An Investigation of the Impact of Size of the Government on Economic Growth: Some New Evidence from OECD-NEA Countries

[1]. They conclude the optimal average government involvement amounts to 41% of GDP in Armey curve standard and using nonparametric Data Envelopment Analysis (DEA) framework.

[2]. The OECD-Nuclear Energy Agency (NEA) members countries in the sample are the following: Australia, Austria, Belgium, Canada, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Japan, Korea, Luxembourg, Mexico, Netherlands, Norway, Poland, Portugal, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, Turkey, United Kingdom, United States.

Thus, it can affect government’s financial position in the long term in these countries, which may lead to considerable share of government spending veer to government investment expenditures to respond these policies’ costs. This is natural that the government consumption expenditures are impressed in its result and also these conversions affect economic growth process of them in the long-run. 

So, this paper provides new evidence about the impacts and consequences of size of the government on economic growth, applying production function developed by Dar & Amir Khalkhali (2002) and non-linear testing for the relationship of variables under investigation. We also examine the validity of positive impacts of government consumption spending - as the result of some linear studies for these set of economies - in Panel Smooth Threshold Regression (PSTR) model which permits a smooth transition as a weak number of thresholds, as for a continuum of regimes.

The structure of the paper is organized as follows: the next section contains the model specification and data of this research. In Section 3, the empirical results are presented and finally, Section 4 concludes the paper.

2. Econometric Methodology and Data Description

Our sample is consisted of data for 30 countries from OECD-NEA members over the 1990–2011. Based on the various studies which precisely debate non-linearity between size of the government and growth, the size of the government as the source of the nonlinear size of the government-growth nexus is known. On the other hand, Armey (1995) implements the Laffer curve to present the nonlinear relationship between size of the government and economic growth that are empirically founded by Sheehey (1993), Vedder & Gallaway (1998), and Chen & Lee (2005), then introduces inverted U-shaped Armey curve.

Moreover, according to the literature review, there are many different studies with different results about the role of government in economic growth. But these studies do not present a unique sign for displaying the impact of size of the government on economic growth. Thus, there is probability of existence of non-linear relationship between these variables that leads us to survey this issue. Analysis of Armey (1995) about this issue is that low government spending can increase economic growth until it reaches a certain level that is called threshold size of the government. But high level of government spending reduces economic growth. Indeed, he believes in the positive impact of government in supplying public goods and infrastructure causes improvement in economic growth, but maintains other government activities in economy, such as additional projects financed by the government that become increasingly less productive, is undesirable to economic growth because of negative impact of excess infrastructure on marginal benefits.

Indeed, the big size of the government contributes in output’s reduction by diminishing the constructive features of government’s intervention through adverse effects of further expansion of government (Herath, 2012).

Note that not only the various studies about analyzing the non-linearity between government expenditure and growth are not precisely stratified in various and or equal income levels economies, but also that they represent different results about inverted U-shape of Armey curve that is much known in this theme. Whereas, the existence of other different linear or non-linear relations between government expenditure and growth that can present best-fit from the relationships of variables is very probable. Regarding the use of linear approaches to delineate the relationship between variables, most of the studies have applied the same models and econometric methodologies (see, e.g., Folster & Henrekson, 2001; Heitger, 2001; Dar & Amir Khalkhali, 2002; Yasin, 2003; Kuştepeli, 2005; Alfonso & Furceri, 2008; Lopez, 2008; Gregoriou & Ghosh, 2009; Wahab, 2011) which have created various conclusions on the issue even for similar economies.

Also, as in some works, the different degrees for size of the government indicator is considered (Anaman, 2004; Kuştepeli, 2005), others have surveyed only the quadratic equation form for their model under investigation in order to answer the inverted U-shape of Armey curve (see, for instance, Pevcin, 2004). But, the downright rest to Armey curve or surveying the different degrees for size of the government indicator will be very cramp, particularly by referring to this considerable mention that some researchers and economists are not unanimous about the existence of U-shape curve for express of this nexus.

On the one hand, most of the studies ignore the high probable existence of heterogeneity in gross domestic production data of various countries and select the individual effects and there are on fixed or random effects approaches- notwithstanding the individual effects and time effects- which causes spurious regressions estimations. However, assuming that coefficients of the observed explanatory variables are identical for all observations will not be rational for decrypting the response of economic growth to variation of government expenditure in various economies, particularly in various income levels and economical structures countries.

Therefore, analyzing the existence of nonlinear nexus between effective variables in output function and thereupon economic growth using the model that includes most of these variables resolves extant main problems of other prior studies and presents reliable results, too. In this study, the introduced integral model by Dar & Amir Khalkhali (2002) applying PSTR models approach has perused. Total factor productivity growth is considered as one of the main variables beside other effective variables such as capital and labor force in this model that can be prevented from biased models specification because of ignoring of effective variables in government consumption-growth nexus. The algebraic form of the model is as follows:

                                                  (1)

                                               (2)

           (3)

where GY is percentage of annual growth rate of GDP (2000 prices base), GS is size of the government (real government consumption spending to real GDP ratio percent), GK refers to the annual growth rate of fixed capital formation to GDP ratio as a proxy of investment growth rate, and GL and GX are the annual growth rate of employment labor force to adult population (+15 years) ratio percent and annual growth rate of real exports, respectively. Moreover, A_it measures the rate of total factor productivity growth. Note that the residual  is assumed to be i.i.d. (0, ) and  refers to the individual fixed sections effect. The subscripts and  index the countries and time periods in the sample, respectively.

The Panel Smooth Threshold Regression (PSTR) model with extreme regimes can be viewed as prominent regime-switching model that is the generalization of the threshold panel model of Hansen (1999). Not only the model permits a smooth transition, as a weak number of thresholds for a continuum of regimes, and does not impose a restrictive function form to relation of variables, but also it can proximate modeling the probable nonlinear relationship between variables, by transition function and base of threshold variables observations. However, heterogeneous time and sectional effects are specified in other simple panel models, the changing the estimated coefficients across individuals and over time is possible in frame of PSTR model which heterogeneity problem of estimated parameters is resolved in this trough.

The PSTR model was proposed by Gonzàlez et al. (2005) and Fok et al. (2004) and then developed by Colletaz & Hurlin (2006). In order to investigate the relationship between variables under investigation the simplest case of this model which suppose the existence of two extreme regimes and a single transition function is as follows:

              (4)

The transition function is a continuous function which depends on the value of threshold variableand is normalized to be bounded between 0 and 1. Gonzàlez et al. (2005), adopting Smooth Transition Autoregressive Regression (STAR) models introduced by Granger & Terasvirta (1993) and Jansen & Terasvirta (1996), specified the logistic form of transition function as follow.

                                                               (5)

where  is as the vector of threshold parameters or locations of occurrence of regime-switching. The parameter  determines the slope of the transition function.

According to theoretical studies, the government consumption spending is different from government investment spending, structurally. Although consumption expenditure besides investment expenditure occurs at about the same time, government consumption expenditure- unlike investment expenditure- can have ineffective impacts on output growth in some times. Since composition and the impact type of them will be different in varying periods, considering the total government expenditure to GDP ratio as a proxy of size of the government to delineate the impact of it on output growth causes the unreliable results to deciding about government financial policies. Thus, we have to use the government consumption spending to delineate its impacts on growth. 

Gonzalez et al. (2005) believe that considering one or two threshold values (m=1 or m=2) will be enough in order to specify the variability of parameters. They stress that for m=1 the model PSTR implies that the two extreme regimes are associated with low and high values of q_it with a single monotonic transition of the coefficients from,,  , to  ,   ,  and  as q_it increases, where the change is centered around c₁. If,   the PSTR model in (4) reduces to the two-regime panel threshold regression (PTR) model of Hansen (1999). Indeed, when the transition function attains the value 1 and 0 otherwise.

For m=2, the minimum of transition function is at (c₁ + c₂)/2 and attains the value 1 both at low and high values of transition variable (q_it). If, the count of regimes raise to a three-regime whose outer regimes are identical and different from the middle regime. But, whenfor any value of m, the transition function becomes constant and the model collapses into a homogenous or linear panel regression model with fixed effects (Gonzalez et al., 2005).

Gonzalez et al. (2005) and Colletaz & Hurlin (2006) have introduced a testing process to investigate the existence or inexistence of non-linear relationship between variables under investigation that presents a context to creating reliable final estimation of PSTR by using Non-Linear Least Squares (NLS) approach that is equivalent of Maximum Likelihood estimator.

Since the surveying of linearity in PSTR under  will have unidentified nuisance parameters, the associated tests will be nonstandard. Therefore, to circumvent the identification problem, it is necessary that  is replaced in (4) by its first-order Taylor expansion around which can be viewed as the testing of equivalent hypothesis in auxiliary regression (Gonzalez et al., 2005). Thus, Taylor expansion for the PSTR model with n threshold is as follow:

                                                                                                   (6)

Due to the  parameters is proportionate with  parameter of transition function, the testing of linearity under  and  is possible. The Wald Lagrange Multiplier, Fischer Lagrange Multiplier and likelihood ratio coefficients are as the criteria in process of testing. The testing of remaining non-linearity on PSTR model to determination of the count of necessary transition functions for specifying of PSTR model is the next stage after support of existence the non-linearity nexus between the variables. The null hypothesis of this test is the existence of one transition functions, while the alternative hypothesis is the existence of at least two transition functions for PSTR model. The second transition function in Taylor expansion access specifies is as follow: