Nonlinear Effects of Changes in Government Expenditures Across Global Economies: Insights from a QNARDL Analysis

Abstract

This study used the QNARDL model to examine how government expenditure affected economic growth in 92 countries during 1992 to 2021. Unlike traditional models, this study decomposes government expenditure into positive and negative changes, estimating their impacts across different growth quantiles. The QNARDL model provides more detailed information than linear models such as the threshold autoregressive model, panel threshold method, quadratic form, and data envelopment analysis. The findings reveal that both positive and negative government expenditure changes hamper economic growth at most quantiles, with only a few positive effects observed at certain quantiles. This study’s novel contribution lies in its methodological approach as it uses the QNARDL model to capture the nuanced and complex relationship between government expenditure and economic growth, providing crucial information for policymakers on the efficacy of fiscal policy across different economic conditions.

JEL Classifications: C01, H50, H54, O40.

Introduction

The role of fiscal policy in driving economic growth has long been a subject of intense debate, particularly as the global economic landscape becomes increasingly interconnected (Sosvilla-Rivero & Rubio-Guerrero, 2022). The potential for government expenditure to either stimulate or stifle growth depends heavily on the economic context and how fiscal interventions are implemented. Fiscal policy is traditionally seen as a double-edged sword: while increased government spending can boost demand and foster growth, it can also burden the economy through increased borrowing costs and taxation (Hajamini & Falahi, 2018). This duality highlights the complexity of assessing the long-term effects of fiscal interventions on economic performance. Thus, considering the impact of fiscal policy is crucial to ensure a country’s future financial sustainability.

Although many studies have examined the impact of fiscal policy on economic activity, the role of government spending remains controversial. Traditional perspectives on the role of fiscal policy in economic dynamics are deeply rooted in contrasting economic theories. Keynesian economics, for instance, champions the idea that government spending can mitigate business cycles, particularly during recessions (Rotheim, 1981); it posits that in times of economic downturn, increased government expenditure can stimulate demand, create jobs, and lead to economic recovery. This perspective underscores the proactive relationship between fiscal policy and economic stabilization and fostering growth. However, classical and neoclassical theories argue that excessive government intervention can have detrimental effects. The crowding out effect suggests that increased public sector borrowing raises interest rates, reducing private sector investment (Alfada, 2019; Asimakopoulos & Karavias, 2016; Barro, 1991; Bin Mohd Aznan et al., 2022; Chiou-Wei et al., 2010; Christie, 2014; Meyer et al., 2017; Hajamini & Falahi, 2018; Vveinhardt & Sedziuviene, 2022). Furthermore, excessive spending can lead to an increased debt burden, which may have long-term fiscal sustainability risks, especially if debt grows faster than the economy (Hajamini & Falahi, 2018; Khan & Reinhart, 1990; Makuyana & Odhiambo, 2018; Mankiw et al., 1992; Suhendra & Anwar, 2014). Resource misallocation and inefficiency can also occur, whereby politically motivated government projects divert resources from more productive uses, reducing overall economic efficiency (Goh and Mohd Aznan, 2023; Sosvilla-Rivero & Rubio-Guerrero, 2022).

Moreover, excessive fiscal spending often leads to higher taxes, which can have distortionary effects on labor supply and investment, further stifling economic growth (Afonso & Sousa, 2012; Alexiou, 2009; Blanchard & Perotti, 2002; Bose et al., 2007; Ghose & Das, 2013; Jawadi et al., 2016; Pata et al., 2023; Romero-Avila & Strauch, 2008). Inflationary pressures are another risk associated with excessive government spending, particularly when this spending is financed by borrowing or monetary expansion. When the government expenditure exceeds the economy’s productive capacity, inflation can erode purchasing power and disrupt economic stability (Afonso & Sousa, 2012; Alexiou, 2009; Blanchard & Perotti, 2002; Bose et al., 2007; Ghose & Das, 2013; Jawadi et al., 2016; Pata et al., 2023; Romero-Avila & Strauch, 2008). Furthermore, sustained government intervention in the economy can create dependency and reduce the private sector’s initiative to innovate and grow independently, as businesses become reliant on public support rather than market competition (Asimakopoulos & Karavias, 2016; Chen & Lee, 2005; Chiou-Wei et al., 2010; De Witte & Moesen, 2010; Hajamini & Falahi, 2018).

Although these concerns highlight the potential drawbacks of government spending, Keynesian economics presents a contrasting view. Keynesians argue that increased government expenditure is essential during economic downturns because it stimulates demand, reduces unemployment, and fosters economic recovery (Rotheim, 1981). This view underscores the countercyclical role of fiscal policy in stabilizing the economy, particularly in times of crisis.

These divergent stances underscore the need for detailed approaches to understanding the relationship between government expenditure and economic growth. A one-size-fits-all fiscal policy is unlikely to succeed across diverse economic conditions (Meyer et al., 2017; Vveinhardt & Sedziuviene, 2022). Thus, the complex, context-dependent nature of fiscal interventions necessitates a tailored approach, one that considers both the short- and long-term effects of government expenditure on growth (Eyraud & Weber, 2013).

Most previous studies have used linear or nonlinear models, such as the threshold autoregressive model, panel threshold method, quadratic form, and data envelopment analysis, to determine the optimal threshold for government spending (Alfada, 2019; Asimakopoulos & Karavias, 2016; Barro, 1991; Bin Mohd Aznan et al., 2022; Chiou-Wei et al., 2010; Christie, 2014; Meyer et al., 2017; Hajamini & Falahi, 2018; Vveinhardt & Sedziuviene, 2022). Nevertheless, these approaches may be unable to capture all the impacts of changes in government expenditure on economic growth.

Thus, to overcome the limitations of previous studies, this paper employs the quantile nonlinear autoregressive distributed lag (QNARDL) model (Cho et al., 2015; Shin et al., 2014) to capture the asymmetric effects of government expenditure on economic growth. As a result, this study makes several major contributions. First, the short- and long-term effects of changes in government expenditure are comprehensively identified at different stages of economic growth. Second, changes in government expenditure are decomposed into positive and negative partial series sums and fully explored across quantiles, highlighting the asymmetric nature of fiscal policy effects, which simpler linear or nonlinear models cannot capture. Unlike conventional models, the QNARDL model allows for assessing how changes in government spending affect the entire growth distribution, uncovering crucial insights that previous approaches could not reveal. In addition, this model offers a comprehensive view of the asymmetric relationship between government expenditure and economic growth, providing valuable information for policymakers seeking to optimize fiscal strategies in a complex global economy. Additionally, to strengthen this study’s findings, we conduct a robustness test using the system generalized method of moments (SGMM) model. To achieve this objective, the study uses a panel dataset of 92 economies, including both OECD and non-OECD countries, covering the period from 1992 to 2021. The findings provide practical knowledge for policymakers, enabling them to design targeted and effective fiscal strategies tailored to specific economic contexts. Thus, the insights generated by this analysis will help policymakers optimize their fiscal interventions to achieve sustainable economic growth.

Literature Review

The debate on the effectiveness of government expenditure remains unsettled, with empirical evidence supporting both Keynesian and classical perspectives under different conditions. Proponents of government intervention argue that strategic public expenditure can stimulate economic growth by investing in infrastructure, education, healthcare, and other areas that may increase productivity and improve the citizens’ quality of life (Afonso & Sousa, 2012; Alexiou, 2009; Blanchard & Perotti, 2002; Bose et al., 2007; Ghose & Das, 2013; Jawadi et al., 2016; Pata et al., 2023; Romero-Avila & Strauch, 2008). On the other hand, critics of government expenditure contend that excessive spending can lead to inefficiencies, resource misallocation, and increased debt, which could harm the economy; they advocate for limited government intervention, emphasizing the role of the private sector in driving innovation and growth through competition and market forces (Afonso & Furceri, 2010; Dar & AmirKhalkhali, 2002; Fatás & Mihov, 2010; Fölster & Henrekson, 2001; Hasnul, 2015). Furthermore, some studies have encountered challenges in establishing a clear relationship between government expenditure and economic growth (da Silva & Vieira, 2017; Onabote et al., 2023) or suggest that government expenditure may have a disproportionate effect on economic growth (Asimakopoulos & Karavias, 2016; Chen & Lee, 2005; Chiou-Wei et al., 2010; De Witte & Moesen, 2010; Hajamini & Falahi, 2018). Table 1 summarizes previous empirical studies.

Table 1.

Summary of Empirical Studies of the Effect of the Government Size on Economic Growth.

Author	Countries	Methodology	Data period	Key summary
Barro (1991)	98 countries	Cross-country linear analysis	1960–1985	There is an inverse relationship between government consumption and economic growth, formalizing the Barro curve in the context of public investment.
Dar and AmirKhalkhali (2002)	19 OECD countries	Linear RCM model	1971–1999	A larger government weakens economic growth by reducing total factor productivity and capital productivity. However, the optimal policy is not necessarily to minimize the government size but to find an efficient balance.
Blanchard and Perotti (2002)	United States	VAR	1947–1997	Positive government spending shocks boost the output, but increases in both spending and taxes hamper investment spending.
Bose et al. (2007)	30 developing countries	Nonlinear analysis	1970–1980	There is a positive correlation between government capital expenditure and economic growth, with educational investment being the most important contributor.
Romero-Avila and Strauch (2008)	15 EU countries	Panel cointegration, distributed lag model	1960–2001	Government consumption and transfers hamper long-term economic growth. Public investment promotes growth. Distortionary taxation negatively impacts private capital accumulation, affecting growth indirectly.
Alexiou (2009)	Bulgaria, Serbia, North Macedonia, Croatia, Bosnia, Albania, Romania	FE, random coefficients	1995–2005	Government spending on capital formation substantially benefits economic growth in transition economies.
Afonso and Furceri (2010)	28 countries	Linear econometric approach	1970–2004	Both a larger government and increased fiscal volatility harm economic growth.
Chiou-Wei et al. (2010)	South Korea, Malaysia, Singapore, Taiwan, Thailand	Smooth transition regression model	1961–2004	Confirms the Barro curve, finding that a government sized above a certain threshold hampers economic growth.
Ghose and Das (2013)	19 countries	Nonlinear dynamic OLS	1970–2006	An increase in government size promotes long-term economic growth, with domestic investment playing a key role.
Christie (2014)	136 countries	Threshold analysis, dynamic panel GMM	1971–2005	The nonlinear hypothesis for government spending and growth is tested, demonstrating threshold effects across different countries.
da Silva and Vieira (2017)	113 countries	Dynamic panel GMM	2001–2012	Monetary policy was countercyclical only for advanced economies before the crisis, whereas fiscal policy was procyclical in the pre-crisis period.
Jawadi et al. (2016)	Brazil, Russia, India, China, South Africa	Nonlinear panel VAR	1990–2013	Strong Keynesian effects from government spending are observed in BRICS economies, suggesting a positive spillover between fiscal and monetary policies
Asimakopoulos and Karavias (2016)	129 countries	Nonlinear panel GMM	1980–2014	The optimal government size that maximizes economic growth is identified, using a novel GMM approach to account for threshold effects.
Meyer et al. (2017)	South Africa	Time-series econometric model	1996–2015	Examines the relationship between economic growth and development, using an economic development index that includes employment and poverty levels.
Hajamini and Falahi (2018)	14 developed European countries	Panel threshold approach	1995–2014	Investigates the nonlinear relationship between government size and economic growth, identifying optimal government expenditure shares.
Alfada (2019)	Indonesia	Threshold model	2004–2015	Assesses the destructive effects of corruption on economic growth in Indonesia, finding a corruption threshold beyond which growth deteriorates.
Bin Mohd Aznan et al. (2022)	Malaysia	Nonlinear threshold approach	1995–2019	Investigates the asymmetric effects of government size on economic growth in Malaysia, finding mixed evidence on optimal government spending levels.
Pata et al. (2023)	Turkey	ARDL	1971–2018	Investigates the effects of life expectancy and hydropower consumption on the ecological footprint, contributing to environmental Kuznets curve discussions.

Methodology

The empirical model used to examine the effect of government expenditure on economic growth was constructed based on a published method (Mehdi Hajamini, 2018). Zaghdoudi et al. (2023) and Kim (2024) argue that macroeconomic shocks, such as the 2008 financial crisis and the COVID-19 pandemic, affect the data structure. Therefore, to better capture the relationship between government expenditure and economic growth, the empirical model in this study incorporates the effects of these two macroeconomic shocks and is expressed as equation (1).

$$\begin{array}{c}\multicolumn{1}{c}{\mathrm{GDP}{G}_{\mathrm{it}}={\mathrm{\alpha }}_0+{\mathrm{\beta }}_1\mathrm{UN}{P}_{\mathrm{it}}+{\mathrm{\beta }}_2\mathrm{IN}{V}_{\mathrm{it}}+{\mathrm{\beta }}_3\mathrm{OPE}{N}_{\mathrm{it}}+{\mathrm{\beta }}_4\mathrm{GC}{E}_{\mathrm{it}}}\\ \multicolumn{1}{c}{+{\mathrm{\beta }}_5{D}_{\mathrm{Crisis}}+{\mathrm{\beta }}_6{D}_{\mathrm{Covid}-19}+{\mathrm{\zeta }}_i+{\mathrm{\epsilon }}_{\mathrm{it}}}\end{array}$$

(1)

Where $\mathrm{GDP}{G}_{\mathrm{it}}$ is the gross domestic product growth of country i. $\mathrm{UN}{P}_{\mathrm{it}}$ is the unemployment rate of country i. The unemployment rate is used to explore how changes in GDP impact employment levels in different countries with varying economic structures and labor market dynamics (Abdul-Khaliq et al., 2014; Dayıoğlu & Aydın, 2020; Levine, 2013; Michael et al., 2016; Nagel, 2015). $\mathrm{IN}{V}_{\mathrm{it}}$ represents the investment variable of country i (a proxy of the ratio of the percentage of gross fixed capital formation to GDP). Investment plays a crucial role in capital accumulation, boosting productivity, complementing private sector efforts, and potentially affecting economic growth through nonlinear threshold effects (Hajamini & Falahi, 2018; Khan & Reinhart, 1990; Makuyana & Odhiambo, 2018; Mankiw et al., 1992; Suhendra & Anwar, 2014). $\mathrm{OPE}{N}_{\mathrm{it}}$ is the trade openness, calculated as the ratio of total exports and imports to the total GDP of country i. Trade openness is an important variable because it contributes to economic growth by promoting access to larger markets, increasing productivity through competition, encouraging technological advancement, and attracting foreign investment (Amna Intisar et al., 2020; Andersen & Babula, 2009; Bahmani-Oskooee & Niroomand, 1999; Eri⋅ & Ula⋅an, 2013; Gundlach, 1997; Kartikasari, 2017; Keho, 2017; Menyah et al., 2014; Raghutla, 2020). $\mathrm{GC}{E}_{\mathrm{it}}$ is the general government final consumption expenditure as a percentage of the GDP of country i. The global financial crisis is represented by the dummy variable $D_{\mathrm{Crisis}}$ , which takes the value of 1 for 2008 to 2010 and 0 for other years. Similarly, the COVID-19 pandemic is represented by the dummy variable $D_{\mathrm{Covid}-19}$ , which takes the value of 1 for 2020 to 2021 and 0 for other years. Country-fixed effects are captured by ${\mathrm{\zeta }}_i$ , and ${\mathrm{\epsilon }}_{\mathrm{it}}$ is the error.

QNADRL Model

This study uses an advanced econometric approach to investigate the asymmetric impact of the $\mathrm{GCE}$ and $\mathrm{GDPG}$ across different quantiles. We used panel data and applied the QNARDL model developed by Cho et al. (2015) and Shin et al. (2014). The QNARDL model has several advantages. First, using the QNARDL model, a partial series sum can be calculated for either positive or negative explanatory variables, capturing both their long- and short-term effects at any desired location in the conditional distribution. Second, the QNARDL model addresses possible endogeneity concerns while performing the calculations.

The basic autoregressive distributed lag (ARDL) model developed by Pesaran et al. (2001) was used to quantitatively express these relationships and is expressed as equation (2):

$$\begin{array}{c}\multicolumn{1}{c}{\mathrm{GDP}{G}_{i,t}=\mathrm{\alpha }+\sum _{j=1}^p{\mathrm{\beta }}_j\mathrm{GDP}{G}_{i,t-j}+\sum _{j=0}^{q_1}{\mathrm{\gamma }}_j\mathrm{UN}{P}_{i,t-j}}\\ \multicolumn{1}{c}{+\sum _{j=0}^{q_2}{\mathrm{\delta }}_j\mathrm{IN}{V}_{i,t-j}+\sum _{j=0}^{q_3}{\mathrm{\eta }}_j\mathrm{OPE}{N}_{i,t-j}+\sum _{j=0}^{q_4}{\mathrm{\theta }}_j\mathrm{GC}{E}_{i,t-j}}\\ +{\mathrm{\lambda }}_1{D}_{\mathrm{Crisis}}+{\mathrm{\lambda }}_2{D}_{\mathrm{Covid}-19}+{\mathrm{\zeta }}_i+{\mathrm{\epsilon }}_{i,t}\hfill \end{array}$$

(2)

This equation allows us to analyze how past values of $\mathrm{GDPG}$ and other variables influence the current state, providing insights into the temporal dynamics of economic growth. We applied the nonlinear autoregressive distributed lag (NARDL) model developed by Shin et al. (2014) to capture the asymmetric effects of the $\mathrm{GCE}$ . The $\mathrm{GCE}$ was divided into the sum of positive ( ${\mathrm{GCE}}_t^{+}$ ) and negative ( ${\mathrm{GCE}}_t^{-})$ partial series and is expressed as follows:

$$\mathrm{GC}{E}_{i,t}=\mathrm{GC}{E}_{i,0}+{\mathrm{GCE}}_{i,t}^{+}+{\mathrm{GCE}}_{i,t}^{-}$$

(3)

Where:

$${\mathrm{GCE}}_{i,t}^{+}=\sum _{k=1}^t\mathrm{\Delta }{\mathrm{GCE}}_{i,k}^{+}=\sum _{k=1}^t\max (\mathrm{\Delta }\mathrm{GC}{E}_{i,k},0)$$

(4)

$${\mathrm{GCE}}_{i,t}^{-}=\sum _{k=1}^t\mathrm{\Delta }{\mathrm{GCE}}_{i,k}^{-}=\sum _{k=1}^t\min (\mathrm{\Delta }\mathrm{GC}{E}_{i,k},0)$$

(5)

Breaking down government expenditures allows us to separate the impact of positive and negative changes on economic growth. Hence, the model was modified to include these asymmetric effects:

$$\begin{array}{c}\multicolumn{1}{c}{\mathrm{GDP}{G}_{i,t}=\mathrm{\alpha }+\sum _{j=1}^p{\mathrm{\beta }}_j\mathrm{GDP}{G}_{i,t-j}+\sum _{j=0}^{q_1}{\mathrm{\gamma }}_j\mathrm{UN}{P}_{i,t-j}}\\ \multicolumn{1}{c}{+\sum _{j=0}^{q_2}{\mathrm{\delta }}_j\mathrm{IN}{V}_{i,t-j}+\sum _{j=0}^{q_3}{\mathrm{\eta }}_j\mathrm{OPE}{N}_{i,t-j}}\\ \multicolumn{1}{c}{+\sum _{j=0}^{q_4}{\mathrm{\theta }}_j{\mathrm{GCE}}_{i,t-j}^{+}+\sum _{j=0}^{q_5}{\mathrm{\sigma }}_j{\mathrm{GCE}}_{i,t-j}^{-}}\\ +{\mathrm{\lambda }}_1{D}_{\mathrm{Crisis}}+{\mathrm{\lambda }}_2{D}_{\mathrm{Covid}-19}+{\mathrm{\zeta }}_i+{\mathrm{\epsilon }}_{i,t}\hfill \end{array}$$

(6)

Where ${\mathrm{\epsilon }}_{i,t}=\mathrm{GDP}{G}_{i,t}-E(\mathrm{GDP}{G}_{i,t}/F_{t-1})$ and $F_{t-1}$ being the smallest $\mathrm{\sigma }$ -field generated by { $\mathrm{GDP}{G}_{i,t}$ , $\mathrm{UN}{P}_{i,t}$ , $\mathrm{IN}{V}_{i,t}$ , $\mathrm{OPE}{N}_{i,t}$ , ${\mathrm{GCE}}_{i,t}^{+}$ , ${\mathrm{GCE}}_{i,t}^{-}$ , $\mathrm{GDP}{G}_{i,t-1}$ , $\mathrm{UN}{P}_{i,t-1}$ , $\mathrm{IN}{V}_{i,t-1}$ , $\mathrm{OPE}{N}_{i,t-1}$ , ${\mathrm{GCE}}_{i,t-1}^{+}$ , ${\mathrm{GCE}}_{i,t-1}^{-}$ }. Furthermore, $p$ , $q_1$ , $q_2$ , $q_3$ , $q_4$ , and $q_5$ represent the lag orders selected based on Schwarz’s information criterion (SIC). Next, we applied the QARDL model developed by Cho et al. (2015). An empirical QNARDL model can be expressed as equation (7):

$$\begin{array}{c}\multicolumn{1}{c}{Q_{\mathrm{GDP}{G}_{i,t}}=\mathrm{\alpha }\left(\mathrm{\tau }\right)+\sum _{j=1}^p{\varnothing }_j\left(\mathrm{\tau }\right)\mathrm{GDP}{G}_{i,t-j}+\sum _{j=0}^{q_1}{\mathrm{\rho }}_j\left(\mathrm{\tau }\right)\mathrm{UN}{P}_{i,t-j}}\\ \multicolumn{1}{c}{+\sum _{j=0}^{q_2}{\mathrm{\omega }}_j\left(\mathrm{\tau }\right)\mathrm{IN}{V}_{i,t-j}+\sum _{j=0}^{q_3}{\mathrm{\theta }}_j\left(\mathrm{\tau }\right)\mathrm{OPE}{N}_{i,t-j}}\\ \multicolumn{1}{c}{+\sum _{j=0}^{q_4}{\mathrm{\psi }}_j\left(\mathrm{\tau }\right){\mathrm{GCE}}_{i,t-j}^{+}+\sum _{j=0}^{q_5}{\mathrm{\xi }}_j\left(\mathrm{\tau }\right){\mathrm{GCE}}_{i,t-j}^{-}}\\ +{\mathrm{\lambda }}_1\left(\mathrm{\tau }\right)D_{\mathrm{Crisis}}+{\mathrm{\lambda }}_2\left(\mathrm{\tau }\right)D_{\mathrm{Covid}-19}+{\mathrm{\zeta }}_i+{\mathrm{\epsilon }}_{i,t}(\mathrm{\tau })\hfill \end{array}$$

(7)

Where ${\mathrm{\epsilon }}_{i,t}=\mathrm{GDP}{G}_{i,t}-Q_{\mathrm{GDP}{G}_{i,t}}(\mathrm{\tau }/F_{t-1})$ ; $Q_{\mathrm{GDP}{G}_{i,t}}(\mathrm{\tau }/F_{t-1})$ is the ${\mathrm{\tau }}^{\mathrm{th}}$ quantile of $\mathrm{GDP}{G}_{i,t}$ , dependent on the information set $F_{t-1}$ (Kim & White, 2003). The lag order $p$ , $q_1$ to $q_5$ was chosen based on SIC to optimize the model. Finally, to capture the differential impacts of each variable across different economic states, the QNARDL error correction model (QNARDL-ECM) was rewritten as follows:

$$\begin{array}{c}\multicolumn{1}{c}{Q_{\mathrm{\Delta }\mathrm{GDP}{G}_{i,t}}=\mathrm{\alpha }\left(\mathrm{\tau }\right)+\mathrm{\eta }\left(\mathrm{\tau }\right)[\mathrm{GDP}{G}_{i,t-1}-{\mathrm{\beta }}_1\left(\mathrm{\tau }\right)\mathrm{UN}{P}_{i,t-1}}\\ \multicolumn{1}{c}{-{\mathrm{\beta }}_2\left(\mathrm{\tau }\right)\mathrm{IN}{V}_{i,t-1}-{\mathrm{\beta }}_3\left(\mathrm{\tau }\right)\mathrm{OPE}{N}_{i,t-1}-{\mathrm{\beta }}_4\left(\mathrm{\tau }\right){\mathrm{GCE}}_{i,t-1}^{+}}\\ \multicolumn{1}{c}{-{\mathrm{\beta }}_5\left(\mathrm{\tau }\right){\mathrm{GCE}}_{i,t-1}^{-}]+\sum _{j=1}^{p-1}{\mathrm{\varphi }}_j\left(\mathrm{\tau }\right)\mathrm{\Delta }\mathrm{GD}{P}_{i,t-j}}\\ \multicolumn{1}{c}{+\sum _{j=0}^{q_1-1}{\mathrm{\rho }}_j\left(\mathrm{\tau }\right)\mathrm{\Delta }\mathrm{UN}{P}_{i,t-i}+\sum _{j=0}^{q_2-1}{\mathrm{\omega }}_j\left(\mathrm{\tau }\right)\mathrm{\Delta }\mathrm{IN}{V}_{i,t-j}}\\ +\sum _{j=0}^{q_3-1}{\mathrm{\theta }}_j\left(\mathrm{\tau }\right)\mathrm{\Delta }\mathrm{OPE}{N}_{i,t-j}+\sum _{j=0}^{q_4-1}{\mathrm{\psi }}_j\left(\mathrm{\tau }\right)\mathrm{\Delta }{\mathrm{GCE}}_{i,t-j}^{+}\hfill \\ \multicolumn{1}{c}{+\sum _{j=0}^{q_5-1}{\mathrm{\xi }}_j\left(\mathrm{\tau }\right)\mathrm{\Delta }{\mathrm{GCE}}_{i,t-j}^{-}+{\mathrm{\lambda }}_1\left(\mathrm{\tau }\right)D_{\mathrm{Crisis}}}\\ +{\mathrm{\lambda }}_2\left(\mathrm{\tau }\right)D_{\mathrm{Covid}-19}+{\mathrm{\zeta }}_i+{\mathrm{\epsilon }}_{i,t}(\mathrm{\tau })\hfill \end{array}$$

(8)

The short-term combined impact of these parameters was calculated by ${\mathrm{\rho }}^{*}(\mathrm{\tau })=\sum _{j=0}^{q_1-1}{\mathrm{\rho }}_j(\mathrm{\tau })$ , ${\mathrm{\omega }}^{*}(\mathrm{\tau })=\sum _{j=0}^{q_2-1}{\mathrm{\omega }}_j(\mathrm{\tau })$ , ${\mathrm{\theta }}^{*}(\mathrm{\tau })=\sum _{j=0}^{q_3-1}{\mathrm{\theta }}_j(\mathrm{\tau })$ , ${\mathrm{\psi }}^{*}(\mathrm{\tau })=\sum _{j=0}^{q_4-1}{\mathrm{\psi }}_j(\mathrm{\tau })$ , and ${\mathrm{\xi }}^{*}(\mathrm{\tau })=\sum _{j=0}^{q_5-1}{\mathrm{\xi }}_j(\mathrm{\tau })$ . Likewise, ${\mathrm{\beta }}_1(\mathrm{\tau })$ , ${\mathrm{\beta }}_2(\mathrm{\tau })$ , ${\mathrm{\beta }}_3(\mathrm{\tau })$ , ${\mathrm{\beta }}_4(\mathrm{\tau })$ , and ${\mathrm{\beta }}_5(\mathrm{\tau })$ are the corresponding long-term parameters and their combined impact is calculated by ${{\mathrm{\beta }}_1}^{*}(\mathrm{\tau })=\frac{{\mathrm{\beta }}_1(\mathrm{\tau })}{\mathrm{\eta }}$ , ${{\mathrm{\beta }}_2}^{*}(\mathrm{\tau })=\frac{{\mathrm{\beta }}_2(\mathrm{\tau })}{\mathrm{\eta }}$ , ${{\mathrm{\beta }}_3}^{*}(\mathrm{\tau })=\frac{{\mathrm{\beta }}_3(\mathrm{\tau })}{\mathrm{\eta }}$ , ${{\mathrm{\beta }}_4}^{*}(\mathrm{\tau })=\frac{{\mathrm{\beta }}_4(\mathrm{\tau })}{\mathrm{\eta }}$ , and ${{\mathrm{\beta }}_5}^{*}(\mathrm{\tau })=\frac{{\mathrm{\beta }}_5(\mathrm{\tau })}{\mathrm{\eta }}$ .

The nonlinear and asymmetric effects of the $\mathrm{GCE}$ on $\mathrm{GDPG}$ in the long and short run were examined using the Wald test. Regarding the short-term parameters, the following are the null and alternative hypotheses:

$H_0:S{\mathrm{\rho }}_j^{*}(\mathrm{\tau })=s$ versus $H_1:S{\mathrm{\rho }}_j^{*}(\mathrm{\tau })\ne s$

$H_0:S{\mathrm{\omega }}_j^{*}(\mathrm{\tau })=s$ versus $H_1:S{\mathrm{\omega }}_j^{*}(\mathrm{\tau })\ne s$

$H_0:S{\mathrm{\theta }}_j^{*}\left(\mathrm{\tau }\right)=s$ versus $H_1:S{\mathrm{\theta }}_j^{*}\left(\mathrm{\tau }\right)\ne s$

$H_0:S{\mathrm{\psi }}_j^{*}(\mathrm{\tau })=s$ versus $H_1:S{\mathrm{\psi }}_j^{*}(\mathrm{\tau })\ne s$

$H_0:S{\mathrm{\xi }}_j^{*}(\mathrm{\tau })=s$ versus $H_1:S{\mathrm{\xi }}_j^{*}(\mathrm{\tau })\ne s$

Regarding the long-term parameters, the null and alternative hypotheses are as follows:

$H_0:F{\mathrm{\eta }}^{*}(\mathrm{\tau })=f$ versus $H_1:F{\mathrm{\eta }}^{*}(\mathrm{\tau })\ne f$

$H_0:S{{\mathrm{\beta }}_1}^{*}(\mathrm{\tau })=S(\mathrm{\gamma }(\mathrm{\tau })/\mathrm{\beta })=s$ versus $H_1:S{{\mathrm{\beta }}_1}^{*}(\mathrm{\tau })=S(\mathrm{\gamma }(\mathrm{\tau })/\mathrm{\beta })\ne s$

$H_0:S{{\mathrm{\beta }}_2}^{*}(\mathrm{\tau })=S(\mathrm{\gamma }(\mathrm{\tau })/\mathrm{\beta })=s$ versus $H_1:S{{\mathrm{\beta }}_2}^{*}(\mathrm{\tau })=S(\mathrm{\gamma }(\mathrm{\tau })/\mathrm{\beta })\ne s$

$H_0:S{{\mathrm{\beta }}_3}^{*}(\mathrm{\tau })=S(\mathrm{\gamma }(\mathrm{\tau })/\mathrm{\beta })=s$ versus $H_1:S{{\mathrm{\beta }}_3}^{*}(\mathrm{\tau })=S(\mathrm{\gamma }(\mathrm{\tau })/\mathrm{\beta })\ne s$

$H_0:S{{\mathrm{\beta }}_4}^{*}(\mathrm{\tau })=S(\mathrm{\gamma }(\mathrm{\tau })/\mathrm{\beta })=s$ versus $H_1:S{{\mathrm{\beta }}_4}^{*}(\mathrm{\tau })=S(\mathrm{\gamma }(\mathrm{\tau })/\mathrm{\beta })\ne s$

$H_0:S{{\mathrm{\beta }}_5}^{*}(\mathrm{\tau })=S(\mathrm{\gamma }(\mathrm{\tau })/\mathrm{\beta })=s$ versus $H_1:S{{\mathrm{\beta }}_5}^{*}(\mathrm{\tau })=S(\mathrm{\gamma }(\mathrm{\tau })/\mathrm{\beta })\ne s$

As suggested by Cho et al. (2015), $F$ and $f$ represent the matrices $h\times \mathrm{ps}$ and $h\times 1$ , respectively. Similarly, S and s represent the matrices $h\times s$ and $h\times 1$ , respectively. These matrices are predefined with h number of restrictions. All short- and long-term parameters are tested for nonlinear characteristics. As an example, consider the parameter ${\mathrm{\beta }}^{*}(\mathrm{\tau })$ , for which the Wald test examines the hypotheses:

$$\begin{array}{c}\multicolumn{1}{c}{H_0:{\mathrm{\eta }}^{*}(0.05)={\mathrm{\eta }}^{*}(0.10)={\mathrm{\eta }}^{*}(0.20)={\mathrm{\eta }}^{*}(0.30)}\\ \multicolumn{1}{c}{=\dots ={\mathrm{\eta }}^{*}(0.90)={\mathrm{\eta }}^{*}(0.95)}\end{array}$$

$H_1:\exists m\backslash {\mathrm{\eta }}^{*}(m)\ne {\mathrm{\eta }}^{*}(n)$ with $m,n\in \{0.05,0.10,0.20,0.30,\dots ,0.90,0.95\}$ , corresponding to 11 quantiles: Q5, Q10, Q20, Q30, Q40, Q50, Q60, Q70, Q80, Q90, and Q95. Similarly, these hypotheses are tested on ${\mathrm{\beta }}_1$ , ${\mathrm{\beta }}_2$ , ${\mathrm{\beta }}_3$ , ${\mathrm{\beta }}_4$ , and ${\mathrm{\beta }}_5$ and on the remaining short-term parameters (i.e., ${\mathrm{\rho }}^{*}$ , ${\mathrm{\omega }}^{*}$ , ${\mathrm{\theta }}^{*}$ , ${\mathrm{\psi }}^{*}$ , and ${\mathrm{\xi }}^{*}$ ).

Data Selection

Consistent with the methodologies employed in the recent econometric literature, the secondary data were sourced from the World Bank’s comprehensive database, ensuring robustness and reliability. This study used yearly observations from 1992 to 2021 for 92 countries (Table A1). This extensive timeframe allows for analyzing long-term trends and assessing the effect of government consumption expenditure on economic growth across various global contexts. This study includes variables such as GDP growth, the unemployment rate, gross fixed capital formation, and government consumption expenditure. Trade openness is calculated as the sum of exports and imports divided by the GDP, with data sourced from both the World Bank (WB) and the Direction of Trade Statistics of the International Monetary Fund (DOTS-IMF) (Table A2). Table 2 presents the descriptive statistics of the variables used in the model.

Table 2.

Descriptive Statistics of the Model’s Variables.

Variables	Mean	Min.	Max.	SD	J–B Stats	Obs.
$\mathrm{GDPG}$	3.275	−54.236	34.500	4.832	45,216.270***	2,760
$\mathrm{UNP}$	7.515	0.140	38.800	5.876	25,540.446***	2,760
$\mathrm{INV}$	22.237	−2.424	81.021	6.702	5,142.818***	2,760
$\mathrm{OPEN}$	0.788	0.030	4.426	0.521	30,935.61***	2,760
$\mathrm{GCE}$	15.289	0.911	36.143	5.457	42,196.460***	2,760

Note. ***p < .01.

Empirical Analysis

Unit Root Test

This study applied the Levin, Lin, and Chu (LLC), Im, Pesaran, and Shin (IPS), augmented Dickey–Fuller (ADF), and Phillips–Perron (PP) tests to search for the unit root. In particular, PP is critical to ensure the validity of the subsequent econometric analysis. Table 3 details the results of the panel unit root tests. The null hypothesis of a unit root for the variables $\mathrm{GDPG}$ , $\mathrm{UNP}$ , $\mathrm{INV}$ , $\mathrm{OPEN}$ , and $\mathrm{GC}{E}^{-}$ was rejected. After taking the first difference, the $\mathrm{GC}{E}^{+}$ variable became stationary at the 1% significance level. Therefore, using the QNARDL approach was appropriate in this case.

Table 3.

Results of the Unit Root Tests.

Variable	Level	LLC	IPS	ADF	PP
$\mathrm{GDPG}$	0	−14.790***	−19.446***	−18.836***	−28.999***
	1	−20.937***	−43.488***	−35.794***	−44.025***
$\mathrm{UNP}$	0	−3.420***	−2.556***	−3.151***	−1.742**
	1	−17.622***	−20.543***	−19.655***	−29.978***
$\mathrm{INV}$	0	−4.602***	−5.197***	−5.273***	−4.243***
	1	−19.947***	−24.115***	−22.796***	−32.915***
$\mathrm{OPEN}$	0	−5.070***	−3.778***	−3.608***	−4.421***
	1	−23.535***	−28.267***	−25.681***	−36.898***
$\mathrm{GC}{E}^{+}$	0	1.738	13.028	12.835	15.136
	1	−22.628***	−22.904***	−21.657***	−33.169***
$\mathrm{GC}{E}^{-}$	0	−4.044***	5.745	5.881	1.809
	1	−22.764***	−25.764***	−22.975***	−30.728

Note. LLC, IPS, ADF, and PP are the Lin and Chu test, Im, Pesaran and Shin test, augmented Dickey–Fuller test, and Phillip–Perron test, respectively.

**

p < .05. ***p < .01.

Results of the QNARDL Model

Table 4 reports the findings from the QNARDL model, which evaluates the threshold effect of the $\mathrm{GCE}$ on $\mathrm{GDPG}$ across various economic growth distribution quantiles, $\mathrm{\tau }$ . Table 4 details the constant term ${\mathrm{\alpha }}_{*}(\mathrm{\tau })$ , the error correction mechanism (ECM) coefficient ${\mathrm{\eta }}_{*}(\mathrm{\tau })$ ., the long-term estimates ${\mathrm{\beta }}_1(\mathrm{\tau })$ to ${\mathrm{\beta }}_5(\mathrm{\tau })$ , and short-term estimates ${\mathrm{\varphi }}_{*}(\mathrm{\tau })$ to ${\mathrm{\xi }}_0(\mathrm{\tau })$ for each quantile. The ECM (−1) coefficient was consistently negative and significant at 1% across all quantiles, indicating that deviations from the long-term equilibrium are corrected over time. This finding is important because it supports the notion that the variables are linked in the long run.

Table 4.

Results of the QNARDL Model.

Quantiles	Constant	ECM (−1)	Long-term estimates					Short-term estimates						$D_{\mathrm{Crisis}}$	$D_{\mathrm{Covid}-19}$
$\mathrm{\tau }$	${\mathrm{\alpha }}_{*}(\mathrm{\tau })$	${\mathrm{\eta }}_{*}(\mathrm{\tau })$	${\mathrm{\beta }}_1(\mathrm{\tau })$	${\mathrm{\beta }}_2(\mathrm{\tau })$	${\mathrm{\beta }}_3(\mathrm{\tau })$	${\mathrm{\beta }}_4(\mathrm{\tau })$	${\mathrm{\beta }}_5(\mathrm{\tau })$	${\mathrm{\varphi }}_{*}(\mathrm{\tau })$	${\mathrm{\rho }}_0(\mathrm{\tau })$	${\mathrm{\omega }}_0(\mathrm{\tau })$	${\mathrm{\theta }}_0(\mathrm{\tau })$	${\mathrm{\psi }}_0(\mathrm{\tau })$	${\mathrm{\xi }}_0(\mathrm{\tau })$	${\mathrm{\lambda }}_1(\mathrm{\tau })$	${\mathrm{\lambda }}_2(\mathrm{\tau })$
0.05	0.307 (1.017)	−0.495*** (−9.635)	−0.148** (−3.066)	0.015 (0.580)	−0.902 (−1.125)	−0.613*** (−6.686)	−0.359*** (−5.437)	−0.184*** (−3.775)	−1.085*** (−6.741)	0.105 (1.422)	2.336 (0.956)	−3.324*** (−8.936)	1.063*** (3.325)	−1.008** (−2.112)	−1.524** (2.172)
0.10	0.360* (1.842)	−0.534*** (−11.128)	−0.109*** (−3.792)	0.073*** (4.567)	−0.624* (−1.770)	−0.415*** (−7.425)	−0.235*** (−5.398)	−0.169*** (−4.009)	−0.882*** (−6.657)	0.163** (2.171)	3.738* (1.837)	−2.936*** (−6.361)	0.467* (1.950)	−0.939** (−2.420)	−2.233*** (−3.691)
0.20	0.147 (1.197)	−0.541*** (−14.730)	−0.047** (−2.193)	0.096*** (8.907)	0.043 (0.153)	−0.271*** (−7.164)	−0.149*** (−4.621)	−0.176*** (−5.554)	−0.769*** (−8.160)	0.147** (2.386)	5.260*** (3.356)	−1.948*** (−4.651)	0.149 (1.061)	−0.969*** (−2.722)	−2.490*** (−4.763)
0.30	0.128 (1.432)	−0.514*** (−13.714)	−0.034** (−2.345)	0.124*** (14.750)	0.095 (0.575)	−0.195*** (−7.261)	−0.115*** (−5.369)	−0.198*** (−5.818)	−0.716*** (−8.844)	0.179*** (3.967)	3.977*** (2.836)	−1.505*** (−6.188)	−0.139 (−0.863)	−0.797** (−2.411)	−2.337*** (−4.524)
0.40	0.104 (1.030)	−0.506*** (−14.195)	−0.029*** (−2.666)	0.147*** (17.895)	0.040 (0.285)	−0.121*** (−3.963)	−0.090*** (−3.694)	−0.207*** (−6.261)	−0.687*** (−8.210)	0.166*** (4.299)	5.157*** (3.696)	−1.064*** (−4.248)	−0.397*** (−4.055)	−0.582* (−1.846)	−1.789*** (−2.851)
0.50	0.145* (1.784)	−0.513*** (−18.988)	−0.031*** (−2.620)	0.172*** (24.448)	0.032 (0.217)	−0.050* (−1.743)	−0.051** (−2.122)	−0.210*** (−6.006)	−0.744*** (−8.455)	0.167*** (4.719)	5.933*** (4.599)	−0.741*** (−4.241)	−0.466*** (−4.676)	−0.387 (−1.608)	−1.096* (−1.800)
0.60	0.212*** (2.623)	−0.521*** (−14.827)	−0.022* (−1.838)	0.197*** (25.576)	0.132 (0.660)	−0.021 (−1.093)	−0.036 (−1.547)	−0.208*** (−5.730)	−0.702*** (−6.887)	0.173*** (5.105)	6.354*** (6.848)	−0.572*** (−4.663)	−0.539*** (−3.572)	−0.387* (−1.937)	−0.228 (−0.408)
0.70	0.279*** (2.928)	−0.541*** (−14.199)	−0.019* (−1.854)	0.213*** (23.325)	0.289 (1.277)	−0.012 (−0.539)	−0.055** (−2.280)	−0.191*** (−5.557)	−0.749*** (−6.543)	0.186*** (4.316)	6.415*** (5.623)	−0.334*** (−2.754)	−0.729*** (−4.107)	−0.329 (−1.328)	0.453 (0.822)
0.80	0.393*** (3.762)	−0.612*** (−15.201)	−0.012 (−0.913)	0.240*** (36.146)	0.435*** (2.614)	0.022 (0.810)	−0.048** (−2.055)	−0.160*** (−5.357)	−0.755*** (6.094)	0.216*** (4.332)	5.634*** (3.463)	−0.188 (−1.174)	−0.891*** (−4.573)	−0.428 (−1.406)	1.009 (1.416)
0.90	0.412*** (2.808)	−0.643*** (−15.182)	0.006 (0.358)	0.275*** (24.942)	0.705*** (3.193)	0.055 (1.450)	−0.061 (−1.511)	−0.138*** (−3.548)	−0.698*** (−5.424)	0.216*** (4.517)	4.729** (2.329)	0.116 (0.866)	−1.579*** (−4.283)	−0.119 (−0.316)	1.388* (1.809)
0.95	0.589** (2.200)	−0.610*** (−11.855)	0.049 (0.985)	0.277*** (13.906)	1.711*** (2.898)	0.102* (1.905)	−0.074 (−1.532)	−0.174*** (−3.403)	−0.783*** (−4.389)	0.161* (1.918)	6.011 (1.403)	0.149 (0.787)	−2.748*** (−3.930)	−0.619 (−1.384)	1.227 (1.217)

Note. The value in parentheses represents the t-stat value.

*

p < .1. **p < .05. ***p < .01.

The $\mathrm{GCE}$ coefficients were both significant and negative across most quantiles for $\mathrm{GC}{E}^{+}$ ( ${\mathrm{\beta }}_4$ ) and $\mathrm{GC}{E}^{-}$ ( ${\mathrm{\beta }}_5$ ), implying that an increase in government expenditure may have an attenuating effect on economic growth in the long run. In particular, the $\mathrm{GC}{E}^{+}$ coefficient was negative and significant at the quantiles $\mathrm{\tau }$  = 0.05, $\mathrm{\tau }$  = 0.1, $\mathrm{\tau }$  = 0.2, $\mathrm{\tau }$  = 0.3, $\mathrm{\tau }$  = 0.4, and $\mathrm{\tau }$  = 0.5, whereas it was positive and significant only at the quantile $\mathrm{\tau }$  = 0.95. Similarly, the $\mathrm{GC}{E}^{-}$ coefficient was negative and significant at the quantiles $\mathrm{\tau }$  = 0.05, $\mathrm{\tau }$  = 0.1, $\mathrm{\tau }$  = 0.2, $\mathrm{\tau }$  = 0.3, $\mathrm{\tau }$  = 0.4, $\mathrm{\tau }$  = 0.5, $\mathrm{\tau }$  = 0.7, and $\mathrm{\tau }$  = 0.8. Furthermore, $\mathrm{GC}{E}^{-}$ had a significantly negative impact on $\mathrm{GDPG}$ across more quantiles than $\mathrm{GC}{E}^{+}$ . The negative effects of both $\mathrm{GC}{E}^{+}$ and $\mathrm{GC}{E}^{-}$ suggest that changes in government expenditure did not boost long-term economic growth in the countries studied.

The ${\mathrm{\psi }}_0(\mathrm{\tau })$ and ${\mathrm{\xi }}_0(\mathrm{\tau })$ coefficients show the short-term impact of $\mathrm{\Delta }\mathrm{GC}{E}^{+}$ and $\mathrm{\Delta }\mathrm{GC}{E}^{-}$ on $\mathrm{GDPG}$ . The ${\mathrm{\psi }}_0$ coefficient was negative and significant at the quantiles $\mathrm{\tau }$  = 0.05, $\mathrm{\tau }$  = 0.1, $\mathrm{\tau }$  = 0.2, $\mathrm{\tau }$  = 0.3, $\mathrm{\tau }$  = 0.4, $\mathrm{\tau }$  = 0.5, $\mathrm{\tau }$  = 0.6, and $\mathrm{\tau }$  = 0.7. Similarly, a significant negative coefficient of ${\mathrm{\xi }}_0$ was found at the $\mathrm{\tau }$  = 0.4, $\mathrm{\tau }$  = 0.5, $\mathrm{\tau }$  = 0.6, $\mathrm{\tau }$  = 0.7, $\mathrm{\tau }$  = 0.8, $\mathrm{\tau }$  = 0.9, and $\mathrm{\tau }$  = 0.95 quantiles, except for the positive coefficient observed at the $\mathrm{\tau }$  = 0.05 and $\mathrm{\tau }$  = 0.1 quantiles. The significance and sign of these coefficients vary across quantiles, suggesting that the impact of changes in the $\mathrm{GCE}$ on $\mathrm{GDPG}$ is not uniform and may differ from the economy’s current state in the short run. In general, government expenditures have a predominantly negative influence on economic growth with a tiny positive impact. In addition, the impact of the two global shocks, $D_{\mathrm{Crisis}}$ and $D_{\mathrm{Covid}-19}$ , on economic growth was negative, as negative coefficients were detected at quantiles $\mathrm{\tau }$  = 0.05, $\mathrm{\tau }$  = 0.1, $\mathrm{\tau }$  = 0.2, $\mathrm{\tau }$  = 0.3, $\mathrm{\tau }$  = 0.4, $\mathrm{\tau }$  = 0.5, and $\mathrm{\tau }$  = 0.6. The remaining quantiles showed insignificant impacts for these shocks, except for a single positive coefficient detected at quantile $\mathrm{\tau }$  = 0.9 for $D_{\mathrm{Covid}-19}$ . This indicates that macroeconomic shocks exerted significant and widespread negative impacts on the studied countries.

The Wald test was used to assess the asymmetries of the parameters across the quantiles examined in the QNARDL model. The Wald test can detect instability for both the intercept and coefficient. Moreover, it can identify structural changes with both known and unknown breakpoints. Table 5 presents the results of the Wald test. The null hypothesis stated that coefficients are equal across different quantiles of the dependent variable, in this case, economic growth.

Table 5.

Results of the Wald Test.

Parameter	Coefficient	Prob.
${\mathrm{\beta }}_1$	3.805	.578
${\mathrm{\beta }}_2$	5.944	.312
${\mathrm{\beta }}_3$	4.203	.521
${\mathrm{\beta }}_4$	20.930***	.001
${\mathrm{\beta }}_5$	17.605***	.004
${\mathrm{\varphi }}_{*}$	10.491*	.063
${\mathrm{\rho }}_0$	5.833	.323
${\mathrm{\omega }}_0$	3.864	.569
${\mathrm{\theta }}_0$	3.497	.624
${\mathrm{\psi }}_0$	20.820***	.001
${\mathrm{\xi }}_0$	5.488	.359
${\mathrm{\eta }}_{*}$	37.475***	.000

*

p < .1. ***p < .01.

Table 5 also presents the parameters’ coefficients ( ${\mathrm{\beta }}_1$ to ${\mathrm{\beta }}_5$ , ${\mathrm{\varphi }}_{*}$ , ${\mathrm{\rho }}_0$ , ${\mathrm{\omega }}_0$ , ${\mathrm{\theta }}_0$ , ${\mathrm{\psi }}_0$ , ${\mathrm{\xi }}_0$ , and ${\mathrm{\eta }}_{*}$ ) alongside the probability values (p-values) that indicate the likelihood of the null hypothesis being true. The results in Table 5 reveal the symmetric relationship between $\mathrm{UNP}$ , $\mathrm{INV}$ , and $\mathrm{OPEN}$ with $\mathrm{GDPG}$ in both the short and long rum. However, the asymmetric relationship between $\mathrm{GC}{E}^{+}$ or $\mathrm{GC}{E}^{-}$ and $\mathrm{GDPG}$ persisted in the long run, as the null hypothesis was rejected at the 1% significance level. Furthermore, in the short run, the asymmetric relationship was observed only for $\mathrm{GC}{E}^{+}$ , while $\mathrm{GC}{E}^{-}$ and $\mathrm{GDPG}$ displayed symmetry. The behavior of the parameter estimates on the quantiles is depicted in Figure 1. The estimated parameters for QNARDL on the quantiles and the basic ARDL estimator fall within the same range.

Figure 1.

Quantile parameter estimates.

Robustness Tests

To ensure the validity of this study’s findings, we conducted a comprehensive robustness check to address potential endogeneity and heterogeneity in the data. We applied the SGMM model developed by Blundell and Bond (1998) for this purpose. SGMM’s advantage was that it provided efficient estimates for dynamic panel data models Adebayo et al. (2023) and overcame issues such as endogeneity, autocorrelation, heterogeneity, and fixed effects (Konstantakopoulou, 2022; Roodman, 2009). The SGMM model estimation results in Table 6 show that both $\mathrm{GC}{E}^{+}$ and $\mathrm{GC}{E}^{-}$ strongly hindered $\mathrm{GDPG}$ , further reinforcing the QNARDL model’s results. Significant negative effects on economic growth were also found for the global shocks $D_{\mathrm{Crisis}}$ and $D_{\mathrm{Covid}-19}$ . Using diagnostic tests such as the Arellano-Bond [AR (2)] test resulted in the acceptance of the null hypothesis at the 5% significance level (p = .410), confirming that the model did not exhibit second-order autocorrelation in the residuals. Additionally, the results of the Sargan and Hansen tests confirmed the instruments’ validity, with p-values above the 5% significance level (.640 and .706, respectively). Overall, the SGMM model’s results are consistent with those of the QNARDL model and complement it. Endogeneity and heterogeneity were addressed, enhancing our findings’ validity.

Table 6.

Results of the SGMM Model.

Variable	Coefficients	SE	z-Statistic	p-Value
$\mathrm{GDP}{G}_{t-1}$	0.124	0.142	0.87	.385
$\mathrm{UNP}$	−0.073***	0.028	−2.59	.009
$\mathrm{INV}$	0.060	0.043	1.38	.168
$\mathrm{OPEN}$	0.311	0.232	1.34	.180
$\mathrm{GC}{E}^{+}$	−0.186***	0.061	−3.03	.002
$\mathrm{GC}{E}^{-}$	−0.188***	0.053	−3.57	.000
$D_{\mathrm{Crisis}}$	−0.638**	0.296	−2.16	.031
$D_{\mathrm{Covid}-19}$	−3.872***	0.537	−7.21	.000
Constant	2.275**	1.088	2.09	.037
AR (1) test	−3.40 (.001)
AR (2) test	−0.82 (.410)
Sagan test	0.22 (.640)
Hansen test	0.14 (.706)

Note. The values in parentheses are the p-values.

**

p < .05. ***p < .01.

Discussion

Few studies have investigated the asymmetric impact of government expenditure on economic growth in terms of quantiles. Thus, this study has made a notable contribution regarding methodology because the QNARDL model was applied to a sample of 92 countries from 1992 to 2021.

The results in Table 4 show that, in the long run, both $\mathrm{GC}{E}^{+}$ and $\mathrm{GC}{E}^{-}$ have a significantly negative effect on $\mathrm{GDPG}$ in most quantiles, except for $\mathrm{\tau }$  = 0.95. Similarly, in the short run, $\mathrm{GC}{E}^{+}$ and $\mathrm{GC}{E}^{-}$ have a significantly negative impact on $\mathrm{GDPG}$ for most quantiles except for $\mathrm{\tau }$  = 0.05 and $\mathrm{\tau }$  = 0.1 in the case of $\mathrm{GC}{E}^{-}$ . These findings indicate that the government’s use of fiscal policy by expanding government expenditure does not stimulate economic growth. Government expenditure is often increased and taxes are reduced to achieve rapid growth as this stimulates aggregate demand, fostering production and economic growth. However, implementing an expansionary fiscal policy increases the budget deficit, and the government must borrow (i.e., public debt rises) to finance the deficit. Debt raises financing costs, gradually substituting private investment, and potentially leading to tax increases. In turn, this may hinder overall economic growth. This result aligns with previous studies (Afonso & Furceri, 2010; Dar & AmirKhalkhali, 2002; Fatás & Mihov, 2010; Fölster & Henrekson, 2001; Hasnul, 2015). Both positive and negative changes in $\mathrm{GCE}$ affect economic growth negatively across most quantiles; this finding contributes to the literature on fiscal policy and economic growth. In particular, the nonlinear effects captured through the QNARDL model across 92 countries over nearly three decades represent a methodological advancement in understanding fiscal policy’s impact.

By comparing our results with previous studies (Afonso & Furceri, 2010; Dar & AmirKhalkhali, 2002; Fatás & Mihov, 2010; Fölster & Henrekson, 2001; Hasnul, 2015), we can conclude that an increase in government size and expenditure hampers economic growth because of inefficiencies and the inhibition of private investment. In other words, excessive government expenditure results in volatility and negative effects on economic stability and growth. This study builds on these perspectives by providing empirical evidence of these negative impacts across various economic conditions, emphasizing the need for a more detailed approach to fiscal policy.

The significant negative coefficients of the error correction mechanism ${\mathrm{\rho }}_{*}$ across all quantiles suggest that the system tends to return to equilibrium after a shock. This reinforces the role of government consumption expenditure as a potential stabilizer in the economic system, helping correct deviations from the long-term growth path.

Furthermore, our results reveal that there are still a few quantiles where there is a positive relationship between government expenditure and economic growth, such as $\mathrm{\tau }$  = 0.95 in the long run, and $\mathrm{\tau }$  = 0.05 and $\mathrm{\tau }$  = 0.1 in the short run. This demonstrates that the impact of government expenditure depends not only on the state of the economy but also on the nature of the expenditure itself. Positive changes in $\mathrm{GCE}$ have a stronger impact on the higher quantiles, whereas negative changes in $\mathrm{GCE}$ have a stronger impact on the mid- and higher quantiles. This dichotomy can be interpreted as evidence of the nonlinear effects of fiscal policy: increasing government expenditure may not always result in enhanced growth and reducing expenditure does not necessarily lead to a contraction.

These findings’ implications are manifold. For policymakers, the challenge lies in identifying the appropriate level and timing of government consumption expenditure to optimize its impact on economic growth. The evidence suggests that a one-size-fits-all approach to fiscal policy is unlikely to be effective. Instead, a context-dependent strategy that considers the current economic conditions and the unique characteristics of the economy in question is required. The threshold effect of government expenditure also has ramifications for economic forecasting and modeling. Traditional linear models may fail to capture the complex dynamics uncovered in this study. Hence, economic models need to incorporate nonlinearities and asymmetries to better predict the outcomes of fiscal policy interventions.

Finally, the analysis of the threshold effect of government consumption expenditure on economic growth reveals a sophisticated interplay between fiscal policy and economic performance. The results highlight the importance of tailoring government expenditure to the prevailing economic conditions and recognizing the nonlinear and asymmetric responses of economic growth to fiscal stimuli. These insights are vital for formulating effective fiscal strategies that can adapt to the fluctuating tides of economic activity and steer countries toward sustainable growth.

Conclusion

We conducted a detailed econometric analysis to elucidate the nonlinear effect of government expenditure on economic growth. The findings obtained with the QNARDL model provide detailed insights into the asymmetric effects of government expenditure on economic growth across different quantiles. Contrary to traditional views of fiscal policy as a consistent stabilizer, our results demonstrate that accumulated changes in government expenditure have a significant and negative impact on economic growth. Furthermore, the short-term impact is more widely observed than the long-term impact. This aligns with classical economic theories that posit that excessive government intervention can lead to inefficiencies, increased borrowing costs, and a crowding-out effect on private investment. The fiscal policy should vary depending on the economic situation, advocating for a more granular policy formulation. The coefficients’ significance across different quantiles demonstrates the varying impact of changes in government expenditure on economic growth. The effect of changes in government expenditure is more pronounced on lower quantiles than on higher quantiles. This study confirms that both positive and negative variations in government expenditure are significant determinants of economic growth, but their effects are not uniformly distributed across different economic conditions. This comprehensive understanding of the threshold effect provides valuable knowledge for policymakers, as it indicates that the effectiveness of government expenditure in stimulating economic growth varies depending on the current economic conditions.

Policy Implications

This study’s policy implications are profound. The results suggest that government expenditure is a potent tool for economic management, capable of influencing growth rates in both expansionary and contractionary phases. However, the effectiveness of government consumption expenditure is contingent upon the economic environment. Thus, policymakers should consider the economy’s position relative to its growth distribution when deciding on the scale and direction of fiscal interventions.

At low quantiles, negative changes in government expenditure can serve as effective stimuli in the short run. By contrast, at higher quantiles, the marginal benefit of additional government expenditure decreases. Furthermore, at the highest quantile, changes in government expenditure may result in a marginal boost to economic growth in the long run, but this effect is limited. This underscores the need for context-specific fiscal strategies. A one-size-fits-all fiscal policy is inadequate because it fails to consider the specific dynamics of different economic environments. Therefore, the proposed approach challenges the efficacy of universal fiscal policies and advocates for tailored interventions that consider each country’s unique economic conditions and characteristics. These findings provide policymakers with a framework for crafting more adaptable and effective fiscal strategies, considering their economies’ unique characteristics and growth stages.

Limitations and Future Research

While this study’s findings offer insightful contributions to our understanding of the relationship between changes in government expenditure and economic growth, the limitations that may influence the interpretation and application of these results should be considered. First, the expansive dataset covering 92 countries over nearly three decades represents a strength. However, the variability in data quality and availability from different sources introduces a layer of complexity to our analysis. Second, this study focuses on the aggregate level of $\mathrm{GCE}$ without dissecting the composition of government expenditure and thus may overlook the differential impacts of capital versus current expenditures on economic growth. This suggests that future research could benefit from a more granular analysis of different spending categories.

Future research should address the abovementioned limitations by incorporating a more diverse set of economic variables, using alternative econometric approaches, and examining the composition of government expenditure. Furthermore, more studies are needed to understand the global factors that influence the effect of government consumption expenditure on economic growth. Dividing the data between OECD and non-OECD countries may produce more detailed information. With these improvements, the economic literature can provide detailed guidance to policymakers seeking to harness fiscal policy for economic stabilization and growth.