Finance Research Letters 59 (2024) 104753

Contents lists available at ScienceDirect

Finance Research Letters

journal homepage: www.elsevier.com/locate/frl

Social protection spending and financial crises

Thanh Cong Nguyen a, *, Vítor Castro b, Justine Wood b
a
Faculty of Economics and Business, Phenikaa University, Hanoi 12116, Vietnam
b
School of Business and Economics, Loughborough University, Epinal Way, Loughborough LE11 3TU, United Kingdom

A R T I C L E I N F O A B S T R A C T

JEL classifications: This paper assesses the impact of distinct types of financial crises on social protection spending
C23 using a panel of 105 countries over the period 1991–2019. The findings show that spending on
D72 social protection increases when financial crises strike, mainly in the aftermath of banking crises.
G01
However, currency and debt crises are detrimental to spending on social protection, threatening
H55
social wellbeing.
Keywords:
Social protection spending
Financial crises
Types of financial crises

1. Introduction

Financial crises have an immediate impact on economic activity. The well-known Keynesian economic theory asserts that free-
market economies have no self-balancing mechanisms to achieve full employment, especially during economic recessions when
aggregate demand is low due to rising unemployment and decline in consumption. Keynesian economists suggest that governments
should step in to foster growth and full employment by creating demand. Keynesians believe that aggregate demand, which is the sum
of spending by households, businesses, and the government, is the primary driving force in an economy. During periods of financial
crises when spending from households and businesses decline, increasing government spending is the main plank of the Keynesian
theory.
But while fiscal expansion policies may help the economy recover relatively quickly, crises have been found to leave deep and long-
lasting effects on people’s lives (Stiglitz, 2013; Nguyen et al., 2021). The social consequences of financial crises may lead to long-term
unemployment, losses in human capital, social exclusion, a poverty trap and even health problems (Fallon and Lucas, 2002; Nikoloski,
2011; Mohseni-Cheraghlou, 2016; Kiendrebeogo et al., 2017; Rewilak, 2017). This motivates us to examine how spending on social
protections responds to financial crises.
Social protection spending can play an important role in preventing the potentially irreversible negative long-run impact of
financial crises on society (Stiglitz, 2013). Social protection provides financial assistance to the sick, disabled, elderly, and families

* Corresponding author at: Faculty of Economics and Business, Phenikaa University, Hanoi 12116, Vietnam.
E-mail address: [email protected] (T.C. Nguyen).

https://doi.org/10.1016/j.frl.2023.104753
Received 11 June 2023; Received in revised form 29 October 2023; Accepted 15 November 2023
Available online 16 November 2023
1544-6123/© 2023 Elsevier Inc. All rights reserved.
T.C. Nguyen et al. Finance Research Letters 59 (2024) 104753

with children as well as housing, social exclusions, and unemployment subsidies. For this reason, increasing spending on social
protection is a measure that alleviates the long-run consequences of financial crises. In practice, this measure is often constrained by
conflicting objectives (Nguyen et al., 2021). On one side, they are bounded by budget constraints to keep public deficits and debt at
sustainable levels; on the other, they have a social commitment to protect their citizens, especially low-income households who are
more vulnerable to financial crises. Nevertheless, reducing social protection in times of crisis for the sake of financial solvency can have
detrimental welfare effects. External aid tied to tight fiscal reforms (e.g., imposed by the IMF) may exacerbate this dynamic.
While prior literature documents significant effects of financial crises on production (Kroszner et al., 2007), trade (Paravisini et al.
2015), human development (Nguyen et al., 2021), and exporter dynamics (Gong et al., 2023), the literature on the impact of financial
crises on social protection spending is scarce (Stiglitz, 2013; Mohseni-Cheraghlou, 2016; Nguyen et al., 2021). This paper aims to fill
this gap in the literature by looking at the effects of different types of financial crises on social protection spending. Grasping the
heterogeneous effects of the different types of financial crises (banking, currency, and debt crises) is critical to shaping appropriate
policy measures to mitigate the long-lasting consequences of financial crises.
The rest of this paper is organised as follows. Section 2 presents the data and empirical methodology. The main findings are
discussed in Section 3. Section 4 concludes.

2. Data and methodology

To assess the impact of financial crises on social protection, we use annual data for a panel of 105 countries over the period
1991–2019. Data for social protection spending as a percentage of GDP are obtained from the International Monetary Fund Gov­
ernment Finance Statistics (IMF-GFS). The following variables are used for financial crises: FinCrises, Banking, Currency, Debt, and
Twin/Triple crises. These dummy variables take the value of 1 for the respective episodes of crises, and 0 otherwise.1
These types of financial crises can have heterogeneous effects on social protection spending. Hence, while we conjecture that
banking crises may have a positive impact on social spending as governments expand their welfare programmes to protect their
citizens against economic shocks, debt crises could be associated with lower social protection due to fiscal pressures. Ambiguities
remain on the effects of currency crises as these crises could hurt the poor and put pressure on debt service, but also benefit exporters
(Nguyen et al., 2021).
According to the standard literature (see Castro and Martins, 2018; Nguyen et al., 2021; Uch et al., 2021; and references therein),
the following set of variables is used to account for economic and political factors: Unemployment rate (Unemployment); Logarithm of
real GDP per capita (LnGDPpc); Trade openness, i.e. the sum of imports and exports to GDP (Openness); Government debt as a per­
centage of GDP (GovDebt); Fiscal balance as a percentage of GDP (GovDeficit); Median age of the population (MedAge); Level of de­
mocracy in a country, which is the POLITY-IV index that ranges from -10 to 10 with higher values indicating higher levels of
democracy (Democracy); Dummy variable that is equal to 1 when the ruling government is left-wing (Left); Dummy variable that is
equal to 1 when the ruling government has a majority in the parliament (Majority); Dummy variable that is equal to 1 in election years
(Election); and a dummy variable that is equal to 1 for developed countries (Developed).
The following dynamic panel data specification is considered to assess the impact of financial crises on social protection spending:
SocialProtectionit = ρSocialProtectionit− 1 + βFinCrisesit− 1 +

+ γEconomicit− 1 + δPoliticalit + αi + τt + εit ,

where i = 1, …, 108 and t = 1991, …, 2019.2 The coefficient on the lag of the dependent variable (ρ) measures its persistence. The
coefficient β captures the effect of financial crises (Banking, Currency, Debt, Twin/Triple), while the vectors γ and δ assess the effect of
economic and political factors described above; αi represents unobserved country-specific effects, while τt captures time-effects like
global shocks, institutional and socio-economic changes or technological progress not accounted for in our set of variables; εit is the
error term, which is assumed to be uncorrelated and have constant expected value and variance, i.e. E(εit) = 0 and V(εit ) = σ2ε .
Nevertheless, the lagged dependent variable is expected to be correlated with the disturbance (Greene, 2020, p. 564).

3. Empirical analysis

To account for the presence of individual effects, we start by using a standard fixed-effects estimator. Due to the presence of the
lagged dependent variable on the right-hand side of the equation and the fact of the time period considered might not be long enough
to mitigate its potential bias, we also rely on the generalised method of moments (GMM) estimator suggested by Arellano and Bover

1
For example, FinCrises dummy variable takes the values of 1 for crisis episodes of any type (banking, currency or debt), and 0 otherwise. These
data are obtained from Nguyen et al. (2022). Twin/Triple crises occur when a crisis coincides or is preceded by another/other types of financial crises
within one year. For further details on variables used in this study, see Table A1 in the Appendix.
2
Economic variables are lagged to account for delays in reporting data and to avoid simultaneity problems.

2
T.C. Nguyen et al. Finance Research Letters 59 (2024) 104753

(1995) and Blundell and Bond (1998). This system-GMM estimator3 not only accounts for the correlation between the lagged
dependent variable and the disturbance, but also allows for higher accuracy/efficiency and reduced finite sample bias (Roodman,
2009).4 The main results of our analysis are reported in Table 1.5
The results are quite similar between the two estimators, and the tests support the GMM specification and the validity of the in­
struments used. They show that governments react to financial crises (FinCrises) by boosting social protection spending.6 Hence,
governments are sensitive to the social effects of financial crises and, in general, accommodate them in their social policy measures.
This is only part of the story, because when we dig deeper into the analysis of the impact of different types of financial crises, we obtain
some striking findings.
Our results show that the positive effect identified above is mainly driven by banking crises. Governments will tend to increase
social spending mainly to neutralize the negative social consequences of banking crises. However, currency and debt crises prompt a
different reaction.7 Depreciation of the domestic currency will increase the prices of imported goods and put pressure on the service of
public debt in foreign currency, which may end up leading to cuts in social protection spending. Moreover, debt crises imply spending
cuts to bring the public deficit and debt under control, also impacting social protection spending negatively. This suggests that welfare
state retrenchment occurs mainly during times of currency and debt crisis. Hence, currency and debt crises undermine social wellbeing
and welfare the most, with nefarious consequences for future generations. This effect can be exacerbated when countries receive
financial aid from external organisations, as they usually require the implementation of drastic welfare state reforms as a condition for
their financial support (Nguyen et al., 2021).
Twin and triple crises seemingly share the negative effect of currency and debt crises; these crises may follow banking crises thus
reverting government policies in such a way that the net result ends up being negative for social protection. For example, the social
measures implemented by Greece and Portugal in the aftermath of the Great Recession had to be retracted when they were hit by the
subsequent debt crises.
Regarding the control variables, the results are in line with our expectations. We observe that social protection spending is on
average higher in developing countries and tends to rise with the unemployment rate, GDP per capita, age of the population, level of
democracy, left-wing governments and during elections. On the contrary, government debt and an increase in fiscal balance are
accompanied by cuts in social protection spending.
To further mitigate potential endogeneity problems, a two-stage least squares (2SLS) approach is employed to examine the effects
of financial crises on social protection spending. We use Crisis wave as instrument for financial crises as financial crises tend to come in
waves (see Laeven and Valencia, 2020; Nguyen et al., 2022). The results are reported in Table A2 in the Appendix. The
under-identification (UIT), weak identification (WIT) and over-identification (OIT) tests confirm the validity of the instrument variable
used. We consistently find that banking crises and financial crises in general lead to higher social protection spending, while currency,
debt, and twin and triple crises appear to reduce social protection spending. Hence, our main findings are robust to potential endo­
geneity issues.
Finally, considering that the effects of financial crises on social protection might be permanent, we use five-year time spans to
further explore their long-term impact. The results are reported in Table 2 and support our main findings that increases in social
protection spending during financial crises occur mainly when banking crises strike, while currency and debt crises end up having a
negative long-run impact on social protection.

4. Conclusions

This study provides new insights into financial crises’ effects on social protection spending. We show that spending increases to
offset the effects of banking crises but reduces when countries are dealing with currency and debt crises. The latter can then have
nefarious repercussions on that layer of the population that rely heavily on social protection.
A long-term welfare development programme may be needed to ensure that social protection is not negatively affected by any type

3
A two-step approach is used in the estimation of this system-GMM estimator. For this estimator to be valid, it requires the stationarity of the
variables and the lack of correlation between the first differences of the instruments and the specific effects (Roodman, 2009). ADF Fisher-type panel
unit root tests rejected the hypothesis that all panels contain unit roots. Those results are not reported here to save space, but they are available upon
request.
4
To avoid over-fitting biases, instruments are collapsed as suggested by Roodman (2009). Moreover, to avoid the proliferation of instruments, we
treat only the lag of the dependent variable and financial crises as endogenous. These variables are instrumented with their lags in the
first-difference and level equations, while the exogenous variables are instrumented with their own values. In additional experiments not reported
here but available upon request, we also treated LnGDPpc as endogenous to mitigate the problem of reverse causality further, but the results
remained qualitatively unchanged.
5
Cross-sectional dependence tests were employed to detect the presence of cross-sectional dependence. Nevertheless, they showed that our
findings are not driven by cross-sectional dependence issues. Moreover, unreported misspecification tests confirm that our main findings do not
suffer from structural breaks, non-linearity, or normality problems. The results of these tests are available upon request.
6
In particular, we observe that social protection spending is on average around 0.39 percentage points higher during financial crises, ceteris
paribus (see column 6). This is the short-run effect, but in the long run, the impact is even higher: 1.07 (=0.39/(1-0.634)). Hence, financial crises
lead to about a 1.1 percentage points increase in social protection spending (as percentage of GDP) in the long run.
7
Our findings remain consistent when we include banking, currency, and debt crisis dummies into a single regression. The results are not reported
here to keep the analysis parsimonious, but they are available upon request.

3
 Table 1
Social protection spending and financial crises 1991–2019.

T.C. Nguyen et al.

FE estimator GMM estimator
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

L.SocialProtection 0.705*** 0.676*** 0.671*** 0.671*** 0.674*** 0.634*** 0.552*** 0.592*** 0.559*** 0.563***
(0.0674) (0.0618) (0.0617) (0.0624) (0.0610) (0.0738) (0.0922) (0.0787) (0.0870) (0.0743)
FinCrises 0.251** 0.388**
(0.112) (0.183)
Banking 0.266*** 0.306**
(0.0947) (0.151)
Currency -0.267** -0.460*
(0.128) (0.236)
Debt -0.413*** -0.632**
(0.145) (0.318)
Twin/Triple -0.482*** -0.691**
(0.171) (0.311)
Unemployment 0.0582*** 0.0523*** 0.0566*** 0.0584*** 0.0583*** 0.0362** 0.0380** 0.0341** 0.0352** 0.0360**
(0.0205) (0.0192) (0.0194) (0.0197) (0.0194) (0.0151) (0.0178) (0.0168) (0.0141) (0.0157)
Openness -0.00573** -0.00516* -0.00451* -0.00467* -0.00478* -0.00568 -0.00689 -0.00607 -0.00751 -0.00696
(0.00282) (0.00281) (0.00268) (0.00269) (0.00272) (0.00484) (0.00668) (0.00594) (0.00523) (0.00638)
LnGDPpc 0.516 0.486 0.501 0.547 0.509 0.315** 0.368** 0.305* 0.357** 0.316*
(0.389) (0.479) (0.485) (0.493) (0.479) (0.134) (0.171) (0.171) (0.177) (0.187)
Gov_Debt -0.00152 -0.00376* -0.00426* -0.00364* -0.00418* -0.00337 -0.00165 -0.00293 -0.00392 -0.00301
(0.00186) (0.00213) (0.00224) (0.00218) (0.00219) (0.00335) (0.00354) (0.00391) (0.00407) (0.00419)
Gov_Deficit -0.0636*** -0.0548*** -0.0630*** -0.0601*** -0.0609*** -0.0644*** -0.0732*** -0.0750*** -0.0730*** -0.0756***
(0.0173) (0.0158) (0.0156) (0.0156) (0.0154) (0.0228) (0.0269) (0.0257) (0.0247) (0.0266)
Med_Age 0.149** 0.187*** 0.192*** 0.191*** 0.193*** 0.195*** 0.256*** 0.220*** 0.237*** 0.236***
4

(0.0600) (0.0581) (0.0583) (0.0579) (0.0583) (0.0501) (0.0666) (0.0608) (0.0582) (0.0606)
Democracy 0.0149 0.0107 0.0177 0.0140 0.0169 0.0845*** 0.0811** 0.0864** 0.0871** 0.0815**
(0.0121) (0.0130) (0.0127) (0.0127) (0.0125) (0.0310) (0.0380) (0.0426) (0.0371) (0.0385)
Left 0.0769 0.0769 0.0734 0.0129 0.0715 0.267* 0.345** 0.328** 0.329** 0.325*
(0.0858) (0.0905) (0.0912) (0.0903) (0.0904) (0.150) (0.154) (0.167) (0.158) (0.167)
Majority 0.00146 0.0211 0.0273 0.0160 0.0173 -0.0188 -0.0222 -0.0319 -0.0301 -0.0348
(0.0414) (0.0383) (0.0401) (0.0394) (0.0403) (0.0971) (0.0914) (0.104) (0.113) (0.0940)
Election 0.111* 0.120* 0.111* 0.115* 0.114* 0.107** 0.116** 0.110* 0.106** 0.127**
(0.0657) (0.0629) (0.0642) (0.0635) (0.0631) (0.0537) (0.0531) (0.0570) (0.0502) (0.0543)
Developed 1.002** 1.275** 1.054** 1.305** 1.146**
(0.424) (0.497) (0.490) (0.600) (0.529)
Time effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 1614 1614 1598 1614 1614 1614 1614 1598 1614 1614
Countries 105 105 105 105 105 105 105 105 105 105

Finance Research Letters 59 (2024) 104753

R2 0.635 0.646 0.636 0.643 0.644
Instruments 108 85 108 100 85
AR(1) 0.004 0.005 0.004 0.006 0.004
AR(2) 0.217 0.277 0.224 0.219 0.280
Hansen J 0.624 0.386 0.519 0.519 0.254
Diff-Hansen 1 0.488 0.323 0.484 0.492 0.212
Diff-Hansen 2 0.522 0.240 0.417 0.334 0.354

Notes: FE and two-step system-GMM estimations using robust standard errors (in parentheses), corrected for finite samples in the case of GMM estimations. Significance levels at which the null hypothesis
is rejected: ***, 1 %; **, 5 %; and *, 10 %. Financial crises dummies and economic variables are lagged. The lag of the dependent variable and financial crises are treated as endogenous in the GMM
estimations; the respective lagged values and the other explanatory variables are used as instruments in the first-difference equation and their differences are used in the levels equation; they were
collapsed to avoid the problem of having too many instruments. The values reported for AR(1) and AR(2) are the p-values of the Arellano-Bond tests for the first and second order auto-correlated
disturbances in the first differences equations. The Hansen test reports the p-value for the null hypothesis of instrument validity; Diff-Hansen1 tests the exogeneity of the instruments used in the level
part (of the system) as a whole. Diff-Hansen 2 tests the exogeneity of the lagged level of social protection spending used as an instrument in the level part.
T.C. Nguyen et al. Finance Research Letters 59 (2024) 104753

Table 2
Five-year time spans.
(1) (2) (3) (4) (5)

L.SocialProtection 0.646*** 0.853*** 0.692*** 0.778*** 0.699***

(0.145) (0.0899) (0.109) (0.0995) (0.125)
FinCrises 1.306**
(0.630)
Banking 1.165**
(0.574)
Currency -0.727**
(0.352)
Debt -1.022*
(0.557)
Twin/Triple -0.490
(1.140)
Unemployment 0.0448** 0.0489** 0.0488*** 0.0461** 0.0553**
(0.0211) (0.0226) (0.0184) (0.0230) (0.0255)
Openness -0.00787* -0.00318 -0.00327 -0.00249 -0.00423
(0.00431) (0.00408) (0.00522) (0.00417) (0.00418)
LnGDPpc 0.231 0.183 0.224 0.212 0.130
(0.159) (0.121) (0.167) (0.138) (0.164)
GovDebt -0.00484 -0.00448 -0.00552 -0.00359 -0.00302
(0.00414) (0.00348) (0.00363) (0.00421) (0.00422)
GovDeficit -0.124*** -0.0922** -0.118*** -0.112*** -0.133***
(0.0401) (0.0381) (0.0383) (0.0414) (0.0401)
MedAge 0.215*** 0.174*** 0.176*** 0.126** 0.167**
(0.0821) (0.0494) (0.0619) (0.0566) (0.0708)
Democracy 0.0668** 0.0685*** 0.0632* 0.0631* 0.0650*
(0.0294) (0.0242) (0.0351) (0.0323) (0.0351)
Left 0.472* 0.409* 0.489** 0.444** 0.405*
(0.271) (0.234) (0.225) (0.203) (0.223)
Majority 0.155 0.282 0.0575 0.275 0.151
(0.308) (0.280) (0.236) (0.278) (0.290)
Election 1.848** 1.686* 1.552* 1.544* 1.444*
(0.853) (0.953) (0.862) (0.858) (0.846)
Developed 1.106* 0.954** 1.036* 0.900* 0.974*
(0.589) (0.439) (0.556) (0.486) (0.546)
Time effects Yes Yes Yes Yes Yes
Observations 306 306 303 306 306
Countries 100 100 98 100 100
Instruments 78 58 74 70 57
AR(1) 0.004 0.003 0.009 0.002 0.007
AR(2) 0.301 0.346 0.336 0.333 0.338
Hansen J 0.175 0.186 0.211 0.391 0.205
Diff-Hansen 1 0.223 0.280 0.165 0.437 0.427
Diff-Hansen 2 0.378 0.241 0.255 0.346 0.338

Notes: See Table 1. Two-step system GMM estimations. Five-year periods: 1990–1994; 1995–1999; 2000–2004; 2005–2009; 2010–2014; 2015–2019.

of financial crisis. This should include governments’ commitment to maintain a robust level of fiscal balance in normal times for them
to be able to finance welfare state expansion programmes during periods of financial crises. Moreover, when external aid is required
that must be designed not to compromise social protection to the more vulnerable citizens. Governments may even consider promoting
private protection (e.g.: private healthcare insurance and private pension provision) to release some of the burden on social protection
when a crisis strikes.
This kind of programme may, however, be easier to implement in countries with a higher level of development. Hence, in future
research, it would be interesting to analyse how developed and developing countries react to these crises and explore possibilities to
design a programme able to cope with their specific social protection needs in times of crises.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Declaration of Competing Interest

None.

5
T.C. Nguyen et al. Finance Research Letters 59 (2024) 104753

Data availability

Data will be made available on request.

Acknowledgments

The authors thank the two anonymous referees, Jan-Egbert Sturm, Sushanta Mallick, Dirk Foremny, Niklas Potrafke, Francisco
Veiga, Ahmad Ahmad, Alistair Milne, Rodrigo Martins, Linda Veiga, Ricardo Sousa and the participants at the 2022 ASSA Annual
Meeting, the 2022 Annual Meeting of the European Public Choice Society, and the 2022 Scottish Economic Society Annual Conference
for their very helpful comments and suggestions.

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.frl.2023.104753.

References

Arellano, M., Bover, O., 1995. Another look at the instrumental variable estimation of error-components models. J. Econom. 68, 29–51.
Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel data models. J. Econom. 87, 115–143.
Castro, V., Martins, R., 2018. Politically driven cycles in fiscal policy: in depth analysis of the functional components of government expenditures. Eur. J. Polit. Econ.
55, 44–64.
Fallon, P.R., Lucas, R.E., 2002. The impact of financial crises on labor markets, household incomes, and poverty: a review of evidence. World Bank Res. Observer 17
(1), 21–45.
Greene, W. H. (2020). Econometric Analysis, 8th Global Ed. Person Education Ltd: London.
Gong, D., Hu, S., Qian, Z., 2023. Banking crises and exporter dynamics. Appl. Econ. forthcoming.
Kiendrebeogo, Y., Assimaidou, K., Tall, A., 2017. Social protection for poverty reduction in times of crisis. J. Policy Modell. 39 (6), 1163–1183.
Kroszner, R.S., Laeven, L., Klingebiel, D, 2007. Banking crises, financial dependence, and growth. J. Financ. Econ. 84 (1), 187–228.
Laeven, L., Valencia, F., 2020. Systemic banking crises database II. IMF Econ. Rev. 68, 307–361.
Mohseni-Cheraghlou, A., 2016. The aftermath of financial crises: a look on human and social wellbeing. World Develop. 87, 88–106.
Nguyen, T.C., Castro, V., Wood, J.A., 2021. Financial crises and human development. Working Paper. https://doi.org/10.2139/ssrn.3802418. Available at:
Nguyen, T.C., Castro, V., Wood, J.A., 2022. A new comprehensive database of financial crisis: identification, frequency and duration. Econ. Modell. 108, 105770.
Nikoloski, Z., 2011. Impact of financial crises on poverty in the developing world: an empirical approach. J. Develop. Stud. 47 (11), 1757–1779.
Paravisini, D., Rappoport, V., Schnabl, P., Wolfenzon, D, 2015. Dissecting the effect of credit supply on trade: evidence from matched credit-export data. Rev. Econ.
Stud. 82 (1), 333–359.
Rewilak, J., 2017. The impact of financial crises on the poor. J. Int. Develop. 30, 3–19.
Roodman, D., 2009. How to do xtabond2: an introduction to difference and system GMM in Stata. Stata J. 9, 86–136.
Stiglitz, J., 2013. The global crisis, social protection and jobs. Int. Labour Rev. 152 (1), 93–106.
Uch, R., Miyamoto, H., Kakinaka, M., 2021. Effects of a banking crisis on credit growth in developing countries. Finance Res. Lett. 43, 102004.