IN INDIA
Inflation and output growth are endogenous variables and both may be argued to respond to change in the other. Therefore, as suggested by Sims (1972), it is important to examine the direction of causality and the legitimacy of their precedence. The scatter plot between inflation and per capita output growth for the period 1971-98 is presented in Figure 4.1, panels a to c. As discussed above, the wholesale price index converted to base years 1993-94 has been chosen as the preferred price index in this study. Economic growth is measured by the growth of per capita real gross domestic product with base years 1993-94. The per capita real output (real gross domestic product) is represented by YPC, while price is represented by P. Logs of the variables are expressed by their lower case letters, and the first difference is expressed by placing a A before the variable. Thus,
Aypc represent the growth rate of real output per capita (economic growth) and Ap
represents the inflation in fractions.
It is clear from Figure 4.1 (panel a), that there is evidence to suggest a negative relationship between contemporaneous inflation and economic growth. Lagged inflation and economic growth appear to be positively related (panel b), while lagged economic growth and inflation appear to be negatively related (panel c). The contemporaneous relationship is also negative (panel a). Thus, while the panel-b result is intuitive, the other two results appear to be counter-intuitive and need further analysis. In addition graphs in Figure 5.1 do not reveal anything about the direction of causality. The relationship between the contemporaneous and lagged or future
variables can be exploited using statistical methods to establish the direction of causality and hence the precedence between these two variables as a first step to further analysis.
Figure 4.1 Scatter plot of inflation (Ap) and per capita real GDP growth (Aypc) (1971-98)
(a) Contemporaneous relationship
0.25 T 0.20 - -
0.15 ••
0.05 ■■
Aypc
(b) Lagged inflation and contemporaneous per capita real GDP growth
0.25 T 0.15 - •0.10 - • -0.08 -0.06 -0.04 -0.02 0.00 Aypc Ap(-1)
(c) Lagged per capita real GDP growth and contemporaneous inflation
0.25 T 0.20 0.15 0.05*- Aypc (-1) Ap 95
There are different forms, in which such causality tests can be undertaken. Hsiao (1979) and Jacobs, et al. (1979) argue that different studies, which use the same data set but different methods of testing, often report results that are not in conformity with one another. Therefore, this study uses the two tests described below before arriving at the conclusions.
Granger Non-Causality Test
The multivariate version of the Granger non-causality test is known as the ‘block Granger non-causality test’, which can be carried out in a VAR set up. Following Hamilton (1994:309), and Pesaran and Pesaran (1997:121-131), the concept of the ‘block Granger non-causality test’ can be described as follows: Consider the
following simple VAR of order p for a m x 1 vector z, of jointly determined
endogenous variables.
z, = a o+ Z ° * .v ,- + M, C4-1)
i = i
Now, suppose the variables of VAR are categorized into two groups as
represented by mx x 1 vector of subset z u and m2 x 1 vector of subset z 2t such that
z, = ( z lf, z 2t) and m, + m 2 = m . Then z, can be decomposed into two blocks as follows.
1=1 1=1
(4.2)
Z 21 ~ a 20 + £ * , 2 1 ^ i,22^2,i-i + U 2t (4.3)
Here, a x0 and a 20 are mx x 1 and m2 x 1 vectors of constants respectively and
O . u , 0 , 12, 0 , 21, a n d 0 ,22 contain autoregressive coefficients. The subset z u is said
to be block exogenous in the time series sense with respect to the variables in subset
z 2t if the elements in subset z 2t are of no help in improving a forecast of any
variable contained in subset z u that is based on the lagged values of all the elements
exogenous when O jl2=0 for i = 1,2,..., p . In other words, the hypothesis that the subset z 2t does not ‘Granger-cause’ zu , which means zu is block exogenous, is defined by
Hg -®12 = ° , w here,012 = ( 0 , 12, 0 2 X 2 p X 2 ). (4.4)
Sims Non-Causality Test
In a seminal paper about the implications of the Granger causality, Sims (1972) proposed an alternative way of testing Granger non-causality. Following Hamilton (1994:304) Sims’ proposition can be stated as follows. Consider following a linear projection of y, on past present and future x ’s, with b. and d. defined as population projection coefficients, that is the value for which E(rjt *r ) = 0 for all
t and z . E is an expectations operator.
yt = c + ^ bjx,-j + ' Z d jx,+j+rlt (4-5)
7=0 7=1
Then, y fails to Granger -cause x if and only if dj= 0 for j = 1,2....
As pointed out in Hamilton (1994:305) a problem with Sims’ form is that the error term p is in general autocorrelated. Thus, a standard F test of the hypothesis that d . =0 for j - 1,2, — in (4.5) will not give the correct answer. To overcome this problem and still be able to carry out a Sims type bivariate-test of Granger non causality Geweke, et al. (1983) (also see Hamilton 1994 c h .ll for more details) suggested the following modified specification.
y, = c ~ ' Z a iy,-j + Y j ß i x.-i + v. <4 -6)
j=1 j =0 7=1
The null hypothesis that y does not Granger-cause xcan be tested by truncating (4.6) at some order p and carrying out a F test of y x- y 2- y p = 0 . The assumption is that y did not cause* can be true if and only if the future values of *
do not have any power to explain the current value of y . In that sense it is also a test of the ability of y in forecasting x.
Considering that in the case of annual data contemporaneous variables can have powerful effects, the outcome of a Sims test, which includes contemporaneous term with the two-sided distributed lags, may give a different result compared to the block Granger non-causality test, which involves only lagged explanatory variables. Further, given the small size of the time series data, the order of the VAR is decided by using the Schwarz Bayesian Criterion (SBC) of model selection. After obtaining the lagged structure, the equations are estimated accordingly.
Both variables are stationary based on the ADF test (see Annexure 4.2 of this chapter) and the order selection criterion of the VAR supports the choice of order 1 for the bivariate system including Ap and Ay p c. As both variables are stationary in the ADF test, the non-causality test in VAR of order 1 would be consistent. The results of the Granger block non-causality test and Sims’ two-sided forecasting ability test are presented in Tables 4.1 and 4.2 respectively.
At the 10% significance level, the Granger block non-causality test (Table 4.1) does not rule out a possibility of bi-directional causation in VAR-1, while Sims’ test (Table 4.2) rejects the unidirectional causation running from per capita output growth to inflation but does not reject causation running from inflation to per capita output growth. Thus, the hypothesis that inflation influences growth is not rejected by either of the tests.
However, in view of the fact that the block Granger non-causality test in VAR-1 suggests the existence of a bi-directional causation, it is important to see which of the two variables has a more persistent effect on the other. In order to verify this, the two tests are repeated such that the regression equations include lags of two, three and four. The estimated equations for each case are presented in Annexure 4.3 and 4.4 corresponding to Granger block non-causality tests and Sims tests, respectively. The results of the hypothesis tests are given in Tables 4.1 and 4.2.
With more lags in the system, the Granger block non-causality test now unambiguously rejects the hypothesis that inflation does not cause per capita output growth, but does not reject the hypothesis that per capita output growth does not cause inflation. A look at the estimated equations of Ap in Annexure 4.3 clearly shows that the coefficients of lagged Aypc are not significant and the VAR-1 result
was not robust. On the contrary, inflation appears to have a more persistent effect on per capita output growth. Importantly, the conclusions from the Sims test remain unaltered. Thus, both tests now support unidirectional causation running from inflation to growth. And, it can be safely concluded that inflation is an important and legitimate explanatory variable in explaining economic growth in the case of India.
Table 4.1 ‘Block Granger non-causality’ test between per capita real GDP growth and inflation in a bi-variate VAR. Direction of non-causality being tested is from x to y (1971-98)
Test variable (lagged x) Ap Aypc
Dependent (y ) Aypc aP
VAR - 1 CHSQ (1) 3.450 [0.063] 4.384 [0.036]
VAR - 2 CHSQ (2) 4.130 [0.130] 2.989 [0.224]
VAR - 3 CHSQ (3) 8.030 [0.045] 1.528 [0.676]
VAR-4 CHSQ (4) 12.74 [0.013] 3.770 [0.438]
Note: P-values for Chi-Square statistics are given in square brackets []. The statistics are to test the null hypothesis in a Wald test that coefficients of the explanatory variables x in the equation of the
y variable are zero according to the equation 4.5. A lower P-value means rejection of the hypothesis and therefore that demonstrates existence of the causal direction. Individual estimated equations are presented in Annexure 4.3 and 4.5.
Table 4.2 Sims test of forecasting ability between per capita real GDP growth and inflation in a bi-variate OLS regression. Direction of non-causality being tested is from y to a: (1971-98)
Test variable (future
X) Ap Aypc
Dependent (y ) Aypc Ap
One past value of x and y and one future value of x
CHSQ (1) 1.620 [0.203] 3.507 [0.061]
Two past value of x and y and one future value of x
CHSQ (1) 0.618 [0.431] 5.460 [0.019]
Three past value of x and y and one future value of x
CHSQ (1) 0.176 [0.675] 8.650 [0.003]
Note: All equations include the contemporaneous value of x in accordance to equation 4.6. P-values for Chi-Square statistics are given in square brackets []. The statistics are to test the null hypothesis in a Wald test that coefficient of the future value of the x variable } in the equation for the y variable
is zero. A higher P-value means that the hypothesis of zero coefficients for future value of x is not rejected and therefore that demonstrates non-rejection of the non-existence of the causal direction from y to x supporting non-existence of causal direction from y to x. Estimation is done
according to equation 5.6.
The regression results of the above test of causal relationships presented in Annexures 4.3 and 4.4 of this chapter clearly show that lagged inflation has a positive influence on economic growth, while contemporaneous inflation has a negative influence. In the subsequent section, the robustness of this relationship is tested by adding more variables related to economic growth.
While it may be simpler to argue that the past experience of higher inflation may be a motivation for higher production, it is not easy to interpret the negative effect of current inflation on growth. As mentioned at the beginning of this chapter, and as also discussed in Chapters 1 and 2, the literature dealing with inflation and output growth provide several explanations, depending upon whether the explanation is provided for stagflation or for cases of high growth with low inflation. In both cases, the standard conviction of the Phillips curve type trade-off is completely at odds. With inflation the purchasing power of the currency falls, and problems in clearing the goods market may arise due to imperfect competition, and or institutional rigidities. Thus, country specific characteristics facilitate in making an argument for the negative impact of current inflation on growth in India.
Drawing on the study of Bruno and Sachs (1985), it was emphasized early in Chapter 2 that a permanent increase in the relative price of a factor of production like raw material could shift the production frontier towards the origin. As capital is fixed in the short-run, the new equilibrium output and profit depend on the relative flexibility of the real wage and legal system regarding labour laws. While there is some kind of wage indexation in most organized sectors, Indian labour law does not allow employers to fire workers easily. With a relatively rigid real wage, unemployment should have increased but has not done so because of the protective labour laws. Where there are such protective labour laws, employment could be forced to remain unchanged in the short-run, which could also be due to technological reasons (for example putty-clay technology). The result is a fall in profit and output. International competitiveness is also reduced resulting in a loss of export demand. The market clearing output is lower, although the potential output based on capital and labour employed has not changed. In the long run if raw material prices revert and expectations are corrected, the system may back track to the original frontier.
As a result of inflation in inputs and intermediate goods, the general price level rises leading to reduced purchasing power of the currency. In the short run loss of
purchasing power is not compensated with a commensurate increase in general income. Market confidence is shaky about investment and output. However, the government may resort to expansionist policy to restore growth. But, as discussed in Chapter 5, if the demand schedule is inelastic, the government intervention may further fuel inflation. Thus, inflation can cause output to fall more sharply under a situation of institutional rigidities and ill-timed expansionist policies in countries like India. Since the implications of inflation can be assigned to both supply and demand side effects, it is rational to examine its effects in a proper framework of fully specified regression for per capita real income growth. More specifically, in such a model, the question is whether inflation is a fundamental determinant of growth as are other variables such as investment, human capital, weather conditions and population growth. The same is attempted in the following section, which also examines the non-linear effects of inflation on growth in the case of India.
Inflation may affect economic growth through its effects on investment by affecting the real interest rate and investors’ expectations about consumer confidence. Therefore, it is also important to examine the effects of inflation on investment and its determinants. Therefore, analysis of this aspect is done later in section 4.5 of this chapter.