This note compares patterns of domestic investment in Serbia with those in other central and eastern European countries, noting the relationships with external balances. The structure of participation and employment rates suggests a need for analysis of the impact of labor market institutions on youth and women. A further focus on redeployment services would be appropriate. The Serbian banking system, the implications of the structure of Serbia’s economy, the operational framework of monetary policy, and the adoption of an inflation targeting regime have been discussed.

Abstract

This note compares patterns of domestic investment in Serbia with those in other central and eastern European countries, noting the relationships with external balances. The structure of participation and employment rates suggests a need for analysis of the impact of labor market institutions on youth and women. A further focus on redeployment services would be appropriate. The Serbian banking system, the implications of the structure of Serbia’s economy, the operational framework of monetary policy, and the adoption of an inflation targeting regime have been discussed.

VII. Inflation23

1. The National Bank of Serbia (NBS) has recently proposed policy shifts anticipating eventual adoption of an inflation targeting regime. Amongst other things, this would require identifying reliable predictors of future inflation. The nominal exchange rate is one candidate. In this regard, understanding exchange rate pass-through is key. The new monetary regime will need to be designed with this behavior in mind, and drawing on the experience of other countries. This note considers the empirical evidence on exchange rate pass-through into core inflation24 and reviews the experience of several other emerging market economies in dealing with various aspects of transition to inflation targeting.

A. Exchange Rate Pass-Through to Core Inflation

General Considerations

2. Several studies have attempted to estimate exchange rate pass-through in Serbia. NBS (2004) employed a one-equation error-correction model and found a pass-through of around 0.32-0.35 in 1997-2004. Gorbanyov (2005) examined VAR and VECM models and found that the pass-through rates vary across model specifications and sample periods. In particular, pass-through into core prices was found to be between 0.7 and 0.9 within a year using 2001-04 data, but dropped to about 0.4 when 1997-2004 data was used. Petrovic and Mladenovic (2005) also report low pass-through estimates using both a single-equation autoregressive distributive lag (ADL) model and a VECM model—about 0.5 using monthly data for 2001:7-2005:9. However, interpretation of these results—and their variety—should take account of a number of limitations in the studies.

3. Data limitations suggest that structural models for Serbia need to be used with care. The short data sets imply significant loss of degrees of freedom in large structural models and they compromise efforts to verify cointegration (Text Table 1).25 And despite the short time period, structural breaks are a significant further difficulty—including the hyperinflation and disinflation between 1999 and 2002 (Text Figure 1), and various changes in the exchange rate regime, including post February 2006 (Text Figure 2). Moreover, the quality of the statistical series—notably those included in unit labor cost data—may be low.26

Text Table 1.

Core Price Index and Exchange Rate, Number of Cointegrating Relations 1/

article image

Test results are based on Johansen’s cointegration test and an assumed 5 percent critical level.

Text Figure 1.
Text Figure 1.

Serbia: Nominal Exchange Rate and Retail Prices Twelve-month growth (m/m(-12)), percent

Citation: IMF Staff Country Reports 2006, 382; 10.5089/9781451833638.002.A007

Text Figure 2.
Text Figure 2.

Serbia: Daily Exchange Rate Dinar/Euro, Feb 01, 2005 - July 21, 2006

Citation: IMF Staff Country Reports 2006, 382; 10.5089/9781451833638.002.A007

Empirical Analysis

4. Analysis can most effectively start with visual inspection of the non-seasonally adjusted data.27 Text Figure 3 plots core inflation and the nominal exchange rate at monthly, three-month, and twelve-month frequencies. Three-month and twelve-month growth series capture movements in the exchange rate that persist over their respective horizons. This is suggestive of asymmetric pass through:

Text Figure 3.
Text Figure 3.

Serbia: Exchange Rate and Core Price Index

Citation: IMF Staff Country Reports 2006, 382; 10.5089/9781451833638.002.A007

  • Pass-through seems to be larger for more persistent exchange rate shocks. Co-movement of the exchange rate and core inflation is less evident in monthly changes—dominated by transitory and seasonal shocks, whereas three- and twelve-month trends exhibit stronger correlation.

  • Pass-through may be slower when depreciation rates decline. This is suggested by the middle and lower charts of Figure 3. The two episodes of persistent decline in depreciation rates are the pre-2003 exchange rate-based disinflation and the period of high capital inflows combined with greater exchange rate flexibility (from late 2005 onwards). In both episodes inflation responds with a lag to the deceleration of depreciation, perhaps reflecting sticky inflationary expectations. In contrast, between roughly March 2003 and September 2005 twelve-month depreciation and inflation were both on an upward, although uneven, trend. The lead time in this period appears shorter and the correlation larger.

  • However, there may be other explanations for the latter apparent asymmetry: prior to 2003, an equilibrating real exchange rate appreciation may have occurred following the hyperinflation and termination of economic sanctions; and from late 2005, a change in policy regime (accommodation of greater exchange rate volatility) and/or feed through to core inflation from rising international energy prices may account for the apparently altered relationship between the exchange rate and core inflation.

5. In light of the various issues noted above, a single-equation autoregressive distributed lag model was used to estimate exchange rate pass-through. Thus, the following equation was estimated:

πt=c+ρπt1+αst+ut

where πt is core inflation and st is nominal exchange rate depreciation. Furthermore, to explore possible asymmetries, two samples were considered:

  • 2003:3 through 2006:5 – full post-hyperinflation sample; and

  • 2003:3 through 2005:9 – the period between the two main structural breaks which also corresponds to a rising trend in depreciation.

Comparison of the parameter estimates from the two data sets allows some insight into the possible different pass through behavior when exchange rate depreciation rates are declining. And insight into the possible dependence of pass through on the persistence of exchange rate shocks is obtained by comparing parameter estimates obtained from monthly series with those based on the three-month series. The results are reported in Text Table 2.

Text Table 2.

Exchange Rate Pass-Through Estimates 1/

article image

Standard errors are reported in parentheses

Sum of future expected changes in inflation following a one-time change in depreciation rate.

Pass-through of a one time permanent change in the level of exchange rate into level of core price index.

6. The findings suggest high pass-through and possible asymmetries. Using monthly growth rates, average exchange rate pass-through to the price level within one year was about 74 percent during the episode of a rising trend in depreciation and no apparent structural breaks, but dropped to just over 45 percent in the full sample. In principle, a lower estimate in the full sample can be attributed either to the structural break or the deceleration trend in depreciation, or both. It would be important to revisit this distinction once more data is available. However, a high noise-to-signal ratio in monthly series and the resulting poor model fit qualifies confidence in the results. Using three-month growth rates considerably strengthens both the model fit and pass-through coefficients which are above unity for both data periods.28 While the above unity estimates are not interpretable as conventional pass-through coefficients,29 they are still useful and suggest that changes in the depreciation trend, as opposed to short-lived changes, could be expected to pass-through almost fully into core inflation.

7. Possible asymmetries of pass-through have important implications for policy. First, policy would need to take account of these—anticipating faster price responses to depreciation than from exchange rate stabilization. Second, high noise-to-signal ratios in monthly inflation warrants caution in policy makers’ reaction to short-term fluctuations. Finally, slower pass-through in times of decelerating depreciation has implications for estimates of the costs of disinflation in terms of loss of competitiveness. Amongst other things, this highlights the importance of a sufficiently supportive fiscal policy in times of disinflation.

8. In sum, technical difficulties are reflected in the wide range of the pass through estimates, requiring care in their interpretation. On one hand, pass through coefficients obtained using three-month series are unrealistically high, due to persistence of the series. On the other hand, structural breaks (and/or asymmetry of pass-through with respect to low depreciation) casts doubt on the validity of low pass-through estimates in the full sample estimates. On balance, twelve-month pass-through of exchange rate shocks still seems high, over 70 percent, at least in times of rising depreciation.

B. Inflation Targeting: Emerging Market Experience30

9. Experience in other emerging market countries which adopted inflation targeting (IT) in the context of uncertainties in their empirical understanding of the determinents of inflation is instructive. Though most of these countries exhibited lower apparent degrees of exchange rate pass through, they generally had similar types of uncertainties on the behavior of inflation. So their experience—including on missing targets, disinflation prior to adopting IT, speed of transition towards IT, formulation of inflation targets, and steps towards developing an efficient forecasting framework—can help policy design in Serbia. The cases of Chile, Brazil, Czech Republic, Poland, Israel, and South Africa are highlighted below.

Escape Clauses and Accountability

10. Given the novelty of the regime in Serbia, there will be—at least in the initial stages of transition towards IT—some likelihood that pre-announced targets/objectives will be missed. Recognizing and explicitly integrating this into the monetary policy framework is important for building credibility. Central banks (CB) sometimes explicitly articulate the circumstances—typically outside of their control—under which such breaches would be tolerated (escape clause) and/or prescribe specific actions in the case of a target miss (explanation or accountability clauses).

11. Of the six countries reviewed here, only the Czech Republic currently employs an explicit escape clause. The clause is published on the central bank’s website:31

  • “In the inflation targeting regime, the need for escape clauses derives from the occurrence of large shock changes in exogenous factors (particularly supply-side shocks) that are completely or largely outside the purview of central bank monetary policy. Attempts to keep inflation on target in these circumstances might cause undesirable volatility of output and employment. If such a shock deflects projected inflation from the target, the CNB does not respond to the primary impacts of the shock. It will apply an exemption (escape clause) from the obligation to hit the inflation target and accept the deviation of the inflation forecast from the target caused in this way. There are a whole range of shocks which can create room for applying such escape clauses. They include, for example, major deviations in world prices of energy raw materials or major deviations in agricultural producer prices. A specific type of exogenous shock is administrative measures that have strong price impacts, in particular major changes in the structure or rates of indirect taxes and major changes in the segment of regulated prices.”

12. South Africa has had an escape clause in the past. It cited unpredictable events that could affect the inflation rate—e.g. significant changes in terms of trade, natural disasters, and interruption of external capital flows—as the circumstances under which the central bank could not be expected to reach the target. In November 2003 the escape clause was replaced with an explanation clause which required the Reserve Bank to explain reasons for deviations from the target and indicate by when the inflation rate was expected to return within the target range (see Merwe (2000)).

13. Explanation clauses are in place in Israel and Brazil as well as in some mature inflation targeters (e.g., Canada, Switzerland, and New Zealand). In Israel the CB is required to offer a public explanation in cases of deviations from the target of more than 1 percent. In Brazil, the CB must issue an open letter to the Ministry of Finance explaining reasons for the target breach and measures taken and the time required to bring inflation back on track.32 Poland and Chile do not employ either of the clauses.

Is Achieving Low Inflation Essential Before Transition Towards Inflation Targeting?

14. The argument for reducing inflation first is that some period of low inflation could help establish the central bank’s track record, thus increasing its credibility. Disinflation might also be important to test the potency of the policy instrument before a regime switch. Nevertheless, the experience of the six countries suggests that this may not be necessary. For example, Chile, Israel and Poland have started their transition with double digit inflation rates (see Text Table 3).

Text Table 3.

Inflation Prior to the Transition to Full-Fledged Inflation Targeting 1/

article image
Source: Schaechter et. al. (2000)

t is time of the start of the transition to full-fledged inflation targeting framework.

How fast have countries moved towards a formal IT regime?

15. The speed of transition has varied widely. Except for South Africa, all countries reviewed here employed a crawling peg before moving to IT (South Africa was targeting broad money growth and inflation and had a flexible exchange rate regime). Brazil and Czech Republic switched to formal inflation targeting immediately. Other countries moved more gradually, from 9 years in Chile to 3 months in Poland (see Text Table 4). In general, the speed of transition should depend on the degree of confidence by the CB in its ability to effectively control inflation.

Text Table 4.

Transition to Inflation Targeting

article image
Source: Schaechter et. al. (2000)

On Specification of Targets

16. Central Banks adopting IT are faced with three closely related decisions:

  • Whether to target a point or a range.

  • Appropriate horizon of the target.

  • Price index to be targeted.

17. The key trade-off with respect to all three decisions is the ease of communication of policy to the public on one hand and the ability to deliver on the other. In particular, it is generally easier to communicate an annual point target for a recognizable price index (such as CPI). From the perspective of implementation, however, the preference would be towards hitting a range, over a longer-term horizon, and for a price index over which the CB has a greater degree of control (such as core CPI).

18. The ability to target headline inflation depends on the importance of prices that are inherently volatile (e.g., fuel) or not market-determined. The six central banks considered here currently target headline inflation. The Czech National Bank had started with targeting net inflation, which excluded administered prices, but switched to targeting headline CPI in 2004. In part, such switch was justified by a relatively low share of administered (10 percent) and fuel (5-6 percent) prices in the CPI. In Poland, the share of administered prices is higher (25 percent), but still well below the level of Serbia, where 45 percent of the prices are regulated, either directly or indirectly.

19. All six central banks currently specify target ranges with a mid-point, rather than single point targets. The width of the ranges has varied over time. For example, Israel and Chile started with a width of 5 percent (range of 15-20 percent) which was gradually brought down to 2 and 1 percent respectively (range of 2-4 percent and 2.5-3.5 percent).

Data Shortcomings and Forecasting33

20. A reliable framework for forecasting inflation is crucial. Such a framework is key for the determination of appropriate targets as well as for enhancing central banks’ understanding of the transmission mechanism, including lags with which variations in the policy instrument translate into changes in inflation. Forecasting is especially challenging for emerging market countries, prompting central banks to employ a wide variety of tools to forecast inflation and formulate reasonable targets. In practice, inflation forecasts are based on a combination of indicator variables and qualitative judgment, time series econometric models and structural quantitative economic models.

21. Use of inflation indicators is important when formal modeling is not feasible. An inflation indicator refers to observable data that have been previously shown to provide useful signals regarding future changes in inflation. Various inflation indicators could be monitored on a regular basis to form a judgmental view of possible future price developments, including aggregate demand and supply variables, monetary aggregates, interest rate and—as discussed above—the exchange rate, past inflation, alternative price measures, and expectations (obtained through surveys).

22. When available data allow modeling, forecasting can be done through either time series or structural models, or both. Time series models impose less economic structure but usually provide better short-term forecasts, as well as a consistency check for the larger structural models. They are easy to produce and allow for frequent assessment of economic conditions. For example, ARIMA models rely mostly on the information within the variable in question and produce forecasts based on internal dynamic properties of the series. Since most macroeconomic series are highly persistent, the forecasts obtained in this fashion are usually accurate at very short horizons, but deteriorate rapidly as the forecast period expands.34 Inclusion of other explanatory variables or more complicated relationships (such as VARs) requires a simultaneous forecast of all variables, with ambiguous net effects on accuracy. Nevertheless, these models are also very popular among central banks.

23. Structural models used by central banks to forecast inflation have important common elements. These include an open economy demand curve, a (New Keynesian) Phillips curve, an international asset market equilibrium condition (some version of uncovered interest parity with adjustment for risk premium) and a monetary policy reaction function. These elements have been used extensively in empirical work on central bank behavior, monetary policy rules, and sacrifice ratios.

24. Emerging economies have traditionally relied less on structural models. This reflects (i) their relatively short experience with inflation targeting that complicates the generation of reliable estimates based on structural models; and (ii) ongoing changes in structural relationships. Nevertheless, at present, Brazil, the Czech Republic, and Israel all work with three or four equation models along the lines of those used by industrial country central banks. There are clear benefits to using structural models. Besides forecasting, they provide an essential framework which helps policy makers think through various policy transmission channels and frame policy discussions within a coherent summary of their understanding of the economy.

Appendix Table A.1.

Serbia: Composition of the Non-Core Price Index, 2006

article image
Source: National Bank and Ministry of Finance of Serbia.

References

  • Gorbanyov, Michael, 2005, “Inflation Determinants in Serbia,” Serbia and Montenegro – Selected Issues and Statistical Appendix, IMF Country Report, No. 05/232 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Merwe A., 2004, “Inflation Targeting in South Africa,” Occasional Paper, No. 19 (South Africa: South African Reserve Bank).

  • Mishkin, Frederic and Klaus Schmidt-Hebbel, 2002, “A Decade of Inflation Targeting in the World: What Do We Know and What Do We Need To Know?” in Loyaza Norman and Raimundo Soto eds. Inflation Targeting: Design, Performance, Challenges.

    • Search Google Scholar
    • Export Citation
  • NBS, 2004, “What Influences Inflation Rate In Serbia?” Economic Review (Serbia: National Bank), pp. 29 –39.

  • Petrović, Pavle and Zorica Mladenović, 2005, “Econometric Modeling of Inflation in Serbia,” (Belgrade: University of Belgrade, manuscript).

    • Search Google Scholar
    • Export Citation
  • Schaechter, Andrea, Mark R. Stone Mark Zelmer, 2000, “Adopting Inflation Targeting: Practical Issues for Emerging Market Economies,” Occasional Paper, No. 202 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Stanic, K., 2005, “Registered Employment and Wages – Statistic Data and Trends 2000–05,” in Quarterly Monitor, (Belgrade: Center for Advanced Economic Studies).

    • Search Google Scholar
    • Export Citation
  • World Bank, 2006, “Serbia: Labor Market Assessment,” (Washington: World Bank).

23

Prepared by Tokhir Mirzoev, with contributions from Janko Guzijan.

24

Core prices comprise about 52 percent of the retail price index and mostly exclude agricultural and regulated prices (see Appendix Table A.1.).

25

For example, Gorbanyov (2005) finds strong evidence of cointegration between prices and exchange rates in their sample, whereas Petrovic and Mladenovic (2005) argue the opposite and suggest a cointegrating relationship between exchange rate, core prices, unit labor costs and their measure of output gap.

26

Stanić (2005) and World Bank (2006) discuss shortcomings of the monthly wage and employment data.

27

The short length of the series weakens efficiency of seasonal adjustment.

28

Using twelve-month changes (not reported here) strengthens this finding even further.

29

Whereas monthly growth rates seem to be dominated by noise, three-month growth rates pick up more persistent changes. However, because of overlapping periods in the three-month data, three-month changes also tend to overstate the degree of persistence and, therefore, estimates of pass-through rates.

30

A more detailed discussion can be found in Mishkin and Schmidt-Hebbel (2002) and Schaechter et. al. (2000).

32

Open letters for target misses in 2002-04 are available at http://www.bcb.gov.br/?INFLATION.

33

See Schaechter et. al. (2000), chapter 5, for further details.

34

The Central Bank of New Zealand, for example, relies on such models to produce two quarters ahead forecasts and uses large scale models for longer-term forecasting.

Republic of Serbia: Selected Issues
Author: International Monetary Fund