Page 123 - GHES-2-4

Global Health Economics and
Sustainability
Fiscal policy shocks and health outcomes

Annual data are used, covering a period of 20 years of mutually orthogonal structural shocks π and π
it
t
(2000 – 2019) from 20 countries in SSA (Appendix A1). represent, respectively, the common structural shocks
Our variables of interest are life expectancy, government shared by all panel members and member-specific

per capita health expenditure, and private per capita health idiosyncratic structural shocks. λ is the member-specific
i

expenditure. Life expectancy refers to the number of years a loading coefficients for the common shocks. Following
newborn infant would live if prevailing patterns of mortality Boiciuc (2015), to estimate a SVAR model, the reduced
at the time of its birth remain unchanged throughout its life. form is determined by multiplying Equation (1) by an
Government per capita health expenditure is the average inverse matrix B 0 − 1 , which produces the following:
amount of money spent by the government to provide
healthcare services per citizen. Private per capita health BBY B AY BZ (II)
1
1
1
expenditure is the average household health expenditure i i it i i it p i iit
for each household member; it is introduced as a control
variable to determine if it moderates the potential negative Arranging Equation (2) in a more compact form yields
impact of fiscal policy shock. Public and private health the following:
expenditures per capita are measured in US dollars.
Y Y it p it (III)
it
2.2. Method
1
−1
The study adopts the panel structural vector autoregression where φ = BA and ψ = BZ . ψ is a 3 × 1
i
it
i
it
i it
i
(SVAR) model developed by Pedroni (2013). The model vector of structural shocks in a reduced form that is
addresses the challenge of insufficient time series data for orthogonal and normally distributed. The relation between
an economic unit and accounts for two major features that structural shocks and reduced-form shocks is expressed as
characterize panel data. First, it accounts for substantial follows:
heterogeneity that is likely to be present across the
individual members of the panel. Second, it accounts for Bψ = Z ε (IV)
it
i
iit
possible cross-sectional dependence that may be present in
the panel as economic units respond not only to their own Estimating Equation (1) may pose some difficulties
idiosyncratic shocks but also to common shocks that affect because of parameter proliferation, that is, the number
all units in the panel. of estimable parameters exceeding the number of
equations. To overcome this challenge and resolve the
The model requires a minimum set of restrictions and problem of identification, we impose restrictions on
1
produces two features – impulse response function (IRF) some parameters of the matrix of contemporaneous
and forecast error variance decomposition, respectively, effects, assuming that the shock to fiscal policy
reflecting the size of the impact and transmission affects health outcomes and private per capita health
mechanism of policy shocks (Cazacu, 2015). We specify a expenditure contemporaneously. Shocks to private
panel SVAR model in the following form: per capita health expenditure may not impact fiscal
BY AY Z (I) operations contemporaneously, and shocks to health
iit
i it
i it p
outcomes affect neither fiscal operations nor private
per capital health expenditure contemporaneously. This
Where B is a 3 × 3 matrix of contemporaneous effects, assumption informs the ordering of the variables.
i
Y is a 3 × 1 vector matrix of estimable endogenous
it
variables (fiscal shock [v], private health expenditure shock We use the recursive approach for identification
[z], and health outcomes [x]). A is a 3 × 3 matrix of restriction, which allows us to set the upper triangular
i
coefficients of endogenous variables, representing the IRFs section of the matrix of contemporaneous effects to zero
of the shocks to the elements of Y . Y is a matrix of and restrict Z to an identity matrix (Caldara & Kamps,
i
it-p
it-p
lagged endogenous variables. Z is a 3 × 3 matrix that 2008), such that the diagonal elements are the loading
i
captures the linear relations between structural shocks and coefficients λ . After the initial period, the variables in the
i
those of the reduced-form model. ε is a 3 × 1 vector matrix system are allowed to interact freely. For example, private
it
of composite structural shocks. These composite shocks are health expenditure shock can affect health outcomes in
distributed independently over time but may be cross- all periods after the one in which the shock occurs. This
sectional dependent. i, t, and p represent the individual unit operation produces the following:
in the panel, time, and optimal lag length, respectively.
1 We use the formula (n -n)/2 to determine the number
2
We consider a common representation of the composite of restrictions to be imposed, where n is the number of
shocks, such that ε = λ π + π ,where the two categories endogenous variables.
it
it
i t
Volume 2 Issue 4 (2024) 3 https://doi.org/10.36922/ghes.3454