## _wp0502

## Source details

**Canonical URL:** [_wp0502](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp0502.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp0502.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp0502.pdf.json)

---

### Introduction and objectives
- Objective of the paper:
  - Present and discuss a new data set on revenue-forecasting practices in 34 low-income and transition countries.
  - Analyze whether key forecasting characteristics lead to more timely and realistic budget estimates.
  - Present a corruption-based model of forecasting interference and test it against the data.
- Key contextual points:
  - Forecasting interference is defined as a significant deviation between the budget forecast and a technical forecast by experts.
  - Sample size: 34 countries; about 80 percent are low-income countries and the rest transition economies (predominantly CIS).
  - Slightly more than half of the countries in the sample were under an IMF program in 2003.

### Data, sample characteristics, and common practices
- Sample composition (regional distribution; format: Region: N, Percent, Average GDP per capita U.S.$ (2000)):
  - Sub-Saharan Africa: 9, 26.4, 756
  - Developing Asia: 7, 20.6, 731
  - Transition CIS including Mongolia: 7, 20.6, 902
  - Middle East, Turkey: 5, 14.7, 1809
  - Western Hemisphere: 6, 17.6, 3975
- Total average GDP per capita (U.S.$ (2000)): 1504
- Common forecasting practices (sample means reported in Table 2):
  - One agency responsible for forecast: 91.1 percent
  - Macroeconomic forecast by one agency: 44.1 percent
  - Forecasting horizon limited to one year: 64.7 percent
  - Budget forecast covers only central government: 47.0 percent
  - Five or fewer staff in charge of forecast: 23.5 percent
  - Methods:
    - Basic extrapolations: 83.9 percent
    - Use of econometric methods: 12.9 percent
    - Use of disaggregate data: 20.0 percent
  - Formality indicators (sample means where reported):
    - Forecasting responsibilities formally defined: 36.6 percent
    - Forecasting formally initiated: 68.7 percent
    - Formal revisions: 64.7 percent
    - Formal documentation: 51.6 percent
  - Organization:
    - Forecasting produced by single agency: 47.1 percent
    - Only one uniform forecast produced: 76.4 percent
    - Macro forecast produced by single agency: 44.1 percent
  - Transparency:
    - Nongovernment agencies participate in forecast: 34.4 percent
    - Information published outside budget document: 36.3 percent
  - Informational content in budget document (percent of countries reporting the item):
    - Aggregate revenue forecast: 82.3 percent
    - Breakdown of forecast into revenue types: 85.3 percent
    - Data on past revenue outturns: 58.8 percent
    - Analysis of past developments and forecasts: 17.6 percent
    - Summary of macro assumptions: 55.9 percent
    - Decomposition of forecast into various effects: 20.6 percent

### Forecasting timeliness and determinants
- Measured indicator: months that the first draft revenue forecast is available prior to budget submission.
- Benchmark: OECD six-month lead time recommended for nonindustrial countries (OECD, 2001).
- Empirical distribution:
  - About two-thirds of the countries prepare the first draft of the revenue forecast less than three months prior to budget submission.
- Regression findings (Table 3; OLS; dependent variable: months before budget submission):
  - Positive associations with timeliness:
    - Transparency: coefficient 0.768 (model 1)
    - Organizational simplicity: coefficient 1.033 (model 2)
    - Higher per capita income improves forecast efficiency (positive effect in combined specifications)
  - Formality: no measurable effect on timeliness (formality parameter not significant).
  - Perceived country corruption: no measurable effect on promptness in these regressions.

### Indices of forecasting practices: definitions and summary statistics
- Indices constructed (see Appendix I):
  - Formality index: unweighted sum of four binary variables (responsibilities formally defined; forecasting formally initiated; formally revised; formally documented). Range 0–4.
  - Simplicity index: unweighted sum of three binary variables (single agency responsible for revenue forecast; single agency responsible for macro forecast; only one unified revenue forecast produced). Range 0–3.
  - Transparency index: weighted sum of eight binary variables (macroeconomic assumptions published outside budget; outside agencies participate; six information items in budget document). Items (3)–(8) combined given equal weight so that aggregate information content equals weight of items (1) and (2).
- Summary statistics for indices (Table 8):
  - Formality:
    - Sample mean: 2.10
    - Median: 2.00
    - Std. Dev: 1.37
    - Significant regional variation: --
    - Significant variation by per capita income: --
  - Simplicity:
    - Sample mean: 1.21
    - Median: 1.00
    - Std. Dev: 0.59
    - Significant regional variation: --
    - Significant variation by per capita income: --
  - Transparency:
    - Sample mean: 1.21
    - Median: 1.25
    - Std. Dev: 0.75
    - Significant regional variation: Yes (high scores in Western Hemisphere)
    - Significant variation by per capita income: Positive

### Forecasting interference: definition, theoretical model, and empirical evidence
- Definition:
  - Forecasting interference is a significant deviation between the budget forecast and a forecast by technical experts, capturing covert and explicit manipulation, conservative assumption adjustments, and the announcement of performance targets as forecasts.
- Incidence in sample:
  - Interference appears to be present in 36.3 percent of the sample countries (reported as 36 percent in text; Table 2 reports 36.3 percent).
- Corruption-based theoretical model (summary of structure and equilibrium properties):
  - Model primitives:
    - Tax revenue T is stochastic and can be either TH or TL with probabilities p and (1-p).
    - The public knows only the distribution p; the government observes the realization ex ante.
    - Government can announce a low forecast when revenue is high and appropriate a residual ∆T for private benefit.
    - Public can audit with intensity α; successful audit recovers lost revenue at linear cost C(α)=α and penalizes government at rate K>0.
    - Government chooses manipulation intensity m (probability of manipulation when revenue is high); public chooses audit intensity α.
  - Equilibrium properties:
    - Government manipulates when expected net gain outweighs expected penalty from audit; manipulation threshold depends on ∆T/K and α.
    - Mixed-strategy equilibria can arise with 0<m**<1 and 0<α**<1 if ∆G is sufficiently small.
    - Interference is welfare deteriorating in the model and can conceal extraction without necessarily increasing ex post forecast errors.
- Empirical tests of the corruption hypothesis:
  - Main empirical finding: country corruption is a robust predictor of forecasting interference.
    - Table 4 (IV specification) reports Corruption coefficient 3.534 (reported as significant).
  - Alternative motives and tests:
    - Revenue administration independence/autonomy: Independence or autonomy of RA does not robustly predict interference (Table 5).
    - Fiscal sustainability/expenditure pressures:
      - Interest/total expenditure and IMF program exposure (IMF-5) are not significant in most specifications (Table 6).
      - Expenditure pressure variable (direct report whether revenue forecasts were adjusted when incompatible with expenditure plans) is positive and significant in one logit specification, supporting an expenditure-pressure motive alongside corruption (Table 6 models 5 and 6).
  - Role of forecasting-practice indices on interference (Table 7):
    - Simplicity: no robust relationship to interference.
    - Formality: negative sign but not statistically significant; introducing formality weakens corruption coefficient.
    - Transparency: nonlinear effects detected in quadratic specifications (inverse U-shaped effect); transparency tends to reduce interference beyond a threshold (turning point reported around 1.29–1.33 on the transparency score). Transparency does not eliminate the corruption effect.

### Principal findings and policy implications
- Principal findings:
  - Forecasting procedures in the sample are generally low standard: widespread reliance on one-year central-government forecasts, simple extrapolation methods, limited use of econometric methods, limited use of disaggregated data, thin human resources.
  - Forecasting cycles are generally short: about two-thirds of countries prepare the first draft less than three months prior to budget submission, falling short of the OECD six-month lead-time benchmark.
  - Organizational simplicity and transparency are associated with more timely forecasts; formality (written rules) does not improve timeliness.
  - Forecasting interference occurs in roughly 36 percent of sample countries and is strongly correlated with measures of country corruption.
  - Expenditure pressures also appear to matter as a motive for interference but do not eliminate the corruption effect.
  - Forecasting transparency reduces interference; more formal or less complex processes alone are not sufficient to reduce interference.
- Policy recommendations emphasized:
  - Give greater emphasis to forecasting transparency: increase information access and the greater participation of nongovernment institutions in revenue forecasting.
  - Focus on streamlining forecasting organization where appropriate to improve timeliness.
  - Be cautious about relying solely on formal rules to improve timeliness or reduce interference; transparency and broader information access are critical, especially given governance constraints.

*Source: _wp0502 - Introduction and "8. Summary Statistics of Indices on Forecasting Practices" (IMF working paper; source PDF: _wp0502).*

### Introduction............................................................................................................

### _wp0502 - Introduction

### Major Themes and Structure
- Introduction (page 3)
- I. Literature Review (page 5)
- II. A New Data Set on Revenue-Forecasting Practices (page 6)
  - A. Sample Characteristics (page 6)
  - B. Description of Forecasting Practices (page 7)
  - C. Forecasting Practices and Forecasting Timeliness (page 13)
- III. Forecasting Interference: Theory and Evidence (page 16)
  - A. Definition of Forecasting Interference (page 16)
  - B. A Model of Government Corruption and Forecasting Interference (page 16)
  - C. Empirical Analysis of Forecasting Interference (page 20)
- IV. Conclusions (page 25)
- References (page 28)

### Data, Tables, and Figures Inventory
- Tables
  - Table 1. Sample Characteristics
  - Table 2. Sample Revenue-Forecasting Characteristics
  - Table 3. Revenue Forecasting Efficiency and Forecasting Practices
  - Table 4. Discretionary Interference and Country Corruption
  - Table 5. Discretionary Interference and Independence of Revenue Authority (RA)
  - Table 6. Discretionary Interference, Fiscal Sustainability, and Expenditure Pressure
  - Table 7. Discretionary Interference and Forecasting Practices
- Figures
  - Figure 1. Forecasting Horizon of the Annual Budget Forecast
  - Figure 2. Score on the Formality Indicator
  - Figure 3. Score on the Simplicity Indicator
  - Figure 4. Score on the Transparency Indicator
  - Figure 5. Timeliness of Revenue Forecast: Number of Months Forecast Available Before Budget Submission
  - Figure 6. Equilibrium Manipulation and Audit Strategies
- Appendices
  - Indices of Forecasting Practices (Appendices and Appendix Tables)

### Focus Areas Covered by the Content Unit
- Review of existing literature on revenue forecasting and related governance issues.
- Construction and description of a new dataset on revenue-forecasting practices, including sample composition and characterization of forecasting methods.
- Analysis of forecasting timeliness and how forecasting practices relate to the timing of budget processes.
- Theoretical development defining "forecasting interference" and a model linking government corruption to interference in forecasting.
- Empirical tests assessing discretionary interference in revenue forecasts and its associations with country corruption, revenue authority independence, fiscal sustainability, expenditure pressure, and specific forecasting practices.
- Conclusions synthesizing theoretical and empirical results on forecasting interference and practices.

*Source: _wp0502 - Introduction (IMF PDF).*

### 8.     Summary Statistics of Indices on Forecasting Practices..................................................27

### 8.     Summary Statistics of Indices on Forecasting Practices

### Introduction and objectives
- Objective of the paper:
  - Present and discuss a new data set on revenue-forecasting practices in 34 low-income and transition countries.
  - Analyze whether key forecasting characteristics lead to more timely and realistic budget estimates.
  - Present a corruption-based model of forecasting interference and test it against the data.
- Key contextual points:
  - Forecasting interference is defined as a significant deviation between the budget forecast and a technical forecast by experts.
  - The sample covers 34 countries; about 80 percent are low-income countries and the rest transition economies (predominantly CIS).
  - Slightly more than half of the countries in the sample were under an IMF program in 2003.

### Data, sample characteristics, and common practices
- Sample composition:
  - Total countries: 34.
  - Regional distribution (N, Percent, Average GDP per capita U.S.$ (2000)):
    - Sub-Saharan Africa: 9, 26.4, 756
    - Developing Asia: 7, 20.6, 731
    - Transition CIS including Mongolia: 7, 20.6, 902
    - Middle East, Turkey: 5, 14.7, 1809
    - Western Hemisphere: 6, 17.6, 3975
  - Total average GDP per capita (U.S.$ (2000)): 1504
- Common forecasting practices (sample means reported in Table 2):
  - One agency responsible for forecast: 91.1 percent
  - Macroeconomic forecast by one agency: 44.1 percent
  - Forecasting horizon limited to one year: 64.7 percent
  - Budget forecast covers only central government: 47.0 percent
  - Five or fewer staff in charge of forecast: 23.5 percent
  - Methods:
    - Basic extrapolations: 83.9 percent
    - Use of econometric methods: 12.9 percent
    - Use of disaggregate data: 20.0 percent
  - Formality:
    - Forecasting responsibilities formally defined (sample mean for this item reported elsewhere): 36.6 percent
    - Forecasting formally initiated: 68.7 percent
    - Formal revisions: 64.7 percent
    - Formal documentation: 51.6 percent
  - Organization:
    - Forecasting produced by single agency: 47.1 percent
    - Only one uniform forecast produced: 76.4 percent
    - Macro forecast produced by single agency: 44.1 percent
  - Transparency:
    - Nongovernment agencies participate in forecast: 34.4 percent
    - Information published outside budget document: 36.3 percent
  - Informational content in budget document (percent of countries reporting the item):
    - Aggregate revenue forecast: 82.3 percent
    - Breakdown of forecast into revenue types: 85.3 percent
    - Data on past revenue outturns: 58.8 percent
    - Analysis of past developments and forecasts: 17.6 percent
    - Summary of macro assumptions: 55.9 percent
    - Decomposition of forecast into various effects: 20.6 percent

### Forecasting timeliness and its determinants
- Measured indicator: months that the first draft revenue forecast is available prior to budget submission.
- Benchmark: OECD six-month lead time recommended for nonindustrial countries (OECD, 2001).
- Empirical distribution:
  - About two-thirds of the countries prepare the first draft of the revenue forecast less than three months prior to budget submission.
- Regression findings (Table 3, OLS; dependent variable: months before budget submission):
  - Positive associations with timeliness:
    - Transparency: coefficient 0.768 (model 1)
    - Organizational simplicity: coefficient 1.033 (model 2)
    - Higher per capita income improves forecast efficiency (positive effect in combined specifications)
  - Formality had no measurable effect on timeliness (formality parameter not significant in regressions).
  - The level of perceived country corruption had no measurable effect on promptness of the forecast in these regressions.

### Indices of forecasting practices: definitions and summary statistics
- Indices constructed (see Appendix I):
  - Formality index: unweighted sum of four binary variables (responsibilities formally defined; forecasting formally initiated; formally revised; formally documented). Range 0–4.
  - Simplicity index: unweighted sum of three binary variables (single agency responsible for revenue forecast; single agency responsible for macro forecast; only one unified revenue forecast produced). Range 0–3.
  - Transparency index: weighted sum of eight binary variables (macroeconomic assumptions published outside budget; outside agencies participate; six information items in budget document). Items (3)–(8) combined given equal weight so that aggregate information content equals weight of items (1) and (2).
- Summary statistics for indices (Table 8):
  - Formality:
    - Sample mean: 2.10
    - Median: 2.00
    - Std. Dev: 1.37
    - Significant regional variation: --
    - Significant variation by per capita income: --
  - Simplicity:
    - Sample mean: 1.21
    - Median: 1.00
    - Std. Dev: 0.59
    - Significant regional variation: --
    - Significant variation by per capita income: --
  - Transparency:
    - Sample mean: 1.21
    - Median: 1.25
    - Std. Dev: 0.75
    - Significant regional variation: Yes (high scores in Western Hemisphere)
    - Significant variation by per capita income: Positive

### Forecasting interference: definition, model, and empirical evidence
- Definition used in the paper:
  - Forecasting interference is a significant deviation between the budget forecast and a forecast by technical experts. This definition is broad and captures covert and explicit manipulation, conservative assumption adjustments, and the announcement of performance targets as forecasts.
- Incidence in sample:
  - Interference appears to be present in 36.3 percent of the sample countries (reported as 36 percent in text; Table 2 reports 36.3 percent).
- Corruption-based theoretical model (summary):
  - Key structure:
    - Tax revenue T is stochastic and can be either TH or TL with probabilities p and (1-p).
    - The public knows only the distribution p; the government observes the realization ex ante.
    - Government can announce a low forecast when revenue is high and appropriate a residual ∆T for private benefit.
    - Public can audit with intensity α; successful audit recovers lost revenue at linear cost C(α)=α and penalizes government at rate K>0.
    - Government chooses manipulation intensity m (probability of manipulation when revenue is high); public chooses audit intensity α.
  - Equilibrium properties:
    - Government manipulates when expected net gain outweighs expected penalty from audit: manipulation threshold depends on ∆T/K and α.
    - Mixed-strategy equilibria can arise with 0<m**<1 and 0<α**<1 if ∆G is sufficiently small.
    - Interference is welfare deteriorating in the model and can conceal extraction without necessarily increasing ex post forecast errors.
- Empirical tests of the corruption hypothesis:
  - Main empirical finding: country corruption is a robust predictor of forecasting interference.
    - Table 4 (IV specification) reports Corruption coefficient 3.534 (reported as significant).
  - Tests of alternative motives and explanations:
    - Revenue administration independence/autonomy:
      - Independence or autonomy of RA does not robustly predict interference (Table 5).
    - Fiscal sustainability/expenditure pressures:
      - Interest/total expenditure and IMF program exposure (IMF-5) are not significant in most specifications (Table 6).
      - Expenditure pressure variable (direct report whether revenue forecasts were adjusted when incompatible with expenditure plans) is positive and significant in one logit specification, supporting an expenditure-pressure motive alongside corruption (Table 6 models 5 and 6).
  - Role of forecasting-practice indices on interference (Table 7):
    - Simplicity: no robust relationship to interference.
    - Formality: negative sign but not statistically significant; introducing formality weakens corruption coefficient.
    - Transparency: nonlinear effects detected in quadratic specifications (inverse U-shaped effect); transparency tends to reduce interference beyond a threshold (turning point reported around 1.29–1.33 on the transparency score). Transparency does not eliminate the corruption effect.

### Principal findings and policy implications
- Principal findings:
  - Forecasting procedures in the sample are generally low standard: widespread reliance on one-year central-government forecasts, simple extrapolation methods, limited use of econometric methods, limited use of disaggregated data, thin human resources.
  - Forecasting cycles are generally short: about two-thirds of countries prepare the first draft less than three months prior to budget submission, falling short of the OECD six-month lead-time benchmark.
  - Organizational simplicity and transparency are associated with more timely forecasts; formality (written rules) does not improve timeliness.
  - Forecasting interference occurs in roughly 36 percent of sample countries and is strongly correlated with measures of country corruption.
  - Expenditure pressures also appear to matter as a motive for interference but do not eliminate the corruption effect.
  - Forecasting transparency reduces interference; more formal or less complex processes alone are not sufficient to reduce interference.
- Policy recommendations emphasized in the paper:
  - Give greater emphasis to forecasting transparency: increase information access and the greater participation of nongovernment institutions in revenue forecasting.
  - Focus on streamlining forecasting organization where appropriate to improve timeliness.
  - Be cautious about relying solely on formal rules to improve timeliness or reduce interference; transparency and broader information access are critical, especially given governance constraints.

*Source: IMF working paper — Summary statistics and analysis reported in the chapter "8.     Summary Statistics of Indices on Forecasting Practices" from the provided PDF content.*

### REFERENCES

### REFERENCES

### Fiscal policy and public finance
- Abed, George T., and others, 1998, Fiscal Reforms in Low-Income Countries: Experience Under IMF-Supported Programs, IMF Occasional Paper No. 160 (Washington: International Monetary Fund).
- Allen, Richard, and Daniel Tommasi, eds., 2001, Managing Public Expenditure (Paris: Organization for Economic Co-operation and Development).
- Lienert, Ian C., and Feridoun Sarraf, 2001, “Systemic Weaknesses of Budget Management in Anglophone Africa,” IMF Working Paper 01/211 (Washington: International Monetary Fund).
- Organisation for Economic Co-operation and Development and World Bank, 2003, “OECD/WB Budget Practices and Procedures Survey,” Public Governance and Territorial Development Directorate (Paris: Organization for Economic Co-operation and Development).
- Tanzi, Vito, and Hamid Davoodi, 2000, “Corruption, Growth, and Public Finances,” IMF Working Paper 00/182 (Washington: International Monetary Fund).

### Revenue forecasting and forecasting bias
- Alt, Robert, 1993, “Revenue Forecasting and Estimation—How It’s Done, State-By-State,” State Tax Notes, Vol. 4 (May), pp. 1038–51.
- Auerbach, Alan J., 1999, “On the Performance and Use of Government Revenue Forecasts,” National Tax Journal, Vol. 52 (December), pp. 767–82.
- Bretschneider, Stuart I., and Wilpen Gorr, 1992, “Economic, Organizational, and Political Influences on Biases in Forecasting State Sales Tax Receipts,” International Journal of Forecasting, Vol. 7, No. 4, pp. 457–66.
- ———, 1987, “State and Local Government Revenue Forecasting,” in The Handbook of Forecasting, ed. by Spyros Makridakis and Steven C. Wheelwright (New York: Wiley, Second Edition).
- Cassidy, Glenn, Mark S. Kamlet, and Daniel S. Nagin,  1989, “An Empirical Examination of Bias in Revenue Forecasts by State Governments,” International Journal of Forecasting, Vol. 5, No. 3, pp. 321–31.
- Danninger, Stephan, 2005, “Revenue Forecasts as Performance Targets,” IMF Working Paper (Washington: International Monetary Fund, forthcoming).
- Golosov, Mikhail, and John King, 2002, “Tax Revenue Forecasts in IMF-Supported Programs,” IMF Working Paper 02/236 (Washington: International Monetary Fund)
- Kyobe, Annette, and Stephan Danninger, 2005, “Revenue Forecasting: How Is It Done? Experiences from Low-Income Countries,” IMF Working Paper (Washington: International Monetary Fund, forthcoming).
- Laster, David, Paul Bennett, and In Sun Geoum, 1999, “Rational Bias in Macroeconomic Forecasts,” The Quarterly Journal of Economics, Vol. 114 (February), pp. 293–318.
- Mocan, H. Nael, and Sam Azad, 1995, “Accuracy and Rationality of State General Fund Revenue Forecasts: Evidence from Panel Data,” International Journal of Forecasting, Vol. 11 (September), pp. 417–27.
- Rodgers, Robert, and Phillip Joyce, 1996, “The Effect of Underforecasting on the Accuracy of Revenue Forecasts by State Governments,” Public Administration Review, Vol. 56 (January–February), pp. 48–56.

### Corruption, governance, and informal activity
- Bardhan, Pranab, 1997, “Corruption and Development: A Review of Issues,” Journal of Economic Literature, Vol. 35 (September), pp. 1320–46.
- Fjeldstad, Odd-Helge, and Bertil Tungodden, 2003, “Fiscal Corruption: A Vice or Virtue?” World Development, Vol. 31, No. 8, pp. 1459–67.
- Friedman, Eric, and others, 2000, “Dodging the Grabbing Hand: The Determinants of Unofficial Activity in 69 Countries,” Journal of Public Economics, Vol. 76 (June), pp. 459–93.
- Hellman, Joel, and Daniel Kaufmann, 2001, “Confronting the Challenge of State Capture in Transition Economies,” Finance & Development, Vol. 38 (September), pp. 31–35.
- Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi, 2003, “Governance Matters III, Governance Indicators for 1996–2002,” (Washington: World Bank).
- Tanzi, Vito, and Hamid Davoodi, 2000, “Corruption, Growth, and Public Finances,” IMF Working Paper 00/182 (Washington: International Monetary Fund).
- Treisman, Daniel, 2000, “The Causes of Corruption: A Cross-National Study,” Journal of Public Economics, Vol. 76 (June), pp. 399–458.

### Forecasting theory, evaluation, and related empirical studies
- Christoffersen, Peter F., and Francis X. Diebold, 1997, “Optimal Prediction Under Asymmetric Loss,” in NBER Econometric Therory, Reseach Paper No. 167 (December), pp. 1–36 (Cambridge, Massachusetts: National Bureau of Economics Research).
- Ehrbeck, Tilman, and Robert Waldmann, 1996, “Why Are Professional Forecasters Biased? Agency Versus Behavioral Explanations,” The Quarterly Journal of Economics, Vol. 111 (February), pp. 21–40.
- Golosov, Mikhail, and John King, 2002, “Tax Revenue Forecasts in IMF-Supported Programs,” IMF Working Paper 02/236 (Washington: International Monetary Fund)
- Harvey, David, Stephen Leybourne, and Paul Newbold, 1997, “Testing the Equality of Prediction Mean Squared Errors,” International Journal of Forecasting, Vol. 13 (June), pp. 281–91.
- Jacob, Brian A., and Steven D. Levitt, 2003, “Rotten Apples: An Investigation of the Prevalence and Predictors of Teacher Cheating,” The Quarterly Journal of Economics, Vol. 118, No. 3, pp. 843–77.
- Ruge-Murciá, Francisco J., 2002, “A Prudent Central Banker,” Staff Papers, International Monetary Fund, Vol. 49, No. 3, pp. 456–69.
- Zellner, Arnold, 1986, “Biased Predictors, Rationality and the Evaluation of Forecasts,” Economics Letters, Vol. 21, No. 1, pp. 45–48.

_ Source: _wp0502 - REFERENCES (Source PDF filename: _wp0502 - REFERENCES; Canonical URL: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp0502.pdf)_

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp0502.pdf_