Journal Issue
Share
Article

Does Inflation Targeting Matter?

Author(s):
Niamh Sheridan, and Laurence Ball
Published Date:
June 2003
Share
  • ShareShare
Show Summary Details

I. Introduction

The performance of inflation-targeting regimes has been quite good. Inflation- targeting countries seem to have significantly reduced both the rate of inflation and inflation expectations beyond that which would likely have occurred in the absence of inflation targets. (Mishkin, 1999, p. 595).

[The U.K. data show] that not only has inflation been lower since inflation targeting was introduced, but that, as measured by its standard deviation, it has also been more stable than in recent decades. Moreover, inflation has been less persistent - in the sense that shocks to inflation die away more quickly - under inflation targeting than for most of the past century. (King, 2002, p. 2).

[O]ne of the main benefits of inflation targets is that they may help to “lock in” earlier disinflationary gains, particularly in the face of one-time inflationary shocks. We saw this effect, for example, following the exits of the United Kingdom and Sweden from the European Exchange Rate Mechanism and after Canada’s 1991 imposition of the Goods and Services Tax. In each case, the re-igniting of inflation seems to have been avoided by the announcement of inflation targets that helped to anchor the public’s inflation expectations and to give an explicit plan for and direction to monetary policy. (Bernanke and others, 1999, p. 288).

Economists have long sought the ideal framework for monetary policy. Since the early 1990s, many have come to believe they have finally found the right approach: inflation targeting. Proponents of this policy cite many benefits. Inflation targeting solves the dynamic consistency problem that produces high average inflation. It reduces inflation variability, and if it is “flexible” it can stabilize output as well (Svensson, 1997). Targeting locks in expectations of low inflation, which reduces the inflationary impact of macroeconomic shocks. For these reasons, many economists advocate inflation targeting for the U.S. Federal Reserve and the European Central Bank.

This paper attempts to measure the effects of inflation targeting on macroeconomic performance. We examine 20 member countries of the Organization for Economic Co-operation and Development (OECD), 7 that adopted inflation targeting during the 1990s and 13 that did not. Not surprisingly, economic performance varies greatly across individual countries, both targeters and nontargeters. But, on average, there is no evidence that inflation targeting improves performance as measured by the behavior of inflation, output, or interest rates.

If we examine inflation-targeting countries alone, we see that their performance improved, on average, between the period before targeting and the targeting period. For example, inflation fell and became more stable, and output growth also stabilized. However, countries that did not adopt inflation targeting also experienced improvements around the same times as targeters. This finding suggests that better performance resulted from something other than targeting.

For some performance measures, both inflation targeters and nontargeters improve over time, but the improvements are larger for targeters. For example, average inflation fell for both groups between the pretargeting and targeting periods, but the average for targeters went from above that of nontargeters to roughly the same. Similar findings have led authors such as Neumann and von Hagen (2002) to argue that inflation targeting promotes “convergence”: it helps poorly-performing countries catch up with countries that are already doing well. Our results, however, do not support even this modest claim of benefits from targeting. For many measures of performance, we find strong evidence of generic regression towards the mean. Just as short people have children who are, on average, taller than they are, countries with unusually high and unstable inflation tend to see these problems diminish, regardless of whether they adopt inflation targeting. Once we control for this effect, the apparent benefits of targeting disappear.

The rest of this paper comprises eight sections. Section II describes the countries and sample periods that we study, and Section III describes our methodology for measuring the effects of inflation targeting.

Sections IV and V present our results concerning inflation and output growth. We estimate the effects of inflation targeting on these variables’ average levels, variability, and persistence. There are occasional hints that targeting has beneficial effects and occasional hints of adverse effects, but overall it appears that targeting does not matter.

Section VI turns to the behavior of interest rates, and presents two main findings. First, inflation targeting has no effect on the level of long-term interest rates, contrary to what one would expect if targeting reduces inflation expectations. Second, targeting does not affect the variability of the short-term interest rates controlled by policymakers. At least by this crude measure, central banks respond neither more nor less aggressively to economic fluctuations under inflation targeting.

Section VII investigates the effects of targeting on several bivariate relations: the slope of the output-inflation trade-off, the inflationary effect of supply shocks (specifically changes in commodity prices), and the effect of inflation movements on expectations (as measured by OECD inflation forecasts). Here the results are imprecise, as it is difficult to estimate these relations over the short periods for which we have observed inflation targeting. However, the results suggest again that targeting has no important effects.

Section VIII compares our results with previous cross-country studies of inflation targeting. Finally, Section IX interprets our results. To be clear, we do not present a case against inflation targeting. We do not find that targeting does anything harmful, and we can imagine future circumstances in which it might be beneficial. Our results suggest, however, that no major benefits have occurred so far.

II. The Sample

This section describes the countries in our sample and the inflation-targeting and non- targeting periods that we examine.

A. Targeters and NonTargeters

We examine major developed, moderate-inflation economies. Specifically, we start with all members of the OECD as of 1990 (thus excluding the emerging-market economies that have joined since then). We delete countries that lacked an independent currency before the Euro (Luxembourg) or have experienced annual inflation over 20% since 1984 (Greece, Iceland, and Turkey). We are left with twenty countries, which are listed in Table 1. Previous macroeconomic studies using the same sample of countries include Layard et al. (1991) and Ball (1997).

Table 1.Starting Dates for Inflation Targeting and Constant Inflation Targeting Periods
CountryInflation

Targeting
Constant

Inflation

Targeting
Rationale for choice of starting dates
AustraliaQ4 1994Q4 1994In September 1994, the Governor of the Reserve Bank of Australia announced that “underlying inflation of 2 to 3 per cent is a reasonable goal for monetary policy.” See Bernanke et al. (1999, pp. 218–220) for further discussion.
CanadaQ1 1992Q1 1994The first target range was announced by the Bank of Canada in February 1991: 2 to 4 percent over 1992 (i.e. December 1991 to December 1992). In December 1993, a range of 1 to 3 percent was established for 1994, and the range has remained constant since then.
FinlandQ1 1994Q1 1994In February 1993, the Bank of Finland stated its intention to “stabilize the rate of inflation permanently at the level of 2% by 1995.” It appears that they were referring to year-over-year inflation measured at the start of 1995; thus the period covered by the first target begins at the start of 1994.
New ZealandQ3 1990Q1 1993A target of 3–5% over 1990 was announced in April 1990. A target of 0–2% for 1993 was announced in February 1991. The target range has remained roughly unchanged since then (but see footnote 2 in the text).
SpainQ2 1995Q1 1994aThe first target, announced in December 1994, was for year- over-year inflation of 3.5–4% “by early 1996.”
SwedenQ1 1995Q1 1995The Riksbank announced in January 1993 that it aimed “to limit the annual increase in the consumer price index from 1995 onwards to 2 percent.” This target applied to inflation over all of 1995, not to year-over-year inflation at the start of 1995 (Svensson, 1995).
United KingdomQ1 1993Q1 1993In October 1992 the Bank of England announced a 2.5% target, beginning immediately.
Non-IT countriesQ3 1993Q1 1994The starting dates were computed as averages of the starting dates for inflation targeting or constant inflation targeting countries.

Spain is an inflation targeter but not a constant inflation targeter. Q1 1994 is the start date of the constant-targeting period for non- constant targcters.

Spain is an inflation targeter but not a constant inflation targeter. Q1 1994 is the start date of the constant-targeting period for non- constant targcters.

Seven of the countries in our sample adopted inflation targeting before 1999: Australia, Canada, Finland, Spain, Sweden, the United Kingdom, and New Zealand. For each country, we define the beginning of targeting as the first full quarter in which a specific inflation target or target range was in effect, and the target had been announced publicly at some earlier time. This definition of targeting is more stringent than that of previous authors, such as Beraanke and others (1999) and Scheater and others (2000). These authors often date the start of targeting at the point when targets were first announced, even if they were implemented with a delay. In other cases, targeting is said to begin when the central bank retrospectively said it did, even though it was not announced at the time. Our view is that many of the intended effects of targeting, such as those working through expectations, depend on agents knowing that they are currently in a targeting regime.

As an example of our dating, consider Sweden. Sweden announced its shift to inflation targeting during 1993, so Bernanke et al. and Scheater et al. date the regime from then. However, the first announced target was 2 percent for inflation over the twelve months to December 1995. We choose the first quarter of this period, 1995:1, as the beginning of the targeting regime. Table I gives the starting dates of targeting for the other countries along with brief explanations for our choices. The starting dates range from 1990:3 for New Zealand to 1995:2 for Spain.

The targeting period lasts through 2001 for all countries except Finland and Spain, where it lasts through 1998 because of the advent of the Euro. For each country, we compare the targeting period to two pre-targeting periods, a longer one that begins in 1960 and a shorter one that begins in 1985. The last quarter of the pre-targeting period is the last full quarter before targeting began (either the quarter before the start of the targeting period or two quarters before, depending on whether targeting began at the start of a quarter or in the middle).

Throughout, we compare the seven inflation targeters to the other thirteen countries in the sample. Two of these countries have adopted inflation targeting recently: Switzerland in 1999 and Norway in 2000. We exclude these countries’ brief targeting periods from our sample and treat Switzerland and Norway as nontargeters. Following our approach for targeters, we compare pre-targeting periods starting in 1960 and 1985 to post-targeting periods. For the nontargeters, we define the post-targeting period as starting at the mean of the start dates for targeters, which is 1993:3. The post-targeting period ends in 1998 for Euro countries and 2001 for non-Euro countries besides Norway and Switzerland. Table 2 gives details of our dating.

Table 2.Sample Periods
CountrySample 1Sample 2Sample 3Sample 4Sample 5Sample 6
Australiastart of sample

end of sample
1960:1

1994:2
1985:1

1994:2
1994:4

2001:4
1960:1

1994:2
1985:1

1994:2
1994:4

2001:4
Canada1960:1

1991:4
1985:1

1991:4
1992:1

2001:4
1960:1

1993:3
1985:1

1993:3
1994:1

2001:4
Finland1960:1

1993:4
1985:1

1993:4
1994:1

1998:4
1960:1

1993:4
1985:1

1993:4
1994:1

1998:4
New Zealand1960:1

1990:1
1985:1

1990:1
1990:3

2001:4
1960:1

1992:4
1985:1

1992:4
1993:1

2001:4
Spain1960:1

1995:1
1985:1

1995:1
1995:2

1998:4
1960:1

1993:3
1985:1

1993:3
1994:1

1998:4
Sweden1960:1

1994:4
1985:1

1994:4
1995:1

2001:4
1960:1

1994:4
1985:1

1994:4
1995:1

2001:4
United Kingdom1960:1

1992:3
1985:1

1992:3
1993:1

2001:4
1960:1

1992:3
1985:1

1992:3
1993:1

2001:4
United States, Japan, Denmark1960:1

1993:2
1985:1

1993:2
1993:3

2001:4
1960:1

1993:3
1985:1

1993:3
1994:1

2001:4
Austria, Belgium, France,

Germany, Ireland, Italy,

Netherlands, Portugal
1960:1

1993:2
1985:1

1993:2
1993:3

1998:4
1960:1

1993:3
1985:1

1993:3
1994:1

1998:4
Norway1960:1

1993:2
1985:1

1993:2
1993:3

2000:4
1960:1

1993:3
1985:1

1993:3
1994:1

2000:4
Switzerland1960:1

1993:2
1985:1

1993:2
1993:3

1999:4
1960:1

1993:3
1985:1

1993:3
1994:1

1999:4
Note: Standard errors are in parentheses.Source: Authors’ calculations
Note: Standard errors are in parentheses.Source: Authors’ calculations

Of the thirteen non-targeting countries, eight joined the euro area in 1999. Previously, these countries were part of the European Monetary System (EMS), so their monetary policies focused on fixing exchange rates and meeting convergence criteria. Two of the nontargeters, Germany and Switzerland (one also in the EMS), followed policies based on money-supply targets. The remaining four countries did not follow any announced rule—they pursued the policy of “just do it” (Mishkin, 1999). In the results we report, we lump all non-targeting countries together and compare them to targeters. We have checked, however, whether there are systematic differences in performance among the non-targeting groups, and fail to find any. We have also performed our comparisons of targeters and nontargeters excluding all Euro countries (which leaves five targeters and five nontargeters). This produces no noteworthy changes in results.2

B. Constant Targeting

In addition to studying inflation-targeting periods, we examine periods in which countries are constant inflation targeters, meaning they have an unchanging target or target range. In some countries the target is always constant, but in others the constant-targeting period is preceded by a transitional period in which the target exceeds its final level. We examine constant-targeting periods because some benefits of targeting might not arise if the target changes. For example, proponents of targeting argue that it reduces the persistence of inflation movements, but a changing target causes permanent changes in inflation.3

Throughout this paper, we compare inflation targeters (IT) to nontargeters (NIT), and constant-inflation targeters (CIT) to non-constant-targeters (NCIT). Spain is an inflation targeter, but its target fell throughout its targeting period; when we split countries into CIT and NCIT, we put Spain in the second group. For both CIT and NCIT countries, we examine periods before and after the start of constant targeting. The start date of the post-targeting period for NCIT countries is the average start date for constant targeting in CIT countries.

Table 2 lists sample periods for each of the twenty countries. We call the two pre- inflation-targeting periods, those starting in 1960 and 1985, samples 1 and 2 respectively. Sample 3 is the post-targeting period. Samples 4 and 5 are pre-constant-targeting periods, and sample 6 is the post-constant-targeting period. While the distinction between IT and CIT is important in principle, our findings about economic performance in the pre- and post-targeting periods are similar in the two cases.

III. Methodology

We want to determine how inflation targeting (or constant targeting) affects dimensions of economic performance such as inflation, output growth, and interest rates. We examine each aspect of performance in turn, using a consistent methodology to measure the effects of targeting. Here we describe the methodology.

Suppose we are interested in how targeting affects a variable X — for example, X might be the average level of inflation or the variance of output growth. We first calculate X for each of our 20 countries in each of our six sample periods. Then, for each period, we calculate the average value of X for inflation targeters and nontargeters (or, for samples 4 through 6, constant targeters and non-constant targeters). These averages show whether X differs systematically across periods or across targeters and nontargeters.

As we have mentioned, many measures of economic performance improved on average between the pre-inflation-targeting and post-targeting periods. In most major economies, the period since the early 1990s has seen low and stable inflation and stable output growth. If we examine inflation targeting countries alone, there are clear economic improvements that one might be tempted to attribute to targeting. However, to learn the true effects of targeting, we must compare improvements in targeting countries to improvements in non-targeting countries.

As a first pass at this comparison, we use a standard “differences in differences” approach. For our sample of twenty countries, we run the regression

where Xpost is a country’s value of X in the post-targeting period, Xpre is the value in the pre- targeting period, and D is a dummy variable equal to one if the country is a targeter. We run several versions of this regression corresponding to different start dates for the pre-targeting period (1960 or 1985) and whether targeting means IT or CIT. The coefficient ai is meant to measure the effect of targeting on the variable X.

This regression can be misleading, however. For some versions of the variable X, the initial value, Xpre, is substantially different on average for inflation targeters and nontargeters. For example, average inflation in the pre-targeting period is higher for targeters. This fact is not surprising: a switch to targeting was most attractive to countries with poor performances under their previous policies. However, a problem arises because of regression to the mean. Poor performers in the pre-targeting period tend to improve more than good performers simply because initial performance depends partly on transitory factors. If inflation targeters are poor initial performers, they will improve more than nontargeters, even if targeting does not affect performance. The coefficient on the targeting dummy can be significant, producing a spurious conclusion that targeting matters.

As an analogy, consider the behavior of Major League batting averages. Suppose a crackpot sports consultant suggests that a hitter will perform better if he sleeps next to his bat at night. In reality, this idea does not work. Most .300 hitters merely chuckle at the consultant, but .220 hitters are desperate enough to try anything, and start taking their bats to bed. Because of regression to the mean, the low-average hitters who sleep with their bats will tend to improve more than the high-average hitters who leave their bats in their lockers. If the sports consultant regresses the change in a player’s average on a bat-in-bed dummy, he will find a significant effect. He will claim incorrectly that the evidence supports his theory.4

For readers who prefer math to baseball, the Appendix to this paper formalizes our argument. We assume that the variable X depends on a country effect, a period effect, a country- period effect, and possibly an inflation-targeting dummy. The presence of the country-period effect generates regression to the mean. If Xpre is correlated with the targeting dummy, as happens in practice, then regression (1) produces a biased estimate of the dummy coefficient.

Fortunately, there is a simple way to eliminate this bias: add the initial value of X to the differences regression. That is, we run

Including Xpre controls for regression to the mean. The coefficient on the dummy now shows whether targeting affects a country’s change in performance for a given initial performance. If ai is significant, then a targeter with poor initial performance improves more than a non-targeter with equally poor initial performance. This difference implies a true effect of targeting.

Once again, the Appendix formalizes our argument. Under the assumptions we make there, regression (2) produces an unbiased estimate of the dummy coefficient.

IV. Inflation

In a recent speech, the next Governor of the Bank of England posed the question “Ten Years of the Inflation Target: what has it achieved?” As quoted at the start of this paper, he suggests that targeting has reduced the average level, variability, and persistence of U.K. inflation. In contrast, we find little evidence in cross-country data that targeting has any of these effects.

A. Average Inflation

Table 3 presents our results concerning the average level of inflation. Inflation is measured by the annualized percentage change in consumer prices from the IMF’s International Financial Statistics (IFS). In Panel A of the table, we show average inflation in each of our twenty countries and six sample periods. For each period, we also show the averages across targeting and non-targeting countries. Panel B reports our estimates of equations (1) and (2) above.

Table 3.Mean Inflation Rate (Annualized)
Panel A
Sample 1Sample 2Sample 3Sample 4Sample 5Sample 6
Australia6.235.382.626.235.382.62
Canada5.354.371.625.163.831.58
New Zealand8.6210.231.948.087.482.00
Sweden6.415.381.016.415.381.01
United Kingdom7.545.502.437.545.502.43
Finland6.904.071.086.904.071.08
Spain9.165.932.499.356.123.06
United States4.823.722.474.803.662.47
Japan5.161.630.125.151.680.09
Denmark6.503.232.216.473.192.23
Austria4.302.721.774.292.721.64
Belgium4.642.531.654.632.531.55
France6.113.051.376.083.011.33
Germany3.402.241.653.402.251.59
Ireland7.853.132.117.823.132.05
Italy8.435.723.298.405.693.18
Netherlands4.411.582.194.401.642.12
Portugal11.9910.643.5411.9610.542.94
Norway6.264.932.206.224.812.28
Switzerland3.893.260.843.873.220.79
Averages
IT7.175.841.88
NIT5.983.721.95
CIT6.725.271.78
NCIT6.203.871.95
Panel BEquation (1)Equation (2)
Dependent Variable: Change in mean inflation between samples
(3) - (l)(3) - (2)(6) - (4)(6) - (5)(3) - (l)(3) - (2)(6) - (4)(6) - (5)
Constant-4.03

(0.46)
-1.77

(0.52)
-4.25

(0.47)
-1.92

(0.46)
0.42

(0.49)
1.12

(0.32)
0.52

(0.50)
1.01

(0.33)
Inflation targeting

dummy
-1.26

(0.78)
-2.19

(0.88)
-0.68

(0.86)
-1.57

(0.84)
-0.38

(0.33)
-0.55

(0.35)
-0.29

(0.33)
-0.51

(0.34)
Initial value-0.74

(0.08)
-0.78

(0.07)
-0.77

(0.07)
-0.76

(0.07)
Adjusted R-squared0.080.21-0.020.120.850.900.850.87
Note: Standard errors are in parentheses.Source: Authors’ calculations
Note: Standard errors are in parentheses.Source: Authors’ calculations

Not surprisingly, there is considerable cross-country variation in average inflation. In sample 2, for example (1985 to start of inflation targeting), average inflation ranges from double digits in New Zealand and Portugal to less than two percent in Japan and Netherlands. In almost every country, average inflation is lower in the targeting periods (samples 3 and 6) than in the pre-targeting periods. The cross-country variation is smaller in the targeting periods, as all inflation rates are under four percent.

Turning to cross-country averages, we see that the IT group had higher inflation than the NIT group before targeting was introduced. (Here and elsewhere, the comparison between the CIT and NCIT groups is similar.) For the shorter pre-targeting sample, average inflation is 5.8% for IT countries and 3.7% for NIT. In the targeting period, by contrast, average inflation is close to 1.9% for both groups. On average, targeters converged to the lower inflation levels of nontargeters.

This convergence result is echoed in the first part of Panel B, where we regress the change in average inflation on the targeting dummy. For the shorter pre-targeting sample, the coefficient on the dummy is –2.2: average inflation fell by 2.2 points more in targeters than in nontargeters. This coefficient is the same as the difference in differences of means between samples 2 and 3. The regression reveals that this inflation-targeting effect is statistically significant (t=2.5).

Inflation targeting is important if it really reduces average inflation by more than two percentage points. However, most of this apparent effect is illusory: it reflects the facts that targeters had high initial inflation, and there is regression to the mean. Panel B shows that regression to the mean is strong: when initial inflation is included in the inflation-change equation, its coefficient is −0.78. Controlling for this effect, the estimated effect of targeting is only −0.55, and its statistical significance is weak (t=1.57, p-value=0.14). Looking ahead, however, we will see that this result is one of our more positive findings about inflation targeting!

Note how much of the variation in inflation changes is explained by initial inflation: including this variable raises the R2’s from 0.2 or below to 0.9. Figure 1 illustrates this point by plotting the change in inflation from sample 2 to sample 3 against the level in sample 2. The Figure shows a tight relationship, confirming the strong role of regression to the mean. The targeting countries tend to have high initial inflation and large decreases, but the decrease for a given initial level looks similar for targeters and nontargeters.

Figure 1.Regression towards the mean

Source: Authors’ calculations

B. Inflation Variability

Tables 4 and 5 examine the variability of inflation, using the same format as the average- inflation table. Table 4 presents standard deviations of quarterly inflation, and Table 5 presents standard deviations of “trend inflation,” defined as a nine-quarter moving average. We examine trend inflation because targeters might stabilize this variable even if they cannot smooth out higher-frequency inflation shocks.5

Table 4.Standard Deviation of Inflation Rate
Panel A
Sample 1Sample 2Sample 3Sample 4Sample 5Sample 6
Australia4.623.513.014.623.513.01
Canada3.341.751.593.351.931.75
New Zealand5.837.421.705.887.211.78
Sweden3.993.621.573.993.621.57
United Kingdom5.702.801.345.702.801.34
Finland4.511.871.164.511.871.16
Spain5.802.001.385.852.071.64
United States3.271.640.943.261.650.96
Japan5.001.761.734.981.761.65
Denmark4.772.140.684.772.120.70
Austria2.701.361.182.691.341.15
Belgium3.311.541.203.311.511.23
France3.771.150.813.781.150.84
Germany2.322.851.022.312.811.05
Ireland6.521.541.046.501.521.06
Italy6.081.551.606.061.541.64
Netherlands3.401.710.753.391.720.71
Portugal9.213.862.509.183.841.52
Norway3.842.521.243.852.571.24
Switzerland2.732.610.892.722.570.89
Averages
IT4.833.281.68
NIT4.382.021.20
CIT4.673.491.77
NCIT4.482.011.16
Panel BEquation (1)Equation (2)
Dependent Variable: Change in the standard deviation of inflation between samples
(3) - (l)(3) - (2)(6) - (4)(6) - (5)(3) - (l)(3) - (2)(6) - (4)(6) - (5)
Constant-3.18

(0.41)
-0.82

(0.34)
-3.31

(0.43)
-0.85

(0.32)
0.50

(0.32)
0.92

(0.24)
0.79

(0.30)
1.01

(0.22)
Inflation targeting

dummy
0.03

(0.70)
-0.78

(0.58)
0.41

(0.78)
-0.87

(0.59)
0.41

(0.23)
0.31

(0.27)
0.59

(0.21)
0.50

(0.26)
Initial value-0.84

(0.07)
-0.86

(0.10)
-0.92

(0.06)
-0.93

(0.09)
Adjusted R-squared-0.060.04-0.040.060.890.830.920.92
Note: Standard errors are in parentheses.Source: Authors’ calculations
Note: Standard errors are in parentheses.Source: Authors’ calculations
Table 5.Standard Deviation of Trend Inflation Rate(9-quarter moving average)
Panel A
Sample 1Sample 2Sample 3Sample 4Sample 5Sample 6
Australia3.802.761.373.802.761.37
Canada2.890.440.532.880.920,53
New Zealand4.433.550.834.484.200.92
Sweden2.632.040.572.632.040.57
United Kingdom4.591.690.344.591.690.34
Finland3.541.260.283.541.260.28
Spain4.660.790.424.650.670.92
United States2.810.810.442.810.820.45
Japan3.711.060.683.701.040.70
Denmark2.850.950.272.870.990.27
Austria1.780.820.491.780.830.41
Belgium2.720.780.212.710.770.21
France3.350.320.373.360.350.39
Germany1.671.330.251.671.420.18
Ireland5.200.410.315.200.430.25
Italy5.350.541.105.340.561.06
Netherlands2.551.300.142.541.310.13
Portugal7.211.370.727.191.470.50
Norway2.511.920.332.531.960.33
Switzerland1.921.680.411.911.650.39
Averages
IT3.791.790,62
NTT3.361.020.44
CIT3.652.140.67
NCIT3.451.020.44
Panel BEquation (1)Equation (2)
Dependent Variable: Change in the standard deviation of trend inflation between samples
(3) - (l)(3) - (2)(6) - (4)(6) - (5)(3) - (l)(3) - (2)(6) - (4)(6) - (5)
Constant-2.92

(0.37)
-0.58

(0.20)
-3.00

(0.36)
-0.58

(0.20)
0.16

(0.18)
0.30

(0.13)
0.14

(0.19)
0.33

(0.13)
Inflation targeting

dummy
-0.25

(0.62)
-0.58

(0.33)
0.02

(0.65)
-0.90

(0.36)
0.15

(0.14)
0.08

(0.16)
0.21

(0.15)
0.10

(0.19)
Initial value-0.92

(0.05)
-0.87

(0.09)
-0.91

(0.05)
-0.89

(0.10)
Adjusted R-squared-0.050.10-0.060.220.950.840.950.85
Note: Standard errors are in parentheses.Source: Authors’ calculations
Note: Standard errors are in parentheses.Source: Authors’ calculations

There is no evidence whatsoever that inflation targeting reduces inflation variability. The standard deviations of inflation and trend inflation fall for all groups of countries during the targeting period. At all times, the standard deviations are lower for nontargeters than for targeters. Equation (1) suggests that targeters experience larger falls in standard deviations, but this result disappears when equation (2) controls for regression to the mean.

In fact, Table 4 suggests that, controlling for regression to the mean, inflation targeting raises the standard deviation of inflation. This effect is sometimes statistically significant. Nonetheless, this perverse result is likely a fluke (given the number of regressions we run, our tests should produce some Type I errors). Our robust finding is that inflation targeting has no beneficial effects.

C. Inflation Persistence

Finally, we examine the persistence of inflation movements. For each country and sample period, we estimate an AR-4 model for quarterly inflation. Then, for each period, we average each AR coefficient across targeting and non-targeting countries. Using these average coefficients, we compute impulse response functions showing the effects of inflation shocks on future inflation.

Figure 2 presents some of our results. We use solid lines for the impulse responses functions in targeting countries and dashed lines for nontargeters. For each group, we present results for the long pre-targeting periods (samples 1 and 4) and the targeting periods (samples 3 and 6). We omit responses for the short pre-targeting samples, which always lie between the responses that we show.

Figure 2.Inflation Persistence

Source: Authors’ calculations

The figure shows that inflation persistence has decreased over time—inflation has become more “anchored.” In the pre-targeting periods, a unit inflation shock in quarter t raises inflation at t+1 by more than 0.4 points, and this effect dies out slowly. For the targeting period, the effect is around 0.2 at t+1, and it disappears in a few quarters. Crucially, this pattern holds for both targeting and non-targeting countries. Once again, there is no evidence that targeting affects inflation behavior.6

V. Output Growth

We now ask whether inflation targeting affects output behavior. We examine the mean and standard deviation of real output growth, using the same methods we applied to inflation behavior. We use annual output data, as reliable quarterly data are not available for all countries in our sample. For each country, we include a year in a given sample period only if all four quarters of the year belong to the sample under our quarterly dating.

A. Average Growth

There is no obvious theoretical reason that inflation targeting should affect average output growth. (It might if it affected inflation behavior and inflation affects growth, but see our negative findings about inflation.) Nonetheless, Mishkin (1999) suggests

“A conservative conclusion is that, once low inflation is achieved, inflation targeting is not harmful to the real economy. Given the strong economic growth after disinflation was achieved in many countries that have adopted inflation targets, New Zealand being one outstanding example, a case can be made that inflation targeting promotes real economic growth in addition to controlling inflation.” (p. 597)

Here we examine this idea, with inconclusive results.

Table 6 presents our results about average growth rates. Average growth increased in inflation targeting countries after targeting began, and it decreased slightly in non-targeting countries. When we control for regression to the mean, our point estimates imply that targeting raises average growth by a substantial amount: from 0.7 to 1.3 percentage points, depending on the specification. However, all the t-statistics are below 1.5, and three of four are below 1.2. Thus the point estimates do not mean much.

Table 6.Mean Annual Growth Rates
Panel A
Sample 1Sample 2Sample 3Sample 4Sample 5Sample 6
Australia3.653.094.593.653.094.59
Canada4.042.523.063.942.303.44
New Zealand3.052.722.792.761.683.42
Sweden2.511.182.822.511.182.82
United Kingdom2.402.692.942.402.692.94
Finland3.151.004.683.151.004.68
Spain4.222.913.254.453.512.94
United States3.402.843.393.402.843.39
Japan5.674.121.175.674.121.17
Denmark2.101.462.812.101.462.81
Austria3.382.872.133.382.872.13
Belgium3.322.562.543.322.562.54
France3.642.552.023.642.552.02
Germany3.444.311.623.444.311.62
Ireland4.174.368.504.174.368.50
Italy3.912.432.013.912.432.01
Netherlands3.992.903.193.992.903.19
Portugal4.104.413.084.104.413.08
Norway3.482.503.503.482.503.50
Switzerland2.552.011.182.552.011.18
Averages
IT3.292.303.45
NIT3.633.022.86
CIT3.071.993.65
NCIT3.693.062.86
Panel BEquation (1)Equation (2)
Dependent Variable: Change in mean annual growth rate between samples
(3) - (1)(3) - (2)(6) - (4)(6) - (5)(3) - (l)(3) - (2)(6) - (4)(6) - (5)
Constant-0.77

(0.47)
-0.17

(0.46)
-0.82

(0.44)
-0.19

(0.43)
2.04

(1.79)
1.64

(1.31)
1.78

(1.83)
1.40

(1.31)
Inflation targeting

dummy
0.93

(0.80)
1.31

(0.77)
1.40

(0.81)
1.85

(0.78)
0.67

(0.78)
0.88

(0.81)
0.97

(0.84)
1.30

(0.88)
Initial value-0.77

(0.48)
-0.60

(0.41)
-0.71

(0.48)
-0.52

(0.41)
Adjusted R-squared0.020.090.100.200.100.150.150.23
Note: Standard errors are in parentheses.Source: Authors’ calculations
Note: Standard errors are in parentheses.Source: Authors’ calculations

Our estimates are imprecise because growth rates vary greatly across individual countries. In our short samples, average growth depends on economies’ cyclical positions when the samples start and end as well as growth in potential output. We need to observe inflation targeting over longer periods to see whether it affects average growth.

B. Output Variability

Some economists argue that “flexible” inflation targeting stabilizes output as well as inflation. Others, such as Cecchetti and Ehrmann (1999), suggest that targeting makes output more variable. Once again, we find that targeting simply does not matter.

Table 7 presents results about the standard deviation of annual output growth. These results mostly echo our findings about the standard deviation of inflation. In the short pre- targeting periods and the targeting periods, output is more stable for non-targeting countries than for targeters. For both groups, output becomes more stable during the targeting period. When we control for regression to the mean, our estimates suggest that targeting raises output variability, but this effect is not statistically significant.

Table 7.Standard Deviation of Annual Growth Rate
Panel A
Sample 1Sample 2Sample 3Sample 4Sample 5Sample 6
Australia2.241.911.732.241.911.73
Canada2.502.601.462.532.481.32
New Zealand2.823.502.282.853.061.93
Sweden2.272.101.362.272.101.36
United Kingdom2.172.330.772.172.330.77
Finland3.233.951.093.233.951.09
Spain3.132.080.733.051.660.68
United States2.381.511.382.381.511.38
Japan4.001.741.284.001.741.28
Denmark2.311.501.262.311.501.26
Austria2.231.170.742.231.170.74
Belgium2.111.130.932.111.130.93
France1.981.280.881.981.280.88
Germany2.793.840.582.793.840.58
Ireland2.081.861.922.081.861.92
Italy2.911.010.662.911.010.66
Netherlands5.531.090.545.531.090.54
Portugal3.591.980.473.591.980.47
Norway1.851.661.701.851.661.70
Switzerland2.771.920.842.771.920.84
Averages
IT2.542.731.45
NIT2.811.671.01
CIT2.552.641.37
NCIT2.831.670.99
Panel BEquation (1)Equation (2)
Dependent Variable: Change in the standard deviation of growth rate between samples
(3) - (l)(3) - (2)(6) - (4)(6) - (5)(3) - (l)(3) - (2)(6) - (4)(6) - (5)
Constant-1.80

(0.32)
-0.65

(0.24)
-1.84

(0.30)
-0.68

(0.23)
1.59

(0.38)
0.95

(0.30)
1.53

(0.34)
1.08

(0.28)
Inflation targeting

dummy
0.52

(0.54)
-0.64

(0.41)
0.66

(0.55)
-0.60

(0.43)
0.29

(0.22)
0.30

(0.28)
0.32

(0.21)
0.43

(0.26)
Initial value-1.20

(0.13)
-0.96

(0.16)
-1.19

(0.11)
-1.06

(0.15)
Adjusted R-squared0.000.070.020.050.830.690.860.75
Note: Standard errors are in parentheses.Source: Authors’ calculations
Note: Standard errors are in parentheses.Source: Authors’ calculations

VI. Interest Rates

We next examine the level of long-term interest rates, which should reflect inflation expectations, and the variability of short-term rates, which might indicate the activism of monetary policy.

A. Average Long-Term Rates

We have seen that inflation targeters and nontargeters have experienced similar reductions in inflation since the early 1990s. Targeting proponents argue, however, that targeting locks in low inflation permanently, while adverse events might reignite inflation under “just do it” policies. If the public believes this argument, then targeting should reduce both expected inflation and inflation uncertainty. As discussed by King (2002), both effects should reduce long- term interest rates.

We look for this effect in OECD data on ten-year government bond rates. The data are annual, so we date our sample periods by year’s, as in our work on output behavior. The data start in 1970, so we begin samples 1 and 4 in that year rather than 1960.

Table VIII presents our results, which are highly reminiscent of our inflation and output results. If we define better performance by lower interest rates, then nontargeters always do better than targeters; both groups improved during the targeting period; the improvement is somewhat larger for targeters; but the effect of targeting disappears when we control for regression to the mean.

Table 8.Long-Term Interest Rates
Panel A
Sample 1Sample 2Sample 3Sample 4Sample 5Sample 6
Australia10.7811.836.8210.7811.836.82
Canada8.7210.197.048.7210.026.72
New Zealand10.7015.157.4410.6513.347.04
Sweden9.2210.996.489.2210.996.48
United Kingdom9.8610.356.629.8610.356.62
Finland9.4610.657.139.4610.657.13
Spain11.7812.246.6611.9012.778.25
United States7.618.436.057.618.436.05
Japan7.015.652.457.015.652.45
Denmark12.0610.176.2812.0610.176.28
Austria8.127.666.188.127.666.18
Belgium8.519.056.338.519.056.33
France9.449.686.269.449.686.26
Germany7.607.326.037.607.326.03
Ireland10.3410.346.9010.3410.346.90
Italy10.4212.458.7710.4212.458.77
Netherlands7.437.436.027.437.436.02
Portugal15.6921.238.3515.6921.238.35
Norway8.5611.656.388.5611.656.38
Switzerland4.675.163.824.675.163.82
Averages
IT10.0711.636.88
NIT9.049.716.14
CIT9.7811.196.80
NCIT9.249.936.29
Panel BEquation (1)Equation (2)
Dependent Variable: Change in mean long-term interest rate between samples
(3) - (l)(3) - (2)(6) - (4)(6) - (5)(3) - (l)(3) - (2)(6) - (4)(6) - (5)
Constant-2.89

(0.47)
-3.57

(0.73)
-2.95

(0.44)
-3.64

(0.69)
2.57

(0.98)
3.38

(0.67)
2.23

(0.96)
3.23

(0.70)
Inflation targeting

dummy
-0.30

(0.80)
-1.18

(1.24)
-0.03

(0.80)
-0.76

(1.25)
0.33

(0.49)
0.20

(0.45)
0.27

(0.49)
0.12

(0.47)
Initial value-0.60

(0.10)
-0.72

(0.06)
-0.56

(0.10)
-0.69

(0.07)
Adjusted R-squared-0.05-0.01-0.06-0.030.630.880.610.86
Note: Standard errors are in parentheses.Source: Authors’ calculations
Note: Standard errors are in parentheses.Source: Authors’ calculations

B. The Variability of Short-Term Interest Rates

In addition to examining economic outcomes, we would like to know whether inflation- targeting central banks move their policy instruments differently from nontargeters. In principle, one can address this issue by estimating reaction functions for short-term interest rates (i.e., Taylor rules). In practice, it appears difficult to get meaningful estimates of these equations with the short samples at hand. We therefore examine a cruder measure of policy behavior, the standard deviation of short-term rates. Differences in policy rules should affect this statistic. For example, if inflation targeters respond more strongly to inflation movements, then short-term rates should become more volatile (unless targeting stabilizes inflation, an effect we fail to find).7

We examine the volatility of short-term rates at the quarterly frequency. Our data are interbank rates from the IFS (Line 60b). We examine only the shorter of our pre-targeting samples, the ones starting in 1985, because consistent data are not available before then. For once, we throw out a few troublesome outliers. For all countries, we delete the three quarters of the Exchange Rate Mechanism (ERM) crisis, 1992:3 through 1993:1, when interest rates jumped to very high levels.

The results, in Table 9, follow the pattern we have seen again and again. Interest-rate volatility is lower for nontargeters than for targeters and falls over time for both groups. The decrease appears larger for targeters if we ignore regression to the mean, but not if we control for it.

Table 9.Standard Deviation of Short-Term Interest Rates
Panel A
Sample 2Sample 3Sample 5Sample 6
Australia4.151.074.151.07
Canada1.871.212.351.20
New Zealand5.242.355.851.79
Sweden2.211.862.211.86
United Kingdom2.100.852.100.85
Finland2.261.102.261.10
Spain2.591.971.991.82
United States1.631.041.750.93
Japan1.620.891.640.75
Denmark1.011.701.031.14
Austria1.941.111.910.78
Belgium1.621.621.611.05
France1.051.601.041.38
Germany2.081.202.060.91
Ireland2.000.772.080.76
Italy1.511.931.592.00
Netherlands1.681.171.660.92
Portugal2.772.542.792.38
Norway1.731.271.971.30
Switzerland2.551.272.511.10
IT2.921.49
NIT1.791.39
CIT3.151.31
NCIT1.831.23
Panel BEquation (1)Equation (2)
Dependent Variable: Change in the standard deviation of the short-term interest rate
(3) - (2)(6) - (5)(3) - (2)(6) - (5)
Constant-0.39

(0.23)
-0.60

(0.24)
1.04

(0.28)
0.96

(0.26)
Inflation targeting

dummy
-1.04

(0.39)
-1.24

(0.44)
-0.13

(0.28)
-0.11

(0.28)
Initial value-0.80

(0.14)
-0.85

(0.12)
Adjusted R-squared0.280.310.760.82
Note: Standard errors are in parentheses.Source: Authors’ calculations
Note: Standard errors are in parentheses.Source: Authors’ calculations

VII. Bivari vte Results

So far we have examined the univariate behavior of inflation, output, and interest rates. In principle, we would like to look more deeply at whether inflation targeting changes the structure of the economy. For our short samples, however, it is impractical to estimate sophisticated structural equations. Here we take one step beyond our univariate analysis by examining several bivariate relations.

A. Methodology

For each country and sample period, we run three regressions:

where y* is the trend level of output (measured by the Hodrick-Prescott filter with smoothing parameter 100); pcom is an index of commodity prices in U.S. dollars, from the IFS; πUS is U.S. inflation; and πfore is an OECD forecast of inflation. All the data are annual.

Equation (3) can be interpreted as an accelerationist Phillips curve: it shows how the output gap affects the change in inflation. Equation (4) measures the inflationary effect of a change in the relative price of commodities, which we interpret as a “supply shock.” The change in the relative price is the change in the U.S. dollar price minus U.S. inflation. Finally, equation (5) shows how expected inflation responds to movements in past inflation. We measure expectations with OECD forecasts, which are produced in consistent ways for all countries.8

Previous authors suggest that inflation targeting should affect the coefficients a, b, and c in these equations. For example, Bernanke et al. argue that targeting “anchors” inflation expectations, so c should fall. They also argue that targeting reduces the effects of supply shocks, so b should fall (see the quote at the start of this paper). The effects on a, the Phillips curve slope, are debatable. This coefficient might fall if inflation becomes more anchored. On the other hand, Corbo et al. (2002) argue that targeting reduces the cost of disinflation, which suggests a rise in a.

We are interested in the averages of a, b, and c for targeting and non-targeting countries. When we estimate these coefficients for individual countries, the standard errors vary greatly. Since there is more noise in some estimated coefficients than in others, a simple average is an inefficient estimator of the true average coefficient. We therefore compute weighted averages, with weights inversely proportional to the variances of the coefficient estimates. Similarly, we estimate our differences regression by weighted least squares, with weights inversely proportional to the standard deviations of the estimated changes in coefficients. We do not add estimates of initial coefficients to the right-hand sides of our regressions, because the measurement error in the coefficients would create bias9

B. Results

Table 10 presents our bivariate results. For the final time, we find that economic behavior has changed over time, but the changes are similar for inflation targeters and nontargeters.

Table 10.Multivariate Results
Panel A: Phillips-Curve Coefficients
Sample 1Sample 2Sample 3Sample 4Sample 5Sample 6
Weighted Averages
IT0.350.100.18
NIT0.270.250.17
CIT0.370.180.14
NCIT0.270.250.18
Equation (1) (Weighted Least Squares)
Dependent Variable: Change in estimated coefficient between samples
(3) - (l)(3) - (2)(6) - (4)(6) - (5)
Constant-0.12

(0.07)
-0.07

(0.09)
-0.11

(0.07)
-0.05

(0.07)
Inflation targeting

dummy
0.13

(0.12)
0.20

(0.12)
0.00

(0.13)
0.07

(0.11)
Panel B: Effect of Commodity-Price Changes on Inflation
Sample 1Sample 2Sample 3Sample 4Sample 5Sample 6
Weighted Averages
IT0.0440.0360.005
NIT0.0540.0680.006
CIT0.0490.0820.014
NCIT0.0530.0650.006
Equation (1) (Weighted Least Squares)
Dependent Variable: Change in estimated coefficient between samples
(3) - (l)(3) - (2)(6) - (4)(6) - (5)
Constant-0.048

(0.010)
-0.050

(0.014)
-0.047

(0.009)
-0.048

(0.013)
Inflation targeting

dummy
0.006

(0.024)
-0.012

(0.031)
0.012

(0.024)
-0.027

(0.034)
Panel C: Response of Expected Inflation to Inflation
Sample 1Sample 2Sample 3Sample 4Sample 5Sample 6
Weighted Averages
IT0.830.710.43
NIT0.830.710.66
CIT0.820.630.45
NCIT0.830.710.63
Equation (1) (Weighted Least Squares)
Dependent Variable: Change in estimated coefficient between samples
(3) - (l)(3) - (2)(6) - (4)(6) - (5)
Constant-0.23

(0.04)
-0.10

(0.06)
-0.25

(0.04)
-0.12

(0.06)
Inflation targeting

dummy
-0.15

(0.10)
-0.13

(0.14)
-0.10

(0.11)
-0.05

(0.15)
Note: Standard errors are in parentheses.Source: Authors’ calculations
Note: Standard errors are in parentheses.Source: Authors’ calculations

There are two significant changes over time: expectations respond less to inflation movements, and inflation responds less to commodity prices. Both results suggest a greater anchoring of inflation. Strikingly, the commodity-price coefficients fall by an order of magnitude. For example, the average coefficient in sample 1 (1960 to the start of IT) is 0.05 for nontargeters. This means that a ten percent rise in the relative price of commodities raises inflation by five tenths of a percentage point. For the IT period (sample 3), the coefficient is 0.006.

In contrast, there is no evidence that inflation targeting affects the coefficients that we consider. In the twelve regressions in Table X, the targeting dummy is never significant at the ten percent level.

VIII. Comparison with Other Studies

The closest study to ours is that of Neumann and von Hagen. Their paper and ours have the same title. Part of their paper, like this one, compares the volatility of inflation, output, and interest rates across time periods and groups of countries. But Neumann and von Hagen’s conclusion differs from ours: “Taken together, the evidence confirms the claim that IT matters” (p. 144).

Our study differs from Neumann and von Hagen in many details, but the crucial difference may be our treatment of regression to the mean. After the sentence quoted above, they continue: “Adopting this policy has permitted IT countries to reduce inflation to low levels and curb the volatility of inflation and interest rates; in so doing, these banks have been able to approach the stability achieved by the Bundesbank” (Neumann and von Hagen’s main example of a non-inflation targeter). We, too, find that targeters have caught up with nontargeters along some dimensions, but this convergence was not caused by targeting.

A number of other studies report evidence that inflation targeting matters. For example, researchers report that targeting steepens the Phillips curve (Clifton et al., 2001); that it dampens movements in expected inflation (Sheridan, 2001); and that it increases the predictability of inflation (Corbo et al., 2002).10 Some of these results may again reflect regression to the mean rather than a true effect of targeting. This possibility is suggested by Corbo et al.’s conclusion that “Inflation targeters have consistently reduced inflation forecast errors (based on country VAR models) toward the low levels prevalent in non-targeting industrial countries” (p. 263).

It is difficult to compare our results directly to previous work, as the methodologies are quite different. We believe, however, that our results cast doubt on earlier findings that inflation targeting affects economic behavior. It seems unlikely that targeting would affect the relationships studied by previous authors and yet, as we find, have no effects on the means or standard deviations of inflation, output, or interest rates.

IX. Conclusion

We find no evidence that inflation targeting improves a country’s economic performance. How should one interpret this result?

One possibility is that targeting and nontargeting countries pursue similar interest rate policies. Research suggests that the policies needed to implement inflation targeting are similar to the Taylor rules that fit the United States and other nontargeters (see, for example, Svensson, 1997; Ball, 1999). Indeed, observers have suggested that the United States is a “covert inflation targeter” (Mankiw, 2001). This view is supported by our finding of similar interest-rate volatility for targeters and nontargeters. If targeting does not change the behavior of policy instruments, it is not shocking that economic outcomes do not change either. This result suggests, however, that the formal and institutional aspects of targeting—the public announcements of targets, the inflation reports, the enhanced independence of central banks—are not important. Nothing in the data suggests that covert targeters would benefit from adopting explicit targets.

Our results do not provide an argument against inflation targeting, for we have not found that it does any harm. In addition, there may be benefits that we do not measure. First, aspects of inflation targeting may be desirable for political rather than economic reasons. Bernanke and others argue that targeting produces more open policymaking, making “the role of the central bank more consistent with the principles of a democratic society” (p. 333).

Second, inflation targeting might improve economic performance in the future. The economic environment has been fairly tranquil during the inflation-targeting era, and so many central banks have not been tested severely. Perhaps future policymakers will face 1970s-size supply shocks or strong political pressures for inflationary policies. At that point, we may see that inflation targeters handle these challenges better than policymakers who “just do it.”

Thus a paper that replicates this study in 25 or 50 years may find ample evidence that targeting improves performance. The evidence is not there, however, in the data through 2001.

APPENDIX I

Methodology

Consider the problem of estimating the effect of inflation targeting on X, some measure of economic performance. For concreteness, we will sometimes refer to X as “average inflation.” We present a simple statistical model of the determinants of X in different countries and periods. In our model, regression (1) in the text, the differences estimator, produces a biased estimate of the effect of targeting if the targeting dummy is correlated with the pre-targeting level of X. Adding the pre-targeting X, as in regression (2), eliminates the bias.

Let Xit be the value of X in country i and period t. The t subscript takes on two values, “pre” and “post.” We assume that Xu is given by

where μi is a country-specific effect, ηt is a period-specific effect, vit is an error term specific to country i in period t, and Qit is a dummy equal to one if country i targets inflation in period t. For all countries, Qi,pre equals zero and Qi,post equals Di; the targeting dummy in the text.

In equation (A1), the Qit term captures the possible effect of inflation targeting. We would like to estimate its coefficient, a1. The other terms are a conventional decomposition of the error term in a panel regression. By construction, the idiosyncratic shock vit is uncorrelated with μi and ηt, and vi,pre and vi,post are uncorrelated with each other.

Differencing equation (A1) over time yields

where we use the fact that Qi,Post–Qi,Pre=Di. Thus, in cross-country data, the change in X depends on a constant (μpost–μpre), the targeting dummy, and a composite error term. We can interpret regression (1), the differences estimator in the text, as an OLS estimator of equation (A2).

Suppose that countries with higher initial inflation, Xi,Pre, are more likely to adopt inflation targeting. The error vi,pre is one component of Xi,pre, so a higher vi,pre makes targeting more likely: vi,pre is positively correlated with the dummy Di. The error in (A2) includes –vi,pre, so the dummy is negatively correlated with the error. This correlation implies that the OLS estimate of the dummy coefficient, ai, is biased downward. Consequently, regression (1) is likely to find that targeting reduces inflation even if there is no true effect.

Now consider what happens when we add the initial level of X to our regression. We can rewrite equation (A2) as

where the true value of a2 is zero. We interpret regression (2) in the text as an OLS estimator of this equation. We now sketch a proof that the estimate of al is unbiased even if Xi,pre affects the likelihood of targeting.

Rather than view vi,pre as part of the error term in (A3), let us interpret it as a variable that is left out when we regress the change in Xi on the constant, Di, and Xi,pre If vi,pre were measured and included in the regression, then OLS would be unbiased, because all right-side variables would be uncorrelated with the remaining error vi,post- We can therefore use standard results to determine the biases that arise when vi,pre is left out (Maddala, 1989, p. 122). Specifically, the bias in the OLS estimate of al is proportional to the expected coefficient on Di in an auxiliary regression of vi,pre on a constant, Di and Xi,pre One can show that this expected coefficient is zero, implying zero bias. Intuitively, vi,pre is correlated with Di, but this correlation works through the effect of vi,pre on Xi,pre When one controls for Xi,pre in the auxiliary regression, there is no relation between vi,pre and Di.

References

    BallLaurence. 1997. Disinflation and the NAIRU. in Reducing Inflation: Motivation and Strategyed. By Christina D.Romer and David H.Romer. (Chicago: University of Chicago Press).

    BallLaurence1999Efficient Rules for Monetary PolicyInternational Finance 2Vol. 1 pp. 6383.

    BernankeBen S.ThomasLaubachFrederic S.Mishkin and Adam S.Posen1999Inflation Targeting: Lessons from the International Experience (Princeton, New Jersey: Princeton University Press).

    CecchettiStephen G. and MichaelEhrmann1999Does Inflation Targeting Increase Output Volatility? An International Comparison of Policymakers’ Preferences and OutcomesNBER Working Paper No. 7426 (Cambridge, Massachusetts: National Bureau of Economic Research).

    CliftonEric V.LeonHyginus and Chorng-HueyWong2001Inflation Targeting and the Unemployment-Inflation Trade-offIMF Working Paper 01/166 (Washington: International Monetary Fund).

    CorboVittorioOscarLanderretche and KlausSchmidt-Hebbel2002Does Inflation Targeting Make a Difference?” in Inflation Targeting: Design Performance Challengesed.by NormanLoayza and RaimundoSaito(Santiaga: Central Bank of Chile).

    GilovichThomas1984Judgemental biases in the world of sports” in Cognitive Sport Psychologyed.by StraubWilliam F. and Jean M.Williams(New York: Sport Science Associates).

    JohnsonDavid R2002The Effect of Inflation Targeting on the Behavior of Expected Inflation: Evidence from an 11 Country PanelJournal of Monetary EconomicsVol. 49 pp. 15211538.

    KingMervyn2002The Inflation Target Ten Years Onspeech delivered to the London School of EconomicsNovember19.

    KuttnerKenneth N. and Adam S.Posen1999Does Talk Matter After All? Inflation Targeting and Central Bank BehaviorFederal Reserve Bank of New York Staff Reports No. 88 (New York)

    LayardRichardStephenNickell and RichardJackman1991Unemployment: Macroeconomic Performance and the Labour Market (New York: Oxford University Press).

    MaddalaG.S1989Introduction to Econometrics (New York: Maxwell Macmillan).

    MankiwN. Gregory2001U.S. Monetary Policy During the 1990sNBER Working Paper No. 8471 (Cambridge, Massachusetts: National Bureau of Economic Research).

    MishkinFrederic S.1999. International Experiences with Different Monetary Policy RegimesJournal of Monetary EconomicsVol 43 pp. 579605.

    MishkinFrederic S.2002. “Does Inflation Targeting Matter? CommentaryFederal Reserve Bank of St. Louis ReviewVol. 84 (July–August) pp. 149-53.

    NeumannManfred J.M. and Jurgen vonHagen. 2002. Does Inflation Targeting Matter?Federal Reserve Bank of St. Louis ReviewVol. 84 (July–August) pp. 12748.

    ScheaterAndreaMark R.Stone and MarkZelmer2000Adopting Inflation Targeting: Practical Issues for Emerging Market Countries IMF Occasional Paper No. 202 (Washington: International Monetary Fund).

    SheridanNiamh2001Inflation Dynamics” (Ph.D. dissertation. Baltimore, Maryland: Johns Hopkins University).

    SvenssonLars E.01995The Swedish Experience of an Inflation Target” in Inflation Targetsed.by LeonardoLeiderman and Lars E.O.Svensson(London: Center for Economic Policy Research)

    SvenssonLars E.01997Inflation Forecast Targeting: Implementing and Monitoring Inflation TargetsEuropean Economic ReviewVol. 41 pp. 111146.

Johns Hopkins University and International Monetary Fund, respectively. We are grateful for research assistance from Witold Czubala, Gergana Danailova-Trainor and Migiwa Tanaka, and for suggestions from Andrew Feltenstein, Mark Gertler, participants at the National Bureau of Economic Research (NBER) Inflation Targeting Conference, January 2003, and participants at a seminar at the IMF Institute.

In addition, we tried adding a Euro dummy to all of our cross-country regressions. This variable is usually insignificant. The only exception is that Euro countries experienced larger falls in the standard deviation of output growth between the pre- and post-targeting periods. Including the Euro dummy never changes our findings about the effects of inflation targeting.

For New Zealand, we date the constant-targeting period from 1993:1 to the end of the sample even though the target range was widened from 0–2% to 0–3% in 1997. The half-point change in the midpoint was smaller (and of the opposite sign) than the target changes during transitional periods in other countries. In our judgement the 1997 episode was not a substantial change in policy.

Baseball statistics exhibit substantial regression to the mean. This fact explains the well-known “sophomore slump”: the tendency of players with strong rookie years to do less well during their second years (e.g. Gilovich, 1984).

In analyzing trend inflation, we include a quarter in a sample only if all quarters that contribute to the nine-quarter average are in the sample.

Note that the impulse responses for targeters in samples 3 and 6 are negative at some lags. We have checked the statistical significance of the negative responses with Monte Carlo experiments, following Sheridan (2001). The only response that is significantly negative is the response for CIT countries in period t+4. We are inclined to dismiss the negative responses as a fluke, because they are not plausible theoretically.

Neumann and von Hagen and Kuttner and Posen (1999) estimate Taylor rules for inflation targeters. For a critique, see Mishkin’s (2002) discussion of Neumann and von Hagen.

Some details: We exclude a constant term from equation (3) because y-y* has a zero mean and we want to rule out a deterministic trend in inflation. In equation (4), the change in relative commodity prices is the same for all countries. We have also estimated equation (4)with y-y* included, which can be interpreted as a Phillips curve augmented with supply shocks. Our results about the coefficient on the change in commodity prices do not change. In addition, we obtain similar results when we replace the change in commodity prices with the change in the relative price of oil. In equation (5), π(–l) is inflation in year −1 as estimated by the OECD in December of that year, when they make forecasts for the following year.

In principle, the optimal estimators of the group means and equation (1) use weights that depend on both the variances of the coefficient estimates and the variances of true coefficients across countries in a group. Using the residuals from our cross-country regressions, we have estimated the variances of true coefficients, and find they are small. We therefore set these variances to zero and derive the optimal weights based on the variances of coefficient estimates. These weights are the ones described in the text.

See also Johnson (2002) and the literature review in Neumann and von Hagen.

Other Resources Citing This Publication