Economic Modelling 42 (2014) 66–76

Contents lists available at ScienceDirect

Economic Modelling
journal homepage: www.elsevier.com/locate/ecmod

Human capital and economic growth: A macroeconomic model

for Pakistan
Faisal Sultan Qadri a,⁎, Abdul Waheed b
a
Government Premier College No 1, North Nazimabad, Karachi, Pakistan
b
Department of Economics, University of Karachi, Karachi, Pakistan

a r t i c l e i n f o a b s t r a c t

Article history: The paper presents a small macro model for Pakistan economy focusing the impact of investment in human
Accepted 6 May 2014 capital on the key macroeconomic variables. The demand side is modeled along the Keynesian lines while the
Available online xxxx supply side is modeled as per neoclassical theory of production. This framework allows analyzing the effects of
investment in human capital on supply side variables (like labor, physical and human capital) and demand
JEL classification:
side variables (like consumption and investment) at the same time.
E24
O40
The model has small forecasting horizon in which three alternative scenarios regarding government spending on
I20 education are evaluated from 2012 to 2016. The model shows that the link between human capital and labor
C51 market is weak however a change in education spending affects output through enhancing productivity and
through multiplier-accelerator principle. Though the model is small in size and forecasting horizon, it can help
Keywords: in evaluating the future paths of key macroeconomic variables associated with education spending.
Human capital © 2014 Elsevier B.V. All rights reserved.
Economic growth
Education
Model construction and estimation

1. Introduction Cooperation and Development). Macroeconomic modeling frame-

The role of human capital in the economic growth of a country is
considered nothing less than necessary. There are numerous studies
based on panel, cross sectional and time series data that found human
capital as one of the most important factors in growth process. Most
of such studies are conducted in the last few decades (Wilson and
Briscoe, 2004) because of the availability of large cross country data
which allowed the researchers to empirically test the models based on
different sets of theories.1 Various cross sectional studies also illuminat-
ed the role of human capital in the income convergence across coun-
tries.2 In the case of Pakistan, the human capital–economic growth
linkages are investigated through various studies conducted using
time series data.3 However, when it comes to modelling this theoretical
and empirically tested relationship in macroeconomic modeling frame-
work, very little work has been done so far.
Macro modeling has been used as an effective framework for policy
analysis and forecasting the paths of key macroeconomic variables for a
country or region (like European Union and Organization for Economic
⁎ Corresponding author. Tel.: +92 3 9278285.
E-mail addresses: faysalsultan@yahoo.com (F.S. Qadri), waheedku@yahoo.com
(A. Waheed).
1
See Sianesi and Reenen (2000) or Temple (2000) for a discussion on such studies.
2
See Nelson and Phelps (1966), Barro (1991), Mankiw et al. (1992) or Qadri and
Waheed (2013) for details.
3
See Abbas (2000 and 2001) or Qadri and Waheed (2011) for details.
work does allow not only the analysis of human capital contribution
in growth process, but also the running of policy simulations to analyze
the effects of investing in human capital on key macroeconomic vari-
ables like output, employment, consumption, investment and prices.
There are various macro models available that are built keeping
different segments of macro economy in consideration however, none
of them explicitly discuss the role of investment in human capital for
economic growth. Spending on education (investment on human capi-
tal) affects output through at least three channels. First, it increases the
productivity of labor so output rises because of high productivity.
Second, an increase in the labor productivity leads to an increase in
the demand for labor and output rises because of more employed
workers. Third, an increase in the stock of human capital attracts phys-
ical capital inflow from other countries (Abbas, 2000, 2001) and output
rises because of increased foreign investment.
1.1. The macroeconomic models for Pakistan economy
There is very limited literature available on macroeconomic model-
ing in the case of Pakistan economy. Mainly, there are large macro
models like Naqvi et al. (1983), including Naqvi and Ahmad (1986)
and Chisti et al. (1992). All of these models are large sized, desegregated
models covering different sectors including production, labor market,
and monetary, fiscal and foreign sector. Another influential work has
been done by SPDC in making a large sized macro model (Pasha et al.,

http://dx.doi.org/10.1016/j.econmod.2014.05.021
0264-9993/© 2014 Elsevier B.V. All rights reserved.
 F.S. Qadri, A. Waheed / Economic Modelling 42 (2014) 66–76 67

1995) connecting the macro economy with the social sector in Pakistan. Table 1
Khan and Din (2011) is a small sized dynamic macro model which Key characteristics of the model.

covered the linkages between aggregate supply, aggregate demand, Number of independent blocks 3
monetary, fiscal and foreign sectors of Pakistan. Hanif et al. (2011) is a Number of recursive blocks 2
small sized macro model analyzing the impact of monetary policy on Number of simultaneous blocks 1
Number of equations 12
key macro economic variables. Since, the supply side is not considered
Behavioral equations 8
in the model building, the model implicitly assumes that a change in Identities 4
output and prices is mainly because of a change in aggregate demand Endogenous variables 12
which is inappropriate in the case of a supply constraint country like Exogenous (policy variables) 3
Exogenous (non-policy variables) 5
Pakistan. Aggregate demand stimulation can cause inflation in the
presence of supply constraints (Hirsch, 1977) instead of increasing the
output and employment in an economy (Mallick, 1999). All the macro
models enlightened the linkages between macroeconomic sectors how- Keynesian model. Key characteristics of the model are presented in
ever; none of the models explicitly discussed the role of investing in Table 1.
human capital in economic growth. This is the first model which illumi- As presented in Table 1, the model classifies exogenous variables
nates the effect of such spending on key macroeconomic variables in into policy and non-policy variables. The exogenous variables on
case of Pakistan. which government enjoys a significant degree of control are exoge-
nous policy variables.6 Government consumption, government in-
vestment and education spending as percentage of GDP are taken
2. Characteristics of the model and data as policy variables in the model.7 The coefficients of all independent
variables in the 08 behavioral equations are the long run equilibrium
The model is a small sized, comprehensive macro model which is values of the variables as per Engle–Granger two step procedure.8
comprised of three blocks. Two of which are recursive while one is si-
multaneous block. There are 12 equations in the model out of which,
08 are behavioral equations and 04 are identities. There are 20 variables 4. Demand block
in total. Out of which, 12 variables are endogenous while 08 are exoge-
nous. The exogenous variables are further classified as policy and non- The demand for products and services comprises of consumption
policy variables. There are 03 policy variables named government con- demand, investment demand, inventory and net exports.
sumption, government investment and spending on education as per-
centage of GDP and 05 non-policy variables.
The data set is annual which is taken from various publications of 4.1. Consumption demand
Pakistan Bureau of Statics (Former, Federal Bureau of Statistics). The
unit for all monetary variables is million rupees in constant price Consumption demand is the sum of private consumption demand
(1999–00) except for world GDP which is measured in million US dollar and government consumption demand. Government consumption de-
in constant price (1999–00), taken from World Development Indicators.4 mand is taken as a policy variable while the estimated coefficients and
The variable capital stock (K) is calculated through perpetual inven- relevant parameters of private consumption are stated below.
tory method. The data set used for model calibration is from 1981 to
2011 except for the private consumption. The data set for private con-
sumption is from 1975 to 2011 in order to get the best fit values for 4.1.1. Private consumption
model calibration. The data set for model's solution is from 1981 to In order to get the best fit consumption function for Pakistan econo-
2011. my, different specifications have been used. In an alternative specifica-
tion, lag of consumption was found to be statistically insignificant
rejecting the existence of adaptive expectations. The relationship be-
3. Structure of the model
tween interest rate and consumption depends on the relative magnitudes
of income and substitution effect. Generally, a negative relationship
This model complements the existing macro models which illumi-
between interest rate and current consumption is expected. However, in-
nates the macro effects of investment in human capital in Pakistan.
terest rate also turned out to be statistically insignificant in the regression
The model follows the framework adopted by Welfe (2005 and 2011)
which implies that interest rate has very limited or no role in the inter-
which used final demand equations along the Keynesian lines but
temporal consumption decision.
the output generation is modeled through neoclassical theory of pro-
duction. This framework allows analyzing the effects of investment
in human capital on supply side variables (like labor, physical and
Equation 1
human capital) and demand side variables (like consumption and in- log(PC) = 6.854 + 0.329 ∗ log(Y) + 0.212 ∗ log(M2) Adjusted R2
vestment) at the same time.5 The pure Keynesian model is generally t-Statistics 5.630 2.795 5.273 0.996
criticized because of its inability to cover the supply side of the economy Prob(t-stats) (0.000) (0.008) (0.000)
which is very important especially in the case of under developed coun- DW statistics 1.624 F-Statistics 4507.348
tries which generally face supply constraints. Moreover, the model does
not cover the role of the money market, relative prices and expectations However, the best fit regression indicates that Keynesian consump-
effectively (Valadkhani, 2004). A pure classical model is also criticized tion function and wealth effect explain the consumption behavior suffi-
because of not covering the demand side of the economy. The model ciently. In Eq. (1), Y represents current income and M2 is taken as a
under Klein et al. (1999), Bodkin et al. (1991) and Whitley (1994) proxy for financial wealth.9
framework overcomes the problems associated with pure classical or
6
See Ra and Rhee (2005) for details.
4 7
http://databank.worldbank.org/ddp/home.do?Step=12&id=4&CNO=2, last accessed For details of variables and their description, see Appendix 1.
8
on July 2012. See Tables in Appendix 1 and 2. Tables show that all the variables in the 08 behavioral
5
See Klein et al. (1999), Bodkin et al. (1991) and Whitley (1994) regarding studies on equations are I(1) and that the residual series obtained from all 08 equations are I(0).
9
similar framework. See Hanif et al. (2011) for a similar specification.
68 F.S. Qadri, A. Waheed / Economic Modelling 42 (2014) 66–76

4.2. Investment demand
Investment demand is the sum of private and public investment,
government investment and inventory investment. Government
investment is taken as an exogenous policy variable while inventory
investment as an exogenous non-policy variable in the model.
4.2.1. Private and public investment
group and chemical group. All of these are positively related to domestic
income with income elasticity greater than unity. The coefficient of par-
tial adjustment effect is found to be statistically significant in the model.
5. Supply block
Supply side of the model represents the neo-classical theory of pro-
duction as discussed in the subsequent subsections.

5.1. Production function

Equation 2
log(INVAB) = −0.911 + 0.945 ∗ log(Y) Adjusted R2 The production function with three variable inputs is used in the
t-Statistics −0.348 5.324 0.952 model namely, labor, physical capital and human capital.
Prob(t-stats) (0.735) (0.000)
DW statistics 1.475 F-Statistics 299.525

Equation 5
Similar to the private consumption equation, different theoretically log(Y) = 4.611 + 0.375 ∗ log(EL) + 0.558 ∗ log(K) + Adjusted R2
acceptable specifications were tested for the second important compo- 0.136 ∗ log(H(−1))
t-Statistics 6.625 2.810 7.088 1.719 0.998
nent of aggregate demand, investment. Interest rate was found to be in-
Prob(t-stats) (0.000) (0.009) (0.000) (0.097)
significant in this regression which rejects the existence of classical DW statistics 1.411 F-Statistics 3203.238
investment demand function. Government investment was also includ-
ed in another specification to check the crowding out phenomenon but
that also turned out to be insignificant. Finally, in the best fit regression, In the function, labor input is captured through employed labor, cap-
the variation in private and public investment can be explained suffi- ital stock is used in order to capture capital input and gross enrolment
ciently by the variation in income as stated in the Keynesian accelerator rate in secondary education is taken as a proxy for human capital
principle. input in the function. All of these are found to be positively and signifi-
cantly related with output as theoretically expected.
4.3. Trade block
5.2. Demand for labor
The final component of aggregate demand, net exports is deter-
mined through trade block.10 This block is comprised of export and Labor demand function is modeled differently from convention.
import functions which are connected through the net export identity. Labor force and the lag of human capital are used as explanatory vari-
ables. As neo-classical theory of labor demand explains, the increased
4.3.1. Product and service export labor force would decrease real wages and thereby increase the quantity
Pakistan's export to the world is taken as a function of world's in- demanded for labor. In a country like Pakistan where labor force has
come which sufficiently explains the variation in Pakistan's exports. been increasing very fast, this variable seems more appropriate in
explaining the labor demand.

Equation 3
log(EXPORT) = −33.679 + 1.513 ∗ log(WGDP) Adjusted R2 Equation 6
t-Statistics −1.890 2.646 0.966 log(EL) = −101.752 + 1.260 ∗ log(L) + 0.063 ∗ log(H(−1)) Adjusted R2
Prob(t-stats) (0.069) (0.013) t-Statistics −0.009 10.119 3.073 0.999
DW statistics 1.970 F-Statistics 432.757 Prob(t-stats) (0.992) (0.000) (0.005)
DW statistics 1.795 F-Statistics 10209.34

In the other alternative specifications, real exchange rate is found to

be insignificant in explaining export variation. Similarly, lag of export
was taken as an independent variable to capture the partial adjustment
effect however, it was also found to be statistically insignificant.
4.3.2. Product and service import
Theory of production and labor demand explains that increased pro-
ductivity of labor would lead to increase in the demand for labor.
Human capital is also expected to be positively related with the demand
for labor because it increases the labor productivity.
5.3. Human capital

In empirical studies, human capital is generally proxied through an

Equation 4
log(IMPORT) = −16.041–0.510 ∗ log(RER) + 1.701 ∗ log(Y) + Adjusted R 2 education stock or flow indicator. At the same time, there are studies
0.412 ∗ log(IMPORT(−1)) which employed health, nutrition and experience as an indicator of
t-Statistics −2.393 −2.344 2.924 2.587 0.899 human capital. There are studies which found the coefficient of human
Prob(t-stats) (0.024) (0.027) (0.007) (0.015) capital as statistically insignificant or even negatively related to the
DW statistics 1.524 F-Statistics 85.529
growth process.12 The empirical studies show that the direction and
magnitude of human capital to affect economic growth are sensitive
In the import demand function, real exchange rate,11 domestic in- with respect to the proxy used. This study used a narrow definition of
come and lag of import are used as explanatory variables. As theoretical- human capital and employed education indicator as a proxy for human
ly expected, the coefficient of real exchange rate is negative. The key capital because of the availability of official data and also because this in-
components of Pakistani imports belong to Petroleum group, machinery dicator is under more direct control of government then any of the broad
indicator like ‘learning by doing’.
10
See Christodoulakis and Kalyvitis(1998) for similar specification.
11 12
RER = IMPPRICE / CPI where IMPPRICE = unit value of imports ∗ exchange rate. See Benhabib and Spiegel (1994) or Pritchett (2001).
 F.S. Qadri, A. Waheed / Economic Modelling 42 (2014) 66–76 69

In line with convention and theory, human capital (gross enrol- based heteroscedasticity. The estimated coefficients of Eqs. (2),
ment rate in secondary education) is modeled as a function of (4) and (8) are obtained through the procedure known as “White
domestic income and government spending on education as percentage heteroscedasticity consistent standard errors and covariance” because
of GDP. these equations were found to have heteroscedasticity.
Chow forecast test is used to see whether the parameters are stable
during the sample period. 1987, 1999 and 2007 are chosen as break
Equation 7 points because these are the transition years from a military person's
log(H) = −1.061 + 0.154 ∗ log(ES) + 0.161 ∗ log(Y) + Adjusted R2 rule to democratic rule or vice versa. The employment parameter is
0.545 ∗ log(H(−1)) found to be affected in the different kinds of regimes.
t-Statistics −2.033 2.199 2.772 4.003 0.933
Prob(t-stats) (0.052) (0.036) (0.010) (0.000)
DW statistics 1.925 F-Statistics 139.859 8. Forecasting performance

The main objective of this model is to predict different paths of mac-

As expected, the coefficients of domestic income and government
roeconomic variables associated with different policy scenarios regard-
spending on education as percentage of GDP are found to be positively
ing government spending on education. Prior to starting the scenario
and significantly related with human capital. The coefficient of partial
experiments, the model is also evaluated for within the sample and
adjustment mechanism is also found to be statistically significant. The
out of sample performance.
other relevant variables like urbanization are found insignificant in
explaining the model.
8.1. Within-sample forecast performance
6. Price block
The within sample forecast performance is evaluated through con-
In line with the theoretical expectations, consumer price index is ventionally used statistics named root mean square percentage error
found to be a positive function of import prices and the broad money (RMSPE), mean absolute percent error (MAPE) and Theil inequality coef-
relative to the domestic income. ficient while out of sample performance is evaluated through stochastic
simulations. Table 3 presents the values of root mean square percentage
error (RMSPE), mean absolute percent error (MAPE) and Theil inequality
Equation 8
coefficient.
Log(CPI) = 2.694 + 0.265 ∗ log(IMPPRICE) + 0.357 ∗ log(M2/Y) Adjusted R2 The table shows that the model performs well in static and dynamic
t-Statistics 6.352 6.066 3.955 0.998 simulations especially in the case of predicting the GDP and employment.
Prob(t-stats) (0.000) (0.000) (0.000) All RMSPE values are less than 05 in the case of static simulation and less
DW statistics 1.496 F-Statistics 5128.391
than 06 in the case of dynamic simulation except the private and public
investment. In the case of MAPE, all statistics are less than 04 in static
Import price is the product of unit price of import and nominal and less than 05 in dynamic simulations other than private and public
exchange rate. If unit price of import increases or domestic direct ex- investment. All values of Theil inequality coefficient are close to zero indi-
change rate increases (or both factors increase simultaneously), the cating satisfactory forecasting ability of the model. Fig. 1 presents actual
domestic prices are expected to increase. In a country like Pakistan, values against the predicted values estimated through static and dynamic
where the biggest category of import is petroleum products, an in- simulations for the key variables in the model. The graphs show that the
crease in import prices is more likely to increase the domestic prices model can track the historical values with remarkable accuracy especially
significantly. in the case of GDP, employment, private consumption and CPI.

7. Diagnostic tests 8.2. Out-of-sample forecasting performance

Prior to solving the model, it is necessary to check whether the Out of sample performance is conventionally evaluated through
coefficients for model calibration are estimated through a statistical- stochastic simulations. The dynamic deterministic simulation predicts
ly reliable procedure or not. There are different tests to check the future path of endogenous variables as a single point at each obser-
whether an individual equation is reliable enough to be chosen for vation however; the procedure of prediction through stochastic simula-
model calibration or not. The findings of such tests are reported in tion is quite different.
Table 2. The stochastic simulations incorporate uncertainty in the predic-
In this section, the diagnostic tests begin with BG/LM which is used tions through adding random shocks into each estimated equation. In
to check if there is an existence of autocorrelation in any equation. None this case, the model predicts a complete distribution of endogenous
of the questions have found to have first or second order autocorrela-
tion. White test is used to check the existence of pure or specification
Table 3
Forecasting accuracy of the key variables.
Table 2 Source: Authors' estimation.
Diagnostic tests.
Source: Authors' estimation. RMSPE Mean absolute Theil inequality
percent error coefficient
Equation BG/LM test Prob White het. test Prob Chow forecast test Prob
Static Dynamic Static Dynamic Static Dynamic
1 0.975 0.388 1.093 0.384 1.121 0.454
2 1.011 0.378 3.632 0.04 0.585 0.81 Y 1.818 2.817 1.495 2.135 0.009 0.016
3 0.019 0.981 1.592 0.221 1.586 0.398 H 4.616 5.724 3.645 4.563 0.023 0.028
4 0.719 0.497 3.743 0.007 0.159 0.994 EL 0.716 1.020 0.467 0.808 0.003 0.004
a a a
5 2.046 0.151 1.022 0.455 1.54 0.572 PC 3.028 2.354 0.016
6 1.057 0.362 1.243 0.319 57.791 0.017 CPI 2.938 5.004 2.447 3.958 0.018 0.031
7 0.058 0.944 0.806 0.616 4.513 0.197 INVAB 8.500 10.759 6.950 9.135 0.044 0.051
8 1.811 0.184 3.521 0.015 7.899 0.118 a
No dynamics in equation.
70 F.S. Qadri, A. Waheed / Economic Modelling 42 (2014) 66–76

Fig. 1. Static and dynamic simulation.

variables at each observation. Since the model is linear, mean and stan-
dard deviations are appropriate statistics to describe the distribution
sufficiently.
The Monte Carlo approach is used to estimate the distribution of en-
dogenous variables in the forecast period. In the Monte Carlo approach,
the model is solved several times with simulated random numbers
substituted for unknown errors at every solution.
For the out-of-sample period, the values for the exogenous variables
are generated through dynamic forecasting by using an autoregressive
component in the equation of INVC, GC, M2 and LF. The values for
IMPPRICE, INVENTORY and WGDP are assumed to grow at the last
05 year's average growth rate. The focused policy variable H (government
spending as percentage of GDP) is assumed to be fixed at around 2.2%
(the average percentage value for the last 10 years).
Fig. 2. presents the graphs for the key variables for out-of-sample pe-
riod. The figure presents three curves for each variable. The curve in the
middle represents the mean of distribution at each observation and the
other two lines show the 95% confidence interval. The figure shows that
the model performance is satisfactory for the out-of-sample period.
9. Scenario analysis
After establishing the baseline path for the key variables, the impacts
of three policy scenarios are analyzed. In all the three scenarios, a one
 F.S. Qadri, A. Waheed / Economic Modelling 42 (2014) 66–76 71

…………… Mean +/- 2 Std.Deviaons ____________ Predicted line

Fig. 2. Stochastic simulation for out-of-sample period.

time or sustained shock would be given to the core policy variable, gov-
ernment spending on education as percentage of GDP. It has been
discussed that the change in spending effects human capital flow in
the economy which effects output in three distinguished ways.
First, it increases output due to higher productivity. Second, it
increases output due to increased employment and third, due to in-
creased foreign capital inflows. In this study, the first two channels
have been modeled.
9.1. Scenario 1: Education spending as 4% of GDP from 2012 to 2016
The first shock is a sustained level of education spending as 4% of GDP
throughout the forecast period. As shown in Fig. 3, this shock increased
enrolment rate by 9.5% from the baseline in the same year. The deviation
of enrolment rate from baseline has been increasing throughout the fore-
cast period and reached 22.28% in 2016 even though the marginal enrol-
ment rate has been declining. The private and public investment
responds to this shock in 2013 and the investment deviates 1.41% from
baseline. This deviation further increased in the later years because of
the multiplier–accelerator process and finally reached 3.18% in 2016.
The increased human capital affects the labor demand and employment
in 2013. A 0.57% increase in employment as compared to the baseline is
recorded in 2013 which increased further in the subsequent years and
reached 1.2% in 2016. The impact of this shock affects GDP in 2013 and
GDP increased by 1.52% from baseline. The deviation of GDP from base-
line increased further due to employment and multiplier–accelerator
principle. In 2016, this deviation reached 3.45% from baseline. An inter-
esting thing is that though the overall economic activities have been
72 F.S. Qadri, A. Waheed / Economic Modelling 42 (2014) 66–76

Base line Scenario 1 Percentage devia on from base line

Fig. 3. Scenario 1.

increasing however, CPI falls relative to the baseline. The deviation of CPI
from baseline has been increasing though with decreasing rate. This
might be because in Pakistan, the CPI depends more on import prices
and money supply relative to GDP and GDP grows faster in scenario 1
therefore money supply relative to GDP declines. The increase in GDP
affects private consumption positively in 2013 and private
consumption deviates from baseline by 0.49%. This deviation reaches
1.12% in 2016.
9.2. Scenario 2: Starting from 3% in 2012, a gradual 0.5% increase in the
education spending as percentage of GDP from 2013 to 2016
Fig. 4 presents the response from key variables to the second shock.
This shock is a gradual 0.5% increase in the education spending as
percentage of GDP throughout the forecast period started from 3% in
2012. This shock increased the enrolment rate by 4.84% from the base-
line in 2012. The deviation of enrolment rate from baseline has been in-
creasing throughout the forecast period and reached 26.75% in 2016.
The private and public investment responds to this shock in 2013. The
investment increased by 0.72% from baseline and ended at 3.17% in
2016. Employment responded in 2013 to this shock. A 0.29% increase
in employment from baseline is recorded in 2013 which reached
1.22% in 2016. This shock increases GDP by 0.78% from baseline
in 2013 which reached 3.43% from baseline in 2016. CPI falls relative
to baseline by 0.27% in 2013 and continued to fall through the forecast-
ing period. In response to the shock, private consumption
increased from baseline by 0.25% in 2013. This deviation reaches 1.11%
in 2016.
 F.S. Qadri, A. Waheed / Economic Modelling 42 (2014) 66–76 73

Base line Scenario 1 Percentage devia on from base line

Fig. 4. Scenario 2.

9.3. Scenario 3: Education spending as 5% of GDP in 2012 than 3% of GDP
throughout from 2013 to 2016
The third shock is a onetime increase in the education spend-
ing as percentage of GDP in 2013 and then a sustained education
spending as percentage of GDP. According to this shock, the edu-
cation spending as percentage of GDP is 5% in 2012 than a
sustained increase of 3% of GDP, throughout the period 2013–
2016. Fig. 5 presents the response from key variables to this
shock. Enrolment rate responded to the shock by an increase of
13.4% from the baseline in 2012. The deviation of enrolment rate
from baseline declines slowly with time and reached 11.91 in
2016. The private and public investment responds to this shock
in 2013. The investment increased by 1.94% from baseline and
this deviation remains quite stable throughout the forecasting
period. A 0.79% increase from baseline in employment is recorded
in 2013 which declined in a smooth manner and reached 0.71% in
2016. This shock increases GDP by 2.1% from baseline in 2013 and
the deviation increases till 2015. In 2016, the deviation from base-
line remains 2.1%. CPI falls relative to baseline by 0.74% in 2013
and this deviation remains almost unchanged throughout the fore-
casting period.
10. Conclusion and policy implications
The paper builds and solves a comprehensive macro model
for Pakistan economy. This is the first model for Pakistan
economy that explicitly discussed the role of education spending in
74 F.S. Qadri, A. Waheed / Economic Modelling 42 (2014) 66–76

Base line Scenario 1 Percentage devia on from base line

Fig. 5. Scenario 3.

determining the key macroeconomic variables. The model
has 12 equations in total out of which 04 are identities. The
demand side of the model is constructed in light of Keynesian
framework and the supply side is modeled in neoclassical lines. The
model differentiates policy variable from ordinary exogenous
variables because these variables are under direct control of
government policies. The core policy variable in the model is spend-
ing on education as percentage of GDP and three scenarios
associated with different paths of education spending are discussed in
the paper.
Calibration of the model is done through ordinary least square
and the model is solved through Gauss–Seidel procedure. The
model performs quite well in tracking the historical paths of the
key variables. The within sample performance is evaluated through
conventional statistics RMSPE, MAPE and Theil inequality coefficient
which are estimated for both the static and dynamic runs of the
model. The out of sample performance is evaluated through stochas-
tic simulations.
The policy scenarios predict different future paths of the key
macroeconomic variables. The simulations show that the link be-
tween education and labor market is quite weak. In the all three
scenarios, eventhough the education spending increased more
than the historical path however, the maximum change in the em-
ployment associated with such spending is 1.22% from baseline in
the 5th year. The maximum output and investment deviation are
3.45% and 3.18% from baseline respectively, which is mainly be-
cause of the productivity effect and through multiplier–accelerator
principle.
For small forecasting span, this model can be useful in taking the
decisions regarding education spending.
 F.S. Qadri, A. Waheed / Economic Modelling 42 (2014) 66–76 75

Appendix 1

Test for stationarity

ADF test

Level First difference

Variables C C&T C C&T

Y 1.622 −1.063 −3.050** −3.676**

L 2.773 −0.552 −5.139* −5.265*
K 2.348 0.000 −3.068** −3.995**
H −0.514 −1.811 −5.374* −5.293*
ES −2.211 −2.165 −3.380** −3.231***
EL 2.452 −0.424 −5.087* −6.476*
PC 2.017 −0.744 −5.699* −4.716*
M2 −0.931 −2.514 −3.777* −4.679*
CPI 1.083 −2.363 −3.199** −3.581***
WGDP 1.463 −2.644 −5.159* −5.369*
INVAB 0.208 −2.540 −6.120* −5.989*
IMPORT −0.746 −2.872 −5.760* −5.672*
EXPORT −0.486 −2.543 −5.780* −5.618*
IMPPRICE 1.700 −3.164 −5.654* −5.789*

Note: Critical values with intercept (C) and intercept and trend (C & T) on level at 10% are −2.625 and −3.225 respectively.
*, ** and *** shows the statistical significance of the variables at 1%, 5% and 10% respectively.
Source: Authors' estimation.

Appendix 2

Variable Description

Exogenous (policy variables)

Spending on education ES Government spending on education as percentage of GDP
Government consumption GC Government consumption in million rupees (constant prices of 1999–2000)
Government investment INVC Government investment in million rupees (constant prices of 1999–2000)

Exogenous(non-policy variables)
Labor force LF Labor force in million
M2 M2 Million rupees
Inventory INVENTORY Inventory in million rupees
Import price IMPPRICE In rupees
World GDP WGDP In constant US dollar (2000)

Endogenous
GDP at constant market price Y GDP at constant prices of 1999–2000
Gross enrolment ratio (Secondary) H Ratio
Employed labor EL In million
Private consumption PC In million rupees (constant prices of 1999–2000)
Private and public investment INVAB In million rupees (constant prices of 1999–2000)
Consumer price index CPI Ratio
Total investment INV In million rupees (constant prices of 1999–2000)
Total consumption TC In million rupees (constant prices of 1999–2000)
Net exports NX In million rupees (constant prices of 1999–2000)
Imports IMPORT In million rupees (constant prices of 1999–2000)
Exports EXPORT In million rupees (constant prices of 1999–2000)
Capital stock K In million rupees (constant prices of 1999–2000)

List of identities

TC = PC + GC…………………………………..…1
INV = INVAB + INVC + INVENTORY.…….…..2
K = K(−1) ∗ 0.95 + INVAB + INVC…..………….3
NX = EXPORT − IMPORT……………….………..4
76 F.S. Qadri, A. Waheed / Economic Modelling 42 (2014) 66–76

ADF test Khan, M.A., Din, M., 2011. A dynamic macroeconometric model of Pakistan's economy.
PIDE Working Papers 2011, 19.
Level Klein, L.R., Welfe, A., Welfe, W., 1999. Principles of Macroeconometric Modeling. North
Holland, Amsterdam.
Variables C C&T
Mallick, S.K., 1999. Modelling Macroeconomic Adjustment With Growth in Developing
U1 −43.553* −40.924* Countries: The Case of India. Ashgate, Farnham.
U2 −5.081* −4.981* Naqvi, S.N.H., Ahmad, A.M., 1986. Preliminary Revised P.I.D.E. Macro-econometric Model
U3 −4.570* −4.462* of Pakistan's Economy. Pakistan Institute of Development Economics, Karachi.
U4 −3.972* −4.348* Naqvi, S.N.H., Khan, A.H., Khilji, N.M., Ahmad, A.M., 1983. The PIDE Macroeconometric
Model of Pakistan's Economy. Pakistan Institute of Development Economics, Karachi.
U5 −4.620* −4.681*
Nelson, R.R., Phelps, E.S., 1966. Investment in humans, technological diffusion, and eco-
U6 −5.309* −4.991*
nomic growth. Am. Econ. Rev. 56, 69–75.
U7 −5.188* −5.108* Pasha, H.A., Hasan, M.A., Pasha, A.G., Ismail, Z.H., Rasheed, A., Iqbal, M.A., Ghaus, R., Khan,
U8 −7.426* −6.833* R., Ahmed, N., Bano, N., Hanif, N., 1995. Integrated Social Policy and Macro-economic
The variables U1 to U8 represent the residual series of Eqs. (1) to (8) respectively. Planning Model for Pakistan. Social Policy and Development Centre, Karachi.
Pritchett, L., 2001. Where has all the education gone? World Bank Econ. Rev. 15, 367–391.
Note: Critical values with intercept (C) and intercept and trend (C & T) on level at 1% are
Qadri, F.S., Waheed, A., 2011. Human capital and economic growth: time series evidence
−3.670 and −4.340 respectively.
from Pakistan. Pak.Bus. Rev. 12–4, 815–833.
* shows the statistical significance of the variables at 1%.
Qadri, F.S., Waheed, A., 2013. Human capital and economic growth: cross-country evi-
Source: Authors' estimation. dence from low-, middle- and high-income countries. Prog. Dev. Stud. 13, 89–104.
Ra, S., Rhee, C.Y., 2005. Nepal macroeconometric model. Working paper series no-1. Nepal
Resident MissionAsian Development Bank.
Sianesi, B., Reenen, J.V., 2000. The Returns to Education: A Review of the Macro-economic
References Literature. Centre for the Economics of Education, London School of Economics and
Political Science, Houghton Street, London WC2A 2AE.
Abbas, Qaiser, 2000. The role of human capital in economic growth: a comparative study Temple, J., 2000. Growth effects of education and social capital in the OECD countries.
of Pakistan and India. Pak. Dev. Rev. 39 (4 part 2), 451–473. OECD Economics Department Working Papers, 263. OECD Publishing.
Abbas, Qaiser, 2001. Endogenous growth and human capital. A comparative study of Valadkhani, A., 2004. History of macroeconometric modelling: lessons from past experi-
Pakistan and Srilanka. Pak. Dev. Rev. 40 (4 Part 2), 987–1007. ence. J. Policy Model 26-2, 265–281.
Barro, Robert, 1991. Economic growth in a cross section of countries. Q. J. Econ. 106–2, Welfe, W. Wladyslaw, 2005. Stylized empirical model of economic growth. In: Welfe, W.,
407–433. Welfe, A. (Eds.), Macromodels 2004. Acta UŁ, Folia Oeconomica, 190. Wydawnictwo
Benhabib, J., Spiegel, M.M., 1994. The role of human capital in economic development: Uniwersytetu Łódzkiego, Łódź, pp. 109–125.
evidence from aggregate cross-country data. J. Monet. Econ. 143–173 (34 pp.). Welfe, Wladyslaw, 2011. Long-term macroeconometric models: the case of Poland. Econ.
Bodkin, R., Klein, L.R., Marwah, K., 1991. A History of Macroeconometric Model Building. Model. 28, 741–753.
B. Elgar, Aldershot. Whitley, J.D., 1994. Course in Macroeconomic Modelling and Forecasting. Harvester/
Chisti, S., Hasan, M.A., Mahmud, S.F., 1992. Macroeconometric modeling and Pakistan's Wheatsheaf, New York.
economy: a vector autoregression approach. J. Dev. Econ. 38–2, 353–370. Wilson, R.A., Briscoe, G., 2004. The impact of human capital on economic growth: a re-
Christodoulakis, N.M., Kalyvitis, S.C., 1998. A four sector macroeconometric model for view. Third Report on Vocational Training Research in Europe: Background
Greece and the evaluation of the community support framework 1994–1999. Econ. ReportOffice for Official Publications of the European Communities, Luxembourg
Model. 15, 575–620. (Cedefop Reference series, 54).
Hanif, M.N., Hyder, Z., Lodhi, M.A.K., Khan, M.H., Batool, I., 2011. A small-size
macroeconometric model for Pakistan economy. Technol. Invest. No-2, 65–80.
Hirsch, A.A., 1977. Analysis of inflation: 1965–1974. Nat. Bur. Econ. Res. 299–332.