This Selected Issues paper analyzes trends in Tunisian inflation. It computes measures of core inflation, and develops a simple framework for forecasting inflation. The paper econometrically estimates the degree of exchange rate pass-through in Tunisia. The methodology relies on time series and panel estimation methods, using both monthly and quarterly data for a basket of 43 consumption products during 1995–2006. It summarizes the results of some empirical studies on developing and emerging economies. It also describes the fluctuations in the nominal effective exchange rate in Tunisia and the resulting effects on inflation.

Abstract

This Selected Issues paper analyzes trends in Tunisian inflation. It computes measures of core inflation, and develops a simple framework for forecasting inflation. The paper econometrically estimates the degree of exchange rate pass-through in Tunisia. The methodology relies on time series and panel estimation methods, using both monthly and quarterly data for a basket of 43 consumption products during 1995–2006. It summarizes the results of some empirical studies on developing and emerging economies. It also describes the fluctuations in the nominal effective exchange rate in Tunisia and the resulting effects on inflation.

I. Inflation in Tunisia: Trends, Core Inflation, and Forecasting1

A. Introduction

1. Tunisia is gradually moving toward full flexibility of its exchange rate and an inflation-targeting framework. Forecasting inflation will become a key task for the Central Bank of Tunisia (BCT) because of the time lags between monetary policy and its effects on the economy, particularly on inflation. Thus, to be able to react on time, the BCT will need to base its monetary policy decisions not on past inflation outcomes but on inflation forecasts. The precision with which inflation can be forecasted is a critical element of the inflation targeting framework. For example, it will determine the amplitude of the deviations that are allowed around the inflation target as well as the period over which average inflation should return to the mid-point. To forecast inflation, it is critical to understand Tunisian inflation dynamics and to explore ways to compute core inflation.

2. The literature suggests several technical requirements as prerequisites for inflation targeting, including:

  • (i) Institutional independence. Under inflation targeting, low inflation is the stated primary goal of monetary policy.

  • (ii) A healthy financial system. In order to minimize potential conflicts with financial stabilization objectives and guarantee effective monetary policy transmission, the banking system should be sound and capital markets well developed.

  • (iii) Good analytical capabilities and infrastructure. Data requirements for inflation targeting are more stringent than for alternative regimes and the monetary authorities must have a well developed capacity to forecast inflation.2

3. Tunisia has taken important steps toward an inflation targeting framework, particularly regarding (i) and (ii). Last year’s amendment to the BCT law strengthened the central bank independence and set price stability as the main objective of monetary policy. To invigorate the financial system, the authorities have recently adopted a number of measures aimed at improving the credit culture, promoting good governance, and reinforcing the legal framework for banks. However, a reliable methodology for forecasting inflation is not yet available.

4. The objectives of this paper are threefold: (a) analyze trends in Tunisian inflation, (b) compute measures of core inflation, and (c) develop a simple framework for forecasting inflation. The paper is organized around these objectives. The paper ends with some conclusions and policy recommendations.

B. Trends in Tunisian Inflation

5. What has been the behavior of Tunisian inflation and the different components of its CPI? This section compares Tunisia’s performance to that of its partners and other countries in the region, and describes the main statistics of the CPI components—weights, mean, median, standard deviation, and trends of each CPI component—over the whole sample period (1991M1-2006M12) and over two sub-periods (1991M1-99M12 and 2000M1-06M12).

6. Tunisia has achieved price stability for more than a decade. Supported by the government’s outward-oriented strategy, including a conscious price liberalization policy, inflation in Tunisia declined from about 10 percent in early 1990s to about 1 percent in 2005.3 Prudent monetary and fiscal policies, combined with structural reforms, played a key role in achieving this goal.

7. Inflation has been moderately volatile, but the volatility has declined since the inflation dropped to the 2-3 percent range. The volatility of headline inflation—measured by the standard deviation—was 1.7 over the whole period (January 1991-December 2006); it was reduced from 1.7 over January 1991-December 1999 to 1.2 over January 2000-December 2006. Since 2000, headline inflation fluctuated between 1 percent and 5 percent.

8. Tunisia outperforms a number of other Middle Eastern countries in terms of low inflation and it compares favorably to trading partners and comparator countries, as indicated in Figures I.1 and I.2.

Figure I.1.
Figure I.1.

CPI Inflation, Compared with Main Trading Partners and Comparator Countries

(1990M1–2006M12)

Citation: IMF Staff Country Reports 2007, 319; 10.5089/9781451837933.002.A001

Source: WEO.
Figure I.2.
Figure I.2.

CPI Inflation, Compared with Neighboring Countries

(1990M1–2006M12)

Citation: IMF Staff Country Reports 2007, 319; 10.5089/9781451837933.002.A001

Source: WEO.

9. To investigate the dynamics of Tunisian inflation, we analyze 43 CPI components.4 The weights and the main statistics of the components and of six groups (food, housing, clothing, health, transport, services and other) are presented in Appendix Tables I.1-I.3. Table I.1 shows weights and selected statistics for the six groups.

Table I.1.

CPI Components, Weights, and Selected Descriptive Statistics

(1991M1–2006M12)

article image
Sources: Tunisian authorities; and IMF staff calculation.

10. Some components of inflation have been much more volatile than others. For example, the volatility ranged from 0.4 to 10.2 over 2000-06; it ranged from 0.9 to 12.7 over 1991-99; and it ranged from 0.9 to 11.8 over 1991-2006 (see Appendix Tables I.1-I.3).

11. However, most volatile components are the same for the two sub-periods (Table I.2). For the sub-period 2000-06, energy and transport items are among the ten most volatile groups, reflecting mainly the increase in international prices for oil. Entertainment (spectacles, shows, and performances) and education—two groups of products for which all prices are administered—are not among the ten most volatile for this period. Their lesser volatility is explained by administrative decisions. The five most volatile components are food items: cooking oil, eggs, fruits, vegetables, and meat and poultry. These five items represent about 20 percent of the CPI basket.

Table I.2.

Ten Most Volatile CPI Components

article image
Sources: Tunisian authorities, and IMF staff calculation.

12. A simple regression confirms the persistence in volatility over the two sub-periods. The regression’s dependent variable is the standard deviation (SD) of the 43 CPI components over 2000-06, the independent variable is their SD over 1991-99 and a constant. Most of the SD differences between CPI components are explained by the previous period’s SD since the adjusted R-squared is above 70 percent. This result shows that CPI components that were volatile during 1991-99 remain volatile during 2000-06. However, the coefficient being less than one implies a reduction in volatility over time.

Table I.3.

Regression of Standard Deviation over 2000-06 on Standard Deviation over 1991-99

article image
Notes: The dependent variable is the Standard Deviation (SD) of the 43 CPI components over 2000-06. The adjusted R-squared is 0.73.

13. Food remains an important item in the total CPI. Food represents 36.5 percent of the total weight compared to 41.2 percent in the 1990 household survey. The second group, housing, accounts for about 18 percent; the other four groups represent between 10 percent and 13 percent.

14. Despite price liberalization, administered prices remain important. Prices of about 32 percent of the basket are still administered. The prices of some sectors are mainly administered (transport and health); others are partially liberalized (food, housing, and service); and clothing is fully liberalized. Analyzing detailed components shows that more than 50 percent of the prices of the following components are administered: tobacco (100 percent), telecom (100), common transport (100), care and treatment (100), energy (97), coffee and tea (87), cereals (71), sugar (59), and education (57).

15. Changes in administered prices are irregular. The month-on-month inflation of these components, shown in Figure I.3, illustrates clearly that inflation changes are discretionary for these items. Thus, Tunisia’s inflation is partially driven by administrative decisions.

Figure I.3.
Figure I.3.

Behaviors of Administered Prices

(month-on-month; 1990–2006)

Citation: IMF Staff Country Reports 2007, 319; 10.5089/9781451837933.002.A001

Sources: Tunisian authorities, and IMF staff calculations.

16. The large share of administered prices constitutes a major impediment to an inflation targeting framework. The BCT does not have any influence on one third of the CPI basket because the government fixes the administered prices. Even if the BCT coordinates with the government, the mechanism for adjusting administered prices introduces a fiscal bias into the control of inflation.5 Monetary policy looses its flexibility and effectiveness in the presence of a large share of administered prices. This is particularly important in the case of an inflation targeting framework where inflation forecasting and the ability to swiftly respond to shocks are crucial.

17. Recent fluctuations in overall CPI inflation have resulted mainly from increases in the prices of food items, transport, and energy. Food and transport have contributed to overall inflation more than their respective weights during 2000-06. Their contribution to overall inflation was about 56 percent compared to their weight of 47 percent (Table I.4 and Appendix Table I.4).

Table I.4.

Average Contribution to Overall Inflation by Different Groups

(In percent)

article image
Sources: Tunisian authorities, and IMF staff calculations.

18. Food and transport contribute significantly to the volatility of headline inflation (Figure I.4). While the standard deviation of inflation was 1.2 during January 2000 to December 2006, that of inflation without food was 0.8, and that of inflation without food and transport was 0.7. Measuring core inflation is the subject of the next section.

Figure I.4.
Figure I.4.

CPI Inflation With and Without Food and Transport Prices

(1991M1–2006M12)

Citation: IMF Staff Country Reports 2007, 319; 10.5089/9781451837933.002.A001

Sources: Tunisian authorities, and IMF staff.

C. Measures of Core Inflation for Tunisia

Based on the analysis of Section B, Section C presents three measures of core inflation. Our preferred measure is based on excluding the five most volatile CPI components.

19. A credible measure of inflation to be targeted is key to the success of inflation targeting. The headline CPI is generally modified for this purpose by excluding the most volatile CPI components such as food (because of weather conditions) and energy (because of supply shocks). Potentially, any volatile component of CPI may be excluded from the “core inflation”. However, the rationale for excess volatility should also be explained.

20. of course, there is no perfect measure of core inflation. There are many alternative approaches to the one based on exclusion.6 The choice of a particular measure of core inflation is country-specific. In some cases, headline inflation is a preferred target if the measure of core inflation is not credible.

21. We present three measures of core inflation based on exclusion. Two measures are based on excluding the most volatile CPI components and one on excluding administered prices. Exclusion-based core inflation measures have many advantages compared to other measures. They are transparent, easy to understand, replicable, credible, and available on a timely basis. For these reasons, exclusion-based methods are often used by countries in their initial stage of inflation targeting.

Excluding most volatile CPI components

22. The first measure eliminates CPI components that are considered to be particularly volatile. Thus, they give implicitly more weight to component price indices that are less subject to shocks. A widespread approach is simply to exclude certain product groups such as food and energy because of their perceived excessive volatility. However, this is problematic in emerging and developing countries, since food items represent a large share in the CPI basket. In the case of Tunisia, food counts for 36.5 percent of the basket. Such exclusions may erode the credibility of the measure.

23. Instead of excluding perceived volatile CPI groups, we examine 43 CPI components to determine which are the most volatile (see Section B). The usual shortcoming of this procedure is that the components that are found to be volatile may become relatively stable over time, and vice versa. However, in Tunisia the ten most volatile components are broadly the same for the two sub-periods. We computed two measures based on the volatility of CPI components, namely by eliminating the five and ten most volatile CPI components (Figure I.5). Over 2000-06, these two measures are respectively 42 and 53 percent less volatile than headline inflation. Nevertheless, they remain highly correlated with headline inflation; the simple correlation is about 0.7 for both measures. These measures remove about 20 percent and 37 percent from the CPI basket, respectively.

Figure I.5.
Figure I.5.

Core Inflation: Excluding Administered Prices; Top Five and Top Ten Most Volatile CPI Components

Citation: IMF Staff Country Reports 2007, 319; 10.5089/9781451837933.002.A001

Sources: Tunisian authorities, and IMF staff calculation.

Excluding administered CPI components

24. The second measure leaves out administered CPI components. Changes in administered prices are irregular and often large and thus have temporary effect on prices, warranting their exclusion. On the other hand, they form part of the inflationary process and should be included. We compute a measure of core inflation, excluding administered items, defined as items where administered prices cover over 50 percent of sub-components (see Figure I.3). Core inflation based on excluding administered CPI components is highly correlated with headline inflation; the simple correlation is 0.98 over 2000-06. However, it is 8 percent more volatile than headline inflation over the same period.7 This measure removes about 30 percent of the CPI basket.

25. Overall, the best measure is the one based on excluding the five most volatile CPI components. The measure excluding administered prices has a major shortcoming, in that it is not operational at a relatively high level of aggregation. In other words, it is not possible to separate the administered part of each component given the aggregation level. The measure excluding the top five is preferable to the one excluding the top ten. Because it removes only 20 percent of the CPI basket while reducing the volatility by 42 percent (compared to respectively 37 percent and 53 percent for the measures based on excluding the top ten). Furthermore, the five most volatile components have remained the same over time8 while some of the top ten components have changed over the two sub-periods.

D. Can Inflation be Forecasted in Tunisia?

26. Among the forecasting models of core and headline inflation, the preferred approach is a leading indicator model. The results show that, even with limited monthly observations, economically meaningful and statistically significant relations can be established between leading indicators and the future direction of inflation. Nevertheless, the results remain fragile due to (a) the relatively small sample period and the lack of relevant data; and (b) the administered prices, an important obstacle to forecast Tunisia’s inflation.

Background, data sample, and lead period

Background

27. There is a large and growing empirical literature on inflation forecasting. Studies of inflation forecasts are mostly on advanced economies. At least three factors may account for this. First, the predominance of agriculture in emerging markets makes inflation more dependent on weather than economic activity (e.g., the Phillips curve). Second, limitations of data quality and frequency are often constraining factors. Third, emerging markets are prone to sudden crises and market gyrations in macroeconomic variables, making it difficult to discern economic regularity. However, a rapidly growing literature has begun to analyze inflation forecasts in emerging markets and to construct leading indicators for inflation.9

28. Theoretical work shows that short- and long-term determinants of inflation are different. In the long term, inflation is a monetary phenomenon with flexible prices and wages and output and employment always at their natural rates. In the short run, inflation also results from real and nominal shocks that affect aggregate demand relative to aggregate supply. Monetary financing of public spending also contributes to inflation (see the literature on the inflation tax). In open economies, inflation can result from movements in the nominal exchange rates. Finally, inflation expectations and their formation affect inflation through price-wage spirals or inertia.

29. The empirical determinants of inflation are consistent with theory. Changes in money growth, nominal exchange rates, price of imports, and exogenous supply shocks, especially to oil and food prices, are the main determinants of inflation in emerging market economies.

Data sample

30. The choice of data sample for forecasting inflation in Tunisia is dictated by data availability. The data sample analyzed here comprises monthly observations of 20 variables from different sectors of the economy from December 1997 to December 2006 (109 observations).10 The sample is reduced because some monetary variables are only available from December 1996.11 We use 12-month growth rates of variables to take into account seasonality. All variables were tested in logarithmic form for nonstationarity using Phillips-Perron and Augmented Dickey-Fuller tests. The unit root hypothesis could not be rejected for some of the variables in levels. However, all variables in first differences (i.e., the growth rate) were found to be stationary. The regressions below are therefore expressed in first differences to avoid spurious correlations associated with nonstationary variables.12

Lead period

31. Forecasting inflation requires determining a “lead period”, defined as the number of months for which leading indicators would predict inflation. We use four lead periods: 3, 6, 9, and 12 months.13

The alternative models

32. We forecast headline inflation and two measures of core inflation—excluding top five and top ten most volatile CPI components.14 We start with two benchmark models, namely the naive model and ARMA model. Then we present other specifications and compare their results to those of the two-benchmark models.

Naïve model

33. The first step—as a benchmark—is to assume that inflation cannot be forecasted. Thus, no other model can beat a random walk, which implies that the best forecast for future inflation is current inflation.15

34. The naïve model performs poorly in the case of Tunisia. This is not surprising, because inflation has been relatively volatile ranging between 1 and 5 percent over 2000M1-06M12. The naïve model performs quite well when inflation fluctuates moderately.

Autoregressive Moving-Average (ARMA) models

35. ARMA models use only past inflation observations to forecast inflation. As a second benchmark, we use the forecast from ARMA models allowing the disturbances to follow an ARMA specification. We estimate the following ARMA(p, q) model that includes both autoregressive and moving average terms:16

πt=c+Σi=1pφiπti+Σj=0qθjɛtj(1)

36. The best specification is an ARMA(2,2) model, which predicts actual inflation reasonably well.17 We started with a large p and q, and then eliminated all lags that were not significant. The results for 12-months forecasting horizons for headline inflation are presented in Figure I.6. The forecast fails to predict the turning points. However, it predicts a small decline in inflation.

Figure I.6.
Figure I.6.

Performance of the ARMA Model

Citation: IMF Staff Country Reports 2007, 319; 10.5089/9781451837933.002.A001

Sources: Tunisian authorities, and IMF staff estimates.
Forecasting inflation by using a formal model: Phillips curve representation

37. The Phillips curve has been used extensively in inflation forecasting, but does not have a strong predictive power for Tunisia. Although the Phillips curve is typically specified in terms of the deviation of unemployment from its natural rate, more generally it is a relationship between inflation and aggregate real activity (deviation of output relative to its potential, i.e., output gap). Following Stock and Watson (1999), we assume that the potential economic activity is constant over the short-term period. Thus, the relationship becomes between inflation and economic activity:

πt+h=α+β(L)Xt+γ(L)πt+ɛt+hh=0,3,6,9,12(2)

Where, πt+h is the inflation h months ahead (henceforth, referred to as h-period inflation); Xt is a proxy of the economic activity; β(L) and γ(L) are polynomials in the lag operator L; and εt+h is the error term.

Following the literature, industrial production index (IP) is taken as a proxy of economic activity. We tried many specifications of the above equations (including h=0); IP was significant only in few cases. However, even in these cases, IP can only explain a small part of the variability of inflation. From these regressions, we conclude that even if the Phillips curve had some explanatory power under certain specifications, still most of the variability of inflation remains to be explained. And it cannot be used alone to produce accurate forecasts of inflation.

Leading indicators

38. Without having a formal model, we analyze the correlation between inflation and a set of variables—including the industrial production index—from different sectors. Then, we combine the best individual leading indicators into an index. In our preferred model, nominal effective exchange rate, a monetary aggregate—namely M4—and the producer price index are found to be good leading indicators. They perform relatively well for 3, 6, 9, and 12 months-lead periods and for the different measures of core and headline inflation.

Individual leading indicators

39. First, we select the best individual leading indicators. We run bivariate regressions of h-period inflation on a large number of potential leading indicators. We estimate equation (2) and fix the number of lags to 3, where Xt is a potential leading indicator for inflation. Only few variables are statistically significant with the expected sign. The main results for the bivariate regressions are presented in Appendix Table I.5. The following variables are significant in one of the models: nominal effective exchange rate (NEER), M2, M4, credit to the economy, industrial production index, producer price index, and Tunis stock market index.

Index of leading indicators

40. Second, we combine the individual leading indicators into an index of leading indicators, because a composite leading index is superior to a single leading indicator. Indeed, a composite index reflects a broader spectrum of the economy, comprising data from several sectors. Moreover, the performance of an individual series may vary over time, making it occasionally a poor leading indicator. In statistical terms, this implies that a composite index reduces the measurement error associated with any given indicator.

41. The selection of variables to be included in the regression is based on a principle of parsimony, also used in Stock and Watson (1989), and Mongardini and Saadi-Sedik (2003). From a generalized model using all potential indicators (Appendix Table I.5), variables were recursively eliminated based on the lowest t-statistic. Attention was also paid to avoid multicollinearity for variables that were close proxies.18

42. The selection procedure outlined above identified the following model that produced consistently superior results:

πt+h=α+β1πt+β2πt1+γ1NEERt+γ2M4t+γ3PPIt+ɛt+hh=3,6,9,12(4)

The three leading indicators are the Nominal Effective Exchange Rate (NEER), a monetary aggregate (M4), and Producer Price Index (PPI). In addition, the current level of inflation and a lag are significant. The results for the three measures of inflation and four lead periods are presented in Appendix Table I.6.

These results are also intuitive from an economic point of view:

  • NEER. This is consistent with the empirical determinants of inflation found in the literature. See also the Chapter II on exchange rate pass-through. Everything being equal, a 10 percent depreciation of NEER is forecast to increase headline inflation of about 1 percent twelve months ahead.

  • M4. The results show that M4 is the best monetary aggregate as a leading indicator for inflation. An increase of 10 percent of M4 is forecasted to increase headline inflation of about 0.4 percent twelve months ahead. Monetary aggregates have not been closely related to inflation because portfolio shifts between aggregates have contributed to their instability. High money growth reflected in part a decline in M3 velocity as the demand for savings and time deposits rose. This was due to investor portfolio shifts away from government securities, which enjoyed a liquidity guarantee from commercial banks (Bons du Trésor Cessibles or BTCs). The broader liquidity aggregate (M4), which includes BTCs, is more stable. However, for the recent period the correlation between M2, M3, and M4 becomes higher (see Figure I.7).

Figure I.7.
Figure I.7.

Inflation and Monetary Aggregates (year-on-year, Monthly Growth Rates)

Citation: IMF Staff Country Reports 2007, 319; 10.5089/9781451837933.002.A001

Source: Tunisian authorities.
  • PPI. This is also consistent with the literature (e.g., Stock and Watson, 1999). An increase of 10 percent of PPI is forecast to increase headline inflation of about 0.8 percent twelve months ahead. The PPI is a comprehensive index of wholesale price index, which is often viewed as an indicator of future retail price index (i.e., CPI index).

43. The explanatory power decreases as the forecast horizon increases. The R-square decreases from about two-thirds for the three months horizon to about one-third for the twelve months. However, for the headline inflation the R-square for twelve months is similar to that of three months.

Pseudo out-of-sample performance19

44. The main conclusion is that the two measures of core inflation produce better pseudo forecasts.

45. The twelve specifications presented in Appendix Table I.6 can be used to measure their pseudo out-of-sample performance based on the last h-months. For example, the pseudo-forecast for the last six months of 2006 (h=6) requires estimating the model using data available through 2006M06, then using this estimated model to produce the 2006M07-2006M12 forecast.

46. Forecasts’ accuracy is measured by the forecasting error. The measure used is Root Mean Square Error (RMSE) computed as the square root of the arithmetic average of the squared differences between actual inflation and predicted inflation over the period for which simulated forecast are constructed (see bottom of Appendix Table I.6). The RMSE indicates a superior forecast results for the two measures of core inflation when compared to the headline inflation.

47. The confidence interval of the forecast is relatively wide. This implies that the mid-point inflation target band should be relatively wide to take into account the imprecision of the inflation forecasts.20 More importantly, however, the statistical tools (including databases) should be developed to improve the accuracy of these forecasts.

E. Conclusions and Policy Recommendations

Conclusions

48. To investigate the nature of the Tunisian inflation process, we analyze the behavior of 43 CPI components and six groups (food, housing, clothing, health, transport, services). We find that some components of inflation have been much more volatile than others. However, the most volatile components remain broadly the same over time. We also find that, despite price liberalization, administered prices remain an important component of the CPI total index (about 32 percent). The prices of some sectors are mainly administered (transport and health); others are partially liberalized (food, housing, and service); and clothing is fully liberalized.

49. Based on the behavior of the 43 CPI components, three measures of core inflation are computed. They are based on the exclusion method. Two measures exclude the most volatile CPI components—namely the top five and top ten most volatile components. The third measure leaves out the components with predominantly administered prices. Our preferred measure is based on excluding the five most volatile CPI components because it removes fewer components, it is transparent, easy to understand, replicable, credible, and available on a timely basis.

50. Assessing the direction of inflation is essential for the formulation of appropriate monetary policies in the context of inflation targeting framework. In this regard, composite indexes of leading indicators for inflation provide useful summary statistics.

51. Although we present several empirical models to forecast Tunisia’s core and headline inflation, our preferred approach is a leading indicator model. In this model, nominal effective exchange rate, a monetary aggregate—namely M4—and the producer price index are found to be good leading indicators. They perform relatively well for 3, 6, 9, and 12 months-lead periods and for the different measures of inflation. These results show that, even with limited monthly observations, meaningful economic and statistically significant relations can be established between leading indicators and the future direction of inflation.

52. Notwithstanding the relatively small sample period, the results seem to be both statistically significant and economically intuitive. However, a larger sample will be needed before the composite indexes for the Tunisian inflation can be relied upon for more than a qualitative assessment. Also, important data for inflation forecasting are not available on a monthly basis (for example, prices for imports). Their availability will increase the precision of inflation forecast.

53. In addition, assessing future inflation will always need to take account of other information that cannot easily be quantified, including geopolitical uncertainties and macroeconomic policy changes. Despite these caveats, a regular updating of these indexes could provide useful information and help monetary policy formulation.

Policy recommendations

  • Strengthen analytical research at the BCT. The BCT should strengthen and broaden its research capacity and develop forecasting skills with respect to inflation and other relevant macroeconomic variables.

  • Administered prices should be reduced. Administered prices are a major impediment to implementing efficiently inflation targeting. The BCT does not have any control on one third of CPI basket. This is an important obstacle to forecasting Tunisia’s inflation. In addition, the mechanism for adjusting administered prices introduces a fiscal bias into the control of inflation in the sense that the inflation target may lead to an implicit accommodation of budgetary slippages.

  • Need of a multivariate model. While the focus of this paper is on inflation, monetary policy makers need to be informed about the future evolution of other variables, such as real GDP growth. Therefore, the forecasting model is typically multivariate, including at a minimum inflation and real GDP forecasts.

  • More data are required. As mentioned above, statistical data are needed to increase forecast accuracy, in particular, monthly import prices, other indicators of economic activity (in addition to the industrial production index) such as retails sales, and forward looking business and consumer confidence indicators.

References

  • Atkeson, Andrew, and Lee Ohanian, 2001, “Are Phillips Curves Useful for Forecasting Inflation?FRB Minneapolis Quarterly Review (Winter), pp. 2 11.

    • Search Google Scholar
    • Export Citation
  • Canales–Kriljenko, Jorge Ivan, Turgut Ki§inbay, Rodolfo Maino, and Eric Parrado, 2006, “Setting the Operational Framework for Producing Inflation Forecasts”, IMF Working Paper No. 06/122 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Chauvet, Marcelle, 2000, “Real Time Leading Indicators of the Brazilian Inflation.Available via the Internet at http://faculty.ucr. edu/~chauvet/mc.htm.

    • Search Google Scholar
    • Export Citation
  • Fisher Jonas D. M., Chin Te Liu, and Ruilin Zhou, 2002, “When Can We Forecast Inflation?FRB Chicago Economic Perspectives (1Q), pp. 3042.

    • Search Google Scholar
    • Export Citation
  • Hamilton, James D., 1994, “Time Series Analysis,” Princeton University Press, Princeton, New Jersey.

  • Heenan, Geoffrey, Marcel Peter, and Scott Roger, 2006, “Implementing Inflation Targeting: Institutional Arrangements, Target Design, and Communications”, IMF Working Paper No. 6/278 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2006, “Inflation Targeting and the IMFIMF Board paper, available via the Internet at: www.imf.org/external/np/pp/eng/2006/031606.pdf.

    • Search Google Scholar
    • Export Citation
  • Leigh, Daniel, and Marco Rossi, 2002, “Leading Indicators of Growth and Inflation in Turkey,” IMF Working Paper No. 02/231 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Marcellino, Massimiliano, 2005, “Leading Indicators: What Have We Learned?IGIER Working Papers No. 286 (Bocconi University).

  • Madhavi, Bokil, and Axel Schimmelpfennig, 2004, “Three Attempts at Inflation Forecasting in Pakistan”, IMF Working Paper No. 05/105 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Mongardini, Joannes, and Tahsin Saadi-Sedik, 2003, “Estimating Indexes of Coincident and Leading Indicators: An Application to Jordan,” IMF Working Paper No. 03/170 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Roger, Scott, and Mark Stone, 2005, “On Target? The International Experience with Achieving Inflation Targets”, IMF Working Paper /05/163 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Silver, Mick, 2006, “Core Inflation Measures and Statistical Issues in Choosing Among Them”, IMF Working Paper /06/97 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Sun, Tao, 2004, “Forecasting Thailand’s Core Inflation”, IMF Working Paper/04/90 (Washington: International Monetary Fund).

  • Stock, James H., and Mark W. Watson, 1989, “New Indexes of Coincident and Leading Economic Indicators,” NBER Macroeconomic Annual Report (Cambridge, Massachusetts: National Bureau of Economic Research), pp. 35194.

    • Search Google Scholar
    • Export Citation
  • Stock, James H., and Mark W. Watson, 1999, “Forecasting Inflation,” Journal of Monetary Economics, Vol. 44, pp. 293335.

  • Stock, James H., and Mark W. Watson, 2003, “Forecasting Output and Inflation: the Role of Asset Prices,” Journal of Economic Literature, Vol. XLI, pp. 788829.

    • Search Google Scholar
    • Export Citation

Appendix Table I.1.

CPI Components, Weights, and Main Descriptive Statistics

(1991M1–2006M12)

article image
Sources: Tunisian authorities, and IMF staff calculations.
Appendix Table I.2.

CPI Components, Weights, and Main Descriptive Statistics (1991M1–99M12)

article image
Sources: Tunisian authorities and IMF staff calculations.
Appendix Table I.3.

CPI Components, Weights, and Main Descriptive Statistics

(2000M1–06M12)

article image
Sources: Tunisian authorities; and IMF staff calculations.
Appendix Table I.4.

Average Contribution to Overall Inflation by Different Components

(In percent)

article image
Sources: Tunisian authorities, and IMF staff calculations.
Appendix Table I.5.

Individual Leading Indicators for Inflation (Bivariate Regressions)

article image
Source: IMF staff estimates.Note. We run bivariate regressions of h-period inflation on a large number of potential leading indicators. We estimate un equation with three lags for the dependend variable (inflation) and a potential leading indicator for inflation. Only significant variables in one of the models are reported here. P-values are in parentheses.
Appendix Table I.6.

Composite Leading Indicators for Inflation (Multivariate Regressions) 1/

article image
Source: IMF staff estimates.

White Heteroskedasticity-Consistent Standard Errors & Covariance. P-values are in parentheses.