Explaining and Forecasting Inflation in Turkey

The growing adoption of an inflation targeting framework in emerging market economies has increased the importance of understanding inflation dynamics and forecasting its future path in these countries. Domac considers the case of Turkey and investigates the performance of models that have some theoretical foundations. To this end, his study focuses on mark-up models, monetary models, and the Phillips curve. The findings suggest that the mark-up models have the best in-sample performance followed by money gap models and the Phillips curve. The empirical results from out-of-sample forecasting performance for the period covering the new economic program (May 2001-December 2002), however, show that the Phillips curve and the money gap models perform better than mark-up models. These findings, in turn, imply that (1) Phillips curves augmented with the exchange rate and money models might provide complementary views in the Turkish context; and (2) the relative importance of output gap and monetary disequilibrium in the inflation process has increased under the floating exchange rate regime. The results underscore the importance of relying on multiple models of inflation in the conduct of Turkish monetary policy. This paper - a product of the Poverty Management and Economic Management Sector Unit, Europe and Central Asia Region - is part of a larger effort in the region to enhance the knowledge on the inflationary process and its policy implications.


Introduction
Why is our money ever less valuable? Perhaps it is simply that we have inflation because we expect inflation, and we expect inflation because we've had it.
-Robert M. Solow 1 Contrary to many emerging market economies, which have experienced noticeable declines in inflation owing to a combination of relatively favorable external factors and implementation of sound domestic policies, Turkey is still attempting to curb chronic high inflation and attain price stability ( Table 1). The absence of a permanent fiscal adjustment coupled with unsatisfactory progress with the implementation of key structural reforms lie at the heart of the chronic high inflation in Turkey. The experiences from a variety of approaches to stabilization adopted by emerging market economies (exchange rate-based orthodox programs and heterodox programs) also confirm that inflation does not stay permanently low in the absence of a permanent fiscal adjustment. 2 Evidence suggests that countries that adopted exchange rate based stabilization arrangements appear to have been more successful in bringing inflation down. This monetary regime, however, is prone to balance of payment crises as countries often fail to implement macroeconomic policies that are consistent with the fixed exchange rate regime. Moreover, the increase in the degree of capital mobility coupled with greater financial openness and liberalization in emerging market economies has made it more difficult for central banks with finite reserves to defend a tight nominal exchange rate commitment. As a consequence, emerging market economies have been moving away from employing the exchange rate as a nominal anchor. In fact, it seems that many emerging market economies, including Turkey, have opted for adopting inflation targeting (IT) or some form of this monetary policy framework, in view of the success of this regime in a number of industrial countries. The adoption of IT, however, entails that the central bank has a good understanding of inflation dynamics and is relatively successful in predicting the future path of inflation.
The vast literature on the causes of inflation includes competing models of the inflation process. To the best of my knowledge, however, there has not been any study investigating the relative performance of the main existing inflation models in Turkey. In light of the envisaged adoption of IT in Turkey, this paper attempts to fill this void by comparing the in-sample and out-of-sample performance of the models in order to shed more light on the inflation dynamics and its forecasting in Turkey. The approach of this study is to consider models that have some theoretical foundations so that models can be useful both for understanding the causes of inflation and for forecasting purposes.
Indeed, comparing the performance of several models is not only justified by the uncertainties inherent in the estimation of any particular model, but also by the approach followed by many central banks in practice. Blinder (1998, page 12) elaborates on the latter point by stating that his approach at the Federal Reserve Board when faced with model uncertainty was to "Use a wide variety of models…My usual procedure was to stimulate a policy on as many of these models as possible…." Judgment is then exercised when evaluating the results from the different models. By the same token, Longworth and Freedman (2000) state that given model uncertainty and a changing environment, it is essential for central banks to rely on a variety of models in conducting policy.
The remainder of the paper is structured as follows. The next section, which focuses mainly on the empirical literature, provides a cursory discussion of the determinants of inflation in emerging market economies including Turkey. Section 3 discuses the main models of inflation. Section 4 presents in-sample estimation results of the models of interest. Section 5 summarizes the out-of-sample performance of these models. Finally, Section 6 concludes the paper.

A Cursory Look at the Determinants of Inflation
There is a consensus that, in the long run, inflation is a monetary phenomenon.
Within the framework of developing countries, however, the nature of the mechanisms underlying the dynamics of inflation has stimulated much discussion between the monetarists and structuralists since the early 1960s. Key aspects of the debate in recent years have been the interactions-and inconsistency-between fiscal, monetary, and exchange rate policies; structural factors (such as the existence of wage and price inertia); credibility problems; and the stance of expectations regarding future policies. The following sub-sections provide a brief overview of the empirical studies focusing on emerging market economies and Turkey.

The Case of Emerging Market Economies
Studies in this strand of the literature suggest that contrary to industrial countries, where real factors emerge as the main determinants of inflation, nominal factors play a more important role in affecting the evolution of inflation in emerging market economies.
For instance, the IMF (1996) shows that the output gap does not play an important role in explaining inflation in developing countries. Instead, changes in money growth and nominal exchange rates have higher explanatory power in explaining inflation. This finding does not suggest that inflation is not a function of excess demand in these countries; it simply implies that the contribution of excess demand is dominated by those of nominal shocks. More precisely, inflation in the medium term is viewed as the result of the government financing its deficit through the creation of money or through time inconsistent monetary policy.
The empirical evidence concerning the link between fiscal deficits and inflation has been rather elusive in spite of the theoretical links. At the level of any particular country, it may be difficult to establish a clear short-term link between fiscal deficits and inflation. In fact, the correlation may be even negative during extended periods of time.
Evidence suggests that the existence of a positive correlation in the long-run is also not a clear-cut phenomenon (Agenor and Montiel, 1999). For instance, Fisher, Sahay, and Vegh (2001) find that the relationship between the fiscal deficit and inflation is only strong in high inflation countries-or during high inflation episodes-but they find no obvious relationship between fiscal deficits and inflation during low inflation episodes or for low-inflation countries.
A recent study by Catao and Terrones (2001), however, was successful in relating long-run inflation to the permanent component of the fiscal deficit scaled by the inflation tax base, measured as the ratio of narrow money to GDP. Their finding suggest that a 1 percent reduction in the ratio of the fiscal deficit to GDP typically lowers inflation by 1.5 to 6 percentage points depending on the size of the money supply.
One of the main focuses of the literature has been central bank independence. It is argued that a lack of central bank independence can lead to succumbing to political considerations that may lead to a monetary policy looser than optimal. For instance, if there is a perception that pursuing expansionary monetary policy can increase output, politicians could put pressure on the central bank-say during the election period-to trade off a boost to growth against higher inflation. In fact, the IMF (1996) finds that inflation performance between 1975 and 1995 in industrial countries is negatively correlated with an index of central bank independence. However, the findings of this study also suggest that the same relationship did not hold over the same period for developing countries. This conclusion may be attributed to imprecision in the measurement of central bank independence arising from a divergence between de jure and de facto central bank independence in these countries. 3 In fact, a recent study by Gutierrez (2003)  Overall, their findings suggest that cost shocks or the output gap are not significant factors affecting the evolution of inflation in these countries.
By contrast, Mohanty and Klau (2001), who study the experience of 14 emerging market economies in the 1980s and 1990s, find that exogenous supply shocks-in particular those to food prices-play an important role in the inflation process. Food prices typically account for a larger percentage of the CPI in emerging market economies than in industrial countries. Furthermore, food prices tend to be very volatile owing to the influence of weather and the presence of trade restrictions. Their results suggest that demand factors, captured by the output gap and excess money, do not play a significant role in the inflation process. Wage growth and exchange rate changes, on the other hand, appear to make important contributions to inflation volatility in many countries. Their findings also indicate that inflation persistence plays an important role in explaining both the average level of inflation and its variation.

The Case of Turkey 5
The Turkish economy has been undergoing chronic high inflation since the 1970s.
Indeed, starting from the 1970s, the inflation rate displayed an upward trend and reached In an attempt to stabilize the Turkish economy, which was plagued by chronic high inflation and real interest rates as well as deteriorating debt dynamics, the authorities In high-inflation countries like Turkey, the question of why prices increase lies at the heart of the debate over which policies should be adopted to stabilize the price level.
As was pointed out earlier, the notion that inflation is ultimately a monetary phenomenon is widely accepted. There is strong evidence that in the medium and long term, there exists a very close correlation between the rate of growth of monetary aggregates and inflation, after changes in output and velocity are taken into consideration. This correlation has been corroborated both in the international 8 and the Turkish experiences Existing studies, see for instance Lim and Papi (1997), Sakallıoğlu and Yeldan (1999), Özcan et. al. (2001), tend to agree upon the importance of the inertia and the exchange rate. However, there seems to be a disagreement over the importance of monetary variables. 11 The above-mentioned indirect transmission channel through which shocks to exchange rates and public prices can influence inflation expectations in Turkey can be illustrated by the following example. Let us assume that the central bank faces an exogenous shock in the form of unanticipated adjustments to administered prices that are not compatible with the central bank's inflation target. In turn, this will generate a higher CPI, thereby raising nominal demand for money. If the central bank increases the supply of base money to match the increase in demand, the central bank would have accommodated the rise in money demand engendered by an exogenous shock to prices caused by the increase in public prices. Under normal circumstances, this would be described as once for all adjustment in the price level, which should not create further problems for the central bank. However, given Turkey's history of high inflation, the dynamics triggered by the increase in public prices are complicated as the public might revise their inflationary

Models of Inflation
A quick glance at the literature points to three classes of models for inflation determination. 12 The first view envisions inflation as a cost-push phenomenon in the context of a long-term constant mark-up over costs. The second view considers inflation mainly as a monetary phenomenon and links changes in monetary variables to those in prices. Finally, the third view envisages inflation as stemming from real factors, imbalances between aggregate demand and supply in particular. A review of the empirical studies suggests that the first type of model has been frequently employed in emerging market economies, while the other two have been typically-but not exclusively-used in industrial countries. The next section will provide a brief discussion of these three classes of models.

Mark-up Models
Goodfriend (1997) provides a comprehensive discussion of the theoretical underpinnings of the mark-up model. These types of models, in which the price level is determined by costs and a given mark-up, take the following form: where P, µ, W, E, P * stand for the domestic level of prices, the mark-up over costs, wages, the nominal exchange rate, and the level of foreign prices, respectively. 13 The above equation arises from the maximization problem of a firm that faces a demand curve expectations upwards, which would lead to increases in wages and non-tradeable goods prices, and thus bringing about inflation and monetary base growth. 12 This section draws largely on Bailliu et al. (2002).
with a negative slope. Under this framework, the firm sells at a price equal to a given mark-up above marginal cost which is determined by the price of domestic inputs, captured by domestic wages, and the price of foreign inputs, reflected in the level of foreign prices.
When expressed in logarithms, the domestic price amounts to a weighted average of nominal wages and foreign prices (expressed in local currency): where lower case letters represent the variables in logarithmic form. The above estimation provides a basis for estimating a long-run relationship between prices, wages, and foreign prices under the assumption that the mark-up is constant or fluctuating randomly around a given long-run value.
In the short-run and medium-run, however, there could be significant and persistent fluctuations in the mark-up depending on the speed of adjustment of price setters to changes in wages or in foreign prices. As a result, it is important to consider more complex dynamics when working with monthly or quarterly inflation rates, which would amount to the following form: The mark-up model has been employed widely to study the inflation process in a number of emerging market economies. For instance, Perez-Lopez (1996) and Garces (1999)

Monetary Models
Monetary models are based on the view that inflation is in essence a monetary phenomenon. Monetarist analysis indicates that a monetary disequilibrium exists if the quantity of money in the economy is greater than what the public desires to hold. Under such a situation, monetary models assert that the price level will increase to re-establish the equilibrium between supply and demand for money. As a consequence, an excess supply of money can lead to inflationary pressure in much in the same way that an excess demand for goods does. Monetary disequlibria is typically captured by using the money gap, which is the difference between the actual money supply and the estimated long-run money demand.
Money gap is usually specified in two different ways in empirical studies. In the case of countries for which the money demand is stable, the following partial adjustment framework can be considered: In the above equations, m s t , m d t , are the natural logarithm of demand for and supply of money, respectively, y is the logarithm of real GDP and i are the nominal short-term interest rates. In the case of countries where a stable demand for money function does not exist, a real money gap variable is constructed by the following expression: where the money gap is defined as the deviation of the actual real money supply from its trend value. In view of the absence of a stable money demand function for Turkey, I rely on the second approach employed by Mohanty and Klau (2001) and consider the following model: where the key variable is the money gap, which aims to capture the impact of the channel for the monetary disequilibrium on the dynamics of price changes. It should be noted that it would not be appropriate to include the exchange rate depreciation as an additional term in the money-gap model. This is because an excess demand for money should translate into higher demand for goods, generating domestic price pressures, and also higher demand for other assets such as foreign currency, leading to a depreciation of the domestic currency. As a result, a theoretically consistent money-gap model should be able to explain change in both tradable and non-tradable goods prices. 15 A number of studies have applied monetary models to industrial countries.
Altmari (2001) is a recent example of such an application. He investigates the performance of a number of monetary models of inflation for the euro area over the period 1998 to 2000. His findings suggest that monetary and credit aggregates contain significant information to forecast inflation in the euro area, particularly at medium-term horizons. Using structural VAR analysis studies by Kasumovich (1996) as well as Fung and Kasumovich (1998) show that a monetary policy shock leads to a persistent money disequlibrium, which is closed as prices adjust over a number of years. 16 Jonsson (1999) and Callen and Chang (1999) are examples of this strand of the empirical literature focusing on emerging market economies. The former study concludes that an increase in the money supply raises domestic prices in South Africa although this effect is somewhat compensated by an increase in domestic interest rates.
The latter study considers two models of inflation in India-one based on the monetary approach and the other using the output gap. Their results suggest that monetary aggregates contain the best information about future inflation and that output gap is not a significant explanatory power.
Nevertheless, monetary models should be used with caution since financial innovations can lead to instability in the money-inflation relationship. Furthermore, in the presence of a high short-run interest elasticity of money with respect to interest rates, it may well be the case that the required interest rate increase to bring money back to its target is not sufficient to reduce spending and inflation. As pointed out by King (2002), however, there are good reasons for considering models based on money growth. First, although models based on monetary aggregates may not perform well in forecasting inflation, they tend to perform better as indicators of long-run inflationary pressures.
Second, expansion of the monetary base may be the only way to relax monetary conditions at low rates of inflation coupled with the possibility of a liquidity trap and interest rates close to zero.

Phillips Curve
The development of contemporary inflation theory was greatly influenced by the development of the Phillips curve model. The original contribution by Phillips (1958) concluded that, on the basis of empirical observations for Great Britain, there is a negative correlation between the rate of change in money wages and the rate of unemployment and that this relationship was stable. The theoretical foundation of the Phillips curve was developed by Lipsey (1960), who derived the Phillips curve from a supply and demand system of a single labor market. In the decade or so following its inception, the Phillips curve was subject to many modifications. For example, the inverse of unemployment rate was substituted by the unemployment/output gap as the proxy for excess demand. Moreover, with Friedman's contribution in 1968, the role of expectations in affecting wage changes was acknowledged and as a result inflation expectations were incorporated into the Phillips curve. Finally, it was converted from a wage equation to a price inflation equation. 16 See also Porter and Small (1991), Hendry (1995), Armour et al. (1996), and Engert and Hendry (1998). As was pointed out by Gordon (1997), the Phillips curve summarizes the dependence of inflation on three basic factors: inertia, demand, and supply. This Phillips curve framework is often referred to as the traditional Phillips curve to separate it from the new Phillips curve. 17 The new Phillips curve in spirit is similar to the traditional Phillips curve as it also links inflation positively with economic activity. However, the new Phillips curve differs from the traditional Phillips curve as it relates inflation to movements in real marginal cost in lieu of the output gap. Proponents of the new Phillips curve argue that this framework is more adequate since it is obtained from a model of staggered nominal price setting by monopolistically competitive firms, and thus has more solid theoretical foundations. Gali et al. (2001), however, show that by imposing certain restrictions on technology and labor market structure, and within a local neighborhood of the steady state, real marginal cost is proportionally related to the output gap.
The traditional Phillips curve's popularity arises from, in part, its relative success as a forecasting tool. Stock and Watson (1999) argue that "As a tool for forecasting inflation, it is widely regarded as stable, reliable and accurate, at least compared to the alternative." In parallel to the widespread use of the traditional Phillips curve for industrial countries, recent studies by Coe and McDermott (1999) and Simone (2000) have considered emerging market economies. The former study estimates Phillips curve based on output gaps for 13 Asian countries. Their findings indicate that the output gap is a significant determinant of inflation in 11 out of 13 countries even after controlling for other variables such as monetary disequlibria. The latter investigation estimates time varying Phillips curves for Chile. His conclusions suggest that although the model, which includes the pre-announced inflation target, displays some autocorrelation, it outperforms the model that excludes this variable in forecasting exercises.
In this paper, I employ a specification for the traditional Phillips curve that links inflation to inflation expectations, some measure of disequilibria and a variable reflecting changes in imported prices since Turkey is a relatively open economy. Under the assumption that expectations are formed adaptively and that the relationship is linear, one can employ lagged inflation as a proxy for inflation expectations and obtain the following specification: where π t is the inflation rate at time t, L is the lag operator, gap t is the output gap at time t, and s t is the nominal exchange rate.

In-Sample Estimation Results of the Three Models for Turkey
This section summarizes the estimation results for the three models of inflation and compares their in-sample performance. Each model is estimated using monthly data covering the period 1990. 01-2002.12. 18 Throughout the investigation, I include two dummy variables, one for 1994.4 and the other for 2001.4, to account for the sharp increases in inflation. As Perron (1989) indicates, it is important to include dummy variables, which allow the coefficient of intercept or trends to shift in response to large shocks. Appendix 1 presents the variables involved in the investigation and Appendix 2 depicts their evolution over the sample period.

Mark-up Model
In an attempt to investigate the short-run and the long-run properties of the markup model, I employ the approach known as the Autoregressive Distributed Lag (ARDL) put forth by Peseran and Shin (1997). The main advantage of this strategy is that it can be applied irrespective of whether the series are I(0) or I(1), and this avoids the pretesting problems related with the unit roots and standard cointegration analysis. The ARDL procedure appears to perform better than the fully modified OLS approach of Phillips and Hansen (1990) in small samples. Moreover, another advantage of the ARDL model is that, in the case of first-difference stationary variables, appropriate augmentation of the order of the regressors is sufficient to simultaneously correct for residual serial correction and the endogeneity problem in the estimation of the long-run parameters.
In practice, the estimation involves two steps. The first stage of the process involves establishing the existence of a long-run relationship between the variables and is tested by considering the joint significance of the coefficients of the lagged level variables in the following equation: The null hypothesis of "non-existence of the long-run relationship" defined by H 0 : δ 1 =δ 2 =δ 3 =0 is tested against the alternative of H 1 : δ 1 ≠δ 2 ≠δ 3 ≠0. The relevant statistic to test the null is the familiar F-statistic. The asymptotic distribution of this F-statistic, however, is non-standard irrespective of whether the variables are I(1) or I(0). Peseran et al. (1986) have tabulated two sets of appropriate critical values. One set assumes that they are all I(0). This provides a band covering all possible classifications of the variables into I(1) and I(0) or even fractionally integrated. If the calculated F-statistic lies above the upper level of the band, the null is rejected, indicating co-integration. If the calculated F-statistic falls below the lower level of the band, the null cannot be rejected, supporting lack of cointegration. If the calculated F-statistic falls within the band, the result is inconclusive. Under this case, following Kremers et al. (1992), the error correction term will be a useful way of establishing cointegration.
Once we have established the existence of cointegration, we move to the second stage of the procedure, which involves estimating the error-correction model. The main objective here is to investigate the short-run dynamics.
In this paper, following Garces (1999), I also include the administered prices (ap) in the mark-up model (hereafter mark-up model II) in view of its importance in the inflation process in Turkey. In the estimations, I consider 8 lags for both versions of the mark-up model. 19 In the case of the mark-up I model, the calculated F-test is 5.16, suggesting that variables of interest are cointegrated. The calculated value of the F-test in the mark-up model II (6.03) also exceeds the upper bound of the critical value band, thus supporting the existence of a long-run relationship among the variables involved. Table 2 reports the estimated long-run models of inflation using the ARDL procedure, in which Akaike's Information Criterion (AIC) is employed to select the lag length. In the case of mark-up model I, empirical results suggest that a 1 percent rise in wages will increase the price level by 0.3 percent in the long-run. The findings indicate that the long-run impact of a rise in foreign price on the domestic price level is considerably higher: a 1 percent increase in (e t +p * t ) will raise the price level by 1.3 percent. These findings document the importance of the exchange rate on the evolution of the price level in Turkey. The empirical findings for the mark-up II are fairly similar to that of mark-up model I, confirming the significance of the exchange rate on the evolution of the price level in Turkey. Table 3 presents the results of the corresponding error correction models to investigate the short-run dynamics. In this stage, I also rely on the AIC to select the lag length of each variable. In the case of mark-up model I, the lagged error-correction term (EC t-1 ) has a size of 0.09, which is negative and statistically significant, suggesting a relatively fast convergence to the long-run equilibrium in the face of shocks. The error-correction term in the mark-up model II also carries its correct negative sign and is statistically significant, supporting the existence of a long-run relationship among the variables involved. The ARDL error-correction term in the mark-up model II, however, has a slightly lower coefficient (-0.05) than the mark-up model I (-0.09), suggesting a somewhat slower return to the long-run equilibrium compared with the mark-up model II.

Money-gap Model
In the case of Turkey, the money gap measure can be constructed for almost any definition of money. As a result, I consider three definitions of monetary aggregates, namely the monetary base (mb), M1, and M2. 20 I construct a real money gap variable, measured as the proportionate deviation of the actual real money supply from its trend value, obtained using the Hodrick-Prescott (HP) filter. The corresponding money gap measures for the three monetary aggregates will be referred to as mgapmb t , mgapm1 t , and mgapm2 t .
I rely on the AIC to select the appropriate lag length for each variable in estimating equation (7). In this respect, the investigation considers all possible combinations up to 8 lags and then employs the combination that minimizes the AIC. 21 Given that the objective of the paper is to examine models that can be employed both to explain and forecast inflation, I decided to use the parsimonious specifications chosen by the AIC. Table 4 reports the results of this analysis. As expected, the impact of an increase in the money gap on inflation is positive and is estimated at around 0.4 percent, 0.081 percent, and 0.11 percent (including both the lagged and contemporaneous coefficients), respectively for the monetary base, M1, and M2 based measures of money gap. 20 M2Y definition of money supply, which includes foreign exchange deposits, is not considered since this aggregate is affected by the evolution of the exchange rate. 21 It should be noted that this exercise involves running 130,305 regressions. The general formula for the number of regressions involved can be written as (2 n -1)(2 n+1 -1) m where n and m are the number of lags considered and number of independent variables involved, respectively.

Phillips Curve
In estimating equation (8), I use the nominal USD exchange rate and a measure of the output gap estimated by Paşaoğulları and Yurttutan (2002). Their study employs Kalman Filter methodology by following the same approach taken by Boone et al. (2002). 22 In the estimation, I rely on the AIC to select the lag length of each variable. In this respect, the investigation considers all possible combinations up to 8 lags and then 22 Output gap measure in original series is quarterly, which are converted into monthly frequency using RATS 5.0 distrib procedure. This procedure maintains the sum of three months equal to the value at the quarter. We have adjusted so that the average of three months, rather than the sum, is equal to the quarter value. For example, the average of output gap measure for the first three months of 2000 will be equal to the output gap at the first quarter of 2000.
employs the combination that minimizes the AIC. 23 Table 5 reports the parsimonious specification chosen by the AIC.  For comparison purposes, it is useful to ferret out how the above-presented insample results of the three models fare vis á vis an AR1. A priori, these models, to be useful in forecasting, should at minimum outperform this simple univariate specification. Table 6 reports the estimation results. A quick glance at the adjusted R 2 and the standard error of the regression (SEE) suggests that all three models, in terms of in-sample-fit, outperform the AR1 model.
Among the three models, the findings indicate that the mark up models have the best insample performance followed by money gap models and the Phillips curve.
The results of the diagnostic tests suggest that in most cases the errors fulfill the classical assumptions. More specifically, the findings suggest that (i) the null hypothesis of normally distributed errors cannot be rejected at 1 percent significance level; (ii) the null hypothesis of no ARCH up to order 12 in the residuals cannot be rejected at 1 percent significance level; (iii) the null hypothesis of no heteroskedasticity cannot be rejected at 1 percent significance level, except for the mark-up models; and (iv) the null hypothesis of no serial correlation up to order twelve cannot be rejected at 1 percent significance level, except for the mark-up II, the Phillips curve, and mgapm2 models. 24,25 The above-presented models are estimated using Newey and West (1987) adjustment methodology to obtain corrected standard errors, which should in general lead to correct inference asymptotically. The inferences are the same when estimated variances and covariances are corrected by using the Newey and West method (results are available upon request). 26 Moreover, CUSUM stability tests, presented in Appendix 3, also indicate that the models are, by and large, stable as evidenced by the graph of the cumulative sum of squares of recursive residuals. Next, I explore the forecasting performance of the models.

Comparison of the Forecasts
In an attempt to examine the out-of-sample forecasting performance of the models considered in this paper, I estimate all the models using dynamic rolling regression

Policy Implications and Conclusion
The successful performance of a number of industrial countries that have adopted inflation targeting (IT) has rendered this monetary policy framework an attractive alternative for emerging market economies (EMs). Indeed, a number of EMs have already instituted IT or some form of this monetary policy framework. The growing attraction of inflation targeting among EMs as a monetary policy framework has, in turn, increased the importance of understanding inflation dynamics and forecasting the future path of inflation in these countries.
This study considers the case of Turkey and examines the in-sample and out-ofsample performance of models for the period January 1990-December 2002. This study focuses on mark-up models, monetary models, the Phillips curve, and the simple AR1 specification. The results from the in-sample estimations suggest that mark-up models perform better than money-gap models and the Phillips curve in explaining movements in Turkish inflation during the period under consideration. The findings also indicate that all three models outperform the AR1 specification, which was selected as a benchmark. 27 The empirical results from the out-of-sample forecasting performance for the period May 2001-December 2002, however, turned out to be quite different. More specifically, the findings show that the Phillips curve and money-gap models have better out-of-sample forecasting performance compared to mark-up models since the inception of the new economic program.
There are two main policy implications emerging from the thrust of the overall findings. First, the findings suggest that the Turkish economy has experienced a change in the dynamics of inflation under the new policy regime, which included, inter alia, the adoption of floating exchange rates and the new Central Bank Law. This, in turn, implies that there has been a change in the relative importance of different determinants of inflation since the introduction of the floating regime. Although the best performing model, based on in-sample results, turned out to be the mark-up models incorporating the exchange rate, wages, and the administered prices, the Philips curve and money-gap models-particularly the one based on M1 measure of the money gap-outperformed the mark-up models when out-of-sample performance was considered. As a result, it can be argued that the relative importance of output gap and monetary disequilibrium in the inflation process in Turkey has increased under the new regime.
Second, monetary policy, by its very nature, is conducted in an environment characterized by uncertainty and change. The uncertainty is likely to be higher in a country like Turkey where the economy has been undergoing significant changes. Based on estimation and forecasting results presented in this study, it can be argued that Phillips curves augmented with the exchange rate as well as money models might provide complementary views in the Turkish context. In light of the experiences of other emerging market economies that adopted IT or some form of this monetary policy framework, it is quite possible that the relative importance of other components in the models of the inflation process is likely to increase as the volatility of the exchange rate and the pass-through decline over time. In summary, the findings highlight the importance of relying on multiple models of inflation in the conduct of Turkish monetary policy.