## _wp04208

## Source details

**Canonical URL:** [_wp04208](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2004/_wp04208.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2004/_wp04208.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2004/_wp04208.pdf.json)

---

### I. Introduction and context
- Credibility is key to the success of a disinflationary program: by reducing inflationary expectations and their impact on wage and price setting behavior, a fully credible policy can deliver a sharp drop in inflation with limited output losses.
- Credibility improves gradually; central banks aim to reduce inflation as quickly as the public can learn about the policymaker’s true intentions (King, 1996).
- Factors that shape policymakers’ ability to steer market confidence:
  - Time-inconsistency delays public learning about true intentions.
  - Perceptions of internal inconsistency in government policies weaken credibility.
  - Credibility can be affected by factors beyond immediate government control or that the government can influence only in the medium term.
- Turkey since the 2001 crisis:
  - Turkey implemented a far-reaching stabilization and reform program complementing disinflation with fiscal discipline and ambitious structural reforms.
  - The approach has delivered the lowest inflation rate in decades.
  - Policy credibility will remain key to further disinflation, especially where the trade-off between the speed of disinflation and output costs may become steeper.
- Main empirical result:
  - The credibility of the disinflation program launched in the aftermath of the 2001 crisis has improved dramatically.

### II. Measurement concept and advantages
- Definition:
  - Credibility = the public’s degree of uncertainty regarding publicly announced policy objectives: the probability, as assessed by agents, that policy outturns coincide with the policy targets.
- Rationale for an econometric measure in emerging markets:
  - Does not require liquid government bond markets or a competitive financial industry.
  - Does not require surveys of inflation expectations or extraction from asset prices.
  - Uses aggregate CPI, embedding information on actual price and wage setting behavior and capturing backward- versus forward-looking expectations embodied in the CPI.
- Limitations of alternative measures noted:
  - Asset-price and survey-derived measures can be problematic in emerging markets due to illiquid bond markets and non-representative financial-market participants.

### III. Methodology
- Econometric model of the disinflation process (AR(1)):
  - (1) ∆π_t = α + β ∆π_{t-1} + e_t, where e_t ~ N(0, σ^2).
- Interpretation of parameters:
  - α = average speed of the disinflation process (average change in the inflation rate).
  - β = persistence of temporary accelerations or decelerations.
  - σ^2 = volatility of shocks producing deviations.
- Mapping policy objectives into model parameters:
  - Use authorities’ announced annual inflation target to construct a projected monthly inflation path (assume a constantly decreasing monthly inflation rate).
  - Estimate (1) on monthly series combining actual data up to program announcement and projected monthly path thereafter to obtain program coefficients α(P), β(P), σ(P)^2.
  - Genuine commitment implies a regime change: α moves from α ≥ 0 to α(P) < 0 during disinflation.
- Bayesian learning implementation:
  - Representative agent has priors on the model parameters and updates beliefs via Bayes’ rule as new inflation realizations are observed.
  - Recursive Bayes’ rule produces posterior distributions for each t = T,..., T+K (T = month of program approval; T+K = last month of projection period).
  - Credibility computed as the posterior probability that α(t) ≤ α(P) at each period t.
- Estimation method:
  - Theil’s (1971) mixed estimator computes Bayes’ rule at each realization by applying OLS to an augmented regression that incorporates prior linear restrictions.
  - Posterior distributions from the mixed estimator are approximately normal, enabling computation of P(α(t) ≤ α(P)).
- Practical notes:
  - By construction credibility cannot reach one at the beginning of the sample even if actual inflation equals projected inflation; the bias is small in practice.
  - Choice of S (length of series entering credibility analysis) reflects agents’ memory span.

### IV. Implementation details and parameter choices
- Required ingredients:
  1. Econometric AR(1) model (equation (1)).
  2. Mapping from authorities’ inflation target to model parameter values (α(P), β(P), σ(P)^2).
  3. Estimation procedure implementing Bayesian learning (Theil’s mixed estimator).
- Procedure summary:
  - Use historical data up to program approval and projected monthly path thereafter.
  - Estimate program coefficients from the augmented sample and compare posterior α(t) to α(P) recursively.
- Agents’ memory span and estimation window:
  - S is set to 24 months (consistent with a standard two-year forecasting horizon under most inflation-targeting regimes).
  - Sensitivity checks use S = 18, 36, 48 months.

### V. Application to Turkey (June 1995–December 2004)
- Empirical setup:
  - Natural starting point: strengthened economic program in May 2001 after the February 2001 financial crisis.
  - Credibility analysis period extends through March 2004, encompassing the IMF-supported program approved in February 2002 and external shocks (September 11, 2001; the Iraq war).
  - Data span: June 1995–December 2004.
- Program coefficient constructions:
  - 2001 program coefficients:
    - Historical data: June 1995–April 2001.
    - Projections: May 2001–December 2002 (end of original three-year program approved December 1999).
  - 2002 program coefficients:
    - Historical data: June 1995–January 2002.
    - Projections: February 2002–December 2004 (end of the new program).
  - Annual inflation targets used are those published in the respective letters of intent with the IMF; monthly projections derived assuming a constantly decreasing monthly inflation rate.
  - Prior coefficients (mean and variance of the initial prior distribution) estimated by fitting equation (1) over June 1995–April 2001.

### VI. Key empirical findings for Turkey
- Program coefficient estimates:
  - The 2001 program coefficients imply a marked acceleration in the disinflation process relative to prior values: historical average decline in inflation was much smaller and statistically indistinguishable from zero, whereas the projected path implied a marked acceleration (α more negative) and reduced uncertainty (higher t-statistics reflecting lower standard errors).
  - The 2002 program coefficients imply a slight deceleration in the disinflation process relative to the 2001 program coefficients, without altering the degree of uncertainty around the projection path.
- Bayesian credibility measure results:
  - Credibility rises from about 35 percent in May 2001 to about 85 percent in March 2004.
  - Credibility is responsive to program implementation, political uncertainty, and external shocks:
    - Steep downward path in second half of 2001 (reflecting skepticism and external shock of September 11).
    - Sizable improvement associated with implementation of the strengthened IMF-supported program since February 2002 and a slightly less ambitious target.
    - Mild slowdown in credibility gains in early 2003 associated with delays in program implementation after end-2002 elections and uncertainty related to the war in Iraq.
  - Credibility, being a probability, varies between 0 and 1.

### VII. Robustness and alternative measures
- Alternative credibility measures considered:
  - Survey-based measure: difference between program inflation and expected inflation in the CBT inflation survey.
  - OLS-based recursive measure using same program coefficients and estimation window S.
  - Both alternatives are more volatile and sensitive to news than the Bayesian measure.
- Sensitivity analyses:
  - Changing the initial prior to a more pessimistic one (including observations up to June 1991) or lengthening the estimation window to S = 36 or 48 months shifts the credibility series downward but does not change its time profile or the end value being well above the initial value.
  - Shortening the agents’ memory span shifts credibility upward.
  - In all sensitivity analyses the marked improvement in credibility over the program period remains.

### VIII. Conclusions
- Since the 2001 crisis the Central Bank of Turkey has acted as a de facto inflation targeter and announced commitment to introducing formal inflation targeting as conditions allow.
- Disinflation has been dramatic; continued success depends on how credible disinflation is perceived to be.
- The paper’s empirical measure of credibility, focusing on agents’ learning, has advantages over measures relying on asset prices or market-participant surveys in emerging markets:
  - Uses aggregate CPI.
  - Embeds actual price and wage setting behavior.
  - Captures backward- versus forward-looking expectations embedded in the CPI.
- Empirical takeaway:
  - Credibility of Turkey’s disinflation program launched after the 2001 crisis has improved dramatically, which bodes well for future disinflation and implementation of inflation targeting in Turkey.

### IX. Appendix (Bayes’ rule summary)
- Model and notation:
  - y = Xβ + ε, ε ~ N(0, σ^2 I); parameters of interest β and σ^2.
- Posterior distribution:
  - p(β, σ^2 | Y) = L(Y | β, σ^2) p(β, σ^2) / p(Y), where L is the likelihood and p(β, σ^2) is the joint prior.
  - Marginal posterior distributions p(β | Y) and p(σ^2 | Y) obtained by integrating the joint posterior over σ^2 and β, respectively.

*Italic: Source — IMF staff paper section and appendix as provided in the content unit.*

### References..............................................................................................................

### _wp04208 - References..............................................................................................................

### I. INTRODUCTION
- Credibility is key to the success of a disinflationary program: by reducing inflationary expectations and their impact on wage and price setting behavior, a fully credible policy can deliver a sharp drop in inflation with limited output losses.
- In practice, credibility improves gradually; when switching from a regime of high and persistent inflation to one of price stability, central banks aim to reduce inflation as quickly as the public can learn about the policymaker’s true intentions (King, 1996).
- Policymakers’ ability to steer market confidence depends on multiple factors:
  - Time-inconsistency—that is, the fact that ex post policies may differ from the ex ante optimal policies—delays public learning about true intentions.
  - Perceptions of internal inconsistency in government policies (for example, pursuing disinflation without measures to limit budget deficits) weaken credibility.
  - Credibility can be affected by factors beyond immediate government control or by factors the government can influence only in the medium term; even time- and internally consistent programs may lack credibility if policymakers lack instruments to implement them or to cope with exogenous shocks.

### Turkey since the 2001 crisis
- Since the 2001 crisis, Turkey has implemented a far-reaching stabilization and reform program that complements disinflation with fiscal discipline and ambitious structural reforms.
- The approach has delivered the lowest inflation rate in decades.
- Policy credibility will remain key to further disinflation, especially where the trade-off between the speed of disinflation and output costs may become steeper.

### Measurement of disinflation credibility
- The paper presents an empirical measure of disinflation credibility and discusses its evolution in Turkey since the 2001 crisis.
- Credibility is unobservable; the paper measures credibility using an econometric approach that mimics how economic agents learn about policymakers’ true intentions and abilities.
- Advantages of the proposed measure:
  - Does not require liquid government bond markets or a competitive financial industry.
  - Does not require surveys of inflation expectations or extraction from asset prices.
  - Suitable for assessing credibility in emerging market economies.
- Main empirical result reported:
  - The credibility of the disinflation program launched in the aftermath of the 2001 crisis has improved dramatically.

### Organization of the paper
- Section II motivates and describes the methodology to quantify disinflation credibility in general terms and illustrates practical application steps to the Turkish disinflation process.
- Supporting material in the document includes:
  - Tables: "1. Priors and Programs Coefficients"
  - Figures: "1. Inflation, 1995-2004" and "2. Disinflation Credibility, 2001-04"

*Source: _wp04208 - References..............................................................................................................*

### Section III applies this methodology to Turkey’s stabilization program over the period since

### _wp04208 - Section III applies this methodology to Turkey’s stabilization program over the period since

### Measuring disinflation credibility (Section II)
- Credibility is defined as the public’s degree of uncertainty regarding publicly announced policy objectives: the probability, as assessed by agents, that policy outturns coincide with the policy targets.
- Credibility is unobservable; existing approaches in the literature include:
  - proxies for credible monetary frameworks (exchange rate variability; central bank independence; inflation persistence),
  - time-varying parameter methods,
  - forecast-error techniques,
  - asset prices,
  - surveys of inflation expectations.
- Limitations of asset-price and survey-derived measures in emerging markets: domestic bond markets may be illiquid; financial market participants may not reveal true expectations; such measures may not represent inflation expectations of actual price setters (workers and firms).
- The paper proposes an econometric measure: the probability that inflation outturns are consistent with announced monetary policy objectives (following Baxter (1985) and Weber (1991)).
  - Advantages claimed:
    - Uses aggregate consumer price index (available and relatively reliable in most emerging markets).
    - Embeds information on actual price and wage setting behavior more fully than financial-market estimates.
    - Captures backward- versus forward-looking expectations embedded in the CPI.
  - Assumes existence of a representative agent in price and wage setting.

### Methodology (Section II.A)
- Econometric model of the disinflation process: first-order autoregressive AR(1):
  - (1) ∆π_t = α + β ∆π_{t-1} + e_t,
  - where e_t ~ N(0, σ^2).
- Interpretation of parameters:
  - α = average speed of the disinflation process (average change in the inflation rate).
  - β = persistence of temporary accelerations or decelerations.
  - σ^2 = volatility of shocks producing deviations.
- Mapping policy objectives into model parameters:
  - Use authorities’ announced annual inflation target to construct a projected monthly inflation path (assume a constantly decreasing monthly inflation rate).
  - Estimate equation (1) on monthly series combining actual data up to the program announcement and the derived projected monthly path from program approval to projection end.
  - Regression yields “program” coefficients denoted α(P), β(P), σ(P)^2 — coefficients consistent with announced disinflation objectives in the absence of shocks.
  - A genuine commitment to taming inflation implies a regime change: α moves from α ≥ 0 to α(P) < 0 during disinflation.
- Bayesian learning implementation:
  - Representative agent has priors on the model parameters and updates beliefs via Bayes’ rule as new inflation realizations are observed.
  - Recursive application of Bayes’ rule produces posterior distributions on the disinflation parameters for each t = T,..., T+K (T = month of program approval; T+K = last month of projection period).
  - Compute posterior probability that estimated α(t) ≤ α(P) at each period t.
- Estimation method:
  - Theil’s (1971) mixed estimator computes Bayes’ rule at each realization by applying OLS to an augmented regression that incorporates prior linear restrictions on the parameter vector.
  - Posterior distributions from the mixed estimator are approximately normal, enabling straightforward computation of P(α(t) ≤ α(P)).
- Practical note:
  - By construction credibility cannot reach one at the beginning of the sample even if actual inflation equals projected inflation; the bias is small in practice.
  - Choice of S (length of series entering credibility analysis) reflects agents’ memory span.

### Implementation details (Section II.B)
- Required ingredients:
  1. Econometric AR(1) model (equation (1)).
  2. Mapping from authorities’ inflation target to model parameter values (α(P), β(P), σ(P)^2).
  3. Estimation procedure implementing Bayesian learning (Theil’s mixed estimator).
- Procedure summary:
  - Use historical data up to program approval and projected monthly path thereafter.
  - Estimate program coefficients from the augmented sample and compare posterior α(t) to α(P) recursively.
  - S is set to 24 months (consistent with a standard two-year forecasting horizon under most inflation-targeting regimes); sensitivity checks use S = 18, 36, 48 months.

### Disinflation credibility in Turkey (Section III)
- Empirical setup:
  - Natural starting point: launch of a strengthened economic program in May 2001 after the February 2001 financial crisis.
  - Credibility analysis period extends through March 2004, encompassing the IMF-supported program approved in February 2002 and external shocks (September 11, 2001; the Iraq war).
  - Data span the period June 1995–December 2004.
  - Two sets of program coefficients:
    - 2001 program coefficients: historical data June 1995–April 2001; projections May 2001–December 2002 (end of original three-year program approved December 1999).
    - 2002 program coefficients: historical data June 1995–January 2002; projections February 2002–December 2004 (end of the new program).
  - Annual inflation targets used are those published in the respective letters of intent with the IMF; monthly projections derived assuming a constantly decreasing monthly inflation rate.
  - Prior coefficients (mean and variance of the initial prior distribution) estimated by fitting equation (1) over June 1995–April 2001.
- Key empirical findings:
  - Estimated program coefficients imply a marked acceleration in the disinflation process with the 2001 program relative to prior values: historical average decline in inflation was much smaller and statistically indistinguishable from zero, whereas the projected path implied a marked acceleration (α more negative) and reduced uncertainty (higher t-statistics reflecting lower standard errors).
  - The 2002 program coefficients imply a slight deceleration in the disinflation process relative to the 2001 program coefficients, without altering the degree of uncertainty around the projection path.
- Bayesian credibility measure results:
  - Credibility rises from about 35 percent in May 2001 to about 85 percent in March 2004.
  - Credibility is responsive to program implementation, political uncertainty, and external shocks:
    - Steep downward path in second half of 2001 (reflecting skepticism and external shock of September 11).
    - Sizable improvement associated with implementation of the strengthened IMF-supported program since February 2002 and a slightly less ambitious target.
    - Mild slowdown in credibility gains in early 2003 associated with delays in program implementation after end-2002 elections and uncertainty related to the war in Iraq.
  - Credibility, being a probability, varies between 0 and 1.
- Robustness checks:
  - Two alternative credibility measures considered:
    - Survey-based measure computed as the difference between program inflation and expected inflation in the CBT inflation survey.
    - OLS-based recursive measure using same program coefficients and estimation window S.
  - Both alternatives are more volatile and sensitive to news than the Bayesian measure.
  - Changing initial prior to a more pessimistic one (including observations up to June 1991) or lengthening the estimation window to S = 36 or 48 months shifts the credibility series downward but does not change its time profile or the end value being well above the initial value.
  - Shortening the agents’ memory span shifts credibility upward; in all sensitivity analyses the marked improvement in credibility over the program period remains.

### Conclusions (Section IV)
- Since the 2001 crisis the Central Bank of Turkey has acted as a de facto inflation targeter and announced commitment to introducing formal inflation targeting as conditions allow.
- Disinflation has been dramatic; continued success depends on how credible disinflation is perceived to be.
- The paper’s empirical measure of credibility focuses on agents’ learning and has advantages over measures relying on asset prices or market-participant surveys in emerging markets:
  - Uses aggregate CPI,
  - Embeds actual price and wage setting behavior,
  - Captures backward- versus forward-looking expectations embedded in the CPI.
- Empirical result: credibility of Turkey’s disinflation program launched after the 2001 crisis has improved dramatically, which bodes well for future disinflation and implementation of inflation targeting in Turkey.

### Appendix: Bayes’ rule (summary)
- Model: y = Xβ + ε, ε ~ N(0, σ^2 I); parameters of interest β and σ^2.
- Posterior distribution given data is obtained via Bayes’ rule:
  - p(β, σ^2 | Y) = L(Y | β, σ^2) p(β, σ^2) / p(Y),
  - where L is the likelihood and p(β, σ^2) is the joint prior.
- Marginal posterior distributions p(β | Y) and p(σ^2 | Y) are obtained by integrating the joint posterior over σ^2 and β, respectively.

*Italic: Source — IMF staff paper section and appendix as provided in the content unit.*

### REFERENCES

### REFERENCES

### References cited
- Agénor, P R and M Taylor, 1993, “Analysing Credibility in High Inflation Countries: A New Approach,” Economic Journal, 103, 329-336.
- Agénor, P.R. and M P Taylor, 1992, “Testing for Credibility Effects,” IMF Staff Papers, Vol. 39, No.3, September.
- Alogoskoufis, G., 1992, “Monetary Accommodation, Exchange Rate Regimes and Inflation Persistence,” Economic Journal, 102, 461-480.
- Baxter, M., 1985, “The Role of Expectations in Stabilisation Policy,” Journal of Monetary Economics, Vol. 15, 343-362.
- Blanchard, O-J., 1984, “The Lucas Critique and the Volcker Deflation,” American Economic Review, Vol. 74, 211-215.
- Celasun O., G. Gelos, and A. Prati, forthcoming, “Would ‘Cold Turkey’ Work in Turkey,” IMF Staff Papers.
- Celasun O., G. Gelos, and A. Prati, forthcoming, b, “Obstacles to Disinflation: What Is the Role of Fiscal Expectations,” Economic Policy.
- Christensen, M., 1987b, “On Interest-Rate Determination, Testing for Policy Credibility, and the Relevance of the Lucas Critique,” European Journal of Political Economy, Vol. 3, 369-388.
- Grilli, V. D. Masciandaro, and G. Tabellini, 1991, “Political and Monetary Institutions and Public Financial Policies in the Industrial Countries,” Economic Policy.
- Kapetanios G., A Pagan, and A. Scott, 2004, “Making a Match: Combining Theory and Evidence in Policy-Oriented Macroeconomic Modeling,” (unpublished manuscript: Australian National University and University of New South Wales)
- King, Mervyn, 1996, “How Should Central Banks Reduce Inflation?,” Bank of England Quarterly Bulletin, November, 434-44.
- Kremers, J., 1990, “Gaining Policy Credibility for a Disinflation,” IMF Staff Papers, Vol. 37, 116-145.
- Mankiw, N. J., Mirron, and D. Weil, 1987, “The Adjustment of Expectations to a Change in Regime: A Study of the Founding of the Federal Reserve,” American Economic Review, Vol. 77, 358-374.
- Theil, H., 1971, Principles of Econometrics, (New York: Wiley).
- Weber, A., 1991, “Reputation and Credibility in the European Monetary System,” Economic Policy, Vol. 12, 57–102.

*Source: _wp04208 - REFERENCES.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2004/_wp04208.pdf_