## _wp11125 - 1. Descriptive Statistics

## Source details

**Canonical URL:** [_wp11125 - 1. Descriptive Statistics](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11125.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11125.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11125.pdf.json)

---

### I. Introduction — purpose and scope
- Studies properties of forecasts of real GDP growth for 46 economies over the period October 1989 to December 2008 using Consensus Forecasts.
- Country coverage: 15 advanced economies (AE) and 31 emerging market economies (EM).
- Data structure: monthly or bi-monthly updates of forecasts of a fixed event (annual real GDP growth), allowing analysis of the speed with which news is absorbed into forecasts.
- Main empirical focus:
  - Sluggishness in revisions of growth forecasts.
  - How sluggishness changes during recessions and banking crises.
  - Linkages between forecast revisions across countries, including a seven-country VAR (G-3: U.S., Germany, Japan; BRICs: Brazil, Russia, India, China).

### II. Data, sample construction, and episodes
- Forecasts:
  - Consensus (simple average) of analysts' monthly forecasts for current and next year.
  - Sequence for each year consists of 24 forecasts indexed h = 24 (January previous year) to h = 1 (December current year).
- Actuals: real GDP growth taken from IMF International Financial Statistics (latest available data as of June 2009 used).
- Frequency notes:
  - Some emerging markets initially bi-monthly; monthly frequency increased over time for many.
  - For bi-monthly gaps, preceding month forecasts used to fill missing values for some emerging markets.
- Recession dating:
  - Based on quarterly changes in level of real GDP using Burns and Mitchell (1946) classical business cycle definitions.
  - Economies classified as in recession in a given month if in recession in respective quarter.
  - Sample includes 45 recession episodes in advanced economies and 61 in emerging and developing economies.
- Banking crises:
  - Dates from Laeven and Valencia (2008) database, extended to 2010.
  - Sample includes 7 banking crises in advanced economies and 22 in emerging markets.
- Forecast error definition: actual minus forecast (negative implies overprediction).

### III. Summary forecast-error statistics (sample averages)
- Mean forecast error: essentially zero (also true for AE and EM separately).
- Mean absolute forecast error: 1.7 percentage points.
- Forecast errors conditional on episodes:
  - During recessions: growth overpredicted by about 2½ percentage points.
  - During banking crises: growth overpredicted by about 4½ percentage points.
- Group differences:
  - Overprediction and absolute errors larger for EM group than AE group.
- Regression evidence (Table 2): absolute forecast errors decline as forecasting horizon draws to a close; errors higher for recession and banking crisis episodes.

### IV. Two statistical tests of information rigidity — methods and core findings
- Test 1 (Coibion and Gorodnichenko (2010) approach)
  - Regress forecast error (AF_t,h - actual_t,h) on forecast revision r_t,h+k.
  - Null under full-information rational expectations: coefficient on forecast revision = 0; positive coefficient indicates information rigidity.
  - Findings (Tables 3 and 4):
    - Coefficient estimates almost all positive and significantly different from zero → reject full-information rational expectations in favor of information rigidity.
    - From 6-month horizon regressions: implied updating frequency in sticky information interpretation is every 5 ½ to 7 months.
    - Coefficients decline monotonically from early to late forecast months (columns 1→5), indicating quicker updating as horizon closes; for AE group the null cannot be rejected in one late-horizon regression.
    - EM coefficients tend to be higher than AE, implying greater information rigidities; in economic terms, updating takes about 1 to 2 months longer in EM group.
    - For 3-month revision horizon, coefficient estimates are larger than for 6-month horizon, but economic differences modest; AE group shows monotonic decline toward no rejection at end-horizon, EM pattern choppier with rigidity persisting at end-horizon.
- Test 2 (martingale / forecast-revision autocorrelation)
  - Under full-information rational expectations, sequence of forecasts for a fixed event must follow a martingale; regress current forecast revision on past revisions.
  - Findings (Table 5):
    - Strong positive correlation between current forecast revision and its first lag → considerable sluggishness.
    - Across alternative horizons and lags, estimated coefficients on lagged revisions indicate informational rigidities persist.

### V. Information rigidity during recessions and banking crises (state dependence)
- Descriptive patterns (Figures 1–3):
  - AE group:
    - April previous-year forecasts are tilted to the right; few forecasts of recessions far in advance.
    - By October current-year, forecasts converge toward distribution of actual values.
    - In recession years: forecasts start close to unconditional average, depart slightly around middle of previous year, with smooth revisions during current year; terminal forecast still slightly above outcome.
  - EM group:
    - April previous-year forecasts tilted right but already include a few recession forecasts.
    - By April current-year forecasts mirror actual distribution better than AE; by October correspondence is quite good.
    - In recession years: deviation from unconditional average appears from start of the horizon, largest revision at start of current year; terminal forecast still underestimates decline.
  - Recessions after banking crises:
    - AE: departure from unconditional forecast starts earlier; terminal forecast vastly underestimates actual decline.
    - EM: recognition starts later than AE but extent of decline more accurately forecast.
- Augmented regressions (Tables 6 and 7) — key statistical findings:
  - Coefficient on recession dummy negative and significant → forecast revisions larger in recession years.
  - Interaction term (recession dummy × lagged revision) negative → information acquisition speeds up during recessions.
  - For both AE and EM groups, cannot reject that sum of coefficients on revision and interaction = 0; for recession years, null of full-information rational expectations not rejected.
  - Recessions associated with banking crises:
    - Information acquisition also speeds up, but result driven by EM group (where null of full-information rational expectations cannot be rejected).
    - AE results weaker (interaction term negative but not significant in one case and positive in another), possibly reflecting small number of AE banking-crisis episodes.

### VI. Cross-country linkages in forecast revisions — VAR evidence
- Framework:
  - Represent forecast revisions as accumulation of news components; write revisions in autoregressive (VAR) form where diagonal elements capture own-country absorption speed and off-diagonals capture absorption of foreign news.
  - Use generalized impulse responses and generalized forecast error variance decompositions (ordering-free, Pesaran and Shin (1998)).
- Empirical specification:
  - Seven-country VAR: United States, Japan, Germany (G-3) and Brazil, Russia, India, China (BRICs).
  - Lag length set at 3 using AIC.
- Key quantitative evidence (Table 8; Figures 4–6):
  - Variance decomposition (normalized to sum to 100): contribution of own-shocks ranges from about 50% (Brazil) to close to 95% (U.S., Russia).
  - Off-diagonal dependence:
    - Japan and Germany forecast revisions show considerable dependence on U.S. revisions.
    - Japan, Germany and India show substantial dependence on Chinese revisions.
  - Generalized impulse responses (own-country shocks; Figure 4):
    - Sluggish absorption of own-country information in all seven cases.
    - Number of months to fully absorb information ranges from about 4–5 months (U.S., Brazil) to about 10 months (Germany, China).
  - Responses to Chinese news (Figure 5):
    - Most countries show sluggish responses to news from China; Chinese news important for several countries’ forecast revisions.
  - Speed of absorption (Figure 6):
    - Immediate absorption of news (at horizon 0) varies across countries from 50% to 90%.
    - By 6 months, in all seven countries 90% of the news has been absorbed into forecasts.

### VII. Conclusions — synthesis of empirical results
- Broad evidence of information rigidity in growth forecasts for 46 economies (1989–2008).
- Quantitative summary:
  - Preponderance of evidence points to 4 to 6 months as the duration it takes forecasters to update forecasts to fully reflect new information (consistent with previous studies for advanced economies).
  - Patterns of information rigidity broadly similar across AE and EM, with some evidence of somewhat faster incorporation of information in advanced economies (differences typically amount to about 1–2 months).
- State dependence:
  - Acquisition of information speeds up during recessions; larger forecast revisions and much lower serial correlation in revisions during recession years.
  - For both AE and EM groups, the null of full-information rational expectations cannot be rejected for recession years.
  - Speeding up also observed during banking crises, stronger evidence for emerging markets than for advanced economies.
- Cross-country linkages:
  - Departures from full-information rational expectations partly arise from slow absorption of news from the U.S. and China into other countries’ forecasts.
  - By 6 months, roughly 90% of information is absorbed for the seven-systemic-economy group.
- Limitation noted:
  - Analysis uses consensus (mean) forecasts, which may introduce aggregation bias and hides individual-forecaster dynamics (herding, group-think). A companion paper studies individual-level forecasts to document information rigidity, state-dependent acquisition, and herding behavior.

### VIII. Appendices — sample lists and data-frequency details (selected items preserved)
- Appendix I: Lists of countries, recessions and crisis episodes for Advanced economies and Emerging economies (country-level episode dates preserved as presented).
- Appendix II: Frequency of Data — start dates of bi-monthly and monthly consensus-forecast availability preserved (examples include ARGENTINA 1993m3 2001m8., AUSTRALIA . 1990m1., BRAZIL 1993m6 1989m11 2001m8, CHINA. 1994m12., U.S.A. . 1989m10., UNITED KINGDOM . 1989m10., etc.).
- Appendix III: Generalized Impulse Responses of Forecast Revisions — figures report country responses (horizons 1–15) with axes and confidence intervals; axis ranges preserved as presented (examples include Response of US to US axis -0.05 to 0.25; Response of Brazil to Brazil axis -0.1 to 0.5; Response of Russia to Russia axis -0.4 to 1.4).

*Source: _wp11125 - 1. Descriptive Statistics (PDF content unit).*

### 1. Descriptive Statistics ......................................................................................... 18

### _wp11125 - 1. Descriptive Statistics ......................................................................................... 18

### Main sections and page references
- 1. Descriptive Statistics ......................................................................................... 18
- 2. Absolute Forecast Errors................................................................................... 18
- 3. Informational Rigidities: Tests Based on Forecast Errors (6-month horizon) .. 19
- 4. Informational Rigidities: Tests Based on Forecast Errors (3-month horizon) .. 20
- 5. Informational Rigidities in Forecast Revisions................................................. 21
- 6. Informational Rigidities during Recessions ...................................................... 22
- 7. Informational Rigidities during Banking Crises ............................................... 23
- 8. Variance Decompositions ................................................................................. 24

### Figures (list and page locations)
- Figure 1. Distributions of Actual and Forecasted Real GDP Growth, 1889-2008 ........... 25
- Figure 2. Actual and Forecasted Real GDP Growth during Recessions .......................... 26
- Figure 3. Actual and Forecasted Real GDP Growth during Banking Crises ................... 26
- Figure 4. Generalized Impulse Responses of Forecast Revisions (Own-country responses; in percentage points) ......................................................................................... 27
- Figure 5. Generalized Impulse Responses of Forecast Revisions (Response to Chinese Revisions; in Percentage Points) ....................................................................... 28
- Figure 6. Speed of Absorption of News ........................................................................... 29

### Appendices and supplementary materials
- Appendix I: Description of Sample ...................................................................................... 30
  - Table 1: List of Countries, Recessions and Crisis Episodes ............................. 30
- Appendix II: Frequency of Data ............................................................................................. 32
- Appendix III: Generalized Impulse Responses of Forecast Revisions .................................... 33
  - Figure1A. Generalized Impulse Responses of U.S. Forecast Revisions (In Percentage Points) ...................................................................................... 33
  - Figure1B. Generalized Impulse Responses of Germany’s Forecast Revisions (In Percentage Points) ...................................................................................... 34
  - Figure1C. Generalized Impulse Responses of Japan’s Forecast Revisions (In Percentage Points) ...................................................................................... 35
  - Figure1D. Generalized Impulse Responses of Brazil’s Forecast Revisions (In Percentage Points) ...................................................................................... 36
  - Figure1E. Generalized Impulse Responses of China’s Forecast Revisions (In Percentage Points) ..................................................................................... 37
  - Figure1F. Generalized Impulse Responses of India’s Forecast Revisions (In Percentage Points) ...................................................................................... 38
  - Figure1G. Generalized Impulse Responses of Russia’s Forecast Revisions (In Percentage Points) ...................................................................................... 39

*Source: _wp11125 - 1. Descriptive Statistics (PDF content unit).*

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

### _wp11125 - References .............................................................................................................

### I. Introduction — purpose and scope
- Studies properties of forecasts of real GDP growth for 46 economies over the period October 1989 to December 2008 using Consensus Forecasts.
- Country coverage: 15 advanced economies (AE) and 31 emerging market economies (EM).
- Data structure: monthly or bi-monthly updates of forecasts of a fixed event (annual real GDP growth), allowing analysis of the speed with which news is absorbed into forecasts.
- Main empirical focus:
  - Sluggishness in revisions of growth forecasts.
  - How sluggishness changes during recessions and banking crises.
  - Linkages between forecast revisions across countries, including a seven-country VAR (G-3: U.S., Germany, Japan; BRICs: Brazil, Russia, India, China).

### II. Data on Consensus Forecasts — dataset features and episode definitions
- Forecasts: consensus (simple average) of analysts' monthly forecasts for current and next year; sequence for each year consists of 24 forecasts indexed h = 24 (January previous year) to h = 1 (December current year).
- Actual real GDP growth taken from IMF International Financial Statistics (latest available data as of June 2009 used in this paper).
- Frequency notes:
  - Some emerging markets initially bi-monthly; monthly frequency increased over time for many.
  - For bi-monthly gaps, preceding month forecasts used to fill missing values for some emerging markets.
- Recession dating: based on quarterly changes in level of real GDP using Burns and Mitchell (1946) classical business cycle definitions; economies classified as in recession in a given month if in recession in respective quarter.
  - Sample includes 45 recession episodes in advanced economies and 61 in emerging and developing economies.
- Banking crises: dates from Laeven and Valencia (2008) database, extended to 2010.
  - Sample includes 7 banking crises in advanced economies and 22 in emerging markets.
- Forecast error definition: actual minus forecast (negative implies overprediction).
- Summary forecast error statistics (averaged over all countries and horizons):
  - Mean forecast error: essentially zero (also true for AE and EM separately).
  - Mean absolute forecast error: 1.7 percentage points.
  - Forecast errors higher during recessions: growth overpredicted by about 2½ percentage points in recessions.
  - Forecast errors higher during banking crises: growth overpredicted by about 4½ percentage points during banking crises.
  - Overprediction and absolute errors larger for EM group than AE group.
- Regression evidence (Table 2): absolute forecast errors decline as forecasting horizon draws to a close; errors higher for recession and banking crisis episodes.

### III. Two statistical tests of information rigidity — methods and core findings
- Test 1 (Coibion and Gorodnichenko (2010) approach)
  - Regress forecast error (AF_t,h - actual_t,h) on forecast revision r_t,h+k.
  - Under full-information rational expectations null, coefficient on forecast revision = 0; positive coefficient indicates information rigidity.
  - Results (Tables 3 and 4):
    - Coefficient estimates almost all positive and significantly different from zero → reject full-information rational expectations in favor of information rigidity.
    - From 6-month horizon regressions: implied updating frequency in sticky information interpretation is every 5 ½ to 7 months.
    - Coefficients decline monotonically from early to late forecast months (columns 1→5), indicating quicker updating as horizon closes; for AE group the null cannot be rejected in one late-horizon regression.
    - EM coefficients tend to be higher than AE, implying greater information rigidities; in economic terms, updating takes about 1 to 2 months longer in EM group.
    - For 3-month revision horizon, coefficient estimates are larger than for 6-month horizon, but economic differences modest; AE group shows monotonic decline toward no rejection at end-horizon, EM pattern choppier with rigidity persisting at end-horizon.
- Test 2 (martingale / forecast revision autocorrelation)
  - Under full-information rational expectations, sequence of forecasts for a fixed event must follow a martingale; regress current forecast revision on past revisions.
  - Results (Table 5):
    - Strong positive correlation between current forecast revision and its first lag → considerable sluggishness.
    - Across alternative horizons and lags, estimated coefficients on lagged revisions indicate informational rigidities persist.

### IV. Information rigidity in recessions and banking crises
- Descriptive evidence (Figures 1–3):
  - AE group: April previous-year forecasts are tilted to the right; few forecasts of recessions far in advance. By October current-year, forecasts converge toward distribution of actual values.
  - EM group: April previous-year forecasts tilted right but already include a few recession forecasts; by April current-year forecasts mirror actual distribution better than AE; by October correspondence is quite good.
  - Graphical implication: forecast recognition of recessions appears somewhat faster for EM than AE.
  - Time profiles (Figure 2):
    - AE recessions: forecasts in recession years start close to unconditional average, depart slightly around middle of previous year, with smooth revisions during current year; terminal forecast still slightly above outcome.
    - EM recessions: deviation from unconditional average appears from start of the horizon, largest revision at start of current year; terminal forecast still underestimates decline.
  - Recessions after banking crises (Figure 3):
    - AE: departure from unconditional forecast starts earlier; terminal forecast vastly underestimates actual decline.
    - EM: recognition starts later than AE but extent of decline more accurately forecast.
- Statistical tests (Tables 6 and 7):
  - Augmented regressions include recession dummy and interaction of recession dummy with lagged forecast revision.
  - Findings:
    - Coefficient on recession dummy negative and significant → forecast revisions larger in recession years.
    - Interaction term negative → information acquisition speeds up during recessions.
    - For both AE and EM groups, cannot reject that sum of coefficients on revision and interaction = 0; for recession years, null of full-information rational expectations not rejected.
    - For recessions associated with banking crises:
      - Information acquisition also speeds up, but result driven by EM group (where null of full-information rational expectations cannot be rejected).
      - AE results weaker (interaction term negative but not significant in one case and positive in another), possibly reflecting small number of AE banking-crisis episodes.

### V. Cross-country linkages in forecast revisions — VAR framework and findings
- Statistical framework (Isiklar, Lahiri and Loungani (2006) approach):
  - Represent forecast revisions as accumulation of news components; write revisions in autoregressive (VAR) form where diagonal elements capture own-country absorption speed and off-diagonals capture absorption of foreign news.
  - Use generalized impulse responses and generalized forecast error variance decompositions (ordering-free, Pesaran and Shin (1998)).
  - Compute percentage of revision variation due to contemporaneous innovations and cumulative percentages over m periods.
- Empirical specification:
  - Seven-country VAR: United States, Japan, Germany (G-3) and Brazil, Russia, India, China (BRICs).
  - Lag length set at 3 using AIC.
- Key quantitative evidence (Table 8; Figures 4–6):
  - Variance decomposition (normalized to sum to 100): contribution of own-shocks ranges from about 50% (Brazil) to close to 95% (U.S., Russia).
  - Off-diagonal dependence:
    - Japan and Germany forecast revisions show considerable dependence on U.S. revisions.
    - Japan, Germany and India show substantial dependence on Chinese revisions.
  - Generalized impulse responses (own-country shocks; Figure 4):
    - Sluggish absorption of own-country information in all seven cases.
    - Number of months to fully absorb information ranges from about 4–5 months (U.S., Brazil) to about 10 months (Germany, China).
  - Responses to Chinese news (Figure 5):
    - Most countries show sluggish responses to news from China; Chinese news important for several countries’ forecast revisions.
  - Speed of absorption (Figure 6):
    - Immediate absorption of news (at horizon 0) varies across countries from 50% to 90%.
    - By 6 months, in all seven countries 90% of the news has been absorbed into forecasts.

### VI. Conclusions — synthesis of empirical results
- Evidence across tests indicates information rigidity in growth forecasts for a broad sample of 46 economies (1989–2008).
- Quantitative summary:
  - Preponderance of evidence points to 4 to 6 months as the duration it takes forecasters to update forecasts to fully reflect new information (consistent with previous studies for advanced economies).
  - Patterns of information rigidity broadly similar across AE and EM, with some evidence of somewhat faster incorporation of information in advanced economies (differences typically amount to about 1–2 months).
- State dependence:
  - Acquisition of information speeds up during recessions; larger forecast revisions and much lower serial correlation in revisions during recession years.
  - For both AE and EM groups, the null of full-information rational expectations cannot be rejected for recession years.
  - Speeding up also observed during banking crises, stronger evidence for emerging markets than for advanced economies.
- Cross-country linkages:
  - Departures from full-information rational expectations partly arise from slow absorption of news from the U.S. and China into other countries’ forecasts.
  - By 6 months, roughly 90% of information is absorbed for the seven-systemic-economy group.
- Limitation noted:
  - Analysis uses consensus (mean) forecasts, which may introduce aggregation bias and hides individual-forecaster dynamics (herding, group-think). A companion paper studies individual-level forecasts to document information rigidity, state-dependent acquisition, and herding behavior.

*Italic source: Content excerpt from _wp11125 - References (PDF) provided above.*

### APPENDIX I. DESCRIPTION OF SAMPLE

### _wp11125 - APPENDIX I. DESCRIPTION OF SAMPLE

### Appendix I — List of Countries, Recessions and Crisis Episodes (Advanced economies)
- Country entries (Country name | Region | Starting date | Banking crisis | Recession) as given:
  - AUSTRALIA* Asia Jan-90 1990Q2-91Q2
  - CANADA* Western Hemisphere Oct-89 1990Q2-91Q1, 2008Q4-09Q2
  - FRANCE* Europe Oct-89 1992Q2-93Q3, 2002Q4-03Q2, 2008Q2-09Q1
  - GERMANY* Europe Oct-89 1992Q2-93Q1, 1995Q4-96Q1, 2002Q4-04Q3, 2008Q2-09Q1
  - GREECE* Europe Jun-93 1990Q2-90Q3,1992Q2-93Q1,1994Q4-95Q2, 2008Q4-09Q3
  - ITALY* Europe Oct-89 1992Q2-93Q3, 1996Q2-96Q4, 2001Q2-01Q4, 2003Q1-03Q2, 2004Q4-05Q1, 2008Q2-09Q2
  - JAPAN* Asia Oct-89 1997 1993Q2-93Q4, 1997Q2-99Q1, 2001Q2-01Q4, 2008Q2-09Q1
  - NETHERLANDS* Europe Nov-89 2008 2008Q2-09Q2
  - NEW ZEALAND* Asia Nov-89 1991Q1-91Q2, 1997Q4-98Q1, 2008Q1-09Q1
  - NORWAY* Europe Nov-89 2002Q3-03Q1, 2008Q3-09Q2
  - SPAIN* Europe Nov-89 1992Q2-93Q2, 2008Q2-09Q3
  - SWEDEN* Europe Nov-89 1991 1990Q2-93Q1, 2008Q2-09Q1
  - SWITZERLAND* Europe Nov-89 1990Q3-93Q1, 1996Q2-96Q3, 1998Q4-99Q1, 2001Q2-03Q1, 2008Q3-09Q2
  - UNITED STATES* Western Hemisphere Oct-89 2007, 2008 1990Q4-91Q1, 2001Q1-01Q3, 2008Q1-09Q2
  - UNITED KINGDOM* Europe Oct-89 2007, 2008 1990Q3-91Q3, 2008Q2-09Q3
- Number of countries or episodes745
- Sources: International Financial Statistics; Claessens, Kose and Terrones (2008); Laeven and Valencia (2008).
- Notes: The classification of countries into advanced, emerging and developing is aligned with Consensus Forecasts publications. Countries for which the dating of recession and recovery episodes is based on quarterly data are marked with an asterisk. Only crises during the time period for which consensus forecasts are available are reported.

### Appendix I — List of Countries, Recessions and Crisis Episodes (Emerging economies)
- Country entries (Country name | Region | Starting date | Banking crisis | Recession) as given:
  - ARGENTINA* Western Hemisphere Mar-93 1995, 2001 1995Q2-96Q1, 1998Q4-02Q4
  - BRAZIL* Western Hemisphere Nov-89 1990, 1994 1990Q2-91Q1, 1992Q2-92Q4, 1995Q4-96Q1, 1998Q4-99Q3, 2001Q4-02Q1, 2008Q4-09Q1
  - BULGARIA Europe Jan-95 1996 1996-97, 2008Q4-09Q1
  - CHILE* Western Hemisphere Mar-93 1998Q4-99Q3, 2008Q3-09Q2
  - CHINA* Asia Dec-94 1998
  - COLOMBIA* Western Hemisphere Mar-93 1998 1998Q3-99Q4, 2008Q3-08Q4
  - CROATIA* Europe May-98 1998 1998Q3-99Q4, 2008Q2-09Q2
  - CZECH REPUBLIC* Europe Jan-95 1996 1997Q3-98Q4, 2008Q4-09Q1
  - ESTONIA* Europe May-98 1999Q1-99Q3, 2008Q1-09Q3
  - HONG KONG* Asia Nov-90 1997Q4-98Q4, 2001Q1-03Q4, 2008Q2-09Q1
  - HUNGARY* Europe Nov-90 1991 1990-93, 2008Q2-09Q2
  - INDIA* Asia Dec-94 1993
  - INDONESIA* Asia Nov-90 1997 1998Q1-99Q1
  - LATVIA* Europe May-98 1993Q1-94Q1, 1996Q4-97Q1, 2008Q4-09Q2
  - LITHUANIA* Europe May-98 1999Q2-99Q4, 2008Q3-09Q2
  - MALAYSIA* Asia Nov-90 1997 1998Q1-99Q1, 2008Q4-09Q1
  - MEXICO* Western Hemisphere Nov-89 1994 1995Q1-95Q4, 2001Q3-02Q1, 2008Q2-09Q2
  - PERU* Western Hemisphere Mar-93 1998Q2-99Q3, 2000Q4-01Q2, 2008Q4-09Q1
  - PHILIPPINES* Asia Dec-94 1997 1998Q2-98Q4
  - POLAND* Europe Nov-90 1992 1990Q1-92Q1, 2008Q4-09Q1
  - ROMANIA Europe Jan-95 1997-1998, 2008Q3-09Q3
  - REPUBLIC OF KOREA* Asia Nov-89 1997 1998Q1-98Q4
  - SINGAPORE* Asia Nov-90 2001, 2008
  - SLOVAK REPUBLIC* Europe Jan-95 1998 1999Q3-00Q1, 2008Q4-09Q1
  - SLOVENIA* Europe Jan-95
  - SOUTH AFRICA* Africa Jun-93 2008Q4-09Q2
  - TAIWAN* Asia Nov-89 2001Q2-01Q4, 2008Q2-09Q1
  - THAILAND* Asia Nov-90 1997 1997Q2-99Q1, 2008Q2-09Q1
  - TURKEY* Europe Jan-95 2000 1999Q1-99Q4, 2001Q1-02Q1, 2008Q2-09Q1
  - UKRAINE Europe Jan-95 1998 2008Q3-09Q1
  - VENEZUELA* Western Hemisphere Mar-93 1994 1993Q1-94Q4, 1996Q2-96Q3, 1998Q3-99Q4,
- Number of countries or episodes2261
- Sources and notes: same as Advanced economies. Countries with quarterly dating are marked with an asterisk.

### Appendix II — Frequency of Data
- Table headings (Country | Start Date of Bi-monthly Data | Start Date of Monthly Data | Second Start of Monthly Data If Data Frequency Was Changed From Monthly to Bi-monthly to Monthly)
- Selected entries as given (preserving exact month-year formats):
  - ARGENTINA 1993m3 2001m8.
  - AUSTRALIA . 1990m1.
  - BRAZIL 1993m6 1989m11 2001m8
  - BULGARIA 1998m6 1995m1 2007m5
  - CANADA . 1989m10.
  - CHILE 1993m3 2001m8.
  - CHINA. 1994m12.
  - COLOMBIA 1993m3 2001m8.
  - CROATIA 1998m5 2007m5
  - CZECH REPUBLIC 1998m6 1995m1 2007m5
  - ESTONIA 1998m5 2007m5.
  - FRANCE . 1989m10.
  - GERMANY . 1989m10.
  - GREECE . 1993m6.
  - HONG KONG . 1990m11.
  - HUNGARY 1998m6 1990m11 2007m5
  - INDIA . 1994m12.
  - INDONESIA . 1990m11.
  - ITALY . 1989m10.
  - JAPAN . 1989m10.
  - LATVIA 1998m5 2007m5.
  - LITHUANIA 1998m5 2007m5.
  - MALAYSIA . 1990m11.
  - MEXICO 1993m6 1989m11 2001m8
  - NETHERLANDS . 1989m11.
  - NEW ZEALAND . 1989m11.
  - NORWAY . 1989m11.
  - PERU 1993m3 2001m8.
  - PHILIPPINES . 1994m12.
  - POLAND 1998m6 1990m11 2007m5
  - ROMANIA 1998m6 1995m1 2007m5
  - RUSSIA . 1995m1.
  - SINGAPORE 1998m6 1995m1 2007m5
  - SLOVAKIA 1998m6 1995m1 2007m5
  - SLOVENIA . 1993m6.
  - SOUTH AFRICA . 1989m11.
  - SPAIN . 1989m11.
  - SWEDEN . 1989m11.
  - SWITZERLAND . 1989m11.
  - TAIWAN . 1989m11.
  - THAILAND . 1990m11.
  - TURKEY 1998m6 1995m1 2007m5
  - U.S.A. . 1989m10.
  - UKRAINE 1998m6 1995m1 2007m5
  - UNITED KINGDOM . 1989m10.
  - VENEZUELA 1993m3 2001m8.
- (Entries preserve original punctuation and formatting, including trailing periods.)

### Appendix III — Generalized Impulse Responses of Forecast Revisions
- Note: The Figure shows confidence intervals for 2 standard deviations.
- Figures and labeled responses included (all axes and labels preserved as presented):
  - Figure 1A. Generalized Impulse Responses of U.S. Forecast Revisions (In percentage points)
    - Response of US to US (axis -0.05 to 0.25; horizons 1–15)
    - Response of US to Japan (axis -0.05 to 0.25; horizons 1–15)
    - Response of US to Germany (axis -0.05 to 0.25; horizons 1–15)
    - Response of US to Brazil (axis -0.05 to 0.25; horizons 1–15)
    - Response of US to Russia (axis -0.05 to 0.25; horizons 1–15)
    - Response of US to India (axis -0.05 to 0.25; horizons 1–15)
    - Response of US to China (axis -0.05 to 0.25; horizons 1–15)
  - Figure 1B. Generalized Impulse Responses of Germany's Forecast Revisions (In percentage points)
    - Response of Germany to US (axis -0.03 to 0.15; horizons 1–15)
    - Response of Germany to Japan (axis -0.03 to 0.15; horizons 1–15)
    - Response of Germany to Germany (axis -0.03 to 0.15; horizons 1–15)
    - Response of Germany to Brazil (axis -0.03 to 0.15; horizons 1–15)
    - Response of Germany to Russia (axis -0.03 to 0.15; horizons 1–15)
    - Response of Germany to India (axis -0.03 to 0.15; horizons 1–15)
    - Response of Germany to China (axis -0.03 to 0.15; horizons 1–15)
  - Figure 1C. Generalized Impulse Responses of Japan's Forecast Revisions (In percentage points)
    - Response of Japan to US (axis -0.05 to 0.25; horizons 1–15)
    - Response of Japan to Japan (axis -0.05 to 0.25; horizons 1–15)
    - Response of Japan to Germany (axis -0.05 to 0.25; horizons 1–15)
    - Response of Japan to Brazil (axis -0.05 to 0.25; horizons 1–15)
    - Response of Japan to Russia (axis -0.05 to 0.25; horizons 1–15)
    - Response of Japan to India (axis -0.05 to 0.25; horizons 1–15)
    - Response of Japan to China (axis -0.05 to 0.25; horizons 1–15)
  - Figure 1D. Generalized Impulse Responses of Brazil's Forecast Revisions (In percentage points)
    - Response of Brazil to US (axis -0.1 to 0.5; horizons 1–15)
    - Response of Brazil to Japan (axis -0.1 to 0.5; horizons 1–15)
    - Response of Brazil to Germany (axis -0.1 to 0.5; horizons 1–15)
    - Response of Brazil to Brazil (axis -0.1 to 0.5; horizons 1–15)
    - Response of Brazil to Russia (axis -0.1 to 0.5; horizons 1–15)
    - Response of Brazil to India (axis -0.1 to 0.5; horizons 1–15)
    - Response of Brazil to China (axis -0.1 to 0.5; horizons 1–15)
  - Figure 1E. Generalized Impulse Responses of China's Forecast Revisions (In percentage points)
    - Response of China to US (axis -0.04 to 0.16; horizons 1–15)
    - Response of China to Japan (axis -0.04 to 0.16; horizons 1–15)
    - Response of China to Germany (axis -0.04 to 0.16; horizons 1–15)
    - Response of China to Brazil (axis -0.04 to 0.16; horizons 1–15)
    - Response of China to Russia (axis -0.04 to 0.16; horizons 1–15)
    - Response of China to India (axis -0.04 to 0.16; horizons 1–15)
    - Response of China to China (axis -0.04 to 0.16; horizons 1–15)
  - Figure 1F. Generalized Impulse Responses of India's Forecast Revisions (In percentage points)
    - Response of India to US (axis -0.10 to 0.30; horizons 1–15)
    - Response of India to Japan (axis -0.10 to 0.30; horizons 1–15)
    - Response of India to Germany (axis -0.10 to 0.30; horizons 1–15)
    - Response of India to Brazil (axis -0.10 to 0.30; horizons 1–15)
    - Response of India to Russia (axis -0.10 to 0.30; horizons 1–15)
    - Response of India to India (axis -0.10 to 0.30; horizons 1–15)
    - Response of India to China (axis -0.10 to 0.30; horizons 1–15)
  - Figure 1G. Generalized Impulse Responses of Russia's Forecast Revisions (In percentage points)
    - Response of Russia to US (axis -0.4 to 1.4; horizons 1–15)
    - Response of Russia to Japan (axis -0.4 to 1.4; horizons 1–15)
    - Response of Russia to Germany (axis -0.4 to 1.4; horizons 1–15)
    - Response of Russia to Brazil (axis -0.4 to 1.4; horizons 1–15)
    - Response of Russia to Russia (axis -0.4 to 1.4; horizons 1–15)
    - Response of Russia to India (axis -0.4 to 1.4; horizons 1–15)
    - Response of Russia to China (axis -0.4 to 1.4; horizons 1–15)

### References (selected list as presented)
- Batchelor, Roy, 2007, Bias in Macroeconomic Forecasts, International Journal of Forecasting, 23, 189–203.
- Benabou, Roland, 2009, “Groupthink: Collective Delusions in Organizations and Markets,” NBER Working Paper No. 14764.
- Bry, Gerhard and Charlotte Boschan, 1971, “Cyclical Analysis of Time Series: Selected Procedures and Computer Programs,” (New York: NBER).
- Claessens, S., A. Kose, and M. Terrones, 2008, “What Happens During Recessions, Crunches and Busts?” IMF Working Paper No. 08/274.
- Laeven, L. and F. Valencia, 2008, “Systemic Banking Crises: A New Database,” IMF Working Paper No. 08/224.
- Pesaran, M. H., and Y. Shin, 1998, “Generalized Impulse Response Analysis in Linear Multivariate Models,” Economics Letters, 58, 17–29.
- (Additional references listed in the Appendix are preserved in the source.)

*Source: _wp11125 - APPENDIX I. DESCRIPTION OF SAMPLE*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11125.pdf_