Dynamic Patterns of Government Spending Allocation in Indian States

Abstract

This study examines the allocation and fluctuation of 12 major categories of government spending at the state level in India. Using panel data from 15 prominent Indian states from 1990 to 2021, a panel autoregressive distributed lag (ARDL) method is employed. The findings reveal that increased central government funding encourages spending on education, agriculture, irrigation, disaster management, nutrition, and rural and urban development. In contrast, economic growth and increased revenue shift focus to energy, healthcare, urban development, transportation, communication, and water supply and sanitation. High public debt adversely affects all spending categories, highlighting the cyclic nature of various expenditures.

JEL Codes: H50, H72, E32, C33

I. Introduction

The effective allocation of public resources through prioritised spending ¹ is a crucial element of fiscal policy that significantly impacts economic development and social welfare. Understanding how governments prioritise spending across different sectors over the business cycle is essential for effective policymaking and resource allocation. While extensive research has explored the cyclicality of public expenditure ² and macro-fiscal factors at the national level (Afonso & Jalles, 2013; Jalles, 2020, 2021), there remains a notable gap in understanding the cyclical patterns and priorities of public expenditure across various sectors in emerging economies. India is one of the fastest-growing emerging economies, and its diverse regional characteristics make it an intriguing case for studying public spending prioritisation dynamics. This study aims to address this gap by providing a comprehensive analysis of public spending priorities across various sectors in Indian states, offering valuable insights for policymakers and practitioners involved in fiscal management and resource allocation.

Existing research suggests that social spending, including areas like health and education, tends to exhibit acyclical behaviour, while spending on social security and welfare shows countercyclical patterns in developed countries (Afonso & Jalles, 2013). Similarly, studies focusing on developing economies have found that government spending on social sectors, such as health, education and social protection, tends to remain relatively stable over time (Jalles, 2020). Recent investigations utilising time-varying measures have revealed that the cyclicality of government spending varies across different expenditure categories in advanced economies (Jalles, 2021). For instance, total government spending, wages, goods and services expenditure, primary expenditure, and interest payments exhibit countercyclical behaviour, indicating a negative relationship with output growth, while public investment tends to be procyclical. Previous studies have examined spending behaviour over the business cycle using macroeconomic determinants such as per capita gross domestic product (GDP), trade openness, government size, financial development and public debt (Jalles, 2021). These findings highlight the importance of understanding the dynamics of public spending priorities in different sectors and their implications for economic stability and growth.

The idea of spending prioritisation stems from the literature on fiscal space for the social sector, particularly the health sector, which has historically received lower budgetary allocations compared to other spending categories (Barroy et al., 2018; Tandon et al., 2014). These studies argue that spending prioritisation is strongly linked to conducive macro-fiscal policies. While a few studies in India have explored spending prioritisation and its macro-fiscal impact at the state level (Behera & Dash, 2019a, 2019b; Behera et al., 2020), these studies have primarily focused on specific sectors, such as the social sector, and have not comprehensively analysed the prioritisation of spending across multiple categories. Our study seeks to fill this research gap by providing a comprehensive analysis of public spending priorities across a range of sectors, including education, agriculture, energy, health, rural development, urban development, social security, irrigation and flood control, transport and communication, water supply and sanitation, and nutrition. Using extensive panel data from 15 major Indian states, we employ a panel autoregressive distributed lag (panel ARDL) approach to assess the cyclicality of state government spending in various categories and to identify the underlying determinants of sectoral spending priorities.

While the literature adopts a variety of methodological approaches to analyse expenditure cyclicality and prioritisation—primarily through time-series and panel data models—our study advances this by incorporating more robust techniques. These include unit root tests (Dickey & Fuller, 1979; Im et al., 2003; Levin et al., 2002), cross-sectional dependence (CSD) tests (Pesaran, 2004), and dynamic panel estimators such as the pooled mean group (PMG) estimator (Pesaran et al., 1999). Furthermore, we apply the cross-sectional ARDL approach proposed by Chudik et al. (2016), which allows for heterogeneous effects and cross-sectional correlations, thereby enhancing the robustness and validity of the results.

To the best of our knowledge, this is the first study to estimate the cyclical patterns of different categories of state government spending in India using this comprehensive framework. Additionally, the analysis incorporates key macroeconomic variables—such as economic growth, aggregate revenue, central transfers, public debt and financial development—to explore the factors influencing subnational spending decisions. These contributions distinguish our work from existing studies and provide novel insights into the nuanced dynamics of public spending prioritisation. The findings are expected to assist policymakers in aligning budgetary allocations with macroeconomic conditions and long-term development goals.

The remainder of the article is structured as follows. Section II presents a review of the relevant literature. Section III outlines the data sources and methodology adopted in the study. Section IV discusses the empirical findings, with a focus on the cyclicality and key determinants of spending prioritisation across Indian states. Finally, Section V concludes with the main insights and policy implications.

II. Literature Review

This section reviews the global and Indian literature on the cyclicality of public expenditure, sectoral spending prioritisation, and the political economy of fiscal policy. Emphasis is placed on insights relevant to India’s federal fiscal structure, with a particular focus on various development sector spending. The review is structured thematically to align with the conceptual framework and empirical objectives of the study.

Cyclicality of Public Expenditure: Global and Developing Country Evidence

A substantial body of empirical work has examined whether public expenditure behaves procyclically—rising during economic booms and contracting during downturns—or countercyclically, in line with Keynesian stabilisation principles. However, evidence from developing and emerging economies suggests a predominance of procyclical behaviour, often attributed to institutional weaknesses and fiscal rigidity.

Bova et al. (2014), in a cross-country International Monetary Fund (IMF) study, report that many developing countries display procyclical fiscal policy due to weak institutional frameworks and the absence of effective fiscal rules. Similarly, Gupta et al. (2010) find that social spending—particularly on health and education—is more vulnerable to procyclical patterns, with sharp increases in spending during booms but relatively mild reductions during recessions. This asymmetric response poses risks to long-term human development objectives.

Expanding on these findings, Afonso and Jalles (2013) and Jalles (2020, 2021) explore the time-varying nature of fiscal cyclicality across social sectors. Their results, drawn from diverse country samples, highlight substantial heterogeneity in cyclicality patterns, shaped by fiscal space, governance, and political incentives. Aizenman et al. (2019) also underscore the centrality of fiscal space in determining governments’ ability to adopt countercyclical policies, while Brueckner and Gradstein (2014) distinguish between responses to temporary versus persistent shocks, revealing the procyclical bias of developing economies under institutional constraints.

Sectoral Spending and Prioritisation: International and Indian Contexts

Sectoral allocation decisions—especially in health, education and infrastructure—are shaped by political, fiscal and developmental considerations. Abbott and Jones (2013, 2021, 2022) find that in Organisation for Economic Co-operation and Development (OECD) countries, the cyclicality of sectoral spending, including health and environmental expenditure, often reflects political incentives and voter pressures rather than stabilisation goals. Liang and Tussing (2019) further demonstrate the implications of cyclical health spending on public health outcomes.

In the Indian context, Goyal and Sharma (2015) analyse the composition and cyclicality of government expenditure and conclude that capital expenditure is more growth-enhancing and less procyclical than revenue expenditure. Ramanjini and Gayithri (2020), focusing on education expenditure across Indian states, identify a strong procyclical tendency, with budget cuts during slowdowns driven by revenue pressures. These findings are consistent with Behera and Dash (2019a, 2019b) and Behera et al. (2020), who emphasise persistent underinvestment in health at the state level, constrained by fiscal transfer mechanisms and limited revenue autonomy.

Bose and Banerjee (2025) critique the growth-centric fiscal paradigm in India and call for a macro-fiscal framework that embeds human development priorities. Similarly, Tandon et al. (2014) and Barroy et al. (2018) argue that reprioritising public expenditure towards health and social infrastructure is essential for achieving the Sustainable Development Goals (SDGs), but fiscal space and inter-sectoral trade-offs remain major hurdles.

Political Economy of Fiscal Policy and Federal Dynamics

Political and institutional factors significantly influence fiscal behaviour and spending priorities. Ayala-Cañón et al. (2022) show that social spending tends to synchronise across countries during crises, revealing a political dimension to expenditure decisions. Mawejje and Odhiambo (2022) demonstrate the influence of political regimes and governance quality on fiscal cyclicality in East Africa.

In India, the federal structure adds another layer of complexity. Behera et al. (2020) suggest that central transfers can moderate procyclical tendencies at the state level, although disparities in fiscal capacity remain pronounced. This is consistent with Xing and Fuest’s (2018) findings in the Chinese context, where decentralised fiscal arrangements exacerbate procyclical spending due to revenue uncertainties and soft budget constraints.

Official policy documents also reflect evolving concerns. The Economic Survey 2021–2022 (Ministry of Finance, 2022) stresses the need to strengthen countercyclical buffers and reprioritise expenditure in health and infrastructure as part of post-pandemic recovery. The Economic Survey 2016–2017 (Ministry of Finance, 2017) highlights that subnational fiscal stress often leads to disproportionate cuts in social sector spending, especially when debt obligations and revenue shocks hit state budgets.

Despite a growing body of work, several gaps remain. Most studies either examine national-level aggregates or focus on isolated sectors, leaving a gap in state-level, cross-sectoral assessments of expenditure cyclicality. Moreover, few empirical studies systematically incorporate political and institutional determinants into models of fiscal behaviour at the subnational level. There is also insufficient alignment between empirical findings and the policy discourse articulated in official surveys and fiscal documents. This study addresses these gaps by conducting a panel analysis of sector-specific expenditure cyclicality across Indian states, using macro-fiscal variables. By grounding the analysis in a more diverse and robust body of literature, the study aims to inform the design of more resilient and equity-oriented fiscal policies at the subnational level in India.

III. Data and Methodology

Data

This study utilises annual panel data from 1990 to 2021, focusing on 15 major Indian states. ³ The data utilised for empirical analysis include a comprehensive range of developmental public expenditures, encompassing sectors such as education, agriculture, energy, health, irrigation and flood control, natural calamities, rural development, urban development, social security and welfare, transport and communication, water supply and sanitation, and nutrition. By considering these 12 key public spending categories, this study covers both social and economic services within the budget classifications. ⁴ In addition to public spending, the analysis incorporates other macro-fiscal variables, namely GSDP per capita, total revenue, grants from the centre, share in central tax, total public debt, and the credit-deposit ratio of all scheduled commercial banks. These data have been obtained from the Economic and Political Weekly Research Foundation (EPWRF) India Time Series database. Detailed information about the variables and their respective sources can be found in Table 1. To measure economic growth, the study employs GSDP per capita as a proxy, considering its relevance in capturing the overall economic performance of the states. Furthermore, the credit-deposit ratio is utilised as a proxy for financial development, providing insights into the strength and efficiency of the banking sector.

Table 1.

Variables and Data Sources.

Variable	Variable Description	Unit of Measure	Source
LEDEX	Log of public expenditure on education	(% of total expenditure)	EPWRF
LAREX	Log of public expenditure on agriculture	(% of total expenditure)	EPWRF
LENEX	Log of public expenditure on energy	(% of total expenditure)	EPWRF
LHEXP	Log of public expenditure on health	(% of total expenditure)	EPWRF
LIFEX	Log of public expenditure on irrigation and flood control	(% of total expenditure)	EPWRF
LNCEX	Log of public expenditure on natural calamities	(% of total expenditure)	EPWRF
LRDEX	Log of public expenditure on rural development	(% of total expenditure)	EPWRF
LUDEX	Log of public expenditure on urban development	(% of total expenditure)	EPWRF
LSWEX	Log of public expenditure on social security and welfare	(% of total expenditure)	EPWRF
LTCEX	Log of public expenditure on transport and communication (including spending on science, technology and environment)	(% of total expenditure)	EPWRF
LWSEX	Log of public expenditure on water supply and sanitation	(% of total expenditure)	EPWRF
LNUEX	Log of public expenditure on nutrition	(% of total expenditure)	EPWRF
LPGSD	Log of per capita state gross domestic product	(Ratio with population)	EPWRF
LTRVG	Log of total revenue	(% of GSDP)	EPWRF
LCGSG	Log of grants from the centre and share in central tax	(% of GSDP)	EPWRF
LPDBG	Log of total public debt	(% of GSDP)	EPWRF
LCDRA	Log of all scheduled commercial banks credit to deposit ratio	(In per cent)	EPWRF

Note: EPWRF: Economic and Political Weekly Research Foundation; GSDP: Gross state domestic product.

Methodology

The study aims to investigate the effect of macro-fiscal variables on selected public spending in Indian states. Specifically, it explores the impact of economic growth, total revenue, grants and tax share from the central government, public debt and financial development on various categories of public spending. Additionally, a separate analysis is conducted to measure the influence of the business cycle, considering economic growth during both good and bad years, along with other relevant variables. The model specification for the econometric analysis is shown below.

$$y_{it}=\mathrm{ }{\complement }_{it}+\mathrm{ }{\alpha }_1{\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D}}_{it}+{\alpha }_2{\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G}}_{it}+\mathrm{ }{\alpha }_3{\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G}}_{it}+{\alpha }_4{\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G}}_{it}+{\alpha }_5{\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}}_{it}+\mathrm{ }{\vartheta }_{it}+{\epsilon }_{it}$$

(1)

In Equation (1), $y_{it}$ represents the selected public spending category, ⁵ LPGSD is the log of GSDP per capita, LTRVG is the log of total revenue as a per cent of GSDP, LCGSG is the log of grants from the centre and the share in central tax as a per cent of GSDP, LPDBG is the log of total public debt as a per cent of GSDP, and LCDRA is the log of all scheduled commercial banks credit to deposit ratio. ${\complement }_{it}$ is the constant term, ${\vartheta }_{it}$ is an unobserved country-specific effect and ${\epsilon }_{it}$ is the error term. The ${\alpha }_1, {\alpha }_2,{\alpha }_3,{\alpha }_4 \mathrm{a}\mathrm{n}\mathrm{d} {\alpha }_5$ represent their coefficients. The expected signs of the selected macro-fiscal variables are ambiguous and need to be empirically estimated.

This study employs a panel ARDL approach to model the dynamic relationship between sectoral government expenditure and macro-fiscal variables across 15 Indian states over the period 1990–2021. The panel ARDL model is chosen due to its suitability for our data set. It can accommodate a mix of stationary variables, including I(0) and I(1), and applies to studies with small sample sizes, providing reliable estimates even with limited observations. With 15 cross-sections and a 32-year time series, which may be considered relatively small for most panel studies, the method adequately addresses the data characteristics. It addresses potential endogeneity issues in panel data analysis by incorporating lagged dependent variables and other exogenous variables. Furthermore, it captures both the short-run dynamics and long-run equilibrium relationships in a heterogeneous panel setting, where states differ in economic structure, fiscal capacity and policy environment.

Given the characteristics of the Indian states, the panel ARDL method, particularly the PMG-ARDL estimation, is adopted in this study. The PMG estimator, proposed by Pesaran et al. (1999, 2001), is based on the assumption that cross-sectional units (here Indian states) share a common long-run equilibrium relationship, while allowing for heterogeneity in short-run dynamics, intercepts and error variances. This makes it particularly appropriate when long-term fiscal responses are expected to be similar across units, but short-term adjustments differ. Equations (1)–(6) can be expressed in the panel ARDL form developed by Pesaran et al. (1999, 2001) as:

$$\text{Δ}{y}_{it}=C_{it}+{\alpha }_i\sum _{i=1}^{\rho }\text{Δ}{y}_{it-i}+{\pi }_i\sum _{i=0}^{\rho }{\text{ΔLPGSD}}_{it-i}+{\varnothing }_i\sum _{i=0}^{\rho }{\text{ΔLTRVG}}_{it-i}+{\omega }_i\sum _{i=0}^{\rho }{\text{ΔLLGSG}}_{it-i}+{\delta }_i\sum _{i=0}^{\rho }{\text{ΔLPDBG}}_{it-i}+{\mu }_i\sum _{i=0}^{\rho }{\text{ΔLCDRA}}_{it-i}+{\beta }_1{y}_{it-1}+{\beta }_2{\text{LPGSD}}_{it}+{\beta }_3{\text{LTRVG}}_{it}+{\beta }_4{\text{LCGSG}}_{it}+{\beta }_5{\text{LPDBG}}_{it}+{\beta }_6{\text{LCDRA}}_{it}+{\vartheta }_{it}+{\epsilon }_{it}$$

(2)

The extended model can be expressed in the following form.

$$\text{Δ}{y}_{it}=C_{it}+{\alpha }_i\sum _{i=1}^{\rho }\text{Δ}{y}_{it-i}+{\text{Ω}}_i\sum _{i=1}^{\rho }{\text{ΔGGSDP}}_{it-i}+{\gamma }_i\sum _{i=1}^{\rho }{\text{ΔBGSDP}}_{it-i}+{\varnothing }_i\sum _{i=0}^{\rho }{\text{ΔLTRVG}}_{it-i}+{\omega }_i\sum _{i=0}^{\rho }\text{Δ}LCGS{G}_{it-i}+{\delta }_i\sum _{i=0}^{\rho }\text{Δ}LPDB{G}_{it-i}+{\mu }_i\sum _{i=0}^{\rho }{\text{ΔLCDRA}}_{it-i}+{\beta }_1{y}_{it-1}+{\text{B˙}}_1{\text{GGSDP}}_{it}+{\text{B˙}}_2{\text{BGSDP}}_{it}+{\beta }_3{\text{LTRVG}}_{it}+{\beta }_4{\text{LCGSG}}_{it}+{\beta }_5{\text{LPDBG}}_{it}+{\beta }_6{\text{LCDRA}}_{it}+{\vartheta }_{it}+{\epsilon }_{it}$$

(3)

As the study uses 12 types of public spending, 12 models are estimated based on Equation (2), and a further 12 models are estimated based on Equation (3). The following models are therefore estimated.

$$\text{LEDEX = F(LPGSD, LTRVG, LCGSG, LPDBG, LCDRA) }$$

(4)

$$\text{LEAREX }=\text{ F(LPGSD, LTRVG, LCGSG, LPDBG, LCDRA) }$$

(5)

$$\mathrm{L}\mathrm{E}\mathrm{N}\mathrm{E}\mathrm{X}=\mathrm{F}\left(\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D},\mathrm{ }\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G},\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}\right)$$

(6)

$$\mathrm{L}\mathrm{H}\mathrm{E}\mathrm{X}\mathrm{P}=\mathrm{F}\left(\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D},\mathrm{ }\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G},\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}\right)$$

(7)

$$\mathrm{L}\mathrm{I}\mathrm{F}\mathrm{E}\mathrm{X}=\mathrm{F}\left(\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D},\mathrm{ }\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G},\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}\right)$$

(8)

$$\mathrm{L}\mathrm{N}\mathrm{C}\mathrm{E}\mathrm{X}=\mathrm{F}\left(\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D},\mathrm{ }\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G},\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}\right)$$

(9)

$$\mathrm{L}\mathrm{R}\mathrm{D}\mathrm{E}\mathrm{X}=\mathrm{F}\left(\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D},\mathrm{ }\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G},\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}\right)$$

(10)

$$\mathrm{L}\mathrm{U}\mathrm{D}\mathrm{E}\mathrm{X}=\mathrm{F}\left(\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D},\mathrm{ }\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G},\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}\right)$$

(11)

$$\mathrm{L}\mathrm{S}\mathrm{W}\mathrm{E}\mathrm{X}=\mathrm{F}\left(\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D},\mathrm{ }\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G},\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}\right)$$

(12)

$$\mathrm{L}\mathrm{T}\mathrm{C}\mathrm{E}\mathrm{X}=\mathrm{F}\left(\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D},\mathrm{ }\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G},\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}\right)$$

(13)

$$\mathrm{L}\mathrm{W}\mathrm{S}\mathrm{E}\mathrm{X}=\mathrm{F}\left(\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D},\mathrm{ }\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G},\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}\right)$$

(14)

$$\mathrm{L}\mathrm{N}\mathrm{U}\mathrm{E}\mathrm{X}=\mathrm{F}\left(\mathrm{L}\mathrm{P}\mathrm{G}\mathrm{S}\mathrm{D},\mathrm{ }\mathrm{L}\mathrm{T}\mathrm{R}\mathrm{V}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{G}\mathrm{S}\mathrm{G},\mathrm{L}\mathrm{P}\mathrm{D}\mathrm{B}\mathrm{G},\mathrm{L}\mathrm{C}\mathrm{D}\mathrm{R}\mathrm{A}\right)$$

(15)

All models from Equations (5) to (15) are estimated using panel ARDL methods. For example, for public expenditure on education, the model can be specified as follows.

$${\text{ΔLEDEX}}_{it}=C_{it}+{\alpha }_i\sum _{i=1}^{\rho }{\text{ΔLEDEX}}_{it-i}+{\pi }_i\sum _{i=0}^{\rho }{\text{ΔLPGSD}}_{it-i}+{\varnothing }_i\sum _{i=0}^{\rho }{\text{ΔLTRVG}}_{it-i}+{\omega }_i\sum _{i=0}^{\rho }{\text{ΔLCGSG}}_{it-i}+{\delta }_i\sum _{i=0}^{\rho }{\text{ΔLPDBG}}_{it-i}+{\mu }_i\sum _{i=0}^{\rho }{\text{ΔLCDRA}}_{it-i}+{\beta }_1{\text{LEDEX}}_{it-1}+{\beta }_2{\text{LPGSD}}_{it}+{\beta }_3{\text{LTRVG}}_{it}+{\beta }_4{\text{LCGSG}}_{it}+{\beta }_5{\text{LPDBG}}_{it}+{\beta }_6{\text{LCDRA}}_{it}+{\vartheta }_{it}+{\epsilon }_{it}$$

(16)

For the extended model, it can be specified as follows.

$${\text{ΔLEDEX}}_{it}=C_{it}+{\alpha }_i\sum _{i=1}^{\rho }{\text{ΔLEDEX}}_{it-i}+{\text{Ω}}_i\sum _{i=1}^{\rho }{\text{ΔGGSDP}}_{it-i}+{\gamma }_i\sum _{i=1}^{\rho }{\text{ΔBGSDP}}_{it-i}+{\varnothing }_i\sum _{i=0}^{\rho }{\text{ΔLTRVG}}_{it-i}+{\omega }_i\sum _{i=0}^{\rho }{\text{ΔLCGSG}}_{it-i}+{\delta }_i\sum _{i=0}^{\rho }{\text{ΔLPDBG}}_{it-i}+{\mu }_i\sum _{i=0}^{\rho }{\text{ΔLCDRA}}_{it-i}+{\beta }_1{\text{LEDEX}}_{it-1}+{\text{B˙}}_1{\text{GGSDP}}_{it}+{\text{B˙}}_2{\text{BGSDP}}_{it}+{\beta }_3{\text{LTRVG}}_{it}+{\beta }_4{\text{LCGSG}}_{it}+{\beta }_5{\text{LPDBG}}_{it}+{\beta }_6{\text{LCDRA}}_{it}+{\vartheta }_{it}+{\epsilon }_{it}$$

(17)

Where ${\alpha }_i,$ ${\pi }_i,{\varnothing }_i, {\omega }_i,\mathrm{ }{\delta }_i,\mathrm{ }{\mu }_i,{\Omega }_i \mathrm{a}\mathrm{n}\mathrm{d} {\gamma }_i$ are the short-run coefficients, while ${\beta }_1$ to ${\beta }_6$ indicate the long-run coefficients. All other variables are explained before. The estimated results and discussion are described in the next section.

IV. Results and Discussion

Descriptive Statistics Results

It is essential to check the descriptive statistics before analysing any raw data series and to provide an overview of the distributional characteristics, central tendencies, and variabilities of the selected variables (Table 2). The variables exhibit different means and medians, indicating variations in their average values and distributions; that is, the mean varies from −0.91 (LNUEX) to 10.93 (LPGSD). Public expenditure on energy has a high standard deviation of 1.77, indicating more dispersed than other variables. Most of the variables are negatively skewed, suggesting a longer left tail. Overall, these variables show variations in their means, medians, minimum and maximum values, indicating differences in their average levels and ranges. Standard deviations vary across the variables, suggesting differences in the dispersion of data points around their respective means. Skewness values indicate the skewness of the distributions, with some variables having longer tails on the left or right side. Kurtosis values differ, reflecting variations in the peakedness or flatness of the distributions.

Table 2.

Descriptive Statistics.

Variables	Mean	Median	Maximum	Minimum	Std Dev	Skewness	Kurtosis	Obs.
LEDEX	2.71	2.73	3.21	−1.58	0.27	−8.46	134.47	480
LAREX	1.52	1.55	2.72	−2.86	0.50	−1.48	14.18	476
LENEX	0.99	1.56	3.34	−9.26	1.77	−1.93	8.00	476
LHEXP	1.49	1.51	2.11	−1.16	0.25	−2.54	27.02	480
LIFEX	1.55	1.64	3.05	−0.43	0.66	−0.41	2.74	480
LNCEX	−0.46	−0.37	2.28	−6.26	1.03	−1.36	7.81	477
LRDEX	1.15	1.25	2.56	−1.94	0.76	−0.89	3.72	480
LUDEX	0.11	0.12	2.11	−6.04	0.97	−0.81	5.99	479
LSWEX	1.50	1.58	3.21	−1.49	0.52	−0.67	5.03	480
LTCEX	1.41	1.40	2.48	0.11	0.43	−0.19	2.77	480
LWSEX	0.66	0.61	2.15	−6.96	0.70	−2.47	31.71	480
LNUEX	−0.91	−0.49	2.09	−7.51	1.68	−1.84	7.08	429
LPGSD	10.93	10.89	12.80	9.07	0.72	0.07	2.62	480
LTRVG	2.58	2.50	3.39	2.04	0.29	0.56	2.25	480
LCGSG	1.45	1.36	3.15	0.07	0.65	0.27	2.16	480
LPDBG	3.35	3.37	4.19	2.55	0.35	−0.07	2.33	480
LCDRA	4.02	4.04	4.91	2.90	0.40	−0.27	2.61	480

Results of Panel Unit Root Tests

Although testing of unit root is not required a priori for estimating panel ARDL models, it is essential to confirm that none of the series is I(2). Therefore, the panel unit root tests were conducted to examine the stationarity properties of the variables used in the model. The tests employed were Levin, Lin and Chu (LLC, 2002), Im, Pesaran and Shin (IPS, 2003), augmented Dickey–Fuller (ADF, 1979), and Phillips–Perron (PP, 1988). The results, as presented in Table 3, show that LEDEX, LAREX, LNCEX, LRDEX, LWSEX and LNUEX are stationary at the level, that is, I(0), while LIFEX, LSWEX, LTCEX, LPGSD and LTRVG are stationary at the first difference, that is, I(1), as satisfied by all the selected unit root tests. LENEX and LHEXP are stationary at the level using IPS, ADF and PP, but not for LLC; it is therefore differenced once to become stationary at I(1). Similarly, LUDEX is stationary at the first difference I(1) using the IPS, ADF and PP tests, but for LLC, it is stationary at the level I(0). LCGSG shows stationary at the first difference I(1) in the case of the LLC, IPS and ADF, but for the PP test, it is level stationary. Although LPDBG is stationary at the level using the LLC, IPS and ADF, but not for PP, it is therefore differenced once to become stationary at I(1). For LCDRA, both the LLC and ADF tests satisfied the stationary condition at level I(0), while the IPS and PP results are stationary at the first difference. Overall, the unit root test results indicate that the selected variables follow a combination of stationary processes, either at level (I(0)) or after differencing once (I(1)). Importantly, none of the variables exhibit second-order integration (I(2)), which satisfies the condition for utilising the Panel ARDL model in further empirical analysis.

Table 3.

Results of the Unit Root Test.

Variables	LLC	IPS	ADF–Fisher	PP–Fisher
LEDEX	I(0)	I(0)	I(0)	I(0)
LAREX	I(0)	I(0)	I(0)	I(0)
LENEX	I(1)	I(0)	I(0)	I(0)
LHEXP	I(1)	I(0)	I(0)	I(0)
LIFEX	I(1)	I(1)	I(1)	I(1)
LNCEX	I(0)	I(0)	I(0)	I(0)
LRDEX	I(0)	I(0)	I(0)	I(0)
LUDEX	I(0)	I(1)	I(1)	I(1)
LSWEX	I(1)	I(1)	I(1)	I(1)
LTCEX	I(1)	I(1)	I(1)	I(1)
LWSEX	I(0)	I(0)	I(0)	I(0)
LNUEX	I(0)	I(0)	I(0)	I(0)
LPGSD	I(1)	I(1)	I(1)	I(1)
LTRVG	I(1)	I(1)	I(1)	I(1)
LCGSG	I(1)	I(1)	I(1)	I(0)
LPDBG	I(0)	I(0)	I(0)	I(1)
LCDRA	I(0)	I(1)	I(0)	I(1)

Notes: I(0) implies stationary at the level, and I(1) represents a stationary process at the first difference. ADF: Augmented Dickey–Fuller; IPS: Im, Pesaran and Shin; LLC: Levin, Lin and Chu; PP: Phillips–Perron.

Results of Panel ARDL

This section sheds light on the factors influencing the allocation of public funds by subnational governments and highlights the importance of central government transfers, economic growth, revenue generation, financial development, and public debt burden in determining public spending priorities (Table 4). Both Akaike information criterion (AIC) and Schwarz information criterion (SIC) are applied to determine the optimal lag length of the selected models. ⁶ The analysis includes both the long-run and short-run effects. Let us discuss the results of the long run, followed by the short run.

The estimated results (Model 1) demonstrate that grants and shared taxes from the central government and financial development have a positive and highly statistically significant influence on public education spending. Conversely, higher economic growth and increased revenue exhibit a negative impact on public spending in the education sector. Furthermore, an escalation in public debt also has an adverse effect on public spending on education. Consequently, when the state government receives greater grants and shared taxes from the central government, they are more inclined to allocate a higher proportion of its budget to education. This trend is similarly observed in regions with greater financial development.

In Model 2, an increase in grants from the central government and financial development led to a rise in public spending on agriculture. Conversely, a higher debt burden and increased revenue tend to reduce public spending on agriculture. However, there is no statistically significant relationship found between economic growth and public spending in the agricultural sector. However, the findings of Model 3 confirm that higher economic growth has a positive and significant impact on public spending on energy. Additionally, both high revenue and financial development exert a favourable influence on public spending in the energy sector. Surprisingly, higher grants from the central government have an adverse impact on public spending on energy.

Upon analysing Model 4, it is observed that high economic growth has a more pronounced effect on public spending on health, indicating that regions with higher growth tend to allocate more resources to the healthcare sector. Conversely, higher public debt exhibits a negative impact on public spending on health. When examining public spending on irrigation and flood control (Model 5), it becomes apparent that only grants from the central government have a positive impact on this sector. The remaining variables exhibit adverse effects. Furthermore, the analysis indicates that public spending on natural calamities increases only when higher grants and tax shares are received from the government (Model 6). High economic growth, revenue and public debt do not significantly influence public spending on natural calamities.

The empirical findings concerning public spending on rural development (Model 7) reveal that, on average, states tend to reduce their public expenditure on rural development in response to higher economic growth, increased revenue collection, and a greater burden of public debt. However, if these states receive more grants from the central government and experience improved financial development, they are more likely to allocate additional resources to rural development. Similarly, for public spending on urban development (Model 8), it is discovered that states tend to increase their spending on urban development when they experience higher economic growth. Furthermore, greater financial development and increased revenue receipts from the central government also contribute to increased spending on urban development. Consequently, higher economic growth leads to a greater emphasis on urban development compared to rural development.

The findings from Model 9 indicate that higher economic growth has a positive and significant influence on public expenditure dedicated to social security and welfare. Similarly, total revenue and financial development also exhibit positive and significant impacts on public spending in this sector. Conversely, a higher burden of public debt results in a decrease in public spending on social security and welfare. Notably, grants from the central government are not utilised for public expenditure in this area. Hence, higher economic growth, increased tax and non-tax revenue, and enhanced financial development are the primary drivers of public spending on social security and welfare. Regarding the long-term outcomes of the transport and communication model (Model 10), it is observed that high economic growth and increased revenue lead to a higher allocation of public funds for transport and communication. Conversely, a rise in the burden of public debt negatively affects public spending in this sector. Thus, the analysis suggests that states prefer to allocate more resources to the development of transport, communication and technology through avenues other than incurring public debt or relying on grants and tax shares. Instead, they prioritise utilising their higher economic growth and revenue collection for this purpose.

In Model 11, the findings suggest that state governments are inclined to increase their expenditure on water supply and sanitation when they experience higher economic growth and achieve high revenue collection. Conversely, higher public debt and increased revenue receipts from the central government discourage spending in this sector. Moving on to Model 12, it is confirmed that higher revenue receipts from the central government have a positive and highly significant impact on public spending on nutrition. However, higher economic growth and increased revenue exhibit an adverse impact on public spending in the nutrition sector.

Analysing the short-run results (Table 4), it is observed that all error correction term (ECT) values are negative, as expected, and highly significant, confirming the existence of a cointegration relationship among the selected variables. The short-run analysis reveals that the coefficients of grants from the central government have a positive and highly significant influence on public spending in several areas, including health, natural calamities, rural development, social security and welfare, and water supply and sanitation. Furthermore, these coefficients also exhibit a positive lagged impact on public spending on transport and communication. Additionally, the analysis finds that financial development has a favourable impact on public spending in the domains of urban development, nutrition, transport and communication. However, it is important to note that high economic growth has a negative and significant impact on public spending in education and health, while higher revenue only adversely affects public spending on health. Therefore, we have re-estimated the selected models by incorporating a dummy variable for the adoption of fiscal responsibility legislation (FRL) by state governments, as this policy reform may have significant implications for the sectoral allocation of public expenditure across states. The results of this extended analysis are presented in Appendix Table A1. The findings remain broadly consistent with our original estimates, thereby reinforcing the robustness of the results. The FRL dummy shows a statistically significant impact on expenditures in several sectors, including energy, health, irrigation and flood control, natural calamities, urban development, social welfare, and transport and communication.

Table 4.

Results of Panel Autoregressive Distributed Lag (ARDL) Models.

Variables

LEDEX

LAREX

LENEX

LHEXP

LIFEX

LNCEX

LRDEX

LUDEX

LSWEX

LTCEX

LWSEX

LNUEX

Long Run

LPGSD

−0.251***

−0.050

0.337***

0.215***

−0.646***

−0.079

−0.138**

0.817***

0.233***

0.100*

0.141*

−0.545***

LTRVG

−0.377***

−0.407***

2.714***

0.073

−0.447***

−0.318

−0.387**

0.042

0.521***

0.655***

0.743***

−1.667***

LCGSG

0.318***

0.393***

−1.289***

−0.168*

0.355**

0.550**

0.676***

0.683***

−0.604***

−0.422***

−0.268**

0.775***

LPDBG

−0.389***

−0.860***

−0.145

−0.725***

−0.483***

0.072

−0.353***

−0.181

−0.685***

−0.280***

−0.431***

−0.198

LCDRA

0.562***

0.226**

0.778***

0.025

−0.483***

−0.401**

0.406**

0.432**

0.988***

−0.209

−0.223

1.349***

Short Run

ECM(t-1)

−0.409***

−0.566***

−0.583***

−0.349***

−0.395***

−0.962***

−0.433***

−0.479***

−0.30***7

−0.304***

−0.288***

−0.322***

D(LPGSD)

−0.429*

−0.749

4.675

−0.359**

−0.120

−0.801

−0.748

−1.281

−0.112

0.122

−1.385

−0.930

D(LPGSD(−1))

0.205

−0.456

D(LTRVG)

−0.140

0.011

2.305

−0.417***

0.374

−0.852

−0.328

−0.760

−0.244

0.470

−0.498

0.197

D(LTRVG(−1))

−0.140

0.062

D(LCGSG)

−0.031

−0.205

−0.253

0.300***

−0.003

0.691**

0.392***

0.418

0.307***

−0.045

0.423*

0.350

D(LCGSG(−1))

0.115

0.324**

D(LPDBG)

0.126

0.222*

0.133

0.047

−0.276*

−1.172

0.214

−0.525

−0.092

−0.336

−0.075

−1.088**

D(LPDBG(−1))

0.174

−0.424**

D(LCDRA)

−0.072

−0.215

−0.327

−0.124

0.121

0.601

−0.279

0.460**

−0.079

−0.243

0.264

1.386**

D(LCDRA(−1))

0.180

0.721**

Constant

2.076***

2.582***

−6.331***

0.549***

4.944***

1.720***

0.985***

−5.270***

−0.868***

0.300***

1.252***

Number of states

15

15

15

15

15

15

15

15

15

15

15

15

Observation (included)

450

457

459

465

465

459

465

464

465

450

465

414

Note: ***, ** and * represent the significance level at 1%, 5% and 10%.

As mentioned earlier, to assess the impact of the business cycle, we have incorporated good-year growth and bad-year growth variables into all respective models, alongside other independent variables. This means that we have re-estimated the Models 4–15, considering both good-year growth and bad-year growth. These findings demonstrate how the business cycle, as captured by good-year growth and bad-year growth variables, influences state governments’ decisions regarding public spending in different sectors (Table 5).

The long-run results reveal that in good years, state governments tend to reduce public spending on education to a lesser extent compared to bad years. Specifically, they are more inclined to decrease public spending on agriculture during bad years in comparison to good years. However, there is relatively little deviation in public spending on energy and health between good and bad years, indicating a similar level of expenditure in both cases. In good years, state governments prefer to reduce public spending on irrigation and flood control more than in bad years. The same trend is observed for public spending on natural calamities and rural development. On the other hand, state governments allocate more funds to urban development during good years compared to bad years, while they allocate more resources to social security and welfare, as well as transport and communication, during bad years in comparison to good years. Concerning nutrition spending, state governments tend to reduce public expenditure on nutrition to a greater extent during good years compared to bad years. The impact of other selected independent variables aligns with the descriptions provided earlier, demonstrating the robustness of the findings.

Table 5.

Results of Panel Autoregressive Distributed Lag (ARDL) Models (Including GOOD and BAD Year GSDP).

Variables

LEDEX

LAREX

LENEX

LHEXP

LIFEX

LNCEX

LRDEX

LUDEX

LSWEX

LTCEX

LWSEX

LNUEX

Long-run Equation

GGSDP

−0.144***

−0.062*

0.266**

0.171***

−0.490***

−0.141**

−0.114**

0.732***

0.215***

0.145**

−0.123**

−0.543***

BGSDP

−0.150***

−0.067**

0.268***

0.170***

−0.486***

−0.134**

−0.103**

0.729***

0.218***

0.151**

−0.102*

−0.506***

LTRVG

−0.472***

−0.386***

2.569***

0.131

−0.410**

−0.478**

−0.260

−0.153

0.215

0.702**

−0.072

−2.539***

LCGSG

0.191***

0.356***

−1.152***

−0.213**

0.371**

0.764***

0.682***

0.721***

−0.480***

0.219

0.298**

1.302***

LPDBG

−0.288***

−0.833***

−0.253

−0.713***

−0.435***

0.087

−0.576***

−0.132

−0.722***

−0.594***

−0.355**

−0.416**

LCDRA

0.251***

0.181**

0.780***

0.069

−0.602***

−0.634***

0.319*

0.255

0.976***

−0.217

−0.501**

1.643***

Short-run Equation

ECM(t-1)

−0.455***

−0.568***

−0.576***

−0.352***

−0.388***

−0.958***

−0.439***

−0.513***

−0.315***

−0.323***

−0.419***

−0.356**

D(GOODGSDP)

−0.563

−0.494

5.074**

−0.910***

−0.107

0.483

−1.754**

−0.543

−0.438

−0.446

−1.485*

−0.181

D(BADGSDP)

−0.563

−0.492

5.075**

−0.912***

−0.108

0.483

−1.758**

−0.538

−0.439

−0.449

−1.494*

−0.181

D(LTRVG)

−0.025

0.016

2.361

−0.427***

0.385

−0.875

−0.302

−0.894*

−0.200

0.373

0.113

0.445

D(LCGSG)

−0.029

−0.172

−0.389

0.295***

−0.026

0.637**

0.357***

0.486

0.287***

−0.171

0.238

−0.112

D(LPDBG)

0.156

0.199

0.191

−0.030

−0.295*

−1.116

0.339

−0.559

−0.191

−0.432**

0.012

−0.715

D(LCDRA)

0.078

−0.257

−0.461

−0.106

0.227

0.843

−0.210

0.548**

−0.041

0.199

0.379*

1.340**

C

2.771***

2.958***

−6.623***

0.265***

5.453***

4.023***

1.548***

−7.038***

−1.038***

−0.149**

2.364***

2.686***

Number of states

15

15

15

15

15

15

15

15

15

15

15

15

Obs (included)

450

457

459

465

465

459

465

464

465

465

465

414

Note: ***, ** and * represent the significance level at 1%, 5% and 10%.

Addressing CSD: Evidence from CS Augmented ARDL (CS-ARDL) Estimation

This section addresses the possibility of CSD across Indian states. CSD arises when states are influenced by common national shocks or coordinated policy responses. Ignoring such dependence can lead to biased and inefficient estimates.

To begin with, we tested for the presence of CSD using both the Breusch–Pagan Lagrange multiplier (LM) test (Breusch & Pagan, 1980) and Pesaran’s CD test (Pesaran, 2004). The results, reported in Table 6, confirm statistically significant CSD in most models, particularly for sectors such as education, health, flood control, social welfare and transport, except for energy and natural calamities spending.

Table 6.

Cross-sectional Dependency Results.

Models	Breusch–Pagan LM	Pesaran CD
LEDEX	434.26***	14.96***
LAREX	298.21***	5.95***
LENEX	268.47***	1.07
LHEXP	672.12***	21.57***
LIFEX	324.04***	5.09***
LNCEX	109.38	1.33
LRDEX	314.36***	4.57***
LUDEX	267.77***	2.96***
LSWEX	420.64***	2.47**
LTCEX	359.75***	7.15***
LWSEX	316.70***	4.06***
LNUEX	472.78***	6.85***

Notes: **Indicates significance at 5% level. ***Indicates significance at 1% level. CD: Cross-sectional dependence; LM: Lagrange multiplier.

To address this issue, we employed the CS-ARDL estimator developed by Chudik and Pesaran (2015). This method augments the standard ARDL framework by incorporating cross-sectional averages of both the dependent and explanatory variables. It helps account for unobserved common factors and corrects for CSD. In addition, CS-ARDL allows for heterogeneity in both short-run and long-run coefficients across states, which is particularly useful in a federal context where states differ in fiscal capacities and policy responses. The model also accommodates variables with mixed orders of integration and the possibility of cointegration relationships.

The estimation results from the CS-ARDL model are presented in Table 7. Economic growth shows a significant positive long-run effect on spending in health, social welfare, transport and nutrition sectors, and a negative effect on irrigation and flood control. Total revenue is negatively associated with education and urban development expenditures, but positively linked with social welfare. Grants-in-aid have a positive and significant impact on most sectors, highlighting their supportive role in enabling state-level expenditure commitments. Public debt has a consistently negative and significant effect on several sectors, such as rural development and agriculture, underscoring the fiscal constraints imposed by debt servicing. Financial development shows weak and inconsistent effects.

Table 7.

Cross-section Augmented Autoregressive Distributed Lag (CS-ARDL) Estimation Results.

Variables	LEDEX	LAREX	LHEXP	LIFEX	LRDEX	LUDEX	LSWEX	LTCEX	LWSEX	LNUEX
Long Run
LPGSD	0.111	−0.072	0.131*	−0.332***	−0.029	0.372	0.173***	0.052	−0.987**	0.694**
LTRVG	−0.505**	−0.314	−0.368	0.224	−0.242	−1.618**	0.456**	−0.293	−1.016	−1.108**
LCGSG	0.445**	0.506**	0.207	0.130	0.844**	1.105***	−0.133	0.296	0.390	0.702**
LPDBG	−0.014	−0.318***	−0.142*	−0.192**	−0.650**	0.855***	−0.231**	−0.341*	−0.433**	−0.108
LCDRA	−0.446	−0.092	−0.134	−0.406**	−0.581**	0.985*	0.062	0.421	0.188	0.127
Constant	−0.768	2.932***	0.987	5.559***	2.553	−8.136***	−1.676**	0.072	−2.672	−0.710
Short Run
D(DEP (−1))	−1.208***	−0.854***	−0.974**	−0.747***	−1.123***	−0.885***	−0.808***	−1.048***	−0.917***	−0.901***
D(LPGSD)	−1.297**	1.208	−0.018	0.123	0.161	0.574	0.451	0.699	−2.085*
D(LPGSD(−1))	1.453	−1.298	0.300	−0.725	−0.250	0.151	−0.095	−0.601	−0.775	1.184**
D(LTRVG)	−0.841***	−0.258	−0.854**	0.285	−0.599	−2.949***	0.443	−0.219	−1.342	−1.416**
D(LTRVG(−1))	−0.226	−0.354	0.080	0.060	−0.112	−0.288	0.351	−0.386	−0.770	−0.564
D(LCGSG)	0.588***	0.347	0.533*	−0.128	1.919**	1.985***	0.083	0.387	0.941	1.043**
D(LCGSG(−1))	0.318	0.613	−0.095	0.385	−0.045	0.224	−0.346**	0.206	−0.015	0.181
D(LPDBG)	−0.446*	0.060	−0.573**	0.012	−1.047	1.047	−0.310	−0.345	−1.120
D(LPDBG(−1))	0.461	−0.674	0.259	−0.333	−0.468	0.495	−0.200	−0.328	1.087	−0.303
D(LCDRA)	−0.211	−0.008	−0.425**	0.047	−1.128	1.068	0.225	0.073	−0.336	0.494
D(LCDRA(−1))	−0.718	−0.218	0.049	−0.682	−0.214	0.901	−0.154	0.820**	0.936	−0.352
Constant	−1.658	5.304***	2.400	9.749***	5.198	−15.631**	−2.758**	0.415		−2.063
R2 (MG)	0.90	0.47	0.83	0.47	0.74	0.16	0.64	0.66	0.60	0.55
Root MSE	0.10	0.28	0.09	0.17	0.19	0.55	0.15	0.14	0.35	0.37

Note: ***, ** and * represent the significance level at 1%, 5% and 10%. Dependent variable (DEP) (−1) refers to the respective dependent variables of the selected models. MG: Mean group; MSE: Mean squared error.

Short-run dynamics are significant in several models, particularly the lagged dependent variable, indicating stable adjustment processes. The results are generally consistent with those obtained under the PMG estimator. Overall, the CS-ARDL results corroborate the key findings from the PMG estimation while offering greater robustness by explicitly accounting for CSD and unobserved heterogeneity. This strengthens the credibility of the empirical results and reinforces the policy implications drawn from the study.

Critical Discussion of Results

The study results explain the cyclical (procyclical, countercyclical and acyclical) response to economic growth on public expenditure categories across Indian states using the panel ARDL model. The study utilises panel ARDL models and incorporates control variables such as total revenue, grants from the centre, share in central tax, total public debt, and the credit-deposit ratio of all scheduled commercial banks to assess their effects on public expenditure at the state level. The uniqueness of the study lies in its analysis of subnational level public expenditure data for various categories, allowing for an assessment of government prioritisation of spending in India.

The long-term findings validate that increased economic growth prioritises expenditures in energy, health, urban development, transportation and communication, water supply and sanitation, while neglecting public spending on education, irrigation and flood control, rural development and nutrition. Likewise, when aggregate revenue rises, subnational governments tend to allocate more public funds towards energy, social security and welfare, transportation and communication, technology, water supply and sanitation over the long run. However, an increase in the central share enhances public spending on education, agriculture, irrigation and flood control, natural calamities, rural development, urban development and nutrition. Financial development also aids in prioritising public spending on education, agriculture, energy, rural development, urban development, social security and welfare, and nutrition at the subnational level in the long run. The findings also confirm that higher levels of public debt have an adverse impact on most of these selected public spending at the subnational level over the long term.

The study reveals that state government spending on agriculture is countercyclical, indicating that spending is reduced during bad years and increased during good years. On the other hand, energy and health spending exhibit acyclical behaviour, implying that government spending remains unchanged regardless of the economic conditions. Spending on irrigation and flood control, natural calamities and rural development is found to be procyclical, with the government reducing spending in good years and increasing it in bad years. In contrast, spending on social security and welfare, nutrition, transport and communication is countercyclical, meaning that state governments spend more on these categories during bad years compared to good years. The study aligns with previous research that suggests government spending on education, health, and social security does not significantly change during economic fluctuations in emerging and developing countries (Afonso & Jalles, 2013; Jalles, 2020). Few studies have argued that developed countries tend to exhibit less spending procyclicality due to better institutions and fiscal stabilisation policies (Jalles, 2021).

Other factors such as public debt, trade openness, government size and political constraints have also been identified in past studies as determinants of government spending cyclicality (Aizenman et al., 2019; Jalles, 2020). Public debt is a robust determinant of procyclicality in government spending, particularly in developed countries (Aizenman et al., 2019). Transitory shocks, such as yearly variations in rainfall, can also influence spending patterns in developing countries (Brueckner & Gradstein, 2014). They argue that government expenditure on consumption shows a persistent effect due to year-to-year variation in the real GDP, as compared to transitory shocks in developing countries. Xing and Fuest (2018) argue that government spending is procyclical during the economic reform period, while less procyclical to output fluctuations after the reform, which was evident in the decentralisation period in China. A few studies also investigate the determinants of the cyclicality of fiscal policy using the panel ARDL model, and they found that government spending is procyclical with real per capita GDP irrespective of growth acceleration or deceleration periods (Mawejje & Odhiambo, 2022).

Earlier studies have also explored the cyclicality of spending on health and social security. Liang and Tussing (2019) find procyclicality of health expenditure in developing countries in a sample of 135 developing countries, which implies that a 1% rise in GDP is positively correlated with 0.61% deviations from government health expenditure. They argue that less procyclicality of health spending leads to shorter life expectancies and higher adult mortality rates. Similarly, Abbott and Jones (2021) argue that the government’s political agenda influences the cyclicality of government spending on public health services. They argue that the vote-maximising government increases health care expenditure during good times, while in bad years, the government provides incentives to sustain expenditure on health care by deprioritising other expenditure that has less priority. On the contrary, some studies argue that social spending serves as a countercyclical tool for stimulating recovery from economic crises in advanced countries (Ayala-Cañón et al., 2022). This is obvious and seen in many developed economies during the 2008 global financial crisis, spending on the social sector was reduced through the process of fiscal consolidation, and this credit crunch badly affected the health sector (Behera & Dash, 2019a, 2019b).

This study has also examined the cyclicality of rural public expenditure, government expenditure on natural calamities (environmental), and spending on infrastructure, which is unique, and the prioritisation of these types of expenditure would positively induce economic growth. It finds that rural public expenditure and government environmental expenditure exhibit procyclical behaviour, with increased spending during good years. Past studies argue that rural public expenditure is procyclical in nature in emerging economies like China (Luo et al., 2020). They argue that the procyclicality of rural public expenditure is related to the degree of development, which means that lower economic development leads to the procyclicality of rural expenditure. Similarly, Abbott and Jones (2022) examined the cyclicality of government environmental expenditure in OECD countries and found that the government increases environmental expenditure during good years and reduces it in the bad years because, during a recession, the government diverts spending from environmental protection to other public services programmes to retain voters’ sentiments. Similarly, a few studies argue that spending on infrastructure, particularly transport and communication, education and health infrastructure, is believed to have a positive effect on economic growth, while spending on agriculture and natural resources infrastructure may have an adverse effect (Babatunde, 2018).

V. Conclusion and Policy Implications

This study provides a unique and comprehensive assessment of spending prioritisation across various social and economic sectors at the state level in India, the largest emerging economy. Over 31 years, we have examined how different categories of public spending respond to macro-fiscal variables during different phases of the business cycle, that is, both good and bad. The findings reveal a mixed response to the impact of macro-fiscal variables, including tax revenue, central government grants, public debt and financial development, on different components of expenditure in the short run and long run.

In the long run, spending prioritisation is observed towards sectors such as energy, health, urban development, transport and communication, and water supply and sanitation. However, public spending on education, irrigation and flood control, rural development and nutrition tends to be neglected. These long-term spending priorities are largely influenced by subnational revenue growth, indicating that central government funding is less necessary to finance these categories in the short run. Additionally, financial development plays a role in prioritising spending on education, agriculture, energy, rural development, urban development, social security and welfare, and nutrition at the subnational level in the long run. On the other hand, higher levels of public debt have an adverse impact on all types of public spending at the subnational level in the long run.

The study underscores that subnational governments tend to prioritise public spending on agricultural development, reducing regional disparities, education, natural calamities and nutrition when they receive more central shares from the central government. Furthermore, higher economic growth and revenue are associated with increased emphasis on public spending in energy, health and hygiene, and social security sectors. The findings of this study have important implications for policymakers and practitioners involved in fiscal management and resource allocation in India. Understanding the cyclical patterns and determinants of public spending priorities can inform the formulation of effective fiscal policies, aid in the identification of areas that require increased investment, and promote sustainable development at both the state and national levels.

However, it is important to acknowledge the limitations of this study. Institutional factors such as political constraints and good governance, as well as open macroeconomic factors like trade openness, globalisation and foreign direct investment, were not included in the empirical estimation, which could be seen as limitations compared to previous studies (Aizenman et al., 2019; Jalles, 2021). Additionally, data limitations at the state level in India pose challenges in incorporating other important factors such as the availability of minerals, mines and forest resources, which may influence the spending prioritisation decisions of regional governments.