Download The Impact of Inflation on Economic Growth: A Study of the Correlation at the OECD Level and more Study notes Literature in PDF only on Docsity!
This PDF is a selection from an out-of-print volume from the National Bureau
of Economic Research
Volume Title: The Costs and Benefits of Price Stability
Volume Author/Editor: Martin Feldstein, editor
Volume Publisher: University of Chicago Press
Volume ISBN: 0-226-24099-
Volume URL: http://www.nber.org/books/feld99-
Publication Date: January 1999
Chapter Title: Does Inflation Harm Economic Growth? Evidence from the
OECD
Chapter Author: Javier Andrés, Ignacio Hernando
Chapter URL: http://www.nber.org/chapters/c
Chapter pages in book: (p. 315 - 348)
8 Does Inflation Harm Economic
Growth? Evidence from
the OECD
Javier And& and Ignacio Hernando
8.1 Introduction
From 1973 until 1984 OECD economies underwent a period of macroeco- nomic distress in which inflation escalated to reach an average rate of 13 per- cent, three times as high as in the previous decade. Since then, achieving low and stable inflation has become the main goal of monetary policy in western economies. This move in monetary policy making rests on the belief, firmly rooted in many economists’ and politicians’ minds, that the costs of inflation are nonnegligible, so that keeping inflation under control pays off in terms of faster sustainable growth in the future. The shortage of theoretical models explicitly addressing the issue of the long-run effects of inflation has not prevented many researchers from trying to estimate the costs of inflation. A series of recent papers have tried to assess the long-run impact of current inflation within the framework of convergence equations. These equations can be derived from a theoretical model of eco- nomic growth, and although the precise channels through which inflation af- fects growth are not always made explicit, they have several advantages for the purposes at hand. First and foremost, an explicit model reduces the risk of omitting relevant variables. Second, convergence equations allow for a variety of effects of inflation, including those that reduce accumulation rates and those that undermine the efficiency with which productive factors operate. Finally, in this framework a clear distinction can be made between level and rute-of
Javier Andrts is professor of economics at the University of Valencia and an economist in the Research Department of the Banco de Espaiia. Ignacio Hemando is an economist in the Research Department of the Banco de Espaiia. The authors are grateful to Palle Andersen, Sean Craig, Juan Dolado, Rafael Dombnech, Angel Estrada, Frederic Mishkin, Teresa Sastre, Javier Vallts, Josb Vifials, and conference participants for their comments and to Francisco de Castro for his excellent research assistance. Javier Andrbs acknowledges financial support by DGICYT grant SEC96- 1435-CO3-01.
315
317 Does Inflation Harm Economic Growth
8.2 Theoretical Framework
8.2.1 International Evidence The negative effects of inflation have been studied in the context of models of economic growth in which the continuous increase of per capita income is the outcome of capital accumulation along with technological progress. I The uncertainty associated with high and volatile unanticipated inflation has been found to be one of the main determinants of the rate of return on capital and investment (Bruno 1993; Pindyck and Solimano 1993). But even fully antici- pated inflation may reduce the rate of return of capital given the nonneutralities built into most industrialized countries’ tax systems (Jones and Manuelli 1995; Feldstein 1997). Besides, inflation undermines the confidence of domestic and
foreign investors about the future course of monetary policy. Inflation also af-
fects the accumulation of other determinants of growth such as human capital or investment in R&D; this channel of influence is known as the accumulation or investment effect of inflation on growth. But over and above these effects, inflation also worsens the long-run macro- economic performance of market economies by reducing total factor produc- tivity. This channel, also known as the eficiency channeZ, is harder to formalize in a theoretical model;2 nonetheless, its importance in the transmission mecha- nism from inflation to lower growth cannot be denied. A high level of inflation induces frequent changes in prices that may be costly for firms (menu costs) and reduces the optimal level of cash holdings by consumers (shoe leather costs). It also generates larger forecasting errors by distorting the information content of prices, encouraging economic agents to spend more time and re- sources in gathering information and protecting themselves against the damage caused by price instability, hence endangering the efficient allocation of re- sources. Although some theoretical models analyze the components of the ef- ficiency channel in more detail, it is difficult to discriminate among them in aggregate empirical growth equations. Thus we shall not pursue this issue fur- ther here. We shall turn our attention to the empirical evidence. Several authors have found a negative correlation between growth and infla- tion. Konnendi and Meguire (1985) estimate a growth equation with cross- sectional data and find that the effect of inflation on the growth rate is negative, although it loses explanatory power when the rate of investment is also in- cluded in the regression. This would indicate that the effect of inflation mainly manifests itself in a reduction in investment but not in the productivity of capi- tal. Grier and Tullock (1989) estimate a model that excludes the rate of invest- ment and includes several measures of nominal instability (such as the inflation
1. See Orphanides and Solow (1990), De Gregorio (1993). and Roubini and Sala-i-Martin
- Briault (1995) surveys the literature on these effects.
(1995) among others.
318 Javier And& and Ignacio Hernando
rate, the acceleration of prices, and the standard deviation of inflation). The results differ according to the group of countries considered, but for the OECD only the variability of inflation seems to have a significant and negative effect on growth. More recently, the study of the long-run influence of inflation has progressed within the framework of convergence equations developed by Barro and Sala- i-Martin (1991).3Fischer (1991, 1993) reports a significant influence of several short-term macroeconomic indicators, in particular inflation, on the growth rate. Cozier and Selody (1992) estimate cross-sectional convergence equations for different samples and find a fairly large negative effect of inflation on in- come at the OECD level. These authors conclude that inflation affects the level rather than the growth rate of productivity and that the impact of inflation vari- ability is weak! This finding coincides with the result obtained more recently for a sample of 120 countries by Barro (1995, 1996), who reports a negative long-run effect of inflation using alternative instruments to correct for the en- dogeneity of inflation. The general conclusion of these and other studies (De Gregorio 1992a, 1992b, 1996; Motley 1994) is consistent with the negative correlation between inflation and income in the long run suggested in the theo- retical literature. However, the consensus in this respect is far from absolute, and several authors have criticized these findings, arguing that the lack of a fully developed theoretical framework makes it difficult to interpret the empir- ical correlations and that even these are not robust to changes in the econo- metric specification. The latter argument is developed by Levine and Renelt (1992), Levine and Zervos (1993), and Clark (1997). Levine and Renelt carry out an exhaustive sensitivity analysis among a broad set of regressors in growth equations and conclude that the statistical significance (and even the sign) of most of these variables (inflation among them) is not invariant under changes in the information set.5 Nor do these results, in turn, escape criticism. Sala- i-Martin (1994) argues that the problem of finding a macroeconomic variable whose effect is invariant under alternative specifications of the convergence equation should not be taken to mean that this influence is absent, but should instead be viewed as a sign of the difficulty of finding indicators that can ade- quately capture this effect for any period and group of countries. Gylfason and Herbertsson (1996) find that the inflation rate is robust to changes in the
3. Several exceptions, however, are worth noting: the studies of Grimes (1991) for the OECD, Smyth (1994) for the United States, Cardoso and Fishlow (1989), who use a panel of five-year averages for 18 Latin American countries, Burdekin et al. (1994), and Bruno (1993). In all these studies, a significant negative effect of inflation on growth is reported. On the other hand, Bullard and Keating (1995) find that the long-run output response to permanent inflation shocks in a struc- tural vector autoregressive model is zero for most advanced economies.
- Judson and Orphanides (1996) measure the variability of inflation as the variance of the quarterly rate for each country and find that it is negatively correlated with growth. These authors also find a significant negative effect of the level of inflation.
- McCandless and Weber (1995) conclude also that the cross-country correlation between in- flation and growth is zero.
320 Javier And& and Ignacio Hernando
omy closes the gap between its current income and its potential or steady state
level.9 This level is, in turn, determined by the parameters of the production
function and by the rates of accumulation of the productive factors:
(3)
where :s is the logarithm of the rate of investment, s? represents the logarithm of the rate of accumulation of human capital, and n* is the growth rate of the
population, all evaluated at their steady state levels; 6 is the depreciation rate
of capital, which will be assumed equal to 3 percent, while the two constants
combine different parameters of the model and the starting level of technol-
This structure allows us to test the different hypotheses considered in this paper. First, the presence of the rates of factor accumulation in equation (3) is useful to discriminate between the two channels through which macroeco- nomic imbalances can affect the growth rate. Thus, if inflation reduces total factor productivity, we could expect a significant coefficient of the rate of in- flation in equation ( 5 ) ,^ below. In this case, the productivity index^ (A,)^ might be assumed to evolve as in equation (4) (Cozier and Selody 1992), which reflects the influence of the inflation rate (IT) and its variability (a):
y; = 0 s + + T + p-l[ols;k+ ys;h- (a + y)log(n? + + + S ) ] ,
ogy (4).
(4) A, = AoexP(+~)exP(F.~,)exP(~.,~,).
The empirical specification is then given by
However, if inflation influenced growth solely through its impact on investment
(SJ, its coefficient in equation ( 5 ) would not be significant.'O Unless it is neces-
sary we shall not impose the parametric restrictions in the previous equations;
we shall focus on the linear version (5') instead:
Yr+T - Y r = + T + (1 - e-")[Q - Y r + +T +
- pzur + p-yas;, + ys:, - (a + y)log(nT + + + S ) ) ].
Second, the exogenous growth model specifies the determinants of both the long-run level of per capita income and the sustained growth rate. Inflation can affect one or the other, although the implications in terms of welfare are
different." According to the specification of equation ( 4 ), the impact of infla-
tion basically impinges on the potential level of income but not on sustained
- This rate can be written as A = (1 - a - _y)(n_* + + + 6).
- In this case, the impact of inflation on growth in the long run should be evaluated by estimat- 1 I. See Thornton (1996) for a discussion of this issue.
ing investment equations.
321 Does^ Inflation Harm Economic^ Growth
growth (represented by +). To examine the latter possibility, we shall also con- sider an alternative specification, (4'), which allows for the influence of infla- tion on the long-run growth rate:'*
(4') A, = A,exp[(+ + +'n)tlexP(Fln)
such that the equation to be estimated would be represented by
yT+T - y T = (+ + +a). + (1 - e-")[O - y r + (+ + +'n)T
- pIT1+ p-'(as& + ys;, - (a + y)log(n? + + + S))].
In section 8.3 we estimate the elasticity of growth with respect to inflation in
models (3,( 5 ' ) , and (6).
Our observations are four-year averages of OECD annual variables. Our data set is described at length by Dabin, DomCnech, and Molinas (1997), who use OECD 1990 purchasing power parities to homogenize OECD national ac- counts from 1960 to 1992. When making real income comparisons among a set of countries, we must be aware of the properties of the data elaborated for that purpose. In particular, the more transitive we want our comparisons to be, the more the reference basket of goods has to depart from the most represen- tative sample of items for each country. Since we restrict the analysis to the OECD, we avoid the use of data sets designed to homogenize information from a much larger set of countries (such as the one in the Penn World Tables, Mark 5, PWT 5; Summers and Heston 1991).
8.3 Estimation of the Effect of Inflation
Tables 8.1 and 8.2 show the instrumental variables estimates of the steady state and convergence equations, using one- and two-period-lagged regressors as instruments. The results are quite robust, both in the linear and in the nonlin- ear specifications, as regards the effect of inflation. Linear models (eq. [ 5 ' ] ) are shown in table 8.1. The models in columns (1) and (2) of table 8.1 corre- spond to different versions of the convergence equation. As predicted by the neoclassical model, the parameter of initial per capita income is negative and highly significant, both when steady state variables are included (conditional convergence) and when they are not (unconditional convergence). In column (2), the coefficients of the input accumulation rates have the expected sign, although the one for human capital is nonsignificant. The estimated parameter of the trend, which according to the theoretical model is approximating the rate of technological progress, has an unexpected negative sign.13On the other
12. This is the specification proposed by Motley (1994). The variability of inflation is excluded in order to simplify the'expression. 13. A possible interpretation for this result is that the trend may be capturing the process of sustained reduction in the rate of growth of per capita income suffered by OECD countries during part of the sample period. We have tried alternative characterizations of technological progress:
323 Does^ Inflation^ Harm Economic^ Growth
Table 8.2 Nonlinear Models: Equation (5) Dependent Variable
Y* AY AY AY (1) (2) (3) (4)
(Y
Y
-2. (2.46)
(1.31)
(6.30) 0.05 1 (3.35)
-0. (4.00)
-4. (2.84)
(2.87)
(3.18)
-0. (2.57)
-0. (2.03)
-1. (1.61)
(2.56)
(2.54)
-0. (1.79)
-0. (3.95)
-0. (0.13)
(3.45)
(1.13)
-0. (1.91)
(2.10) -0. (2.38)
Note: Estimation method: see note to table 8.1. Numbers in parentheses are absolute t-ratios.
of the accumulation rates in the steady state equation (col. [l]) are quite far from those usually obtained in the empirical literature, the low value of ci being particularly remarkable. The effect of inflation is negative and significant. The
estimated coefficients in the convergence equation (col. [ 2 ] ) look^ more reason-
able, pointing toward a technology of similar factor shares ({ 1/3, 1/3, 1/3}), with an implicit rate of convergence around 2.7 percent. Again, the effect of inflation is negative and significant. Additionally, some tests of the sensitivity of the inflation coefficient to the sample definition have been performed in order to ascertain whether the nega- tive correlation between inflation and income is driven by the presence of some high-inflation countries. The most noticeable change in the estimated coeffi- cient takes place when Iceland is excluded from the sample. In this case (col. [3] of table 8.2) the correlation between inflation and growth is almost twice as high as when it is included. This is not surprising because Iceland, the coun- try with the second highest average inflation within the OECD, is also a high- income fast-growth economy, which may be generating a downward bias in the absolute value of the growth-inflation correlation. We have proceeded to estimate the model for different subsamples according to their average infla-
324 Javier And& and Ignacio Hernando
I
-0.1 \‘
-0.12 I^1 I^ I^ I^ I^ I^ I^1 1 2 3 4 5 6 7 8 9 Sample definition
Fig. 8. Nofe: The figure depicts the estimated coefficient (pl) for inflation in model (5) as well as the 95 percent confidence intervals (? 1.96 standard deviation band) for different sample definitions. Sample Dejinition: 1 -High-inflation countries (above OECD average). 2-High-inflation coun- tries (excluding Iceland). 3-OECD. 4-OECD excluding Turkey. 5-OECD excluding Turkey and Iceland. 6-OECD excluding Turkey, Iceland, and Portugal. 7-OECD excluding Turkey, Iceland, Portugal, and Greece. 8-OECD excluding Turkey, Iceland, Portugal, Greece, and Spain. 9-Low-inflation countries (below OECD average).
Sensitivity of inflation coefficient to sample definition: basic model
tion. The results, depicted in figure 8.1, indicate that if anything, the coefficient of inflation in the convergence equation is higher (in absolute value) and more significant for low-inflation countries. The negative effect of inflation on per capita income seems to be robust both in the steady state and in the convergence equation. Although the negative influence of inflation on per capita income is well established, its effect on the sustainable growth rate is less clear. If the inflation rate is a determinant of steady state per capita income ( y * ) it should also appear in the convergence equation. But it is not clear whether the negative coefficient in this equation points to an effect on the level or on the growth rate of output. To discriminate between these effects we have estimated equation (6), allowing for an effect of inflation both on the steady state level of income (k) and on the permanent
component of the growth rate (4’). Both these coefficients are negative and
significant when they are introduced individually, but when they are jointly included in the model (col. [4] of table 8.2) the effect on the trend component takes an unexpected positive sign. This would indicate that the negative effect of inflation impinges on the level of per capita income but not on the sustain-
326 Javier And& and Ignacio Hernando
Table 8.3 Convergence Equation with Individual Effects: Equation (5')
Dependent Variable: Ay
R U
JI, (LI)'
4% (HI)'
JI, (HI-ICL)'
-0. (1.22) -0. ( 2. 96 ) -0. (5.15)
( 2. 34 )
-0. ( 2. 34 ) -0. ( 1. 90 )
36
053
54 (2.03)
( 3. 16 ) -0. (4.70) -0. ( 1. 54 ) -0. ( 0. 28 )
- 21 ( 0. 98 ) -0. ( 1. 16 )
-0. ( 2. 01 ) -0. ( 0. 84 ) -0. (1.83)
( 6. 44 )
( 3. 26 )
( 5. 44 )
-0.
-0. ( 2. 29 )
24 ( 2. 95 )
( 3. 04 ) -0. ( 5. 45 ) -0. ( 0. 84 ) -0. ( 0. 25 ) -0. ( 0. 07 ) -0. ( 2. 47 )
-0.01 1 ( 5. 18 ) -0. ( 0. 87 ) -0. ( 1. 88 )
( 3. 22 )
( 3. 26 ) -0. ( 6. 99 ) -0. ( 1. 24 ) -0. (0.70) -0. (0.11) -0. ( 3. 99 )
- 048
Note: Estimation method in col. (1) is random effects (instrumental variables); instruments are first and secdnd lags of the regressors. Estimation method in cols. ( 2 )-( 5 ) is country-dummies instrumental variables; instrumentsare as in table 8. 1 plus country dummies and inflation variabil- ity. Dummy variables included: Cols. ( 2 )and ( 3 ), one for each country except Australia. Col. 4, one for each of the following countries: Canada, Switzerland, Germany, Spain, the United King- dom, Finland, Greece, Ireland, Iceland, Luxembourg, New Zealand, Portugal, Turkey, and the United States. Col. ( 5 ), one for each of the following countries: Iceland, Spain, Greece, and Tur- key; and one for each of the following country groups: Ireland and Portugal; Canada and Germany; Switzerland,Luxembourg, and the United States; and Finland, New Zealand, and the United King- dom. Numbers in parentheses are absolute r-ratios. "HI-sample of six countries with inflation rates above the OECD average; LI-OECD excluding HI countries; HI-ICL-HI countries excluding Iceland.
the reasons to include country-specific effects in the model suggest that the assumption of noncorrelation among these and the regressors might not be appropriate in this setting. Thus, in what follows, we focus on the fixed effects estimates, which we compute including dummies in the linear convergence equation. All the models have been estimated by instrumental variables. When we add a dummy variable for each country (col. [ 2 ] ) the explanatory power of most regressors changes, as compared with the models in the previous section. In particular, while inflation still has a negative effect on income, its t-statistic
327 Does Inflation Harm Economic Growth
is now lower (-1.16).” The changes in the rest of the model are far more radical, though. First, whereas the negative trend coefficient was an unappeal- ing feature of the models in section 8.3, this coefficient now becomes positive and significant, with a reasonable point estimate of 0.04. Second, the point estimates of the technological coefficients are now either nonsignificant or wrongly signed. In fact, excluding the accumulation rates from the equations, the negative correlation between growth and inflation becomes highly signifi- cant with a f-statistic of -2.29 (col. [3]). Finally, several country dummies are not different from zero, which means that the model might be overparame- trized. The search for a more parsimonious specification proceeds along the follow- ing steps. Starting from the model with a dummy variable for each country, the nonsignificant dummy variables are removed, setting aside the one with the lowest t-statistic each time. As^ a second step, these excluded variables are
added again, one at a time, retaining those with f-ratios greater than 1.5.16Ev-
ery time a dummy variable is added back into the model, the process is reini- tiated. This procedure does not involve the analysis of every single possible specification according to all the combinations of country-specific constants. However, it provides a model selection procedure that allows us to test, at least twice, the marginal significance of each dummy variable: first against a more general model (with all the country-specific dummies) and next against a more
restricted one. The model in column (4) summarizes the final outcome of this
specification process. The results do not change very much from those in col- umn (l), except in that now the coefficient of the inflation rate is negative and significant and its size is similar to that obtained for the model without individ- ual effects. Furthermore, this result is quite robust to the set of country-specific dummies included in the regression. The same search process has also been carried out for different subsamples with different average inflation rates. The point estimates of the inflation coefficient, along with its confidence interval, are depicted in figure 8.2. The coefficient of inflation turns out to be larger and more significant whenever high-inflation countries are not considered. Hence, as was the case in models without country dummies, the estimated correlation between inflation and growth (or income) does not depend on the presence of a group of high-inflation countries in the sample. Taking column (4) of table 8.3 as a starting point, in the model in column (5) individual dummies are clustered into country-group dummy variables. The
- It must be noted that the fixed effect estimate of the coefficient of inflation is still significant at the 5 percent level if we focus on the low-inflation countries (LI). This coefficient is lower and weakly significant for the subsample of all countries but Iceland (HI-ICL) with inflation above the OECD average.
- If the threshold level of the t-ratio is 2.0, the final specification is more parsimonious. Never- theless, the estimated long-run coefficient of inflation does not depart very much from that in col. (4).
329 Does Inflation Harm Economic Growth
rate. The estimated individual effects reveal a systematic pattern that, if ig- nored, could have led to a bias in the estimated effect of inflation. The individ- ual effect is strongly correlated with the level of per capita income achieved at the end of the sample period. Thus, omitting the individual effect, the model would underestimate the growth of the richest countries and overestimate that of the poorest countries. Since there is a negative correlation at the OECD level between per capita income in 1993 and the average inflation rate, excluding the individual effects is a source of potential upward bias in the estimation of the effect of inflation. Indeed, although the estimated coefficient of inflation re- mains largely unchanged, compared with that in table 8.1, there is nevertheless a significantchange in the point estimate of the long-run effect of inflation once country-specific dummies are included in the model. The coefficient of initial GDP is now almost five times as large as the one in tables 8.1 and 8.2, thus the estimated long-run cost of inflation is now lower. A permanent increase of 1 percentage point leads to a 0.75 percent permanent fall in output. This time, though, the transition period is much shorter because a higher coefficient of initial GDP means that convergence to the steady state is much faster too.
. Although OECD economies have certain common institutional features, their inflation performances are rather different. Once we have a more accurate estimate of the long-run cost of inflation we can address the issue of whether this cost varies according to the level of inflation or not. The different perspec- tives adopted to analyze the linearity of the inflation effect have led to contra- dictory results. For instance, Barro (1993, estimating different coefficients for different levels of inflation, finds a greater effect of inflation on growth the greater the inflation 1 e ~ e l. l ~Motley (1994), estimating the growth model for different subsamples, concludes the opposite. We have tried these two ap- proaches in equation (5’) and found that they also yield somewhat different results for the OECD, although the coefficients of inflation in different sub- samples were not very precisely estimated. In general, though, the coefficient corresponding to lower inflation rates tends to be higher, although with a lower r-ratio. This would indicate that the benefits of lower inflation are indeed higher at low rates, although the functional form might be inappropriate to capture this result. As an alternative, we have estimated the basic model allowing for a nonlinear effect of inflation on growth. When IT and IT^ are included, both coefficients are significant while the positive coefficient on IT^ indicates that the marginal cost of inflation is positive but decreasing with its level. Two alternative specifications that allow for a falling marginal cost of inflation have also been tried. In these, inflation is represented by log IT and the ratio a/( 1 + a),2o respectively. In all the specificationstried (with country dummies, exclud- ing Iceland, and so on) these equations perform better than the ones with the level of inflation.
- Although the null of linearity cannot be rejected (see also Barro 1996).
- Gylfason and Herbertsson (1996) propose this nonlinear transformation.
330 Javier And& and Ignacio Hernando
Table 8.4 Linearity of the Inflation Effect
Elasticity of Income with Respect to Inflation in Estimates of Linear Version of Inflation Level Convergence Equation A. Whole Sample Estimates with Specific Inflation Coeficients” Low inflation -0. (2.33) Medium inflation -0. (2.58) High inflation -0. (4.23) B. Subsample Estimatesb Low inflation -0. (2.59) High inflation -0. (2.15) Very low inflation -0. (1.82) Very high inflation -0. (1.61)
Note: Estimation method: see note to table 8.1. Numbers in parentheses are absolute t-ratios. ‘Low inflation-observations with inflation lower than 6 percent; medium inflation-observations with inflation between 6 percent and 12 percent; high inflation-observations with inflation greater than 12 percent. b L o ~inflation-countries with average inflation lower than the median; high inflation-countries with average inflation greater than the median; very low inflation-eight countries with the lowest inflation; very high inflation-eight countries with the highest inflation.
A further test for linearity has been carried out in the model in log T. In panel A of table 8.4, a different coefficient is allowed for log IT depending on its level. These elasticities are always negative and significant but not statisti- cally different. As an alternative approach, the homogeneity assumption may be relaxed by estimating the convergence equation for different subsamples. This approach allows all the parameters, and not only the coefficient of infla- tion, to vary across subsamples. The results are summarized in panel B of table 8.4. The effect of inflation is negative and significant both for low (and very low) as well as for high (and very high) inflation countries, and the coefficient of log IT is similar across different subsample specifications.21The results of
these two approaches lead us to conclude that the elasticity of income with
respect to inflation does not change significantly with the level of inflation. If anything, this tells us that it pays more in a low-inflation country than in a high-inflation one to reduce the inflation rate by a given amount. By the same
- The coefficient of initial GDP is also similar across the specifications in table 8.4, panel B. Thus the hypothesis of homogeneity in the long-run elasticity cannot be rejected either.
332 Javier And& and Ignacio Hernando
Table 8.6 Per Capita Income Gain from Reducing Inflation: Steady State and Present Value (percent) Basic Model Country Effects Steady state per capita income gain 2.5 0.
x 2.5 13
Half-life per capita income gain 1.2Y 0.375b Present value:' discount rate 4% 0.32 0. Present value:' discount rate 5% 0.23 0.
aHalf-life is 30 years. bHalf-life is^ 7 years. <Discounted present value of half-life gain (expressed in percentage points of steady state per capita income).
estimated in the basic model (col. [3], table 8.2) with that obtained in the model with country-specific effects (col. [4], table 8.3). The estimated benefit from a permanent reduction in the inflation rate by 1 percentage point is higher in the former (2.5 vs. 0.75 percent). Nevertheless, since this is a steady state effect and the convergence rates also differ across models (2.5 vs. 13 percent), the relevant comparison should be made in present value terms, which makes the outcome depend on the discount rate. According to the figures in this example, for discount rates slightly above 4 percent the benefit of disinflation is larger in models with faster dynamics, despite the lower coefficient of the inflation rate in the convergence equation. Hence, the present value of the per capita income gain might well be within the range of those found in other studies.
8.5 Analysis of Causality
The models studied in previous sections can generate a nonnegligible bias in the estimation of the influence of inflation on growth by focusing on the con- temporaneous correlation between these two variables. Inflation and growth are the joint outcome of the way an economy responds to different shocks. If demand shocks predominate, a positive association between GDP growth and inflation can be expected, whereas the association will be negative in response
to supply shocks. Also, even if we consider the possibility of a true influence
of one variable over the other, the theoretical literature presents arguments in favor of causality in both directions. For this reason, the contemporaneous cor- relation between growth and inflation may not be very informative as to the existence and magnitude of a real cost associated with inflation. In fact, it might be the case that the estimated negative correlation between inflation and growth is driven by the predominance of negative supply shocks during the sample period. To test this possibility we have estimated the linear version of the convergence equation for two periods: 1961-72 and 1989- 92, during which demand shocks predominated, and 1973-88, during which
333 Does Inflation Harm Economic Growth
Table 8.7 Inflation Effect for Different Periods
Estimation
Coefficient of Inflation in Linear Version of Convergence Equation' A. Demand Shocks Predominance Period: 1961-72 and 1989- Ordinary least squares -0. (2.33) Instrumental variablesb -0. (2.52) B. Supply Shock Predominance Period: 1973- Ordinary least squares -0. (3.85) Instrumental variablesb -0. (2.14)
Nore: Numbers in parentheses are absolute f-ratios. "Eq. ( 5 ' ) excluding the trend. bInstmments are constant, first- and second-order lags of the regressors, and second lag of the dependent variable.
supply shocks were probably more ~ignificant.~~The results of this split are shown in table 8.7, where we present only the coefficient on inflation for both the ordinary least squares (OLS) and the instrumental variables (IV) estima- tions. As expected, the IV coefficient is higher (lower), in absolute value, than the OLS coefficient for the first (second) period given the nature of the ex- pected bias in each case. But in all cases, the coefficients are negative and significant, meaning that the negative supply shocks that hit the OECD econ- omies during most of the second half of the sample period are not primar- ily responsible for the estimated negative correlation between inflation and growth. If this had been the case, we ought to find a positive coefficient for the first period, at least in the OLS estimation. The finding of negative coefficients for both periods strengthens the view that there is indeed a genuine negative effect of inflation on growth that does not rely on the existence of supply shocks determining simultaneously inflation and growth. In order to pursue this issue more thoroughly, this section analyzes the statis- tical causality, as formulated by Granger, of inflation to growth and vice versa. This perspective is broader than that of convergence equations in several ways. First, the analysis of causality focuses on the study of noncontemporaneous effects of one variable on the other. This is precisely the influence of inflation on growth predicted by the theoretical models: an influence that does not oper- ate in the short run but that takes time to show instead. Second, in using a more flexible specification, we avoid the imposition of the parametric restrictions of the neoclassical growth model, which might make the correlation that concerns
- Similar results were obtained when we split the period up in other two parts: 1961-76, for demand shock predominance, and 1977-92, for supply shock predominance.
