## Information Rigidities in Economic Growth Forecasts: Evidence from a Large International Panel (Section 1–3)

## Source details

**Canonical URL:** [Information Rigidities in Economic Growth Forecasts: Evidence from a Large International Panel (Section 1–3)](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1356.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1356.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1356.pdf.json)

---

### Abstract — main findings
- Sample: individual forecasts from 30 advanced and emerging economies during 1989–2010.
- Main empirical findings:
  - Evidence does not support the validity of the sticky information model (Mankiw and Reis, 2002) for describing the dynamics of professional growth forecasts.
  - Empirical evidence is more in line with implications of ―noisy‖ information models (Woodford, 2002; Sims, 2003).
  - Information rigidities are more pronounced in emerging economies than advanced economies.
  - Nonlinearities exist in forecast smoothing: smoothing is less pronounced in the tails of the distribution of individual forecast revisions than in the central part.

### Motivation, hypotheses, and testing framework
- Context and hypotheses:
  - Forecast efficiency as analog to rational expectations; strong-form efficiency implies forecast errors orthogonal to all relevant information.
  - Nordhaus (1987) weak efficiency tests: forecast error independent of past forecast revisions; today’s forecast revision independent of past forecast revisions.
  - Theoretical mechanisms producing correlated forecast revisions include:
    - Sticky information (Mankiw and Reis, 2002): infrequent updating due to fixed costs.
    - Noisy/imperfect information (Woodford, 2002; Sims, 2003): continual but noisy updating.
    - Behavioral explanations (Tversky and Kahneman, 1981): slow incorporation of surprises.
    - Non-standard/asymmetric loss functions.
- Value added:
  - Use of individual-forecaster data from 30 countries avoids aggregation bias.
  - Broad coverage permits comparison between advanced and emerging economies.
  - Degree of smoothing from individual data is far lower than from average (consensus) forecasts.
  - Evidence favors noisy information models; large-tail revisions are less correlated with subsequent revisions.

### Data and descriptive statistics
- Data source and coverage:
  - Consensus Economics Inc. cross-country survey.
  - Survey started in October 1989.
  - Monthly frequency; forecasts collected during first two weeks and published mid-month.
  - Panel includes public and private institutions (mostly banks and research institutes).
- Sample and cleaning:
  - Data set comprises 188,639 individual forecasts from 30 different countries, of which 104,894 are from 14 advanced economies.
  - Forecasts cover target years between 1989 and 2011.
  - For each target year the data set contains a sequence of 24 forecasts of each panelist made between January of the year before the target year and December of the target year.
  - On average, the data set includes nearly 16 individual forecasts per period for each country.
  - Inclusion rules: only countries with individual forecasts reported; only forecasters with at least 10 forecasts included; name variants concatenated; continuity maintained through evident mergers/acquisitions.
- Descriptive patterns:
  - Average root mean squared forecast error (RMSFE) declines with the forecast horizon.
  - RMSFEs for emerging economies are, on average, more than twice as high as for advanced economies for large forecast horizons and still almost 75 percent higher at the end of the target year.
  - For advanced economies, revisions larger around the turn of the year; average size of revisions does not vary much with horizon.
  - For emerging economies, revisions much smaller for very long horizons and much higher during the target year (h<=12); at h=1 the average revision in emerging economies is about twice as large as for advanced economies.
  - Distributions of revisions are unimodal and bell-shaped; emerging-economy distributions more flattened; advanced economies show significantly negative skew for most horizons; emerging economies symmetric for large horizons and positively skewed for smaller horizons.
  - Forecasts cluster as horizon shrinks; advanced economies show very few deviations larger than half a percentage point; emerging economies show substantially larger dispersion even at h=1.

### Methodology for testing forecast smoothing
- General approach:
  - Test whether sequences of forecasts for the same event (annual real GDP growth) follow a martingale under full-information rational expectations.
  - Regress contemporaneous revision on past forecast revisions (Nordhaus 1987 analogue).
- Specifications and estimation:
  - Revisions computed over k*=k=3 months throughout.
  - Average-forecast regression: contemporaneous revision of the average forecast regressed on past forecast revisions; under sticky or imperfect information the regression coefficient maps to degree of informational rigidity.
  - Horizon dependence studied via interaction terms between forecast horizons and lagged revisions (equation (5)); coefficients expected positive and rising with horizon.
  - Estimation: fixed-effect panel OLS with Driscoll and Kraay (1998) standard error corrections; Arellano-Bond/Arellano-Bover (ABB) GMM used as robustness check.
  - Individual-forecast analogue (equation (6)): revision of forecaster j for country i, target year t, horizon h regressed on past individual revisions; autocorrelation coefficient λ measures forecast smoothing but should not be directly mapped to theoretical parameters.
  - Non-parametric sticky-information measure: fraction of individuals who revise their forecasts at least once during the 3 months prior to a given point (Andrade and Le Bihan, 2010).
  - Nonlinearity assessed using threshold models to compare tails versus central distribution behavior.

### Empirical findings — Average (consensus) forecasts
- Estimation horizons: h=1, 4, 7, 10, 13 and 16; revisions computed with k=3.
- Evidence of information rigidities:
  - Strong positive correlation between current forecast revision and its first lag for all country groups and estimation methods; coefficients highly statistically significant.
  - Coefficient on lagged revisions at very short horizons:
    - Emerging economies: 0.41
    - Advanced economies: 0.37
  - Information rigidities larger around the turn of the year (horizons between 13 and 10); interaction terms positive and statistically significant for horizon 10 (emerging economies) and horizon 13 (advanced economies).
- Aggregate degree of informational rigidity:
  - Without conditioning on horizon, degree of informational rigidity estimated on average forecast revisions equals 0.5 for both advanced and emerging economies.
  - Interpretation in sticky information framework: with quarterly revisions, forecasters update about every six months on average.
  - Interpretation in imperfect information framework: implies a weight of about 0.5 assigned to past forecasts in constructing current forecasts.
  - Comparison: Coibion and Gorodnichenko estimate a weight of 0.14 for the United States.

### Empirical findings — Individual forecasts
- Information stickiness (frequency of updating):
  - Share of forecasters who updated at least once in the three months prior to a given forecast horizon ranges between 0.8 and 0.9 over horizons.
  - Average fractions for advanced economies tend to be higher than for emerging economies.
  - Slight tendency for higher update shares as horizon shrinks; hump around the turn of the year (about h=13).
  - Comparison with aggregate-implied update probabilities:
    - Coefficients on lagged revisions estimated using average forecast data yield an implied probability to update a forecast in a given quarter of only about 0.06 to 0.75 (depending on horizon).
    - Individual data indicate shares of 0.8-0.9, implying a higher frequency of updating than suggested by aggregate regressions.
  - Interpretation: high share updating consistent with imperfect information theory; publication rounding conventions can explain observed share near 0.8-0.9 (Andrade and Le Bihan, 2010).
- Forecast smoothing (persistence in individual forecasts):
  - Regression results show strong evidence of forecast smoothing; coefficient on lagged revision positive and statistically significant across specifications.
  - Degree of rigidity smaller in individual forecasts than in consensus forecasts.
  - Ratio of coefficients on past revisions (individual to average) indicates persistence in individual revisions is about halved relative to consensus forecasts—suggesting averaging induces additional stickiness.
  - Cross-country and group differences:
    - OLS example: coefficient on lagged revisions higher for emerging economies than advanced economies (0.23 versus 0.13).
    - Consistent with stronger smoothing in emerging economies, possibly due to data lags and weaker statistical quality.
  - Horizon effects:
    - Non-monotonic horizon effects; some positive interaction terms at medium horizons (e.g., horizons 7, 10, 13), especially for advanced economies; largest at h=13 under some standard errors.
    - Using Driscoll-Kraay standard errors often weakens significance of horizon interaction effects.
  - Cross-country distribution:
    - Estimated persistence parameters vary substantially across countries.
    - Mean persistence lower for individual data than for consensus data.
    - Mean persistence higher for emerging economies than for advanced economies.

### Empirical findings — Nonlinearity
- Threshold model results:
  - Forecast smoothing is less pronounced in the tails of the distribution of individual forecast revisions than in the main body.
  - Coefficients on past revisions in the 90th and 10th percentiles have negative signs and are statistically significant for both country groups.
  - For emerging economies the effect in the 10th percentile is much larger than in the 90th percentile:
    - Interpretation: after large downward revisions forecast smoothing declines more than after large positive revisions; positive news shocks are fully incorporated more slowly than negative news in emerging economies.
  - No significant nonlinear effects based on relative position of forecasts in the distribution of all available forecasts.
  - Aggregate-level finding: degree of stickiness significantly lower after large downward revisions of the average forecast—updating more synchronized during downturns.
  - Robustness: results broadly robust when using ABB standard errors.

### Selected quantitative summaries (from tables and figures)
- Sample and counts:
  - Number of target years: 23 23 23
  - Number of countries: 36 14 22
  - Number of individual forecast observations: 188 639 104 894 83 745
  - Average number of forecasts per country per target year: 15.5 17.2 13.7
- Average forecast:
  - Mean: 3.2 2.1 4.6
  - Median: 3.0 2.4 4.8
- Average forecast errors:
  - Mean: 0.0 -0.1 0.1
  - Median: 0.2 0.1 0.4
- Variance of Deviation from Consensus (h=18 h=12 h=6 h=1)
  - Full sample: 0.29 0.26 0.19 0.10
  - Advanced economies: 0.24 0.22 0.15 0.07
  - Emerging economies: 0.36 0.32 0.28 0.15
- Mean Absolute Deviation from Consensus (h=18 h=12 h=6 h=1)
  - Full sample: 0.44 0.47 0.27 0.09
  - Advanced economies: 0.21 0.16 0.09 0.03
  - Emerging economies: 0.77 0.88 0.52 0.17
- Information rigidity coefficients (OLS with Driscoll-Kraay Robust Errors):
  - Average Forecasts (Past revision)
    - Full Sample: 0.404*** (t-statistic 6.7)
    - Advanced Economies: 0.370*** (t-statistic 6.1)
    - Emerging Economies: 0.409*** (t-statistic 5.6)
  - Individual Forecasts (Past revision)
    - Full Sample: 0.203*** (t-statistic 6.3)
    - Advanced Economies: 0.127** (t-statistic 2.7)
    - Emerging Economies: 0.223*** (t-statistic 6.1)
  - Ratio of coefficients (individual/average)
    - First estimator set: 0.50 0.32 0.55
    - Second estimator set (Arellano-Bond GMM): 0.58 0.41 0.61
  - Note: reported numbers of observations in regressions shown as 34081698 1710355782 105414524 in source.
- Nonlinear (Table 4, individual forecasts, OLS with Driscoll-Kraay):
  - Past revision: Full Sample 0.628***; Advanced Economies 0.419***; Emerging Economies 0.721***
  - Past revision (90th percentile): Full Sample 0.060; Advanced -0.050; Emerging 0.103
  - Past revision (10th percentile): Full Sample -0.274**; Advanced -0.105; Emerging -0.397***
  - Number of observations reported: 293315091424359962133614660 (as shown in source).

### Main conclusions and implications
- Consensus-based estimates suggest forecasts updated on average every 6 months, but individual-forecaster data indicate much higher updating frequency (fractions of forecasters updating in 3 months between 0.8 and 0.9).
- Evidence casts doubt on the sticky information model as the primary description of professional growth-forecast dynamics; sticky-information estimates based on averages overstate true individual inattentiveness.
- Findings are consistent with imperfect/noisy information models once publication rounding conventions and averaging effects are considered.
- Averaging across forecasters induces additional stickiness; individual-level smoothing is substantially smaller than consensus-level smoothing.
- Information rigidities and forecast uncertainty are more pronounced in emerging economies than advanced economies.
- Nonlinearities matter: smoothing weaker in distribution tails and stickiness lower after large downward aggregate revisions (greater synchronization during downturns).

### Suggested directions for future research (from source)
- Investigate herding behavior and its interaction with forecast smoothing.
- Explore implications of uncertainty for the dynamics of macroeconomic forecasting.
- Study the evolution of forecast rigidities over the business cycle.

*Source: IMF Working Paper WP/13/56 — Information Rigidities in Economic Growth Forecasts: Evidence from a Large International Panel (Jonas Dovern, Ulrich Fritsche, Prakash Loungani, Natalia Tamirisa), February 2013.*

### Section 1

### Information Rigidities in Economic Growth Forecasts: Evidence from a Large International Panel (Section 1)

### Abstract — main findings
- Sample: individual forecasts from 30 advanced and emerging economies during 1989–2010.
- Main empirical findings:
  - Evidence does not support the validity of the sticky information model (Mankiw and Reis, 2002) for describing the dynamics of professional growth forecasts.
  - Empirical evidence is more in line with implications of ―noisy‖ information models (Woodford, 2002; Sims, 2003).
  - Information rigidities are more pronounced in emerging economies than advanced economies.
  - Nonlinearities exist in forecast smoothing: smoothing is less pronounced in the tails of the distribution of individual forecast revisions than in the central part.

### Introduction — context, hypotheses, and motivations
- Role of expectations and forecasts: Forecast efficiency as the natural analog of rational expectations; strong-form efficiency implies forecast errors orthogonal to all relevant information.
- Practical testing challenge: Forecasters’ information sets may not be publicly known; Nordhaus (1987) introduced weak efficiency — forecast errors orthogonal to information in the forecaster’s set of past forecasts.
- Nordhaus’s two tests of weak efficiency for fixed-event forecasts:
  - Forecast error should be independent of past forecast revisions.
  - Today’s forecast revision should be independent of past forecast revisions.
- Theoretical explanations for correlated forecast revisions (not mutually exclusive):
  - Sticky information (Mankiw and Reis, 2002): infrequent updating due to fixed costs of acquiring information.
  - Noisy/imperfect information (Woodford, 2002; Sims, 2003): continual updating but receiving noisy signals about the state of the economy.
  - Behavioral explanations (Tversky and Kahneman, 1981): slow incorporation of surprises.
  - Non-standard or asymmetric loss functions: e.g., reluctance to produce a “jumpy” forecast or asymmetric costs for over/under forecasting.
- Value added of the paper:
  - Use of individual-forecaster data from 30 countries avoids aggregation bias and allows direct testing at the individual level.
  - Broad country coverage permits comparison of forecast smoothing between advanced and emerging economies.
  - Finding: degree of smoothing estimated from individual data is far lower than estimates from average (consensus) forecasts.
  - Evidence favors noisy information models over sticky information models for individual professional growth forecasts.
  - Nonlinearities: large negative and positive revisions (tails) are less correlated with subsequent revisions than average revisions.

### Methodology for testing forecast smoothing
- General approach:
  - Use sequences of forecasts for the same event (annual real GDP growth) and test whether the sequence follows a martingale under full-information rational expectations.
  - Estimate regression of contemporaneous revision on past forecast revisions; equivalently regress forecast errors on past revisions (Nordhaus 1987).
- Average forecasts specification:
  - Equation (1): contemporaneous revision of the average forecast regressed on past forecast revisions.
  - Revisions computed over k* months, with k*<=k; paper assumes k*=k=3 throughout.
  - Under sticky information (Reis, 2006) the average forecast is a weighted average of the lagged average forecast and current rational expectation; the regression coefficient maps to degree of information rigidity.
  - Under imperfect/noisy information (Woodford, 2002; Sims, 2003; Coibion and Gorodnichenko, 2012) the parameter from the same regression also reflects informational rigidity.
  - To study horizon dependence, interaction terms between forecast horizons and lagged revisions are added (equation (5)); coefficients on interaction terms are expected to be positive and rising with forecast horizon.
  - Estimation: fixed-effect panel OLS with Driscoll and Kraay (1998) standard error corrections for complex cross-sectional and temporal correlations. Nickel bias noted but likely modest since bias is order 1/T.
  - Robustness: Arellano-Bond/Arellano-Bover (ABB) GMM estimator as a check; caveat that under the null of full information current and past revisions are uncorrelated making instruments weak.
- Individual forecasts specification:
  - Individual-forecast analogue is equation (6): revision of an individual forecaster j for country i, target year t, horizon h regressed on past individual revisions.
  - Revisions computed over k*=k=3 months.
  - Interpretation: autocorrelation coefficient λ is a general measure of forecast smoothing reflecting behavioral features or deviations from efficiency but must not be directly mapped to theoretical model parameters.
  - Under sticky information, at the individual-agent level there should be no correlation between current and last period’s revisions because agents either do not update or move directly to rational expectations when they update.
  - Under imperfect information, OLS estimator may be biased because the regression error can be correlated with current revision; IV approaches are difficult due to lack of good instruments.
  - Non-parametric measure of sticky-information updating: fraction of individuals who revise their forecasts at least once during the 3 months prior to a given point in time (Andrade and Le Bihan, 2010); comparable to coefficients from equation (5) using k*=3 months.
- Nonlinearity:
  - Use threshold models to examine differences in forecast smoothing in the tails versus the central portion of the distribution of revisions.
  - Rationale: large revisions may reflect full incorporation of new information into the next forecast rather than gradual smoothing; position within distribution of available forecasts may influence smoothing behavior.

### Data and descriptive statistics
- Data source: Consensus Economics Inc. cross-country survey data set for annual GDP growth forecasts.
- Coverage and frequency:
  - Survey started in October 1989.
  - Monthly frequency.
  - Includes public and private institutions (mostly banks and research institutes).
  - Survey process identical across countries: forecasters send responses during the first two weeks of each month; data published in the middle of each month.
  - Implication: panelists are likely aware of competitors’ forecasts from one month prior when making their forecasts.

*Source: IMF Working Paper WP/13/56 — Information Rigidities in Economic Growth Forecasts: Evidence from a Large International Panel (Jonas Dovern, Ulrich Fritsche, Prakash Loungani, Natalia Tamirisa), February 2013.*

### Section 2

### _wp1356 - Section 2

### Data and sample
- The data set comprises 188,639 individual forecasts from 30 different countries, of which 104,894 are from 14 advanced economies.
- Forecasts cover target years between 1989 and 2011.
- For each target year the data set contains a sequence of 24 forecasts of each panelist made between January of the year before the target year and December of the target year.
- On average, the data set includes nearly 16 individual forecasts per period for each country.
- Inclusion rules and cleaning:
  - Only countries for which Consensus Economics Inc. reports individual forecasts are included.
  - Only forecasters that reported their growth forecasts at least 10 times were included.
  - Forecaster name variants were concatenated when they correspond to the same forecaster (examples provided in the source text).
  - Continuity of forecast series was maintained through mergers/acquisitions where it was evident which entity continued producing forecasts.

### Descriptive patterns of forecast errors, revisions, and dispersion
- RMSFE and horizons:
  - Average root mean squared forecast error (RMSFE) declines with the forecast horizon (errors smaller as h approaches 1).
  - RMSFEs for emerging economies are, on average, more than twice as high as for advanced economies for large forecast horizons and still almost 75 percent higher at the end of the target year.
- Forecast revision patterns:
  - For advanced economies, revisions are larger around the turn of the year and the average size of revisions does not vary much with the forecast horizon.
  - For emerging economies, revisions are much smaller for very long forecast horizons and much higher during the target year (h<=12).
  - At h=1 the average revision in emerging economies is about twice as large as for advanced economies.
  - Interpretation: higher uncertainty about actual data in emerging economies just before the end of the forecasting horizon, possibly owing to lags in statistical data collection and poor quality of initial data releases.
- Distributional features:
  - Forecast revisions are frequently small (high density around zero); distributions are unimodal bell-shaped at all horizons.
  - Distributions for emerging economies are more flattened out; forecasts revised less frequently but with larger revisions.
  - Skewness differs by group:
    - Advanced economies: significantly negative skew for most horizons.
    - Emerging economies: distributions are symmetric for large horizons and positively skewed for smaller horizons (negative revisions more frequent but smaller than upward revisions).
- Forecast clustering and dispersion:
  - Forecasts become more clustered as the forecast horizon shrinks.
  - Deviations from the average follow a unimodal distribution with most forecasts close to the average.
  - Advanced economies: very few deviations larger than half a percentage point.
  - Emerging economies: considerably larger dispersion and considerable disagreement across forecasters even at h=1.

### Empirical findings — Average forecasts
- Estimation setup:
  - Revisions calculated over horizon k=3 (quarterly frequency of updating forecasts).
  - Horizons used for estimation: h=1, 4, 7, 10, 13 and 16 (results robust to alternative horizon choices).
- Evidence of information rigidities:
  - Strong positive correlation between current forecast revision and its first lag for all country groups and estimation methods; coefficients on lagged revisions are highly statistically significant.
  - Coefficient on lagged revisions at very short horizons:
    - Emerging economies: 0.41
    - Advanced economies: 0.37
  - Information rigidities larger around the turn of the year (horizons between 13 and 10); interaction terms positive and statistically significant for horizon 10 (emerging economies) and horizon 13 (advanced economies).
- Aggregate degree of informational rigidity:
  - Without conditioning on horizon, the degree of informational rigidity estimated on average forecast revisions equals 0.5 for both advanced and emerging economies.
  - Interpretation in sticky information framework: with quarterly revisions, forecasters update about every six months on average.
  - Interpretation in imperfect information framework: implies a weight of about 0.5 assigned to past forecasts in constructing current forecasts.
  - Comparison: Coibion and Gorodnichenko estimate a weight of 0.14 for the United States.

### Empirical findings — Individual forecasts
- Information stickiness (frequency of updating):
  - The share of forecasters who updated at least once in the three months prior to a given forecast horizon ranges between 0.8 and 0.9 over horizons.
  - Average fractions for advanced economies tend to be higher than for emerging economies.
  - Slight tendency for higher update shares as horizon shrinks; hump around the turn of the year (about h=13).
  - Comparison with aggregate-based implied update probabilities:
    - Coefficients on lagged revisions estimated using average forecast data yield an implied probability to update a forecast in a given quarter of only about 0.06 to 0.75 (depending on horizon).
    - Individual data indicate shares of 0.8-0.9, implying a higher frequency of updating than suggested by aggregate regressions.
  - Interpretation: high share updating consistent with imperfect information theory; rounding conventions in published forecasts can explain observed share near 0.8-0.9 (Andrade and Le Bihan, 2010).
- Forecast smoothing (persistence in individual forecasts):
  - Regression results on individual forecasts show strong evidence of forecast smoothing; coefficient on lagged revision positive and statistically significant across specifications.
  - Degree of rigidity smaller in individual forecasts than in consensus (average) forecasts.
  - Ratio of coefficients on past revisions (individual to average) indicates persistence in individual revisions is about halved relative to consensus forecasts—suggesting averaging induces additional stickiness.
  - Cross-country and group differences:
    - Coefficients on lagged revisions higher for emerging economies than advanced economies (example OLS: 0.23 versus 0.13).
    - Consistent with forecast smoothing being stronger in emerging economies, possibly due to data lags and weaker statistical quality.
  - Horizon effects:
    - Non-monotonic horizon effects; some positive interaction terms at medium horizons (e.g., horizons 7, 10, 13) especially for advanced economies, largest at h=13 under some standard errors.
    - Using more conservative Driscoll-Kraay standard errors often weakens statistical significance of horizon interaction effects.
  - Cross-country distribution:
    - Distribution of estimated persistence parameters across countries shows substantial variation.
    - Mean persistence lower for individual data than for consensus data.
    - Mean persistence higher for emerging economies than for advanced economies.

### Empirical findings — Nonlinearity
- Threshold model results:
  - Forecast smoothing is less pronounced in the tails of the distribution of individual forecast revisions than in the main body.
  - Coefficients on past revisions in the 90th and 10th percentiles have negative signs and are statistically significant for both country groups.
  - For emerging economies the effect in the 10th percentile is much larger than in the 90th percentile:
    - Interpretation: after large downward revisions forecast smoothing declines more than after large positive revisions; positive news shocks are fully incorporated more slowly than negative news in emerging economies.
  - No significant nonlinear effects based on relative position of forecasts in the distribution of all available forecasts (no additional effect for forecasts far in tails or close to average).
  - Aggregate-level finding:
    - Degree of stickiness significantly lower after large downward revisions of the average forecast.
    - Interpretation: updating of forecasts is more synchronized during downturns than in normal times, reducing persistence caused by staggered processing of news across forecasters.
  - Robustness: results broadly robust when using ABB standard errors.

### Conclusion — key takeaways
- The paper analyzes a large panel of individual forecasters across 30 advanced and emerging market economies for 1989–2011; the data set is larger and more country-diverse than previous panels of individual forecasts.
- Main contributions and findings:
  - Confirmation of persistence in average forecast revisions consistent with existing literature documenting smoothing/rigidity.
  - Evidence against the sticky information model as the main description of professional growth forecast dynamics:
    - Average-forecast-based estimates overstate true forecasters' inattentiveness.
  - Evidence consistent with the imperfect (noisy) information theory:
    - High shares of individual updating (0.8–0.9 in three-month window) and observed patterns align with imperfect information predictions once publication rounding conventions are accounted for.
  - Forecast smoothing exists at the individual level but is substantially reduced relative to consensus averages; averaging induces additional stickiness.
  - Stronger smoothing and higher uncertainty in emerging economies relative to advanced economies (higher RMSFEs, larger revisions near h=1, higher individual persistence coefficients).
  - Nonlinearities: forecast smoothing weaker in distribution tails and stickiness lower after large downward aggregate revisions (greater synchronization in downturns).

*Source: _wp1356 - Section 2*

### Section 3

### _wp1356 - Section 3

### Main findings on information rigidities and forecast updating
- When consensus forecasts are used, estimates suggest that forecasts are updated on average every 6 months.
- Analysis of fractions of forecasters who update their forecasts points to a higher frequency of updating than implied by consensus-based measures.
- The evidence based on fractions suggests a small role of sticky information in explaining the overall degree of information rigidity in economic forecasts.
- The predictability of individual forecast revisions casts doubt on the validity of the sticky information theory, which implies that revisions are unpredictably (though infrequent).

### Cross-country differences
- Information rigidities are more pronounced in emerging economies than advanced economies.
- Possible explanations noted in the source:
  - Greater uncertainty about cyclical positions and transmission of shocks in emerging economies.
  - Weaker quality of economic and financial statistics in emerging economies.
  - Fewer resources devoted to monitoring emerging economies and producing up-to-date forecasts for them.

### Nonlinearities in forecast smoothing
- Evidence of nonlinearities in forecast smoothing is found.
- Forecast smoothing is less pronounced in the tails of the distribution of individual forecast revisions than in the main body of the distribution.

### Suggested directions for future research
- Herding behavior and its interaction with forecast smoothing deserve closer examination.
- Implications of uncertainty for the dynamics of macroeconomic forecasting warrant further exploration.
- The evolution of forecast rigidities over the business cycle is identified as an important topic for future work.

### Key figures and notes (selected)
- Figures refer to forecast horizon denoted by h (examples shown for h=1, h=6, h=12, h=18).
- Figure 6 note: "Fractions show how many forecasters on average revised their forecasts at least once three months prior to the forecast horizon indicated on the horizontal axis."

### Selected descriptive statistics (Table 1)
- Number of target years: 23 23 23
- Number of countries: 36 14 22
- Number of individual forecast observations: 188 639 104 894 83 745
- Average number of forecasts per country per target year: 15.5 17.2 13.7
- Average forecast
  - Mean: 3.2 2.1 4.6
  - Median: 3.0 2.4 4.8
- Average forecast errors
  - Mean: 0.0 -0.1 0.1
  - Median: 0.2 0.1 0.4

### Revisions and deviations from the average forecast (Table 2)
- Horizon (in months) — Variance of Deviation from Consensus (h=18 h=12 h=6 h=1)
  - Full sample: 0.29 0.26 0.19 0.10
  - Advanced economies: 0.24 0.22 0.15 0.07
  - Emerging economies: 0.36 0.32 0.28 0.15
- Mean Absolute Deviation from Consensus (h=18 h=12 h=6 h=1)
  - Full sample: 0.44 0.47 0.27 0.09
  - Advanced economies: 0.21 0.16 0.09 0.03
  - Emerging economies: 0.77 0.88 0.52 0.17

### Information rigidity and forecast smoothing (Table 3, selected coefficients)
- Ordinary Least Squares with Driscoll-Kraay Robust Errors — Average Forecasts (Past revision)
  - Full Sample: 0.404***
  - Advanced Economies: 0.370***
  - Emerging Economies: 0.409***
  - Numbers below the coefficients are t-statistics (for Full Sample: 6.7; Advanced: 6.1; Emerging: 5.6)
- Ordinary Least Squares with Driscoll-Kraay Robust Errors — Individual Forecasts (Past revision)
  - Full Sample: 0.203***
  - Advanced Economies: 0.127**
  - Emerging Economies: 0.223***
  - Numbers below the coefficients are t-statistics (for Full Sample: 6.3; Advanced: 2.7; Emerging: 6.1)
- Ratio of coefficients on past revisions (defined as the quotient of the baseline rigidity parameter for individual revisions and the equivalent for the revisions of the average forecast)
  - First estimator set: 0.50 0.32 0.55
  - Second estimator set (Arellano-Bond GMM): 0.58 0.41 0.61
- Number of observations reported in regressions: 34081698 1710355782 105414524 (as shown in source)

Note: Numbers below the coefficients are t-statistics. Asterisks indicate significance: *** 1 percent, ** 5 percent, * 10 percent. Regressions include fixed effects; constants are identified by restricting the sum of fixed effects to equal 0. Results skip December 2008 forecasts for 2009 growth to avoid heavy influence of the Great Recession adjustments (including these observations would increase the effect of "Past revision*Horizon 13" to about 0.84, implying a total rigidity parameter above 1, which the authors note is not consistent with any theory of informational rigidities).

### Nonlinear effects in forecast rigidities (Table 4, selected coefficients)
- Individual Forecasts (OLS with Driscoll-Kraay standard errors)
  - Past revision: Full Sample 0.628***; Advanced Economies 0.419***; Emerging Economies 0.721***
  - Past revision (90th percentile): Full Sample 0.060; Advanced -0.050; Emerging 0.103
  - Past revision (10th percentile): Full Sample -0.274**; Advanced -0.105; Emerging -0.397***
  - Number of observations: 293315091424359962133614660

Source: Authors' estimates.

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1356.pdf_