoub Zeraatkish , Mohammad Javad Rajabalipour and Reza Moghaddasi

Published/Copyright: November 25, 2024

Published by

Become an author with De Gruyter Brill

Submit Manuscript Author Information Explore this Subject

From the journal Open Agriculture Volume 9 Issue 1

Abstract

Developing countries have made significant economic progress in the recent decades, but agriculture sector has not produced enough value add for the economy due to the low productivity in this sector, which contributes to poverty and delays in the growth of the entire economy in such countries. The aim of this research is to evaluate Abram’s theory in the agricultural sector of Iran using the nonlinear autoregressive distributed lag (NARDL) approach over the years 1981–2021. To this end, the effect of government size, agricultural sector wages, value add of the agricultural sector, inflation rate, mechanization coefficient, and the ratio of agricultural sector wages to other economic sectors on the unemployment rate of the agricultural sector were investigated. In the first step, stationarity of variables was examined by applying the generalized Dickey–Fuller and Phillips–Peron methods. Outcomes of the unit root test showed that the degree of cointegration between variables was zero and one, and there were no limitations to the application of the NARDL approach. Results showed that the Abrams curve would not be confirmed for the agricultural sector in Iran, because the coefficient obtained for the size of the state was negative. Also, the findings suggest that the government’s size affects the unemployment rate in the agricultural sector, while real wages in Iran’s agricultural sector impact unemployment.

Keywords: Abram’s theory; NARDL approach; ECM; Iran

1 Introduction

One of the most important sectors of Iran’s economy is the agricultural sector because it constitutes a significant share in Gross Domestic Product (GDP) and employment. Considering the progress of economic activity in many developing countries in recent decades including Iran, agriculture is still far behind, and low productivity in this sector is one of the main causes of poverty and delays in the development of the entire economy of such countries. Obviously, this sector requires the government aids due to various reasons such as its high risk, poverty of farmers, and the inability of the private sector to invest in infrastructure [1]. Therefore, the government should provide support to combat the low productivity in agricultural sector [2].

According to Ansari Samani and Khilkordi, one of the variables that can be influenced by the extent of government intervention is employment and, consequently, the unemployment rate [2].

Therefore, here, we try to discuss the effect of government intervention on the unemployment rate for the agricultural sector of Iran relying on the study of Abram’s curve. This curve was first described by Abram [3], and then a study by Christopoulos et al. showed that there is a positive relationship between government size and unemployment rate in developed countries and this relationship became known as the Abram’s curve [4].

While research on the relationship between unemployment and the government size is not entirely clear, a wide range of evidence supports the idea that government size has a negative impact on the growth rate of production. Meanwhile, Abrams realized the existence of a strong and positive relationship between these two variables, which was later introduced as the Abram’s curve [3].

2 Literature review

Various studies were done to show the effect of this curve, for instance, Ansari Samani and Khilkordi examined the Abrams curve in developing and developed countries according to annual data from 2000 to 2013 and relying on data panel econometric method and error correction model. The results of long-term estimations in both groups of countries indicated a significant correlation between government size and unemployment rate. This relationship is negative and significant in developing countries, meaning that reducing unemployment requires increasing the size of government. Likewise, based on the research results, in developed countries, government size has a positive and significant relationship with the unemployment rate, so results showed that the Abrams curve is confirmed in developed countries, but for developing countries, the situation is reversed [2]. Sunan, examined the impact of government and private investment on the rate of employment and claimed that government and private investments have a significant and positive impact on this rate [5]. This means that increasing government spending and private investment will also increase employment rate.

Afonso et al. investigated the relationship between the size of the government and two other macroeconomic variables, namely, unemployment and inflation in eight major economies of the world between 1980 and 2015. There is a positive correlation between government size and both unemployment and inflation, according to the results. There are two aspects that stand out from their analysis. Government size has a significant impact on unemployment and inflation depending on how it is measured. The government size is significantly and positively correlated to both unemployment and inflation when government consumption spending is used as a proxy measure. Additionally, indirect taxes, like government consumption spending, are positively and statistically associated with unemployment. Nevertheless, direct taxes alone exert a strong influence on inflation in the countries studied [6]. Mufeed et al. investigated the relationship between government size and unemployment rate in 17 MENA countries from 2003 to 2017 using seemingly unrelated regression models (SURs). They found that government size negatively affecting the labour market has a negative and statistically significant impact on the labour market. A dampening effect was also found on the labour market by total government expenditures as well as investment expenditures. They argued that government size and unemployment rate are causally related in a two-way manner [7]. Montazeri Shoorekchali and Zehed Gharavi investigated the effect of government size on unemployment using the Markov-Switching approach in Iran for the period 1979–2018. Their findings showed that during recessions, expansionary fiscal policy had a significant negative effect on unemployment. While in the non-recession period (years with lower unemployment), there was no evidence of a significant effect of government size on unemployment in Iran. Finally, consistent with Okun’s law, their findings showed that real economic growth has a significant negative effect on the unemployment rate [8]. Fiaz et al. conducted a research project to explore the potential uneven impact of currency fluctuations on the agricultural industry by employing a nonlinear autoregressive distributed lag (NARDL) model. The findings from the NARDL analysis indicate that favourable shifts in exchange rates have a smaller influence compared to unfavourable changes in the agricultural sector, both in the short term and over an extended period [9].

Sama et al. examined the relationship between the Human development index (HDI), GDP, inflation, and CO₂ emissions in relation to crude oil production (COP) in Cameroon over the period from 1977 to 2019. They utilized stationary tests such as Augmented Dicky–Fuller and Zivot–Andrews, as well as Autoregressive distributed lag model (ARDL) and NARDL modelling, alongside the Toda–Yamamoto causality test. The outcomes suggest that CO₂ emissions and GDP impact COP negatively in the long term, while HDI and inflation have a positive effect in the short term. Additionally, there is a nonlinear relationship between GDP and HDI in the short term, as well as between inflation, CO₂ emissions, and COP in the long term [10].

Agriculture is the centre of growth and development in Iran but due to the lack of strong institutions in Iran that can bring an appropriate share of government funds and expenses to this sector and the government’s inadequate attention to this sector, the share of government funds and expenses in this sector is sufficient and proportionate to the importance of this sector. So far, several studies have been conducted regarding the size of the government and employment in different economic sectors of countries as well as the agricultural sector. But the important thing that has been neglected is the neglect of Abrams’ theory on Iran’s agricultural sector, on the government size, real wages on the unemployment rate, inflation on the unemployment rate, agricultural value add on the unemployment rate, and mechanization coefficient on the unemployment rate.

3 Data and methodologies

The present study was practical in terms of its purpose, because it used the background and cognitive context and information provided through basic research. In terms of method and nature, this study was both descriptive and correlational. In this study, we applied NARDL approach to analyze the model in the agricultural sector of Iran for the period 1981–2017. In fact, the NARDL model, an asymmetric (nonlinear) single equation method, was utilized in this study. In other words, the NARDL is an extended model of the ARDL model. The Eviews11 software was used to evaluate and test hypotheses. The proposed research model is as follows:

(1) U t = α 0 + β 1 Gov t + β 2 RW t + β 3 Y t + β 4 P t + β 5 M t + β 6 W t + ε t ,

where U represents the unemployment rate in the agricultural sector, Gov is the size of government (ratio of government spending to GDP), RW is the real wage of the agricultural sector, Y is the value add of the agricultural sector, P is the inflation, M is the mechanization coefficient, and W is the ratio of wages in the agricultural sector to wages in other economic sectors.

3.1 Description of tests used:

3.1.1 Generalized Dickey–Fuller unit root test

For the unit root test of Dickey and Fuller [11], the following three regression equations were considered [12]:

(2) Δ y t = γ Δ y t − 1 + ε t ,

(3) Δ y t = a 0 + γ y t − 1 + ε t ,

(4) Δ y t = a 0 + γ y t − 1 + a 2 t + ε t .

In the listed equations, the coefficient γ will be tested. If γ = 0, then the series { y t } has a single root. We calculate the t-statistic for coefficients γ in the equations and then compare it with the critical values of the Dickey–Fuller table. Dickey and Fuller named the critical values of the coefficients γ in the equations τ, τμ, and ττ, respectively. Therefore, if the computational values of t are less than the critical values of the table, the null hypothesis that there is a unit root is accepted. Dickey and Fuller generalized their regression equations and added autoregressive components to these equations. In this case, the critical values of the table τ, τμ, and ττ remain unchanged.

In addition to the γ = 0 test, Dickey and Fuller [11] introduced three other tests. These new tests applied multiple constraints to the coefficients of the equations. These tests are measured by statistics ϕ ₁, ϕ ₂, and ϕ ₃.

Hypothesis γ = a ₀ = 0 is tested using ϕ ₁, hypothesis γ = a ₀ = a ₂ = 0 using ϕ ₂, and hypothesis γ = a ₂ = 0 is tested using ϕ ₂. The general formula for ϕ _i statistics is as follows:

(5) ϕ i = [ SSR ( restricted ) − SSR ( unrestricted ) ] r SSR ( unrestricted ) ( T − k ) ,

where SSR ( restricted ) and SSR (unrestricted) represent the sum of the squared residuals of the restricted and unrestricted model, r = the number of constraints, T = the number of usable observations, and k is the estimated parameters of the unrestricted model. Null hypothesis also represents the restricted model. Therefore, if the computational value ϕ _i is less than the critical value of the table, the null hypothesis cannot be rejected. As a result, you accept the constrained model.

3.1.2 Phillips and Peron unit root test

Phillips and Perron [13] proposed a non-parametric unit root test known as PP for serial correlation control. The PP method estimates the regression of the generalized Dickey–Fuller test and with heteroscedasticity and autocorrelation consistent covariance matrix so that the serial correlation does not affect the asymptotic distribution of the test statistic.

(6) ∆ y t = α y t − 1 + x t ′ δ + ϵ t ,

where x _t represents the vector of exogenous variables as well as the y-intercept or time trend. ϵ _t is assumed to be a white noise. The statistical hypotheses of this test are defined as follows:

H 0 : α = 0 ,

H 1 : α < 0 .

Under the null hypothesis, the time series is non-stationary and the opposite hypothesis implies the stationary of the time series. The ratio t of the coefficient α in the Dickey–Fuller test is calculated as follows:

(7) t α = α ˆ ( se ( α ˆ ) ) ,

where α ˆ is an estimate of the coefficient α and se ( α ˆ ) is the standard error of this coefficient. PP test is based on t ∼α statistic.

(8) t ∼ α = t α γ 0 f 0 1 / 2 − T ( γ 0 − f 0 ) ( se ( α ˆ ) ) 2 f 0 1 / 2 s ,

where s is the standard error of the test regression and γ ₀ represents the consistency estimate of the variance of the regression equation error (∆y _t). f ₀ is also an estimate of the residual spectrum at the frequency source. The asymptotic distribution of the t ∼α statistic is the same as the distribution of the generalized Dickey–Fuller statistic.

3.1.3 ARDL

The ARDL model was used by Pesaran and Shin [14] and Pesaran et al. [15] as a self-correlation approach with distributive interruptions. They proved that if the coexistence vector obtained by applying the ordinary least squares method in a self-correlated pattern with well-defined distribution intervals, in addition to having a normal distribution, in the sample, smaller ones were less biased and more efficient.

This approach has special advantages compared to previous methods. First, this approach distinguishes between dependent and explanatory variables and solves the problem of endogeneity. Second, it estimates long-term and short-term components simultaneously and solves the problems of omitted variables and auto-correlation. Third, it is one of the methods in which the degree of stationary of the variables does not have to be the same, and only by determining the appropriate intervals for the variables, the appropriate model can be selected.

Fourth, avoiding the existing defects of other models, including the presence of bias in small samples and the inability to test statistical hypotheses, led us to more appropriate methods for analyzing long-term and short-term relationships between variables, including the approach leads.

In general, a dynamic pattern is a pattern in which the intervals of variables are also entered, such as the following pattern [16]:

(9) Y t = α X t + β X t − 1 + δ Y t − 1 + U t .

However, to reduce the bias associated with estimating pattern coefficients in small samples, it is best to use a pattern that takes into account the large number of interrupts for the variables as much as possible [16].

(10) ∅ ( L . P ) Y t = ∑ i = 1 k b i ( L . q i ) X it + c W t + u t .

The above pattern is called an autocorrelation pattern with distributive interrupts, which is represented by ARDL in which

(11) ∅ ( L . P ) = 1 − ∅ 1 L − ∅ 2 L 2 − … − ∅ p L p ,

(12) b i ( L . q i ) = b i 0 + b it L + … + b iq L q . i = 1 , 2 . … . k ,

where L is the interrupt operator, W is the vector of fixed variables such as y-intercept, virtual variables, time trend, or exogenous variables with fixed interrupt. Microphite software estimates the equation for all cases and for all possible arrangements of values, to number ( m + 1 ) k + 1 . m is the maximum interval determined by the researcher.

In the next stage, one of these equations is selected based on the criteria of Akaike, Schwarz-Bayesian, or adjusted coefficient of determination. Usually in samples less than 100, the Schwarz-Bayesian criterion is used, so as not to lose much degree of freedom [17]. The same dynamic model is used to calculate the long-term coefficients of the model. The long-term coefficients related to the X variables are obtained from the following equation:

(13) θ i = b i ( L . q i ) ∅ ( L . P ) = b i 0 + b it + … + b iq 1 − ∅ 1 − ∅ 2 − … − ∅ p i = 1 , 2 . … . k .

Now, to verify that the long-term relationship obtained from this method is not false, the following hypothesis is tested:

(14) H 0 = ∑ i = 1 p ∅ i − 1 ≥ 0 ,

(15) H a = ∑ i = 1 p ∅ i − 1 < 0 .

Hypothesis zero indicates the absence of co-existence or long-term relationship, because the condition for a short-term dynamic relationship to tend toward a long-term equilibrium is that the sum of the coefficients is less than one. For the above test, it is sufficient to calculate the difference of one from the sum of the coefficients with the interval of the dependent variable and divide it by the sum of the standard deviation of the mentioned coefficients. If the absolute value of t is greater than the absolute value of the critical values presented by Banerjee et al. [18], we reject the null hypothesis and accept the existence of a long-term relationship.

(16) Y t = α X t + β X t − 1 + δ Y t − 1 + u t .

In this regard, the principle of simplicity of explanatory variables dictates that a model should be considered as simple as possible. This indicates that in order to get the basis of the phenomenon under study, only important variables should be entered according to the theoretical framework and theoretical analysis and the work done in the analysis.

3.2 NARDL

The growing popularity of nonlinear modelling in the field of long-term all-encompassing relationships would lead to the proliferation of regime change models. Among existing studies, nonlinearity is typically limited to error-correction mechanism, and estimates using the Error correction model (ECM) mechanism of the Markov-switching threshold or using the error correction mechanism of gentle regressions [19]. However, the common assumption that the co-integration relationship may be represented as a linear combination of non-static variables is likely to be very limited. So, several solutions to these problems were proposed in the field of static regression model (including dynamic regression model) [19].

In this framework, a dynamic and flexible parameter was considered, with which utilizes in models that express short-term and long-term hybrid asymmetry. Therefore, the following nonlinear ARDL model was considered.

(17) y t = ∑ j = 1 p φ j y t − j + ∑ j = 0 q ( θ j + ′ x t − j + + θ j − ′ x t − j − ) + ε t ,

where x _t is a vector k × 1 defined by several regresses, such that φ_j is an autoregressive parameter, θ j + and θ j − are asymmetric distribution interval parameters, and ε t is a process of independent random variables with equal distribution and zero mean constant, σε ₂.

Throughout this study, we focused on the case where x _t is decomposed into x j + and x j − around the zero thresholds, thereby distinguishing positive and negative changes in the growth rate of x _t. The resulting partial aggregation processes in their various applications would provide interesting and economically meaningful interpretations.

Therefore, according to the study of Pesaran and Shin [14],

(18) ∆ y t = ρ y t − 1 + θ j + ′ x t − j + + θ j − ′ x t − j − + ∑ j = 1 p − 1 γ j ∆ y t − j + ∑ j = 0 q ( θ j + ′ x t − j + + θ j − ′ x t − j − ) + ε t ,

(19) ∆ y t = ρ ξ t − 1 + ∑ j = 1 p − 1 γ j ∆ y t − j + ∑ j = 0 q ( θ j + ′ x t − j + + θ j − ′ x t − j − ) + ε t .

3.3 Data and descriptive statistics

In this section, the descriptive statistics of each variable in the study were examined. Unemployment rate of the agricultural sector, the government size, the real wages of the agricultural sector, value add of the agricultural sector, inflation, mechanization coefficient, and the ratio of wages of the agricultural sector to other economic sectors were selected as variables of interest in this study. All this information were received from the Ministry of Agriculture of Iran. The results of these indicators are shown in Table 1.

Table 1

Descriptive indicators of the studied variables

Variable	Minimum	Maximum	Average	Standard deviation	Kurtosis	Skewness
Unemployment rate in the agricultural sector (U)	6.600	14.600	9.985	2.422	1.735	0.309
Size of Government (Gov)	0.096	0.279	0.140	0.039	5.216	1.319
Real wages in the agricultural sector (RW)	1.147	2.132	1.616	284	1.755	0.029
Value add of the agricultural sector (Y)	106.052	513.695	276.846	113.880	2.084	0.322
The inflation rate (P)	6.900	49.400	20.340	9.775	3.538	0.963
Mechanization coefficient (M)	0.350	2.200	0.827	0.572	2.733	1.096
Wage ratio of agricultural sector to other economic sectors (W)	0.180	0.505	0.290	0.075	3.393	0.580

4 Empirical findings and discussion

4.1 Stationary test

Since non-stationary in variables can lead to spurious results, it is necessary to verify the stationary of all research variables before model estimation. In order to evaluate the stationary of a time series process, the unit root test was applied. It is also important to note that, in the current study, generalized Dickey–Fuller unit (ADF) and Phillips and Peron (PP) tests were used to evaluate the stationary of the variables. The results of this test are listed in Table 2. According to the details of the table, the variables of the government size, inflation rate, and the ratio of wages of the agricultural sector to other economic sectors were stationary at the current level. In other words, they are cointegrated at I(0). On the other hand, variables including the mechanization coefficient, the real wage of the agricultural sector, the unemployment rate of the agricultural sector, and the value add of the agricultural sector were non-stationary at levels and had remained stationary after first difference. In other terms, they required one differencing to achieve stationary. Thus, according to the results of the unit root tests, there were no limitations to the application of the NARDL approach.

Table 2

Results of unit root test for each variable

Variable	Test	Statistic	Significance level	First order difference statistics	Significance level	Condition
LnGov	ADF	−3.185	0.028	^—	—	I (0)
LnGov	PP	−3.342	0.019	—	—	I (0)
LnM	ADF	−1.469	0.822	4.881	0.001	I (1)
LnM	PP	−1.363	−0.856	4.181	0.011	I (1)
LnP	ADF	−4.671	0.000	—	—	I (0)
LnP	PP	−2.778	0.070	—	—	I (0)
LnRW	ADF	−2.002	0.284	0.000	−6.548	I (1)
LnRW	PP	−2.029	0.273	0.000	−6.619	I (1)
LnU	ADF	−3.006	0.143	0.000	−5.324	I (1)
LnU	PP	−3.064	0.128	0.000	−6.543	I (1)
LnW	ADF	−3.482	0.056	—	—	I (0)
LnW	PP	−4.971	0.001	—	—	I (0)
LnY	ADF	−3.027	0.137	0.000	−6.775	I (1)
LnY	PP	−2.848	0.189	0.000	9.428	I (1)

4.2 NARDL model results

4.2.1 Short-term relationship estimation

Estimation of short-term model according to NARDL model is shown in Table 3. Based on these results:

The first interruption of the unemployment rate in the agricultural sector: The estimated coefficient was equal to 0.438, which was significant at the level of 99%. This means that the unemployment rate of the previous period affected the unemployment rate of this year.
Positive government size shock: The estimated coefficient was around −0.396, which was not significant. Meaning that with a 1% increase in government size, the unemployment rate in the agricultural sector would decrease by 0.396%.
Negative government size shock: The estimated coefficient was about −0.932, which was significant at 95% level. That is to say, the occurrence of a negative shock in the size of the government, the unemployment rate in the agricultural sector would decrease by 0.932%.
Real wage of agricultural sector: The estimated coefficient was equal to −0.677, which was significant at the level of 99%. Meaning that with a 1% increase in real wages, the unemployment rate in the agricultural sector would decrease by 0.677%.
Value add of agricultural sector: The estimated coefficient around −0.138 which was not significant.
Inflation rate: The estimated coefficient was about −0.231, which was significant at the level of 99%. This means that with a 1% increase in inflation, the unemployment rate in the agricultural sector would decrease by 0.231%.
Mechanization coefficient: The estimated coefficient was very little and considered ignorable (0.029), which was not significant.
The first difference of mechanization coefficient: the estimated coefficient was equal to 0.547, which was not significant.
Wage ratio of agricultural sector to other economic sectors: the estimated coefficient was around 0.372, which was significant at the level of 95%. This means that with a 1% growth in the wage ratio, the unemployment rate in the agricultural sector would grow by 0.372%.

Table 3

Estimation of short-term model based on NARDL model

Variable	Coefficient	Standard deviation	t-Statistic	Significance level
lnU (−1)	0.438	0.119	3.657	0.001
lnGov_Pos	−0.396	0.794	−0.498	0.621
lnGov_Neg	−0.932	0.433	−2.152	0.040
lnRW	−0.677	0.237	−2.851	0.008
lnY	−0.138	0.344	−0.402	0.690
lnP	−0.231	0.050	−4.580	0.000
lnM	−0.029	0.435	0.967	0.946
lnM (−1)	0.547	0.381	1.432	0.163
lnW	0.372	0.135	2.751	0.010
y-intercept	9.741	3.669	2.655	0.012
Time trend	−0.060	0.035	−1.700	0.100
Statistic F 18		Durbin–Watson 1.983	Coefficient of determination 0.865	Adjusted coefficient of determination 0.817

Apart from the above descriptions, according to the calculations, the coefficient of determination was equal to 0.865, which indicates that the independent variables were able to explain 86.5% of changes in the unemployment rate of the agricultural sector. Additionally, the adjusted coefficient of determination was equal to 0.817. Another statistic was the Durbin–Watson statistic which was about 1.983, due to its slight difference with the number 2 indicator (number 2 indicates the absence of autocorrelation), it can be said that the current model did not face any problems of autocorrelation. Eventually, the F statistic and its significance level were also expressed. Considering the value of this statistic (∼18), which was significant at the level of 99%, it can be said that the estimated regression model is statistically significant.

4.3 Study of the long-term relationship

According to the results of Table 4, the value of the test statistic is approximately 6.070, which is greater than all critical values at the level of 1, 5, and 10%. As a result, the null hypothesis was rejected and we can say that there is a long-term relationship.

Table 4

Bands test for linear model

I (1)	I (0)	Significance level (%)	Value	Test statistic
4.920	2.668	10	6.070	Test F
4.564	3.121	5
5.965	4.31	1

4.4 ECM estimation

Based on the results presented in Table 5, the ECM value was estimated to be around −0.561. This coefficient shows the speed of error correction and the movement from short-term equilibrium to long-term equilibrium.

Table 5

ECM model estimation

Variable	Coefficient	Standard deviation	t-Statistic	Significance level
y-Intercept	9.741	1.249	7.795	0.000
Time trend	−0.060	0.007	−7.727	0.000
D(lnM)	0.029	0.302	0.097	0.923
ECM	−0.561	0.072	−7.791	0.000
Statistic F 20.425	Durbin–Watson 1.983	Coefficient of determination 0.636		Adjusted coefficient of determination 0.605

4.5 Long-term model estimation

The model was also used for a long-term estimation and its results are presented in Table 6, as well as following conclusions are reached based on the results from the table:

A positive shock of government size has a negative and non-significant effect on the unemployment rate in the agricultural sector. The estimated coefficient was about −0.706 which was not significant.
The negative shock of government size had a significant negative effect on the unemployment rate in the agricultural sector. The estimated coefficient was equal to −1.661, which was significant at the level of 90%. This means that with the occurrence of a negative shock in government size, the unemployment rate in the agricultural sector would decrease by 1.66%.
Real wages in the agricultural sector had a significant negative effect on its unemployment rate in the agricultural sector. The estimated coefficient was equal to −1.207, which was significant at the 99% confidence level. This means that with a 1% increase in real wages in the agricultural sector, the unemployment rate in the agricultural sector would decrease by 1.207%.
The value add of the agricultural sector had a negative and non-significant effect on the unemployment rate of the agricultural sector. The estimated coefficient was around −0.246, which was not significant.
Inflation had a significant negative effect on the unemployment rate in the agricultural sector. The estimated coefficient was equal to −0.411, which was significant at the level of 99%. Meaning that with a 1% increase in inflation, the unemployment rate in the agricultural sector would decrease by 0.411%.
In the long-term, the mechanization coefficient had a positive and significant effect on the unemployment rate in the agricultural sector. The estimated coefficient was equal to 1.027, which was significant at the 95% level. This means that with a 1% increase in the mechanization coefficient, the unemployment rate in the agricultural sector would increase by 1.027%.
The ratio of wages in the agricultural sector to other economic sectors had a positive and significant effect on the unemployment rate in the agricultural sector. The estimated coefficient was equal to 0.663, which was significant at the 95% level. This means that with a 1% increase in the wage ratio, the unemployment rate in the agricultural sector would increase by 0.663%.

Table 6

Results of long-term model estimation

Variable	Coefficient	Standard deviation	t-Statistic	Significance level
lnGov_Pos	−0.706	1.403	−0.503	0.618
lnGov_Neg	−1.661	0.835	−1.988	0.056
lnRW	−1.207	0.385	−3.134	0.004
lnY	−0.264	0.624	−0.395	0.695
lnP	−0.411	0.101	−4.038	0.000
lnM	1.027	0.429	2.389	0.023
lnW	0.663	0.281	2.357	0.025

4.6 Structural stability test

Based on Figures 1 and 2, the null hypothesis states that no structural break is accepted.

Figure 1

CUSUM structural stability test.

Figure 2

CUSUMQ structural stability test.

4.7 Study of symmetry of positive and negative shocks

Finally, the Wald test was used to investigate the long-term asymmetry of positive and negative shocks of government size variable. According to the obtained results presented in Table 7, the null hypothesis of the test was not rejected, which indicates that the positive and negative shocks of government size were not asymmetric in the long-term.

Table 7

NARDL model Wald test

Variable	Test statistic	Value	Degrees of freedom	Significance level
lnGov	Statistic t	0.509	28	0.614
	Statistic F	0.259	(28.1)	0.614
	Chi-square statistic	0.259	1	0.610

5 Conclusion and policy recommendations

This study focuses on the validity of Abrams’ theory for the agricultural sector of Iran. According to the NARDL results, the following conclusions were drawn; (1) The Abrams curve cannot be confirmed for the agricultural sector in Iran, as the coefficient obtained for the size of the state was negative, which is a contradiction. (2) The findings suggest that the government’s size affects the unemployment rate in the agricultural sector in Iran. (3) There was evidence that real wages in Iran’s agricultural sector have an impact on unemployment. (4) It seems that the inflation rate affects the unemployment rate in the agricultural sector in Iran. (5) The added value of the agricultural sector does not appear to influence its unemployment rate. (6) There is a correlation between the mechanization coefficient and the unemployment rate in Iran’s agricultural sector. (7) Importantly, the ratio of wages in the agricultural sector to the wages of other economic sectors affects the unemployment rate in the agricultural sector in Iran. This study’s results can be compared with Feldmann’s study which confirmed that a rise in government expenditures increases unemployment [20]. According to his study, a large share of government expenditures and a large share of transfers and subsidies affect GDP negatively. Similarly, Afonso et al. found that the size of the government has a positive relationship with inflation [6], which is in line with our findings. Soliman et al. investigated the effects of inflation in the UK energy, agriculture, and consumer sectors on agricultural output, using monthly data between February 2015 and October 2022. Existing studies on agricultural inflation explore the impact on economic activity vis-a-vis unemployment, consumption, interest rates, and agricultural production. Their approach adopted a NARDL, Structural vector auto regression model, including impulse response analysis. Their results showed that the increase in energy inflation, agflation, and CPI adversely affects agricultural output, while decrease in energy inflation, agflation, and CPI positively affects agricultural output in the UK [21]. These results are consistent with the findings of the present study.

Based on the results of this research, the following suggestions are made:

(1) Strengthening the domestic economy and adhering to the resistance economy and strengthening them in order to reduce dependence. (2) The capital market and banking financial support for productive agricultural activities. (3) Investing and developing in the agricultural sector and its resources and equipment. (4) Planning to increase investment in the agricultural sector with government policies. (5) Ease of access to the foreign market and establishing conditions for the export of agricultural products. (6) Meeting the needs of the agricultural sector inside the country and reducing dependence on imports. (7) For economic stability, the government should act through the creation of an investment fund in the direction of projects with high returns in the field of agriculture. (8) The main factor of increasing inflation is the growth of liquidity. The government should adopt a policy to be able to correct the price of agricultural products in a gradual and step-by-step manner. This will help to reduce liquidity and basic control of money. (9) Due to the existing sanctions, the government should leverage this opportunity to increase non-oil exports in the future and increase non-oil exports, especially in the agricultural sector, which has a high foreign exchange rate. And thereby prevent dependence on one of the basic needs of the country, which is food.

Funding information: Authors state no funding involved.
Author contributions: All authors have accepted responsibility for the entire content of this manuscript and consented to its submission to the journal, reviewed all the results and approved the final version of the manuscript. SYZ and MR designed the experiments and carried them out. MR developed the model code and performed the simulations. SYZ and MR prepared the manuscript with contributions from all co-authors. SYZ and RM have participated in guiding and supervising the writing process, and their ideas and points of view have been used.
Conflict of interest: Authors state no conflict of interest.
Data availability statement: The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. All data generated or analysed during this study are included in this published article.

References

[1] Lotfalipour M, Azarinfar Y, Mohammadzadeh R. The impact of government expenditure on the overall economic growth and the growth of the agricultural sector in Iran economic growth in Iran. J Agric Econ Dev. 2012;26(2):86–96.Search in Google Scholar

[2] Ansari Samani H, Khilkordi R. The Effect of Government Size on Unemployment: Study of Abrams Curve in Developed and Developing Countries. J Econometric Model. 2017;2(1):113–34.Search in Google Scholar

[3] Abrams BA. The effect of government size on the unemployment rate. Public Choice. 1999;99(3/4):395–401. 10.1023/A:1018349411246.Search in Google Scholar

[4] Christopoulos D, Loizides J, Tsionas E. The Abrams curve of government size and unemployment: evidence from panel data. Appl Econ. 2005;37(10):1193–9.10.1080/00036840500109274Search in Google Scholar

[5] Sun’an M. The effect of government expenditure and private investment on work opportunities and unemployment rate in Indonesia. Russ J Agric Socio-econ Sci. 2018;76(4):92–100. 10.18551/rjoas.2018-04.10.Search in Google Scholar

[6] Afonso A, En H, Kaya A. Government size, unemployment, and inflation nexus in eight large emerging market economies. SSRN Electron J. 2018;1–39. 10.2139/ssrn.3174414.Search in Google Scholar

[7] Mufeed A, Marwan D, Mustafa H. Does government size explain unemployment in MENA countries? Appl Econ Q. 2020;66(3):223–38.10.3790/aeq.66.3.223Search in Google Scholar

[8] Montazeri Shoorekchali J, Zehed Gharavi M. The Effect of government size on unemployment in Iran: new evidence from the Markov-switching model. Iran Econ Issues. 2021;8(11):333–59.Search in Google Scholar

[9] Fiaz A, Malik SW, Khurshid N, Ul A. Empirical evidence from NARDL analysis: an asymmetric effect of exchange rate volatility on agricultural sector in Pakistan. Asia-Pac Soc Sci Rev. 2021;21(2):171–83.10.59588/2350-8329.1375Search in Google Scholar

[10] Sama JMS, Sapnken FE, Mfetoum IM, Tamba JG. Unravelling the dynamics of human development and economic growth on crude oil production based on ARDL and NARDL models. MethodsX. 2023;11:102404. 10.1016/j.mex.2023.102404.Search in Google Scholar PubMed PubMed Central

[11] Dickey DA, Fuller WA. Distribution of the estimators for autoregressive time series with a unit root. J Am Stat Assoc. 1979;74(366):427. 10.2307/2286348.Search in Google Scholar

[12] Enders W. Applied Econometric Time Series, 4th edn, UK: Wiley; 2014.Search in Google Scholar

[13] Phillips PCB, Perron P. Testing for a unit root in time series regression. Biometrika. 1988;75(2):335–46. 10.1093/biomet/75.2.335.Search in Google Scholar

[14] Pesaran HH, Shin Y. Generalized impulse response analysis in linear multivariate models. Econ Lett. 1998;58(1):17–29. 10.1016/S0165-1765 (97)00214-0.Search in Google Scholar

[15] Pesaran MH, Shin Y, Smith RJ. Bounds testing approaches to the analysis of level relationships. J Appl Econom. 2001;16(3):289–326. 10.1002/jae.616.Search in Google Scholar

[16] Noferesti M, Jalouli M. Analyzing the impact of the removal of basic commodity subsidies on the main macroeconomic variables within a structural macro econometric framework. J Econ Modeling. 2010;1(1):81–105.Search in Google Scholar

[17] Bauldry S. Structural equation modeling. International Encyclopedia of the Social & Behavioral Sciences, Elsevier. 2015; p. 615–20. 10.1016/B978-0-08-097086-8.44055-9.Search in Google Scholar

[18] Banerjee A, Dolado J, Mestre R. Error‐correction mechanism tests for cointegration in a single‐equation framework. J Time Ser Anal. 1998;19(3):267–83. 10.1111/1467-9892.00091.Search in Google Scholar

[19] Saridakis G. Violent crime in the United States of America: a time-series analysis between 1960–2000. Eur J Law Econ. 2004;18(2):203–21. 10.1023/B:EJLE.0000045082.09601.b2.Search in Google Scholar

[20] Feldmann H. Government size and unemployment: evidence from developing countries. J Dev Areas. 2009;43(1):315–30. 10.1353/jda.0.0047.Search in Google Scholar

[21] Soliman AM, Lau CK, Cai Y, Sarker PK, Dastgir S. Asymmetric effects of energy inflation, agri-inflation and CPI on agricultural output: evidence from NARDL and SVAR models for the UK. Energy Econ. 2023;126:106920. 10.1016/j.eneco.2023.106920.Search in Google Scholar

Received: 2023-11-20

Revised: 2024-02-23

Accepted: 2024-03-11

Published Online: 2024-11-25

This work is licensed under the Creative Commons Attribution 4.0 International License.

Articles in the same Issue

https://doi.org/10.1515/opag-2022-0278

Keywords for this article

Abram’s theory; NARDL approach; ECM; Iran

Creative Commons

BY 4.0