WORKING PAPER HAS INFLATION PERSISTENCE IN INDONESIA CHANGED? WP/10/2007. Tri Yanuarti

Transcription

1 WORKING PAPER WP/10/2007 HAS INFLATION PERSISTENCE IN INDONESIA CHANGED? Tri Yanuarti October 2007

2

3 Has Inflation Persistence in Indonesia Changed? By Tri Yanuarti 1 Abstract This paper measures the degree of inflation persistence in Indonesia and examines whether there have been changes in inflation persistence since The results show high inflation persistence in Indonesia if we estimate over the entire sample period from 1990 to However, once we control for the crisis period during , we find that inflation tends to have low persistence. In the post-crisis period (2000 onwards), both aggregate inflation and its components seem to be less persistent. The source of declining inflation persistence stems from the housing, food, and transportation components. This finding of lower persistence in the post-crisis period probably partially relates to declining inflation expectations. As a result, in the post-crisis period monetary policy may only require small adjustments to achieve the inflation target. The anchoring of inflation expectations to the target level and the ability of inflation to quickly revert to its expected level are important consequences of increased credibility, which may largely owe to the adoption of an inflation targeting monetary framework. JEL classification: E31, C11 Keywords: Inflation persistence, structural change 1 Economist in the Directorate of Economic Research and Monetary Policy, Bank Indonesia. I would like to thank the Reserve Bank of Australia, particularly staff in Economic Group for technical assistance and helpful comments. I am grateful to Firman Mochtar, Fadjar Majardi, Akhis R. Hutabarat and other staffs in Bank Indonesia for excellent comments and discussion. I also thank to Pandri Hanandya Rama and Hesti Werdaningtyas for providing data for this study. The views expressed in this paper are those of the author and do not necessarily reflect those of Bank Indonesia. : tri_yanuarti@bi.go.id

4 Table of Contents I. Introduction... 1 II. Development of Inflation... 1 III. Literature Review on Inflation Persistence... 3 III.1. Definition of Inflation Persistence... 3 III.2. Measuring Inflation Persistence... 4 IV. Empirical Model... 6 IV. 1. Model... 6 IV.2. Data... 7 V. Results... 8 V. 1. Estimated Persistence Parameter... 8 V. 2. Temporal Aggregation Effect... 9 V. 3. Change in Inflation Persistence V. 4. Why has Inflation Become Less Persistent? VI. Conclusion References Appendix A: Estimated Model for Aggregate Inflation Appendix B: Estimated Model for Disaggregated Inflation ii

5 1. Introduction As Bank Indonesia moves towards inflation targeting, information and in-depth knowledge on the factors that explain the behavior of inflation have become increasingly important. One key aspect relating to inflation behavior is inflation persistence, which informs us about how quickly inflation reverts to its long-run equilibrium path following a shock. The ability to measure inflation persistence is important in determining how preemptive Bank Indonesia should be in responding to exogenous shocks in order to curb inflationary pressure and maintain price stability. A number of structural changes occurred around the crisis period in , such as, a shift to a floating exchange rate regime, the adopting of inflation targeting and changes in subsidies and administered prices policies. As a result, the recent inflation process probably differs from the pre-crisis period. As described in the recent literature, changes in the monetary policy framework and exchange rate regime may influence price-setting behaviour, particularly if the price setter has a better understanding about expected future inflation and the response of monetary policy to shocks in the economy. This is also likely to change the level of inflation persistence. So far, previous Indonesian studies have mostly focused on how inflation responds to its determinants. The previous study on inflation persistence (Inflation Team, 2006) measured the persistence of core inflation in Indonesia, but provided only limited insight into the change in persistence over time, because it covered a short time period. The purpose of this study is to measure the degree of inflation persistence in Indonesia and to determine whether inflation persistence has changed over This study could be helpful in conducting more effective monetary policy. The reminder of the paper is organised as follows. Section 2 briefly overviews inflation developments in Indonesia. Section 3 discusses inflation persistence measures that have been employed in the literature. Section 4 describes the empirical model and data. Section 5 explains the results. Finally, Section 6 concludes. 2. Development of Inflation Figure 1 plots the month-on-month Consumer Price Index (CPI) in Indonesia. As the Figure indicates, the development of inflation over the last two decades can generally be divided into three main periods. The first period is from 1990 to 1997, where inflation was generally stable and below 3 per cent. The dynamics of inflation during this period have often been linked to seasonal demand, particularly in conjunction with Islamic festivity days such as the Ramadhan fasting month and Ied Fitr. The average monthly inflation rate was around 2 per cent during these seasons. The highest impact of Ied Fitr

6 on inflation was found in 1990, where inflation reached 2.3 per cent. Changes in administered prices such as the petrol price and transportation costs, as well as wage increases also contributed to movements of inflation during this period. Among the administered prices, changes in petrol prices usually generated the most significant effect on inflation; for example, in January 1993, when the government increased petrol prices, inflation rose to 2.9 per cent. There was also a period of deflation in This resulted primarily from low food inflation reflecting improvements in food production and distribution; a result of market intervention by the government. Figure 1: Inflation in Indonesia % (month to month) Crisis Period Pre-crisis Period Post-crisis Period Source, BPS The second period is from 1998 to 1999 and is often referred to as the crisis period. During this period, the average monthly inflation rate rose to 2.6 per cent and volatility increased substantially. The change in inflation behavior was mainly generated by a significant depreciation in the Rupiah, following the collapse of the fixed exchange rate regime. This depreciation, combined with other factors, such as rumors of food shortages, disturbance in food distribution and Islamic festivity days, led to inflation increasing by more than 10 per cent in February In July 1998, inflation again rose sharply, because lower production, harvest failure and interruption to distribution networks resulted in an inadequate supply of goods. However, inflation fell sharply once the distribution of essential items re-commenced and control of the monetary aggregate was regained. 2

7 Table 1: Summary Statistics for Inflation Period Before Crisis ( ) Crisis ( ) After Crisis ( ) Monthly Mean Standard Deviation Quarterly (average monthly) Mean Standard Deviation Quarterly (end of period) Mean Standard Deviation Source: BPS, calculated. The third period covers 2000 to Over this period, the mean inflation rate has been similar to that observed before the crisis. This period has been characterized by a change in the monetary policy regime; since June 2005, Bank Indonesia has adopted an inflation targeting framework. Compared to the first period, the volatility of inflation has been higher. This stems from several factors including, increased volatility in the exchange rate, because Indonesia adopted a floating exchange rate system just before the crisis period, and changes in petrol prices owing to reduced government subsidies. The large rise in inflation in October 2005 was caused by the increase in petrol prices, which also caused second-round effects through higher transportation costs (Figure 1). 3. Literature Review on Inflation Persistence 3.1. Definition of Inflation Persistence Batini (2002) gives three definitions of inflation persistence, namely: (1) positive serial correlation in the inflation series; (2) lags between systematic monetary policy and their peak effect on inflation; and (3) lagged responses of inflation to policy shocks. The first type of persistence refers to a reduced-form property of inflation that manifests simultaneously the underlying process, the conduct of monetary policy, and the expectations formation process of price-settings agents (Batini, 2002, p 11). Changes in these factors will influence the autocorrelation properties of inflation. The second type of persistence was pioneered by Friedman (1972) and refers to a delay between a monetary policy action and its maximum effect on inflation. Finally, the third type of inflation persistence, applied by Christiano, Eichenbaum, and Evans (2001), relates to the number of lags it takes for inflation to respond to a policy action. 3

8 In the context of an autoregressive process of order p (AR(p)), Dias and Marques (2005) define inflation as being highly persistent if following a shock to the disturbance term, inflation converges slowly to its mean. The mean of inflation is sometimes used to represent the equilibrium level of inflation, as suggested by Marques (2004). In the context of this parametric representation, the concept of persistence appears as intimately linked to the impulse response function (IRF) of the AR(p) process. However, as the impulse response function is an infinite-length vector it is not useful measure of persistence. (Dias and Marques, 2005, p 10) Inflation Persistence Measurement Measures of inflation persistence, as described in many studies, can basically be divided into two distinct approaches. One method is denoted as the univariate approach since it measures inflation persistence in the context of a simple univariate time-series representation of inflation. Under this approach, a simple autoregressive model for inflation is assumed and shocks are measured in the white noise component of the autoregressive processes. The other approach is denoted as the multivariate approach. This approach assumes a causal relationship between inflation and its determinants, which is usually represented using a Phillips Curve or a structural VAR model. Inflation persistence in this context refers to the duration of a shock s effect on inflation. 2 Marques (2004) highlights the important differences between the multivariate and univariate method. Under the multivariate approach, we can analyse different kinds of shocks influencing inflation and determine those that have a specific affect on persistence. In contrast, in the univariate approach, the shocks to inflation do not have an economic interpretation because they are generally regarded as a summary measure of all the shocks influencing inflation in a given period. Under the univariate approach, Pivetta and Reis (2006) summarise three scalar functions that have been widely used as measures of persistence, namely (i) the largest autoregressive root (LAR), (ii) the sum of the coefficients in the autoregressive process and (iii) the half-life. Although all these measures are typically used in measuring persistence, they are not without their draw-backs. Phillips (1991) and Andrews (1993) in Pivetta and Reis (ibid) highlights that the main problem in using LAR as a measure of persistence is that it ignores the effect of the other roots of the autoregressive process in the overall persistence series. Andrew and Chen (1994) demonstrate this problem with an impulse response function of two series, using the same LAR but different magnitudes for the other roots. 2 This section draws heavily on Marques (2004). Also see Taylor, and Ceccheti and Debelle (2005) for an extensive literature review on inflation measures. 4

9 A major disadvantage of measuring persistence using the sum of the coefficients of the autoregressive process is that ρ could imply two series are equally persistent, even if they exhibit completely different patterns of mean reversion (Andrew and Chen, ibid). The main draw-back of the half-life measure, which is defined as the number of periods in which inflation remains above 0.5 following a unit shock 3, is it may be an inappropriate measure in cases where the impulse response function is oscillating (Murray and Papell, 2002 in Paya, et al 2005). In addition, Pivetta and Reis (ibid) also suggest that the half-life can not be used to compare two different series if one impulse response function is more convex to the origin than the other, for reasons unrelated to persistence. Marques (ibid) suggests an alternative measure of persistence, which explores the relationship between persistence and the degree of mean reversion. In the context of mean reversion, if a persistent series converges slowly to its equilibrium level after a shock, then such a series must cross its mean only infrequently. Similarly, a nonpersistent series must revert to its mean frequently. Diaz and Marques (ibid) suggest that this measure is important for the central bank because it helps to identify how frequently inflation reverts to the inflation target. However, this measure can be very sensitive to the assumption used for the long-run mean of inflation. In sum, despite the limitations, in this study we focus on the sum of the autoregressive coefficients as a useful scalar statistic because it is simple to calculate and easy to understand. However, persistence in this sense should be interpreted as the average speed with which inflation converges to equilibrium after a shock. A higher value of the average speed indicates that inflation is more persistent. Most of the previous studies, such as Batini (ibid), Levin and Piger (ibid), and Cecchetti and Debelle (ibid), use the sum of the autoregressive coefficients and find that persistence is relatively high. Importantly though, when a mean break is taken into account, Levin and Piger (ibid), Cecchetti and Debelle (ibid) and Poonpatpibul, et.al (2004) found low persistence of inflation. In addition, Clark (2003) and Hondroyiannis and Lazaretou (2004) employed rolling persistence estimates to present the structural break in persistence and found a tendency of persistence to decline. Similarly, other studies such as Levin and Piger (ibid), Marques (ibid) and Poonpatpibul, et.al (ibid) identified changes in mean inflation by introducing an intercept dummy variable to account for changes in the mean of inflation. 3 See Pivetta and Reis (2006) for detail of the measures. 5

10 4. Empirical Model Model In this study, we estimate an univariate autoregressive process to measure inflation persistence as we are interested in the inflation process. Following Ceccheti and Debelle (2005), we assume that inflation follows a stationary autoregressive process of order p (AR(p)). The model can be represented as follows: p π = α + β π + ε (1) t j t j t j= 1 where π t is the inflation rate in period t, β i is the autoregressive coefficient representing persistence, and ε t is the white noise error term. The degree of persistence is measured by the sum of the autoregressive coefficients (Σβ j ). The constant (α) plays a key role in estimating the persistence regression because it controls for the mean of inflation. Equation (1) can be reparameterized as: where p Δ π = α + δδ π + ( ρ 1) π + ε (2) ρ t t j t 1 t j= 1 p p = β, and δ j j = j= 1 i=+ 1 j β. In the context of Equation (2), the level of persistence i depends on how fast inflation converts to its mean following a shock. In order to estimate Equation (1), we determine the appropriate lag by employing the Akaike Information Criterion (AIC) and the Schwarz Criterion (SC). In this study we measure inflation persistence using a monthly inflation series. However, to gauge the effect of temporal aggregation in persistence measures, we also estimate Equation (1) using quarterly data. We derive the quarterly series in two ways: an average of the monthly inflation series and the end-of-quarter inflation rate. To compare the results between using monthly and quarterly data, we calculate the true estimate based on the monthly estimates, following Taylor (2000). In the case of P- period temporal aggregation, the autoregressive coefficient generally has an expected value given by 4 : Cov (x,x ) ρ ρ ρ = (3) * * * * s s-1 (P, ) = E( ) * Var(x s-1) When the quarterly series is derived using monthly averages, the value of the autoregressive coefficient is defined as: 4 See Taylor (2000) for a detailed derivation. 6

11 * ρ = (4/9 + 11/9 ρ + 14/9 ρ + 19/9 ρ + 16/9 ρ + 10/9 ρ + 4/9 ρ + 1/9 ρ ) (19/9 + 13/9 ρ+ (2 + 2/9) ρ + 8/ 9ρ + 2/9 ρ ) (4) and, when the quarterly series is derived using the end-of-period inflation rate, the value of the autoregressive coefficient is defined as: * ρ = ( ρ + 2 ρ + 2 ρ + 2 ρ + ρ ) 2 (3+4ρ+ 2 ρ ) (5) where ρ is the persistence measure, generated from monthly data, and ρ* is the true estimate of persistence on a quarterly basis. Since there is a possibility of a structural break during (the crisis period), estimating inflation persistence without controlling for the break in the mean of the time series would potentially generate an exaggerated degree of inflation persistence, as suggested by Perron (1990). Thus, it is important to obtain formal econometric evidence about the presence of structural breaks in the series. In this study, we use three approaches to identify the presence of a break, using monthly analysis. First, we use the following equation: p p p 0 + α1d1 + β jπ t j + β j D1π t j + β j Dcrisisπ t j ε t (6) j= 1 j= 1 j= 1 π = t α + where D 1 is a dummy variable: equal to 0 before the crisis and 1 thereafter, and D crisis is a dummy variable for the crisis: equal to 1 for 1998, -1 for 1999, and 0 otherwise. Second, if the results provide evidence of a change in persistence, then we estimate two separate equations for the and sub-periods. Finally, an alternative way of analysing the influence of breaks on the persistence parameter is to use rolling regressions to estimate the parameters in Equation (1). Following Stock (2001), Pivetta and Reis (2001) and Levin and Piger (2004), we estimate the constant and the persistence parameter for every sub-sample using a 48-period window (fixed length). To determine the source of changes in inflation persistence we also examine disaggregated CPI data in both the pre- and post-crisis periods. Since the current CPI measure consists of seven main groups, we re-categorise the previous CPI data from four groups into seven groups Data The inflation series used in this study is defined as the percentage change, periodto-period, in the headline Consumer Price Index (CPI). We also use non-oil CPI, which is defined as headline minus the oil price components, since we are interested in estimating the impact of fuel prices on inflation persistence. 7

12 In addition, we also use disaggregated CPI data, consisting of seven main subcomponents, namely: (i) food; (ii) processed food, beverages and tobacco; (iii) housing; (iv) clothing; (v) health (medical care); (vi) education, recreation and sport; and (vii) transportation and communication. The sample period is from 1990 to 2006, due to the availability of the disaggregated CPI data. 5. Results 5.1. Estimated Persistence Parameter The estimated persistence parameters, as presented in Table 2, indicate that inflation (both headline and non-oil) in Indonesia tends to be quite persistent. The persistence parameter is 0.61 for headline inflation and 0.63 for non-oil inflation. Since the parameter of persistence is below 1, this means that a shock to inflation in Indonesia has only a temporary effect on inflation. After the shock, inflation will revert to its mean. The results in Table 2 show that the persistence of headline inflation is generally lower than for non-oil inflation. This implicitly indicates that fuel price shocks lower the persistence of headline inflation. However, based on the Wald test, the persistence parameter for headline inflation is not significantly different from the non-oil persistence parameter. This insignificant difference is probably owing to the fact that a sudden increase in the domestic fuel price generally also affects other prices in the CPI basket. For example, when the fuel price rose sharply in October 2005, transportation costs also rose and this would have consequently affected the distribution of goods and services. As a result, headline inflation and non-oil inflation reached 8.7 per cent and 5.5 per cent, respectively (Figure 3). Hence, excluding only fuel price items from the headline CPI does not substantially decrease the rate of inflation or its persistence. Table 2: Estimated Persistence Parameter Model Degree of persistence ( ρ ) Including crisis period Excluding crisis period **) Without dummy crisis With dummy crisis *) Headline CPI inflation Non-oil inflation *) We introduce dummy crisis and October 2005 to control the breaks in estimation **) We exclude crisis period and October 2005 in estimation 8

13 - Introducing a dummy variable to control for the crisis period generates relatively lower persistence than the previous results. In general, the degree of persistence is below 0.6. However, when we alternatively exclude the crisis period and October 2005 from our sample period completely, our results suggest that neither the headline nor non-oil CPI series are very persistent (Table 2). These results are not surprising and are consistent with those previous studies, such as Fuhrer and Moore (ibid), Levin and Piger (ibid) and Ceccheti and Debelle (ibid) who all find that controlling for breaks in the mean of inflation substantially reduces the estimates of persistence. Figure 2 shows how inflation behaves in response to a 1 per cent shock. The persistence parameter is estimated at 0.61 for the overall sample period and we find that headline inflation tends to revert to its mean after 12 months, following a 1 per cent shock. The response of non-oil inflation to a shock is similar to headline inflation, consistent with the fact that the degree of persistence between the two measures is not significantly different. Figure 2: Decay of Inflation Following A Shock Figure 3: Headline and Non-Oil Inflation Headline inflation Non-oil inflation 0.3 Non-oil inflation Headline Inflation Month Temporal Aggregation Effect Although in this study we focus on persistence in monthly inflation, it is important to examine whether these results are robust to different data frequencies. The results in Table 3 suggest that the estimates of persistence lie within a close range across different frequencies. However, we find that the quarterly models generate an upward bias in the persistence estimates. This means that the estimated persistence always exceeds the true estimate. Of our two quarterly measures monthly average and end-ofperiod the end-of-period measure appears to be less biased. 9

14 Table 3: Quarterly Inflation Persistence Estimates Monthly Quarterly I Quarterly II Headline Non-oil Headline Non-oil Headline Non-oil Persistence estimates True estimates (ρ*) *) Quarterly I model is estimated using monthly average data **) Quarterly II model is estimated using end of period data Furthermore, we calculate the impulse response function for each of the quarterly and the monthly series to examine the path of inflation following a 1 per cent shock. The impulse response functions in Figure 4 are calculated using a quarterly frequency for both the monthly and quarterly models. As the Figure shows, the impulse response functions of the monthly and quarterly II models are quite similar. The impact of a 1 per cent shock tends to dissipate after 4 quarters. However, in the quarterly I model, the impact of a shock on inflation takes longer to dissipate, consistent with the higher estimated persistence. Figure 4: Impulse Response Function Monthly Quarterly I Quarterly II Monthly Quarterly I Quarterly II Headline Inflation Non-Oil Inflation 5.3. Changes in Inflation Persistence To identify whether inflation persistence has changed during , we estimate equation (6). The results, presented in Table 4, provide strong evidence of changes in both the mean and persistence of inflation over this period. After the crisis, the mean of inflation is slightly lower than in the pre-crisis period, as indicated by the expected mean of 0.48 (α / (1-β1 -β2)). In contrast, the mean of inflation during the crisis period is 2.5 (α / (1-β1 -β3)). 10

15 Table 4: Testing for Changes in Inflation Persistence Overall Period Specification πt = α + β 1 πt-1 + β2 (D1 *πt-1) + β3 (Dcrisis*πt- 1) + γ 1 D1+ γ2 Dfuel2005 α 0.46** (0.16) β1 0.42** (0.12) β2-0.30** (0.13) β3 0.33* (0.10) γ (0.16) γ2 8.04** (0.06) T 198 Adjsted R Std. error reg LM AR(p): T*R p (values) (0.12) α/(1- β1) 0.79 α/(1- β1 β2) 0.48 α/(1- β1 β3) 2.50 Notes: Standard errors reported in parentheses. *, ** denote significance at the 5% and 1% confidence levels π is inflation rate. Dcrisis is 1 for 1998, -1 for 1999, and 0 otherwise.dfuel2005 is dummy for October 2005 D1 is 1 for 2000 onward and 0 otherwise We find that the degree of inflation persistence falls significantly in the post-crisis period, as given by the sign of β 2 (d 1 * π t-1 ). Comparing β1 and β2 in Table 4, we find that the degree of persistence falls from 0.42 to 0.12 in the post-crisis period. During the crisis period we find that inflation persistence is high, estimated at 0.75 (β 3 ). Since there is evidence of a change in inflation persistence, we also estimate two separate regressions for the pre- and post-crisis sub-periods. Splitting the sample to take into account for the crisis period confirms the above finding of a substantial fall in persistence. Persistence is estimated at 0.23 in the pre-crisis period and 0.12 in the postcrisis period. Based on the Wald test, we find this difference in persistence to be statistically significant. 11

16 Table 5: Comparison of Inflation Persistence in the Pre and Post-crisis Periods Pre-crisis Post-crisis Specification πτ = α + β1π t-1 + β2 π t-2 + β3 π t-3 π t = α + β1π t-1 + γ2 Dfuel2005 Estimate of persistence (β 1 ) Headline inflation Sample (monthly) Notes: π is inflation rate, Dfuel93 is dummy for January 1993, Dfuel2005 is dummy for October 2005 An alternative way to illustrate how the persistence parameters have changed over time is by rolling estimates (Figure 5). The results show there was a marked step-up in persistence in However, after the crisis period, the persistence parameter falls substantially to This result is again consistent with the previous results (Tables 4 and 5). The rolling estimates of the intercept shows an increase during the post-crisis period, especially following the reduction in the government fuel subsidies in October Figure 5: Rolling Estimates of the Persistence and Intercept Parameters Crisis-period Pre-crisis Period Post-crisis Period Pre-crisis Crisis Post-crisis Persistence estimates Intercept estimates One way to explore reasons for the change in inflation persistence is to examine the persistence of disaggregated CPI components (Table 6). We find the degree of persistence varies considerably across components of the CPI. The degree of persistence is relatively high in the housing, food, and processed food components during the precrisis period. High persistence in housing probably reflects the cost of housing items, household operations, and fuel. It may also reflect the fact that changes in contracts for 5 We employ rolling regression (fixed 48-month windows, moving the start date from 1990 to 2006). 12

17 rent are generally indexed to the past inflation rate. In the post-crisis period, we find that the components with high persistence are education, food and processed food. The relatively high persistence in the education component is due to the fact that since the crisis, education fees have been set on an annual basis (July September) and are most likely indexed to the CPI. Table 6: Disaggregate Inflation Persistence in the Pre- and Post-crisis Periods Subgroup Weight of subgroup in CPI expenditure Inflation persistence parameter ( ρ ) Aggregate (weighted average Disaggregate persistence of CPI expenditure) Changes Pre-crisis Post-crisis Pre-crisis Post-crisis Pre-crisis Post-crisis Food Processed food, beverages, 0.17 and tobacco Housing Clothing Health Education, recreation, and 0.09 sports Transportation and 0.13 communication Disaggregate (weighted average of CPI expenditure) We find that the housing, food and transportation components appear to be the main drivers of the change in aggregate persistence. High volatility for the prices of fuel items, which are included in the housing and transportation components, and sudden large shocks, generally brought about by government policy changes to administered prices, may have resulted in lower persistence of those components after the crisis. In addition, more volatile food prices (especially rice) probably resulted in the decline in food price persistence in the post-crisis period. The degree of persistence in the weighted average components of the CPI is lower than that in the aggregate CPI. This result could reflect a number of factors. First, idiosyncratic shocks at the sub-component level may wash out in the aggregate CPI. As a result, shocks to the aggregate CPI will largely be common shocks. Second, higher persistence is found for sub-components with a relatively larger expenditure weight, such as food, housing and processed food, thereby resulting in higher persistence in the aggregate CPI. This result is, to some extent, similar to results found for the US (Clark, 2003). Our estimated results for persistence of both headline inflation and its components, in the post crisis period, are significantly lower than those found for core 13

18 inflation by the previous Bank Indonesia study (Inflation Team, 2006). Measured persistence of the core inflation components are all greater than 0.7, as shown in Table 7. These estimates of higher persistence for core inflation probably stem from several factors. First, core inflation excludes both the volatile food and administered prices components. Second, the previous study examines monthly year on year percentage changes, rather than monthly or quarterly percentage changes. Finally, the method used is time-varying mean reversion, which depends on the assumption of a long-run equilibrium level of inflation. The time-varying mean is constructed by using a Hodrick-Prescott filter. Table 7: Core Inflation Persistence CPI Components Degree of persistence Food 0.87 Housing 0.86 Clothing 0.75 Education 0.81 Health (Medical care) 0.88 Transport Why Has Inflation Become Less Persistent? The results above have shown that inflation persistence tended to decline after the crisis. Since the model used in this study is very simple (AR(p)) and does not describe the structural relationship between inflation and its explanators, our models do not determine the factors that might have influenced changes in persistence over time. There are a number of reasons why inflation in Indonesia may have become less persistent recently. First, price setters have been facing more unpredictable or uncertain economic environment during post crisis period than before crisis when depreciation rate was predictable and industrial fuel price was fixed for a quite certain period of time. Second, changes in government policy, especially one-off changes in fuel prices, has probably led to a large decrease in persistence for components such as housing, food, and transportation, as we also described in the previous analysis. Third, changing expectations may have contributed to the decline in inflation persistence. The consumer expectations survey suggests that there has been a declining trend in inflation expectations over (Figure 6). Moreover, a previous Bank Indonesia study (Mochtar and Wibowo, 2005) found that during the effect of past inflation on expectations started to decline by the end of the sample period. Although Bank Indonesia s inflation target is yet to significantly influence inflation expectations, the signal of monetary policy through the SBI rate has influenced inflation expectations. 14

19 Figure 6: Consumer Expectation on Inflation Consumer expectation Trend of consumer expectation Source: BI 6. Conclusion The results of this paper can be summarized as follows. First, using a simple univariate approach this study shows that the previous finding of high inflation persistence in Indonesia is not robust. Once we control for breaks, such as the crisis period and October 2005, inflation tends to exhibit low persistence. It indicates that high inflation persistence is not intrinsic in the Indonesian economy. Second, the results are robust to various measures of persistence. Using different frequencies of the inflation series, as well as disaggregate components of the CPI to measure persistence produces reasonably similar results. Third, by comparing the pre- and post-crisis periods using separate sub-period equations, we find that inflation persistence has changed during In the post-crisis period (2000 onwards) inflation in Indonesia seems less persistent than in the pre-crisis period ( ). A finding of a decline in inflation persistence is quite robust to various estimation methods, such as the inclusion of a dummy variable to capture changes in the persistence estimates, separating the sample periods, and rolling regressions. Using disaggregate CPI data also provides similar evidence of a decline in inflation persistence after the crisis. Lower inflation persistence in the post-crisis period partially relates to declining inflation expectations. Less persistent inflation in the post-crisis period implies that after a shock, inflation tends to return more quickly to its pre-shock level. Hence, in the post-crisis period monetary policy may only require small adjustments to achieve the inflation target. Moreover, the anchoring of inflation expectations to the target will be enhanced by the increasing credibility of the inflation targeting framework. This will also help increase the speed at which inflation will adjust to its expected level following a shock. 15

20 Since our findings are generated from a very simple model, the results of this study should be interpreted carefully. In future work it would be interesting to use a multivariate approach in order to analyze the factors that have caused the changes in inflation persistence. It would be especially useful to analyze the extent to which the change in the monetary policy regime has contributed to the decline in inflation persistence and to examine other factors that influence the dynamic behavior of inflation. 16

21 References Andrews, D and W Chen (1994). Approximately median-unbiased estimation of autoregressive models. Journal of Business and Economic Statistics, 12, pp Andrews, D (1993). Exactly Median-Unbiased Estimation of First Order Autoregresive/Unit Root Models. Econometrica, Vol. 61, No. 1.. Batini, N Euro area inflation persistence. ECB Working Paper No Cecchetti, S., and Debelle, G Has the inflation process changed? BIS Working Paper no Clark. T.E Disaggregate evidence on the persistence of consumer price inflation? Federal Reserve Bank of Kansas. Gali. J. Has inflation process changed? A Comment. BIS Working Paper no Hutabarat, A Determinan inflasi di Indonesia. Working Paper. Bank Indonesia. Honroyiannis G, and Lazaretou, S Inflation persistence during periods of structural change: an assessment using Greek data. ECB Working Paper No Levin, A.T., and Piger, J.M Is inflation persistence intrinsic in industrial economies? Working Paper, Federal Reserve Bank of St. Louis. Marques,C.R Inflation persistence: facts or artefacts? ECB Working Paper No Miskhin, F.S Inflation Dynamics. NBER Working Paper No Mochtar, F and Wibowo,W.A Determinan ekspektasi inflasi: Peran target inflasi masih tidak signidikan, sedangkan pengaruh suku bunga mulai meningkat. Tidak dipublikasikan. Poonpatpibul C., Vongsinsirikul P., and Chantanahom. P Exploring inflation in Thailand through sectoral price setting behaviour and underlying trend. Bank of Thailand Discussion Paper. Paya, I, Duarte A., and Holden K On the relationship between inflation persistence and temporal aggregation. Working Peper No. 37. Lancaster University Management School. Taylor, A.M Potential Pitfalls for the purchasing-power-parity puzzle? Sampling and specification biases in mean-reversion test of the low of one price. NBER Working paper series No Tim Inflasi (Inflation Team) Persistensi inflasi inti. Isu Strategis. Bahan Rapat Dewan Gubernur 4 Januari, Bank Indonesia. Willis, J.L. (2003). Implication of structural changes in the U.S. Economy for pricing behaviour and inflation dynamic. Federal Reserve Bank of Kansas. Economic Review, First Quarter

22 Appendix A: Estimated Models for Aggregate Inflation Table A1: Estimated of Persistence for Headline inflation Including crisis period Without dummy crisis With dummy crisis Specification π t = α + β 1 π t-1 π t = α + β 1 π t-1 + γ1dcrisis + γ 2 Dfuel γ3 Dfuel93 Excluding crisis period π t = α + β 1 π t-1 α 0.37** 0.47** 0.56** (0.10) (0.09) (0.06) β ** 0.44** 0.15** (0.07) (0.10) (0.05) γ1 1.49* (0.64) γ2 7.92** (0.07) γ3 2.16** (0.07) ρ T Adjsted R Std. error reg LM AR(p): T*R p (values) (0.14) (0.00) (0.27) Notes: Standard errors reported in parentheses. *, ** denote signifi cance at the 5 and 1 per cent confidence levels. π is inflation rate, Dfuel93 is dummy for January 1993, Dfuel2005 is dummy for October 2005 Dcrisis is 1 for 1998, -1 for 1999, and 0 otherwise Table A2: Estimates of Persistence for Non-oil Inflation Including crisis period Without dummy crisis With dummy crisis Specification π t = α + β 1 π t-1 π t = α + β 1 π t-1 + γ1dcrisis + γ 2 Dfuel γ3 D971 Excluding crisis period π t = α + β 1 π t-1 α 0.35** 0.46** 0.54** (0.10) (0.09) (0.06) β ** 0.46** 0.20** (0.05) (0.06) (0.06) γ1 1.43** (0.25) γ2 4.71** (1.00) γ3 4.95** (1.00) ρ T Adjsted R Std. error reg LM AR(p): T*R p (values) (0.27) (0.00) (0.05) Notes: Standard errors reported in parentheses. *, ** denote significance at the 5 and 1 per cent confidence levels π is inflation rate, Dfuel93 is dummy for January 1993, D971 refers to January 1997 Dcrisis is 1 for 1998, -1 for 1999, and 0 otherwise, Dfuel2005 is dummy for October

23 Specification Table A3: Estimates of Persistence for the Pre- and Post-crisis Periods Pre-crisis Post-crisis πτ = α + β1π t-1 + β2 π t-2 + β3 π t-3 + γ1 π t = α + β1π t-1 + γ2 Dfuel2005 Dfuel93 α 0.46** 0.58** (0.12) (0.08) β ** 0.12** (0.10) (0.06) β (0.10) β ** (0.10) γ1 2.25** (0.08) γ2 8.04** (0.62) ρ T Adjsted R Std. error reg LM AR(p): T*R p (values) (0.28) (0.55) Notes: Standard errors reported in parentheses. *, ** denote significance at the 5 and 1 per cent confidence levels. π is inflation rate, Dfuel93 is dummy for January 1993, Dfuel2005 is dummy for October

24 Appendix B: Estimated Models for Disaggregate Inflation Specification Table B1: Estimates of Persistence for Food Prices Pre-crisis Post-crisis πt = α + β1π t-1 + γ1 Dpros + γ2 Dpros2 + γ3 Dpros3 + γ4 Dpros4 π t = α + β1π t-1 + β2 π t-1+ γ5 Dpros5 + γ6 Dpros6 + γ7 Dfuel2005 α 0.32** 0.46** (0.04) (0.08) β ** 0.22* (0.06) (0.08) β 2 γ1 1.06** (0.27) γ2 1.52** (0.36) γ3 1.04** (0.36) γ4 0.94** (0.18) γ5 0.94** 1.49** (0.18) (0.17) γ6 0.94** -0.78** (0.18) (0.13) γ7 0.94** 2.49** (0.18) (0.07) ρ T Adjsted R Std. error reg LM AR(p): T*R p (values) (0.58) (0.29) Notes: Standard errors reported in parentheses. *, ** denote significance at the 5 and 1 per cent confidence levels. π is inflation rate, Dpros is seasonal dummies for processed food, beverage, and tobacco, Dfuel2005 is dummy for October 2005 Table B2: Estimates of Persistence for Housing Prices Pre-crisis Post-crisis Specification πt = α + β1π t-1 + γ1 Dhouse π t = α + β1π t-1 + γ2dfuel γ3 Dhouse2 α 0.38** 0.59** (0.07) (0.05) β ** 0.08 (0.09) (0.04) γ1 2.01** (0.19) γ2 6.77** (0.35) γ3 1.12** 0.84 ρ T Adjsted R Std. error reg LM AR(p): T*R p (values) (0.58) (0.04) Notes: Standard errors reported in parentheses. *, ** denote significance at the 5 and 1 per cent confidence levels. π is inflation rate, Dhouse is dummies for housing, Dfuel2005 is dummy for October

25 Table B3: Estimates of Persistence for Processed Food, Beverages, and Tobacco Prices Specification Pre-crisis Post-crisis πt = α + β1π t-1 + γ1 Dpros + γ2 Dpros2 + γ3 Dpros3 + γ4 Dpros4 π t = α + β1π t-1 + β2 π t-1 + γ5 Dpros5 + γ6 Dpros6 + γ7 Dfuel2005 α 0.32** 0.46** (0.04) (0.08) β ** 0.22* (0.06) (0.08) β 2 γ1 1.06** (0.27) γ2 1.52** (0.36) γ3 1.04** (0.36) γ4 0.94** (0.18) γ5 0.94** 1.49** (0.18) (0.17) γ6 0.94** -0.78** (0.18) (0.13) γ7 0.94** 2.49** (0.18) (0.07) ρ T Adjsted R Std. error reg LM AR(p): T*R p (values) (0.58) (0.29) Notes: Standard errors reported in parentheses. *, ** denote significance at the 5 and 1 per cent confidence levels. π is inflation rate, Dpros is seasonal dummies for processed food, beverage, and tobacco Dfuel2005 is dummy for October

26 Table B4: Estimates of Persistence for Clothing Prices Pre-crisis Post-crisis Specification πt = α + β1π t-1 + γ1 Dcloth π t = α + β1π t-1 + β2 π t-2 + γ2 Dcloth2 α 0.34** 0.56** (0.05) (0.09) β ** (0.06) (0.04) β2-0.23** (0.09) γ1 1.71** (0.12) γ2 3.10** (0.58) ρ T Adjsted R Std. error reg LM AR(p): T*R p (values) (0.56) (0.54) Notes: Standard errors reported in parentheses. *, ** denote significance at the 5 and 1 per cent confidence levels. π is inflation rate, Dcloth is dummies for clothing, Dfuel2005 is dummy for October 2005 Table B5: Estimates of Persistence for Health Prices Pre-crisis Post-crisis Specification πt = α + β1π t-1 + β2π t-2+ β3π t-3 π t = α + β1π t-1 + β2 π t-2 + γ1 Dhealth + γ2 Dhealth2 α 0.85** 0.31** (0.15) (0.04) β * 0.28** (0.07) (0.07) β2-0.21** (0.07) β3 0.44** (0.10) γ1 0.88** (0.10) γ2 2.07** (0.24) ρ T Adjsted R Std. error reg LM AR(p): T*R p (values) 0.00 (0.81) Notes: Standard errors reported in parentheses. *, ** denote significance at the 5 and 1 per cent confidence levels. π is inflation rate, Dhealth is dummies for health 22

27 Table B6: Estimates of Persistence for Education Prices education Pre-crisis Post-crisis Specification πt = α + β1π t-1 + γ1 Dedu + γ2 Dedu2 πt = α + β1π t-1 + γ3 Dedu3 + γ4 Dedu4 α 0.40** 0.27** (0.07) (0.08) β * 0.21** (0.04) (0.04) γ1 1.79** (0.13) γ2 7.63** (3.99) γ3 3.56** (0.23) γ4 9.00** (0.06) ρ T Adjsted R Std. error reg LM AR(p): T*R p (values) (0.20) (0.81) Notes: Standard errors reported in parentheses. *, ** denote significance at the 5 and 1 per cent confidence levels. π is inflation rate, Dedu is dummies for education Table B7: Estimates of Persistence for Transportation and Communication Prices Pre-crisis Post-crisis Specification πt = α + β1π t-1 + γ1 Dfuel1 + γ2 Dfuel2 πt = α + β1π t-1 + γ2 Dfuel γ3 Dfuel g4 Dfuel5 α 0.12** 0.52** (0.05) (0.09) β ** 0.01 (0.02) (0.02) γ1 6.98** (0.19) γ2 1.58** 28.0** (0.29) (0.74) γ3 9.51** (0.74) γ4 4.72** (0.43) ρ T Adjsted R Std. error reg LM AR(p): T*R p (values) (0.18) (0.01) Notes: Standard errors reported in parentheses. *, ** denote significance at the 5 and 1 per cent confidence levels. π is inflation rate, Dfuel is dummies for fuel increase 23