Professional Documents
Culture Documents
Journal of Development Economics: Shu Lin, Haichun Ye
Journal of Development Economics: Shu Lin, Haichun Ye
A R T I C L E I N F O A B S T R A C T
Article history: We evaluate the treatment effect of inflation targeting in thirteen developing countries that have adopted
Received 4 October 2007 this policy by the end of 2004. Using a variety of propensity score matching methods, we show that, on
Received in revised form 13 March 2008 average, inflation targeting has large and significant effects on lowering both inflation and inflation
Accepted 28 April 2008
variability in these thirteen countries. However, the effectiveness of inflation targeting on lowering inflation
is found to be quite heterogeneous. The performance of a given inflation targeting regime can be affected by
JEL classification:
country characteristics such as government's fiscal position, central bank's desire to limit the movements of
E4
E5
exchange rate, its willingness to meet the preconditions of policy adoption, and the time length since the
policy adoption.
Keywords: © 2008 Elsevier B.V. All rights reserved.
Inflation targeting
Treatment effect
Propensity score matching
Developing countries
0304-3878/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.jdeveco.2008.04.006
S. Lin, H. Ye / Journal of Development Economics 89 (2009) 118–123 119
Table 1 Table 2
Inflation targeting developing countries and starting years Control group countries
Table 3
Probit estimates of propensity scores
Baseline model No hyperinflation episodes Using conservative starting years Post-1990 sample Adding DEBTGDP Adding CBTOR5
CPIG_1 −8.484⁎⁎⁎ −8.463⁎⁎⁎ −14.497⁎⁎⁎ −7.778⁎⁎⁎ − 3.519⁎⁎⁎ −3.815⁎⁎⁎
(1.492) (1.499) (2.057) (1.555) (1.151) (0.882)
OPEN −0.007⁎⁎⁎ −0.007⁎⁎⁎ −0.007⁎⁎⁎ −0.008⁎⁎⁎ − 0.005⁎⁎⁎ −0.003
(0.001) (0.001) (0.002) (0.002) (0.002) (0.002)
BMG −1.763⁎ −1.704⁎ −1.926⁎ −1.859⁎ − 2.989⁎⁎⁎ −0.481
(1.003) (1.007) (1.148) (1.057) (0.850) (0.540)
FIX −1.635⁎⁎⁎ −1.634⁎⁎⁎ −2.020⁎⁎⁎ −1.727⁎⁎⁎ − 1.582⁎⁎⁎ −0.844⁎⁎⁎
(0.169) (0.168) (0.195) (0.176) (0.220) (0.239)
GDPPCG 2.616 2.460 −0.241 2.205 4.856⁎⁎ 2.483
(1.772) (1.830) (1.840) (1.949) (2.448) (1.975)
DEBTGDP 0.0006
(0.003)
CBTOR5 −0.516
(0.494)
No. of observations 831 723 831 508 421 495
Pseudo R2 0.35 0.31 0.44 0.28 0.30 0.15
Notes: Constant terms are included but not reported. Robust standard errors are reported in parentheses. ⁎, ⁎⁎, and ⁎⁎⁎ indicate the significance level of 10%, 5%, and 1%, respectively.
measure of openness to trade in this group.8 Since exchange rate on lowering both inflation and inflation variability in developing
targeting is more attractive to countries that have already adopted this targeting countries.
policy and to countries that are more open to trade, we expect to see
negative coefficients on these variables. 3.3. Robustness checks
The results are reported in the first column of Table 3. All estimated
coefficients have the expected signs. We find that the lagged inflation This subsection checks the robustness of our empirical results.
rate, broad money growth, openness, and exchange rate regime Since twenty-six countries (including both targeting and non-
systematically affect a country's targeting decision. The estimated targeting) in our dataset have experienced hyperinflation (defined
coefficients on these variables are all negative and significant, as an annual inflation rate of 40% or higher), one may suspect that our
indicating that countries with higher previous inflation, higher results might be driven by those extreme values of inflation. Our first
money growth, higher level of openness to trade, or fixed exchange robustness check, therefore, is to drop those hyperinflation episodes.
rate regimes are less likely to adopt inflation targeting. Real GDP per The new probit estimates of the propensity scores are shown in the
capita growth is insignificant. The overall fit of the regression is second column of Table 3.The results are very similar to those reported
reasonable with pseudo R2 around 0.35. in the first column. CPIG_1, BMG, OPEN, and FIX are all negative and
significant while GDPPCG is still insignificant. The pseudo R2 of the
3.2. Results from matching new regression is about 0.31. The new matching results are reported in
the second rows of Tables 4 and 5. Excluding hyperinflation episodes
Before applying the matching methods, we want to make sure that does not change our results. The estimated ATTs on inflation and
our treated and control units share the same support so that they are inflation variability are all negative and significant. The average
comparable. We sort all the observations by their estimated estimated ATT on inflation across different matching methods is
propensity scores and then discard all the control units whose about-2.64% in terms of the annual inflation rate.
estimated propensity scores are lower than the lowest score among Our second robustness check is to examine if our results are
the treated units. sensitive to different starting time of inflation targeting using Rose's
The matching results based on the new sample are presented in (2007) conservative starting years. The probit estimates of propensity
the first rows of Tables 4 and 5.9 Table 4 reports the estimated average score are shown in the third column of Table 3, and the matching
treatment effect on the treated (ATTs) on the level of inflation, and results are presented in the third rows of Tables 4 and 5. We obtain
Table 5 reports the estimated ATTs on inflation variability. The first very similar results: negative and significant ATTs on inflation and
two columns of each table show the results from one-to-one-nearest- inflation variability with an average ATT on inflation of −2.70%.
neighbor and three-nearest-neighbor matching. The next three Third, to check whether our results are robust to different sample
columns report the results from radius matching, with radii ranging periods, we also apply the propensity score matching analysis to the
from 0.01 to 0.04. Local linear regression matching and kernel post-1990 sample. Column 4 of Table 3 reports the results from the
matching results are shown in the last two columns of each table. probit regression, and the fourth rows of Tables 4 and 5 present the
The results are strong and robust. The estimated ATTs in the first results from matching. Dropping the pre-1990 observations does not
rows of Tables 4 and 5 are all found to be negative and statistically change our results either. The estimated ATTs are all negative and
significant. The average estimated ATT across different matching statistically significant.
methods is large in magnitude, around −2.97% in terms of annual Last, we want to check whether our results are sensitive to
inflation rate. The evidence suggests that, on average, inflation alternative specifications of the probit model. In the last two columns
targeting has quantitatively large and statistically significant impacts of Table 3, we add government debt to GDP ratio (DEBTGDP) and a
five-year central bank governor turnover rate (CBTOR5) as an inverse
8
Our primary data source for de facto exchange rate regimes is Reinhart and Rogoff proxy of central bank independence to the probit regressions,
(2004), which is available till 2001. We use the classifications reported by the IMF, respectively. Adding these variables, however, comes at a large cost.
which has switched from reporting de jure classifications to reporting de facto
Due to limited data availability, the sample sizes in the last two
classifications since 1999, after 2001. A fixed regime is defined as either a hard peg or a
soft peg.
columns are reduced almost by half. The problem is more severe in the
9
Matching estimates are obtained by using Stata command PSMATCH2 developed last regression. Since the five-year central bank governor turnover rate
by Leuven and Sianesi (2003). is available only till 1999, we now only have 25 targeting observations
S. Lin, H. Ye / Journal of Development Economics 89 (2009) 118–123 121
Table 4
Matching estimates of treatment effect on the level of inflation
Matching Methods
Notes: A 0.06 fixed bandwidth and an Epanechnikov kernel are used for kernel and local linear regression matching. Bootstrapped standard errors are reported in parentheses. They
are based on 500 replications of the data. ⁎, ⁎⁎, and ⁎⁎⁎ indicate the significance level of 10%, 5%, and 1%, respectively.
(which have dependent variables equal to one). It turns out that Table 6 reports the estimated treatment effect on level of inflation
neither of these two variables is significant in the probit regressions. based on the control function regression approach. In the first column,
Moreover, the overall fit of the last probit regression is substantially we simply run an OLS regression of inflation on a targeting dummy
worse compared to other probit specifications: the pseudo R2 falls to within the common support. The estimated coefficient, which catches
0.15 when the central bank governor turnover rate is included. the difference in mean inflation between targeting and non-targeting
Nevertheless, our matching results are quite robust to these countries, is negative and statistically significant. We then include the
alternative specifications. According to the last two rows of Tables 4 estimated propensity score obtained from our baseline probit model
and 5, most of the estimated ATTs on inflation and inflation variability as a control function in the second column. The estimated coefficient
remain negative and statistically significant. on the propensity score is statistically significant at the 1% level, which
All in all, the results tell a very consistent story: on average, the is strong evidence for the presence of self-selectivity. The coefficient
adoption of inflation targeting has significant effects on lowering both on the targeting dummy is still statistically significant but is much
inflation and inflation variability in these thirteen developing smaller in magnitude. The estimated average treatment effect after
targeting countries. controlling for self-selection is about 2.6 percentage points in terms of
annual inflation rate, which is close to the average treatment effect
4. Exploring the heterogeneity in treatment effects obtained from matching.
The heterogeneity feature of the treatment effect is explored in the
Unlike developed countries, developing countries usually exhibit next four columns of Table 6. In the third column, in addition to the
substantial heterogeneity in economic and institutional development. propensity score, we also include an interaction term of the targeting
Quite a few studies in the literature (e.g., Carare and Stone, 2006; dummy and the difference between the estimated propensity score
Fraga et al., 2003; Fry et al., 2000; Masson et al., 1997; Minella et al., and its sample average. This specification allows for varying treatment
2003; Mishkin, 1996; Mishkin, 2004; Mishkin and Savastano, 2001; effect.13 The coefficient on the targeting dummy gives an estimate of
Svensson, 2002) argue that the performance of an inflation targeting the treatment effect at the mean of the propensity score, and a
regime can be affected by these idiosyncrasies. Therefore, in addition significant interaction term is evidence of heterogeneity. The results
to evaluating the average treatment effects, it is also important to show that treatment effect at the mean of the propensity is significant
explore the heterogeneity feature of the effectiveness of adopting and is about −1.8% in terms of annual inflation rate. More interestingly,
inflation targeting in developing countries.10 the estimated coefficient on the interaction term is found to be −0.054
In this section, we explore four possible sources of heterogeneity and significant at the 5% level, implying that inflation targeting is
employing a control function regression approach. Our first investigation more effective in countries that have higher estimated propensity
is to test whether inflation targeting performs better in countries that have scores (better meet the preconditions of policy adoption). Other
higher estimated propensity scores (countries that better meet the things equal, inflation targeting leads to an additional reduction in
preconditions of policy adoption). Second, we examine whether, due to inflation by 0.54 percentage point for every 10 percentage point
the lagged effects of monetary policy, the time since the adoption of an increase in the propensity score.
inflation targeting regime affects the outcome.11 We then test if The fourth column tests the role of the time length since the policy
government fiscal position plays a role in the functioning of an inflation adoption by introducing an interaction term of the targeting dummy
targeting regime. Last, we also examine whether putting too much focus and the time length since the policy adoption.14 We find that the
on limiting exchange rate movements makes inflation targeting less interaction term is negative and significant at the 1% level, which
effective in lowering inflation (variability).12 confirms our suspicion that the performance of an inflation targeting
regime is deeply affected by the time length since the policy adoption.
10
The authors would like to thank an anonymous referee for this suggestion.
11
See Masson et al. (1997), Minella et al. (2003), and Fraga et al. (2003).
12 13
See Mishkin (2004) and Mishkin and Savastano (2001) for detailed discussions on See Wooldridge (2002) for detailed discussions of this specification.
14
how government fiscal position affects the functioning of inflation targeting and the As a general rule, both interacted variables should be included in the regression
problems with too strong a focus on limiting exchange rate movements under an individually. We do not include time in the regression because the interaction term is
inflation targeting regime. the same as the time variable.
122 S. Lin, H. Ye / Journal of Development Economics 89 (2009) 118–123
Table 5
Matching estimates of treatment effect on inflation variability
Matching methods
Notes: A 0.06 fixed bandwidth and an Epanechnikov kernel are used for kernel and local linear regression matching. Bootstrapped standard errors are reported in parentheses. They
are based on 500 replications of the data. ⁎, ⁎⁎, and ⁎⁎⁎ indicate the significance level of 10%, 5%, and 1%, respectively.
For one additional year of policy adoption, the treatment effect on Table 7 shows the estimated treatment effect on inflation
lowering inflation becomes 0.5 percentage point larger. variability. Although the treatment effect on inflation variability is
In the fifth column, we investigate the necessity of a fiscal still negative and significant in most cases, it is less heterogeneous
discipline to assure the success of the inflation targeting regimes by compared to the treatment effect on level of inflation. IT ⁎ TIME is the
including a debt to GDP ratio and its cross product with the targeting only interaction term that is statistically significant. Furthermore, we
dummy. As expected, the interaction term is found to be positive and observe that self-selection is a less important issue in evaluating the
statistically significant, meaning that inflation targeting policy is less treatment effect on inflation variability for the propensity score
effective in lowering inflation in countries that have higher debt to variable becomes insignificant in most of the regressions.
GDP ratios. The magnitudes of the estimated coefficients on the
targeting dummy and the interaction term suggest that inflation 5. Conclusions
targeting helps lower inflation only in countries with a debt to GDP
ratio below 71.4%. Previous work by Lin and Ye (2007) shows that inflation targeting
Finally, in the last column, we include a fixed exchange rate has no significant impacts on either inflation or inflation variability in
dummy variable FIX and an interaction term of IT ⁎ FIX to test whether developed countries. In this study, we evaluate the treatment effect of
putting too much focus on limiting exchange rate movements makes inflation targeting in thirteen developing countries that have adopted
inflation targeting less effective. We find that, interestingly, although a this policy by the end of 2004. Using a variety of propensity score
fixed exchange rate itself helps lower inflation, limiting exchange rate matching methods, we show that the average treatment effect of
movements under an inflation targeting regime actually makes inflation targeting on inflation (variability) is quantitatively large and
inflation targeting less effective in lowering inflation. In fact, the statistically significant in targeting countries. On average, the
results imply that the treatment effect on inflation would be zero if a adoption of inflation targeting has led to a fall in the level of inflation
country attempts to maintain a hard or soft peg under an inflation by nearly 3 percentage points. Our results suggest that the credibility
targeting regime. gain from an explicit announcement of an inflation target is more
Table 6 Table 7
Exploring heterogeneity in the treatment effect on level of inflation Exploring heterogeneity in the treatment effect on inflation variability
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
IT −0.038⁎⁎⁎ −0.026⁎⁎⁎ −0.018⁎⁎ 0.001 − 0.025⁎⁎ −0.030⁎⁎⁎ IT −0.018⁎⁎⁎ −0.018⁎⁎⁎ −0.017⁎⁎⁎ −0.012⁎⁎⁎ − 0.0065 − 0.020⁎⁎⁎
(0.006) (0.007) (0.008) (0.009) (0.013) (0.007) (0.003) (0.004) (0.003) (0.004) (0.0064) (0.005)
PS −0.046⁎⁎⁎ −0.031 −0.044⁎⁎⁎ − 0.043⁎⁎ −0.295⁎⁎⁎ PS 0.0003 0.004 0.0007 0.014 − 0.060⁎⁎⁎
(0.017) (0.022) (0.016) (0.019) (0.026) (0.012) (0.017) (0.012) (0.015) (0.0017)
P P
IT ⁎ (PS − PS) −0.054⁎⁎ IT ⁎ (PS − PS) −0.011
(0.028) (0.018)
IT ⁎ TIME −0.005⁎⁎⁎ IT ⁎ TIME −0.001⁎⁎⁎
(0.001) (0.0004)
DEBTGDP − 0.00004 DEBTGDP 0.00007
(0.0001) (0.00007)
IT ⁎ DEBTGDP 0.00035⁎⁎ IT ⁎ DEBTGDP − 0.0001
(0.00018) (0.0001)
FIX −0.148⁎⁎⁎ FIX − 0.036⁎⁎⁎
(0.010) (0.006)
IT ⁎ FIX 0.030⁎⁎⁎ IT ⁎ FIX 0.007
(0.010) (0.007)
Notes: Constant terms are included but not reported. Robust standard errors are Notes: Constant terms are included but not reported. Robust standard errors are
reported in parentheses. ⁎, ⁎⁎, and ⁎⁎⁎ indicate the significance level of 10%, 5%, and 1%, reported in parentheses. ⁎, ⁎⁎, and ⁎⁎⁎ indicate the significance level of 10%, 5%, and 1%,
respectively. respectively.
S. Lin, H. Ye / Journal of Development Economics 89 (2009) 118–123 123
substantial in the developing world. We also find that the effective- Johnson, D., 2002. The effect of inflation targeting on the behavior of expected inflation:
Evidence from an 11 country panel. Journal of Monetary Economics 49, 1521–1538.
ness of inflation targeting on lowering inflation is quite heterogeneous Kuttner, K.N., Posen, A.S., 1999. Does Talk Matter After All? Inflation Targeting and
within developing countries. The performance of a given inflation Central Bank Behavior. Federal Reserve Bank of New York. Staff Report 88.
targeting regime can be affected by country characteristics such as the Leuven, E., Sianesi, B., 2003. PSMATCH2: Stata module to perform full Mahalanobis and
propensity score matching, common support graphing, and covariate imbalance
time length since the policy adoption, government's fiscal discipline, testing. http://ideas.repec.org/c/boc/bocode/s432001.html Version 3.0.0.
central bank's desire to limit the movements of exchange rate, and its Lin, S., Ye, H., 2007. Does inflation targeting really make a difference? Evaluating the
willingness to meet the preconditions of policy adoption. treatment effect of inflation targeting in seven industrial countries. Journal of
Monetary Economics 54, 2521–2533.
Masson, P., Savastano, M., Sharma, S., 1997. The scope for inflation targeting in
Acknowledgments developing countries. IMF Working Papers 97/130, October.
Minella, A., Freitas, P., Goldfajn, I., Muinhos, M., 2003. Inflation targeting in Brazil:
Constructing credibility under exchange rate volatility. Journal of International
The authors would like to thank Lant Pritchett and two anonymous
Money and Finance 22, 1014–1040.
referees for their valuable comments. Any remaining errors are ours. Mishkin, F.S., 1996. Understanding financial crises: A developing country perspective.
In: Bruno, M., Pleskovic, B. (Eds.), Annual World Bank Conference on Development
References Economics. World Bank, Washington, DC, pp. 29–62.
Mishkin, F.S., 1999. International experiences with different monetary policy regimes.
Journal of Monetary Economics 43, 579–605.
Ammer, J., Freeman, R.T., 1995. Inflation targeting in the 1990s: The experiences of New
Mishkin, F.S., 2004. Can inflation targeting work in emerging market countries? NBER
Zealand, Canada, and the United Kingdom. Journal of Economics and Business 47 (2),
Working Papers 10646.
165–192.
Mishkin, F.S., Posen, A.S., 1997. Inflation targeting: Lessons from four countries. Federal
Ball, L., Sheridan, N., 2005. Does inflation targeting matter? In: Bernanke, B.S.,
Reserve Bank of New York Economic Policy Review, pp. 9–110.
Woodford, M. (Eds.), The Inflation-targeting Debate. University of Chicago Press.
Mishkin, F.S., Savastano, M., 2001. Monetary policy strategies in Latin America. Journal
Bernanke, B.S., Launach, T., Mishkin, F.S., Posen, A.S., 1999. Inflation Targeting: Lessons
of development economics 66 (2), 415–444.
from the International Experience. Princeton University Press, Princeton, NJ.
Reinhart, C., Rogoff, K., 2004. The modern history of exchange rate arrangements: A
Carare, A., Stone, M., 2006. Inflation targeting regimes. European Economic Review 50,
reinterpretation. Quarterly Journal of Economics 119, 1–48.
1297–1315.
Rose, A., 2007. A stable international monetary system emerges: Inflation targeting is
Fraga, A., Goldfajn, I., Minella, A., 2003. Inflation targeting in emerging market
Bretton Woods, reversed. Journal of International Money and Finance 26, 663–681.
economies. NBER Working Papers 10019.
Svensson, L., 1997. Inflation forecast targeting: Implementing and monitoring inflation
Fry, M., Julius, D., Roger, S., Mahadeva, L., Sterne, G., 1994. Key issues in the choice of
targets. European Economic Review 41, 1111–1146.
monetary policy framework. In: Mahadeva, L., Sterne, G. (Eds.), Monetary Policy
Svensson, L., 2002. Inflation targeting: Should it be modeled as an instrument rule or a
Frameworks in a Global Context. Routledge.
targeting rule? European Economic Review 46, 771–780.
Ghosh, A., Gulde, A., Wolf, H., 2003. Exchange Rate Regimes: Choice and Consequences.
Truman, E., 2003. Inflation Targeting in the World Economy. Institute for International
MIT Press, Cambridge, MA.
Economics, Washington, DC.
Goncalves, C., Salles, J., 2008. Inflation targeting in emerging economies: What do the
Wooldridge, J., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press,
data say? Journal of Development Economics 85 (12), 312–318.
Cambridge, MA.
Groenveld, J.M., 1998. Inflation Patterns and Monetary Policy: Lessons for the European
Central Bank. Edward Elgar, Cheltenham.