Turkey has had a long history of persistently high inflation. Inflation started to take off during the 1970s and peaked at more than 100 percent in the mid-1990s. Following an unsuccessful exchange rate-based disinflation attempt in 2000–01, inflation did not decline until only recently, under a money-based stabilization program (Figure 3.1). Against this backdrop, this chapter analyzes the reasons behind Turkey’s successful disinflation of recent years. Building on earlier work, the chapter finds that inflation expectations—rather than backward-looking indexation mechanisms—dominate price-setting behavior, and have increased in importance in recent years.1 It also finds that inflation expectations are in turn heavily influenced by fiscal variables.

Inflation: The Long View
(In percent)
Source: IMF, World Economic Outlook database.
Inflation: The Long View
(In percent)
Source: IMF, World Economic Outlook database.Inflation: The Long View
(In percent)
Source: IMF, World Economic Outlook database.The Importance of Inflation Expectations
A structural price-setting model is used to test the importance of inflation expectations. Following Gali and Gertler (1999) and Celasun, Gelos, and Prati (2003), a model that nests two types of price-setting behavior is used. Backward-looking agents are assumed to update prices by the most recently observed inflation rate, while forward-looking price setters use current and expected future pricing conditions, particularly real marginal costs. This results in the following equation for inflation dynamics for the inflation rate in period t, πt:
where δf + δb = 1, Etπt+1 denotes expected inflation in period t + 1 in period t, mct denotes real marginal costs, α, δ, and β are unknown parameters, and ut is the disturbance term.
The model is estimated through early 2004 to capture recent progress on disinflation, and is based on 12-month consumer price index (CPI) inflation over the period from 1995:01 to 1994:02 (see equation 2 below). Real marginal costs are proxied with the real effective exchange rate, given the importance of imported inputs. Inflation expected one year ahead is obtained from survey findings of Consensus Economics, an international economic survey organization, and the Central Bank of Turkey. The estimation frequency is monthly or bimonthly, depending on the frequency at which one-year-ahead inflation is observed. Since the level of the real exchange rate is endogenous to the current disturbance to inflation, ut, the equation is estimated by two-stage least squares, using the real exchange rate in the previous year, the price of public sector wholesale prices relative to the CPI in the previous year, and current external demand (the GNP of industrialized trade partners) as instruments for the real effective exchange rate.2
The following key findings emerge:
Inflation expectations dominate the price-setting process. The coefficient of expected future inflation is highly significant, and statistically indistinguishable from 1. In contrast, the coefficient of past inflation is statistically indistinguishable from zero. The real effective exchange rate is also estimated to have a statistically significant coefficient, suggesting that the evolution of foreign prices (measured in domestic currency terms) is a significant determinant of Turkish prices. These findings are not altered when the coefficients of lagged and expected future inflation are constrained to add up to 1—a model restriction that is not rejected in the unconstrained version.
Inflation expectations have become more important over the estimation period. Recursive estimates are used for samples ending at the start of the current program supported by the International Monetary Fund right through to early 2004 (Figure 3.2). The coefficient of the real exchange rate is kept fixed to economize on degrees of freedom, but no restrictions are imposed on the lagged and expected future inflation coefficients. The main finding—that the estimated weight on expected future inflation increases while that on past inflation diminishes over time—is robust to allowing the coefficient on the real exchange rate to also vary with the sample, but the confidence bands on the estimates are considerably wider in this case.
The results are found to be robust when allowance is made for shorter-duration price contracts. If the duration of price contracts is less than one year—quite possible in a country with high and persistent inflation up until recently—the use of quarterly, as opposed to 12-month, inflation measures is a useful robustness test (even allowing for the shorter period of availability of quarterly inflation expectations data and their inherent noisiness).3,4 The results are, however, very similar using quarterly inflation, with inflation expectations predominating and becoming more significant over the sample period (Figure 3.3).

Recursive Coefficients of Equation (1)
(Recursive two-stage least squares estimates using annual inflation data)

Recursive Coefficients of Equation (1)
(Recursive two-stage least squares estimates using annual inflation data)
Recursive Coefficients of Equation (1)
(Recursive two-stage least squares estimates using annual inflation data)

Additional Recursive Coefficients of Equation (1)
(Recursive generalized methods of moments estimates using quarterly inflation data)

Additional Recursive Coefficients of Equation (1)
(Recursive generalized methods of moments estimates using quarterly inflation data)
Additional Recursive Coefficients of Equation (1)
(Recursive generalized methods of moments estimates using quarterly inflation data)
These results are also consistent with the limited evidence of indexation in Turkey. While some elements of public price- and wage-setting behavior involve ex post inflation indexation, there is little evidence of widespread indexation in the private sector. Collective wage bargaining is limited, with private sector wage determination adopting flexibly to labor market conditions. Indeed, Shiller (1997) singles out Turkey as a surprising example of a country that has experienced persistently high inflation without moving to widespread indexation.
What Determines Inflation Expectations?
With inflation expectations dominating the price-setting process, anchoring expectations is crucial for successful disinflation. The role of the central bank is key in this respect, with operational independence, a transparent monetary policy, and a clear anchor all playing important roles. But can other factors play a supporting role?
Regression results indicate that fiscal policy can help support the disinflation process by anchoring inflation expectations. To test the role of fiscal policy, a multivariate regression of expected one-year-ahead inflation is run on the primary balance. The regression includes control variables such as past inflation, the exchange rate, short-term money market rates, the real effective exchange rate, and real unit labor costs. The estimation results confirm a strong negative link between expected inflation and the primary balance (Table 3.1). While past inflation is also important, its coefficient is well below 1, suggesting little evidence of a rigidly adaptive expectation formation process. Not surprisingly, exchange rate movements and labor costs also play an important role in driving inflation expectations.5
Determinants of Inflationary Expectations
Determinants of Inflationary Expectations
Dependent Variable: Expected Inflation (12-month ahead) Sample: 1995:01–2004:01 | |||
---|---|---|---|
Variable | Coefficient | Std. error | t-statistic |
Constant | 0.172 | 0.040 | 4.233 *** |
Inflation (–2) | 0.498 | 0.059 | 8.392 *** |
Primary balance/GDP (–2) | −2.485 | 0.698 | −3.557 *** |
Exchange ratedepreciation (–2) | 0.222 | 0.033 | 6.725 *** |
Money market interest | |||
rate (–4)–inflation (–4) | −0.0003 | 0.0009 | −0.375 |
Real exchange rate (–2) | −0.0574 | 0.2269 | −2529 ** |
Real unit labor cost (–2) | −0.128 | 0.052 | −2.488 * |
Number of observations: | 90 | ||
Adjusted R2 | 0.90 |
Determinants of Inflationary Expectations
Dependent Variable: Expected Inflation (12-month ahead) Sample: 1995:01–2004:01 | |||
---|---|---|---|
Variable | Coefficient | Std. error | t-statistic |
Constant | 0.172 | 0.040 | 4.233 *** |
Inflation (–2) | 0.498 | 0.059 | 8.392 *** |
Primary balance/GDP (–2) | −2.485 | 0.698 | −3.557 *** |
Exchange ratedepreciation (–2) | 0.222 | 0.033 | 6.725 *** |
Money market interest | |||
rate (–4)–inflation (–4) | −0.0003 | 0.0009 | −0.375 |
Real exchange rate (–2) | −0.0574 | 0.2269 | −2529 ** |
Real unit labor cost (–2) | −0.128 | 0.052 | −2.488 * |
Number of observations: | 90 | ||
Adjusted R2 | 0.90 |
Recursive regressions are used to evaluate the evolving impact in recent years of the following key variables (Figure 3.4):
Lagged inflation. Price setters appear to place a higher weight on lagged inflation in forming price expectations when the inflation path is relatively stable; the opposite holds in periods of uncertainty. For instance, the coefficient on lagged inflation is estimated to have declined sharply during the postcrisis period. This is not surprising, given the uncertainties following the forced float in February 2001, when past manifestations of inflation might not have been expected to convey much information about future inflation trends. In contrast, the weight attached to past inflation increased during periods of relatively predictable disinflation (for instance, beginning in early 2002 as the disinflation process began to take hold).
Fiscal outcomes. The impact of fiscal outcomes is significant for most of the sample period, but was at its highest in 2000, accompanying a sizable improvement in the primary balance. The apparently nonlinear effect of the primary balance on expected inflation suggests that primary fiscal balances play an important signaling role, with large fiscal adjustments signaling the commitment of the government to fiscal sustainability, which in turn curbs inflationary expectations. This is consistent with the fact that the primary balance is significantly linked to inflationary expectations during periods of disinflation from chronically high inflation rates (Celasun, Gelos, and Prati, 2004).
Exchange rate. The exchange rate is estimated to have a statistically insignificant coefficient in samples ending before 2003, but its weight has increased significantly since then. These results suggest that the strong Turkish lira played an important role in aiding the disinflation process in 2003.

Determinants of Inflationary Expectations: Recursive Coefficients
(Recursive two-stage least squares estimates)

Determinants of Inflationary Expectations: Recursive Coefficients
(Recursive two-stage least squares estimates)
Determinants of Inflationary Expectations: Recursive Coefficients
(Recursive two-stage least squares estimates)
Conclusions and Challenges
Turkey’s inflation process is quite flexible, with inflation expectations playing a large and increasingly important role in determining inflation rates. This suggests that disinflation need not be too costly as long as inflation expectations are properly anchored. While central bank credibility plays a key role in anchoring these expectations—a topic not addressed here—the chapter suggests that fiscal variables play an important supporting role.
Looking ahead, an important unresolved issue is whether disinflation will become more difficult with inflation moving to single digits. A possible hint of this is suggested by the fact that past inflation appeared to play a larger role in the formation of expectations under stable conditions. The examination of cross-country evidence for similar disinflation episodes elsewhere may offer useful lessons and will be the subject of future research. For now, at least, inflation inertia does not appear to be a major obstacle to further disinflation. As long as central bank credibility and fiscal discipline are maintained, Turkey should be in a position to continue with its disinflation objectives.
An earlier version of a similar analysis, which includes more detailed technical discussions, can be found in Celasun, Gelos, and Prati (2003).
The real exchange rate and the GNP of industrialized trade partners deviate from a trend estimated for 1990:01 to 2004:01.
Expectations of one-quarter-ahead inflation are constructed using monthly inflation expectations surveyed bimonthly by Consensus Economics for 1998:05 to 2004:02. Consensus Economics has published average monthly inflation expectations for the six months ahead every other month since 1998:05. For any given month t, we construct a three-month ahead inflation forecast as: (1 + πt+1)(1 + πt+2)(1 + πt+3)–1, where πs is the consumer price index inflation forecast for month s.
In order to take into account the potential serial correlation of the residuals in the quarterly specification, the equation is estimated with the generalized method of moments.
The coefficients on these two variables are negative, which is surprising. A possible explanation is that price setters expect reversals in these variables at the annual frequency. If marginal costs have been above trend in a given 12-month period, below-trend marginal costs are expected in the year ahead, leading to lower inflationary expectations.