## wp17219

## Source details

**Canonical URL:** [wp17219](https://www.imf.org/-/media/files/publications/wp/2017/wp17219.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17219.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17219.pdf.json)

---

### Introduction: motivation, approach, and dataset
- Background and motivation:
  - Heightened economic uncertainty contributed to the 2007-09 recession and shaped the speed of the subsequent recovery; eight years after the end of the recession there is still no sign of a complete global recovery.
  - Advanced economies face uncertainty about effects of monetary policy normalization; emerging market economies face uncertainty about growth challenges.
- Core contribution:
  - Survey-based uncertainty measure built from professional forecasters with a two-component decomposition: common uncertainty (perceived variability of future aggregate shocks) and idiosyncratic uncertainty (forecaster disagreement).
  - Three-layer aggregation:
    1. Variable-specific uncertainty for eight nominal and real indicators.
    2. Country-specific uncertainty as weighted average of standardized variable-specific components (equal weights ݓ_j = 1/8 in the paper).
    3. Global uncertainty as PPP-weighted average of country-specific uncertainties across 45 countries (covering more than 90 percent of the world economy); market GDP weights yield very similar results.
- Dataset and forecast processing:
  - Coverage: 45 advanced and emerging market economies.
  - Eight variables per country: GDP, consumption, investment, industrial production, inflation, unemployment rate, short-term interest rate, long-term interest rate.
  - Monthly forecasts from panels of 10-30 professional forecasters per country; forecast source: Consensus Forecasts (Consensus Economics Inc.), monthly surveys in the first week of each month.
  - Forecast transformations:
    - All variables except interest rates: fixed event forecasts (current and next calendar year) transformed into approximated 12-month-ahead fixed-horizon forecasts using a weighted average method.
    - Interest rates: twelve-month ahead forecasts used.
  - Actual series sources: Global Data Source of IMF, Haver Analytics, OECD Analytical databases, country statistical offices.
  - Sample/time coverage examples: variable-specific estimates sample between 1989M11-2014M7.

### Key methodological features and estimation choices
- Decomposition and measures:
  - Volatility of a “typical” forecaster decomposed into common volatility (empirical variance of consensus forecast) and idiosyncratic volatility (forecast disagreement).
  - Idiosyncratic uncertainty measured by interquartile range of survey forecasts (robust to outliers); median forecast errors used when estimating common uncertainty.
  - Variable-specific uncertainty defined as sum of min-max normalized common and idiosyncratic components (each scaled between 0 and 1, so sum bounded between 0 and 2).
- Estimation:
  - Common uncertainty estimated using a stochastic volatility model (preferred over GARCH because it permits shocks to the second moment independent of first-moment innovations).
  - Estimation implemented via Markov Chain Monte Carlo (MCMC) methods as in Kim et al. (1998).

### Main empirical summary findings (from introduction)
- All uncertainty measures are countercyclical.
- At variable, country, and global layers, combined uncertainty (common + idiosyncratic) is more countercyclical than either component alone.
- Shocks to the proposed uncertainty measures are associated with large and persistent drops in real activity at national and global levels.
- Decomposition evidence: common uncertainty shocks produce large and persistent responses in real activity; idiosyncratic uncertainty shocks have negligible contributions.

### Properties of Economic Uncertainty — variable-specific (VSU) and components
- General patterns:
  - VSU estimated for eight indicators is countercyclical for most economies.
  - For the United States, pairwise correlations across most VSU estimates are "quite high."
- Common versus idiosyncratic:
  - Pairwise correlations between VSU estimates (except long-term interest rate) are higher for the common component than for the idiosyncratic component (example: inflation–investment growth correlation is 0.27 for idiosyncratic uncertainty and 0.78 for common uncertainty).
  - Using only forecast disagreement yields many spikes; using only common component yields series that are too smooth; combined estimate better reflects whole-economy uncertainty.
  - For all countries, common uncertainty is less volatile and on average higher than idiosyncratic uncertainty; few peaks in common uncertainty and those peaks are usually around recessions.
  - Regional recessions can produce higher uncertainty peaks than global recessions for some countries (example: Indonesia and South Korea around the 1997 Asian financial crisis).

- Selected U.S. variable-specific correlations (Table 1.A — sample 1989M11-2014M7):
  - Output–Inflation: 0.57
  - Output–Consumption: 0.79
  - Output–Investment: 0.77
  - Output–Industrial production: 0.82
  - Output–Unemployment rate: 0.77
  - Output–Short-term interest rate: 0.43
  - Output–Long-term interest rate: 0.41
- Selected U.S. idiosyncratic correlations (Table 1.B):
  - Output–Inflation: 0.38
  - Output–Consumption: 0.60
  - Output–Investment: 0.54
  - Output–Industrial production: 0.56
  - Output–Unemployment rate: 0.49
  - Output–Short-term interest rate: 0.19
  - Output–Long-term interest rate: 0.35
- Selected U.S. common correlations (Table 1.C):
  - Output–Consumption: 0.83
  - Output–Investment: 0.75
  - Output–Industrial production: 0.87
  - Output–Unemployment rate: 0.71
  - Output–Short-term interest rate: 0.54
  - Output–Long-term interest rate: 0.20

### Country-specific uncertainty (CSU), explained variation, and cross-country patterns
- CSU is strongly countercyclical; almost all countries show CSU peaks around 2009 even if they did not experience a recession (examples: China and Australia).
- For some emerging markets earlier recessions produced higher uncertainty than the latest global recession (examples: Argentina peak around 2001-2002; Hong Kong around 1997-1998).
- Proportion of VSU explained by CSU (selected statistics from Table 2):
  - Entire sample: explanatory power of CSU for VSU almost the same during recessions (ܴଶൌ0.585) and expansions (ܴଶൌ0.576).
  - Advanced economies: higher during recessions (ܴଶൌ0.51) than expansions (ܴଶൌ0.46).
  - Heterogeneity by variable and country: many ܴଶ statistics are much lower than one, indicating substantial idiosyncratic variation across variables.
- Selected country R-squares (Table 2 — Full sample averages):
  - United States average: 0.606 (Output 0.822, Consumption 0.678, Investment 0.763, Industrial production 0.734, Unemployment rate 0.690, Inflation 0.506, Short-term interest rate 0.348, Long-term interest rate 0.303)
  - United Kingdom average: 0.617
  - France average: 0.483
  - Germany average: 0.455
  - Euro Zone average: 0.760 (Long-term interest rate N/A)

### Global uncertainty and relation to CSU
- Global uncertainty (PPP-weighted average across 46 economies) is strongly countercyclical and rises during global recessions of 1991 and 2009 (Kose and Terrones (2015) dating).
- Proportion of CSU variation explained by global uncertainty (selected Table 3 entries):
  - Lithuania: R2 0.925, β 1.351***
  - Latvia: R2 0.904, β 1.346***
  - Bulgaria: R2 0.886, β 1.396***
  - Hong Kong example: R2 0.435
  - Thailand example: R2 0.079
  - United States: R2 0.782, β 1.212***
  - Bottom coefficients: Indonesia R2 0.017, β 0.208***; Venezuela R2 0.001, β 0.039
- Global uncertainty amplifies CSU for almost half the sample where estimated coefficient > 1; amplification less evident for common component than idiosyncratic component.
- Sum of components (common + idiosyncratic) better reflects worldwide uncertainty than any single component.

### Comparison with other uncertainty measures (correlations and standardized behavior)
- Selected correlations with U.S. measures (Table 4):
  - Country-specific uncertainty — Jurado et al. (2015): 0.79
  - Country-specific uncertainty — Economic policy uncertainty (Baker et al., 2016): 0.18
  - Country-specific uncertainty — News-based policy uncertainty: 0.19
  - Country-specific uncertainty — Scotti (2016): 0.38
  - Country-specific uncertainty — VXO: 0.54
  - Common uncertainty — Idiosyncratic uncertainty: 0.54
  - Economic policy uncertainty — News-based policy uncertainty: 0.90
  - Jurado et al. (2015) correlations: 0.79 with country-specific, 0.75 with common, 0.59 with idiosyncratic
- Standardized-comparison behavior (Figure 3):
  - For almost all twelve countries compared, increases in the paper’s uncertainty measures are more persistent during recessions than alternative measures.
  - U.S. specifics:
    - All measures countercyclical.
    - Only this paper’s measure and the policy uncertainty measure exceed the 1.65 standard deviation line for all three recession periods; policy uncertainty exceeds it many more times after the latest recession.
    - Jurado et al. (2015) exceeds the line only once during the latest recession.
    - VXO exceeds the line six times, covering three recessions and three non-recession periods.
  - Country examples: United Kingdom, Canada, France, Germany, Japan, China, South Korea, Russia—paper’s measure tends to peak during recessionary episodes and remain low in expansions, while news-based indexes spike more in non-recessionary episodes.

### VAR evidence: uncertainty shocks and macroeconomic activity
- Motivation and channels:
  - Elevated macroeconomic uncertainty linked to slow recovery through demand-side "wait-and-see" effects (consumption, investment) and supply-side effects (credit provision, productivity).
- U.S. VAR (eight-variable model: log(S&P500), uncertainty, Federal funds rate, log(wages), log(CPI), hours, log(employment), log(industrial production)):
  - VXO or EPU shocks show overshooting: industrial production and employment increase above initial levels by middle of third year after a VXO shock.
  - Jurado et al. (2015) shock: employment and industrial production decline sharply and persist for five years.
  - Decomposition:
    - Idiosyncratic (forecast disagreement) shocks: less significant impacts on industrial production and employment.
    - Common uncertainty shocks: large and long-lived impacts peaking after two to three years.
    - Country-specific (combined) shocks: industrial production and employment responses not significant in first nine months; negative impact on industrial production starts around 10 months and remains persistent, keeping industrial production below initial level until middle of third year.
  - Interpretation: "wait-and-see" mechanism observed mainly in common component rather than idiosyncratic component.
- Global VAR (monthly, seven variables: stock prices, global uncertainty, short term interest rate, oil prices, food prices, unemployment rate, industrial production):
  - Industrial production responses (Figure 5 Panel A):
    - Global uncertainty shock: immediate decline that dissipates within a few months.
    - Idiosyncratic shock: short-lived impact with overshooting after six months.
    - Common shock: long-lived impact with peak after two years.
  - Unemployment responses (Figure 5 Panel B):
    - Global uncertainty shocks produce sizable and protracted increases in unemployment; significant increases dissipate almost after 30 months.
    - Idiosyncratic shocks: high initial response and overshooting after 30 months.
    - Common shocks: more persistent high unemployment rates peaking around 30 months.
  - Interpretation: employers may "wait-and-see" before laying off and then hire later during recovery.
- Joint common and idiosyncratic VAR:
  - Common uncertainty shocks: large and persistent impacts on industrial production across United States, France, Italy, Japan, United Kingdom.
  - Idiosyncratic uncertainty shocks: short-lived and negligible effects on industrial production in those countries.
  - Canada and Germany: both components show limited and short-lived effects.

### Conclusion: contributions, empirical findings, and implications
- Main contributions:
  - Decomposition of a typical forecaster’s uncertainty into common and idiosyncratic components.
  - New monthly index combining perceived variability of future aggregate shocks and forecaster disagreement using actual market analyst forecasts.
- Properties and empirical findings:
  - Country-specific measures have fewer peaks than other leading measures, but peaks are larger and more persistent when they occur.
  - Shocks to country-specific uncertainty associated with sizable and protracted declines in real activity.
  - Global uncertainty shocks have long-lived effects on industrial production and unemployment.
  - Common uncertainty shocks account for a large fraction of fluctuations in economic activity at business cycle frequencies; idiosyncratic shocks play a small role.
- Implications and future research directions:
  - Persistence and size of uncertainty peaks are relevant for theories where uncertainty drives recessions.
  - Future research should quantify economic effects of different types of uncertainty and analyze transmission of uncertainty shocks across countries.

*Source: wp17219*

### 1. Introduction

### 1. Introduction

### Background: uncertainty and the post-2007-09 recovery
- Heightened economic uncertainty contributed to the 2007-09 recession and shaped the speed of the subsequent recovery.
- Eight years after the end of the recession, there is still no sign of a complete global recovery.
- Advanced economies are uncertain about the effects of monetary policy normalization and emerging market economies are uncertain about the growth challenges ahead.
- Economists face challenges in understanding origins of economic uncertainty and analyzing causal impacts on the real economy.

### Existing proxies for economic uncertainty and their limitations
- Common proxies and key limitations:
  - Implied or realized stock market volatility (e.g., VIX): may reflect leverage or financial stress and not Main Street uncertainty; see Bloom (2009) and Bekaert et al. (2013).
  - Jurado et al. (2015): measure based on common variation in uncertainty across hundreds of economic series; reflects uncertainty around objective statistical forecasts and focuses on common, not idiosyncratic, uncertainty.
  - News-based measures (e.g., Baker et al. (2016)): depend on reporters/editors mentioning policy-related uncertainty and may miss events not written about.
  - Cross-sectional disagreement (dispersion in directional or point forecasts; e.g., Bachmann et al. (2013)): captures interpersonal dispersion but omits intra-personal volatility of aggregate shocks (Lahiri and Sheng (2010)).

### Our contribution: a comprehensive survey-based uncertainty measure
- Approach highlights:
  - Use surveys of professional forecasters to capture perceived uncertainty (similar to Jo and Sekkel (2015), Rossi and Sekhposyan (2015), Scotti (2016)), but with a two-component decomposition: common uncertainty and idiosyncratic uncertainty.
  - Decomposition parallels Campbell et al. (2001) market/firm volatility approach: common = perceived variability of future aggregate shocks; idiosyncratic = disagreement among professional forecasters.
  - Three-layer aggregation:
    1. Variable-specific uncertainty for eight nominal and real economic indicators.
    2. Country-specific uncertainty as weighted average of standardized variable-specific components.
    3. Global uncertainty index constructed from a large set of countries (covering more than 90 percent of the world economy).
- Coverage:
  - Dataset of 45 advanced and emerging market economies.
  - Eight variables per country: GDP, consumption, investment, industrial production, inflation, unemployment rate, short-term interest rate, long-term interest rate.
  - Monthly forecasts from panels of 10-30 professional forecasters for each country.

### Key methodological features
- Decomposition logic:
  - Volatility of a “typical” forecaster is decomposed into common volatility (empirical variance of consensus forecast) and idiosyncratic volatility (forecast disagreement).
  - Forecast disagreement is used as the measure of idiosyncratic uncertainty.
- Estimation choices:
  - Common uncertainty estimated using a stochastic volatility model (preferred over GARCH because it permits shocks to the second moment that are independent of first-moment innovations).
  - Estimation implemented via Markov Chain Monte Carlo (MCMC) methods as in Kim et al. (1998).
  - Median forecast errors used (instead of means) to reduce impact of outliers when estimating common uncertainty.
  - Forecast disagreement measured by the interquartile range of survey forecasts (rather than standard deviation) to mitigate outlier effects.
  - Variable-specific uncertainty is the sum of standardized common and idiosyncratic components; both components are min-max normalized and thus scaled between 0 and 1, so the sum is bounded between 0 and 2.
- Aggregation:
  - Country-specific uncertainty: weighted average of eight variable-specific uncertainty estimates; equal weights ݓ_j = 1/8 are used in the paper, with principal component alternatives explored.
  - Global uncertainty: PPP-weighted average of country-specific uncertainties across 45 countries (covering more than 90 percent of the world economy). Market GDP weights yield very similar results.

### Data highlights and forecast processing
- Forecast source: Consensus Forecasts (Consensus Economics Inc.): monthly surveys conducted in the first week of each month.
- Forecast structure:
  - Forecasts for all variables except interest rates are fixed event forecasts (current and next calendar year); transformed into fixed-horizon forecasts for consistency using a weighted average method to approximate 12-month-ahead fixed-horizon forecasts.
  - For interest rates, twelve-month ahead forecasts are used to match horizons.
- Actual values:
  - Monthly series available for industrial production, inflation, unemployment, short-term and long-term interest rates.
  - Real GDP, consumption, and investment are quarterly; quarterly value assigned to each month in the quarter.
  - Main sources: Global Data Source of IMF, Haver Analytics, OECD Analytical databases, and country statistical offices.
- Notes on forecast properties:
  - Mincer-Zarnowitz regressions indicate some forecasts are biased and inefficient in incorporating new information.
  - Despite inefficiencies, survey forecasts are used because surveys often provide more accurate forecasts than models and reflect market participants’ perceptions—crucial for capturing experienced uncertainty.

### Main empirical findings (summary)
- All uncertainty measures are countercyclical.
- At all layers (variable, country, global), combined uncertainty is more countercyclical than either its common or idiosyncratic component alone.
- Comparison with alternative country-specific measures:
  - The proposed measures exhibit fewer peaks (all around recessions) and show persistent and heightened uncertainty during recession episodes.
- Macroeconomic impacts:
  - Shocks to the proposed uncertainty measures are associated with large and persistent drops in real activity at both national and global levels.
  - Decomposition results: common uncertainty shocks produce large and persistent responses in real activity; contributions of idiosyncratic uncertainty shocks are negligible.

### Paper structure (as provided)
- Section 2: Methodology on measuring uncertainty.
- Section 3: Data used in this paper.
- Section 4: Properties of economic uncertainty measures.
- Section 5: Dynamic relationship between uncertainty and economic activity.
- Section 6: Conclusion.
- Online appendix: detailed dataset information, alternative country-specific uncertainty measures using principal component analysis, and regional measures of uncertainty.

*Source: wp17219 - 1. Introduction*

### 4. Properties of Economic Uncertainty

### 4. Properties of Economic Uncertainty

### Overview
- Variable-specific uncertainty (VSU) is estimated for eight indicators.
- For most economies in the sample, the VSU is countercyclical for all series.
- Some VSU estimates are highly correlated across variables; for the United States, pairwise correlations are described as "quite high" for most VSU estimates.

### Variable-specific uncertainty and its components
- Common versus idiosyncratic components:
  - Pairwise correlations between all VSU estimates except long-term interest rate are higher for the common than for the idiosyncratic component.
  - Example: correlation between inflation and investment growth is 0.27 for idiosyncratic uncertainty, but 0.78 for common uncertainty.
- Trade-offs in measurement:
  - Using only forecast disagreement leads to too many uncertainty spikes due to idiosyncratic shocks.
  - Using only the common component yields series that are too smooth.
  - A combined estimate (common + idiosyncratic) better reflects uncertainty in the entire economy than any individual component.
- Cross-country patterns:
  - For all countries, common uncertainty is less volatile and on average higher than idiosyncratic uncertainty.
  - Very few peaks in common uncertainty; those peaks are usually around recessions.
  - Example: in the United States, uncertainty for output, consumption, investment, unemployment rate and short-term interest rates increases during all three recession periods covered in the sample of 1989-2014.
  - Regional recessions can produce higher uncertainty peaks than global recessions in some countries (e.g., Indonesia and South Korea around the 1997 Asian financial crisis).

### Country-specific uncertainty (CSU) and explained variation
- CSU is strongly countercyclical; almost all countries show CSU peaks around 2009 even if the country did not experience a recession (examples: China and Australia).
- For some emerging markets, earlier recessions produced higher uncertainty than the latest global recession (examples: Argentina peak around 2001-2002; Hong Kong around 1997-1998).
- Proportion of VSU explained by CSU (Table 2 highlights):
  - For the entire sample: explanatory power of CSU for VSU is almost the same during recessions (ܴଶൌ0.585) and expansions (ܴଶൌ0.576).
  - For advanced economies: higher during recessions (ܴଶൌ0.51) than expansions (ܴଶൌ0.46).
  - Shorter time coverage of emerging markets complicates phase comparisons.
  - Heterogeneity in which variable CSU explains most:
    - Eight out of fifteen advanced economies: CSU explains output growth uncertainty the most.
    - Examples for emerging markets where ܴଶ is highest for specific variables:
      - Industrial production uncertainty: China, Poland, Czech Republic.
      - Consumption uncertainty: Argentina, Brazil, Colombia, Peru, South Korea, Philippines, Lithuania, Romania.
      - Investment uncertainty: Bulgaria, Croatia, Russia.
  - Many ܴଶ statistics are much lower than one, indicating substantial idiosyncratic variation across variables.

### Global uncertainty and its relation to CSU
- Global uncertainty constructed as weighted average of country-specific uncertainties (Figure 2).
- Global uncertainty is strongly countercyclical and rises during the global recessions of 1991 and 2009 (Kose and Terrones (2015) dating).
- Proportion of CSU variation explained by global uncertainty (Table 3 examples):
  - Hong Kong: ܴଶ = 0.435.
  - Thailand: ܴଶ = 0.079.
  - Lithuania: ܴଶൌ0.925.
  - Latvia: 0.904.
  - Bulgaria: 0.886.
- Global uncertainty amplifies CSU for almost half the sample where estimated coefficient > 1; amplification is less evident for the common component than the idiosyncratic component.
- The sum of both components (common + idiosyncratic) better reflects worldwide uncertainty than any individual component.

### Comparison with other uncertainty measures
- Correlations (Table 4, U.S.):
  - Our uncertainty measure correlation with Jurado et al. (2015): 0.79.
  - Correlation with Baker et al. (2016) economic policy uncertainty (EPU): 0.18.
  - Jurado et al. (2015) correlation with common uncertainty: 0.75; with idiosyncratic uncertainty: 0.59.
  - Our U.S. measure weakly correlated (0.38) with Scotti (2016) (note: Scotti starts May 2003; Table 4 covers November 1989 - July 2014 for other measures).
- Standardized-comparison behavior (Figure 3):
  - For almost all twelve countries compared, increases in our uncertainty measures are more persistent during recessions than alternative measures.
  - U.S. findings:
    - All uncertainty measures are countercyclical.
    - Only our measure and the policy uncertainty measure exceed the 1.65 standard deviation line for all three recession periods; policy uncertainty exceeds this line many more times after the latest recession.
    - Jurado et al. (2015) exceeds the line only once during the latest recession.
    - VXO exceeds the line six times, covering three recessions and three non-recession periods.
  - Country examples:
    - United Kingdom: policy uncertainty exceeds 1.65 line five times; our measure exceeds it in two recessions out of three.
    - Canada, France, Germany: our measure usually outperforms policy uncertainty in capturing recessionary episodes.
    - Japan: Scotti (2016) does not exceed 1.65 line around 2007-09 recession, whereas our measure does.
  - News-based (Baker et al. (2016)) indexes often spike during non-recessionary episodes; our measures peak during most recessionary episodes and remain low during expansions.

### Uncertainty and macroeconomic activity (VAR evidence)
- Motivation:
  - Elevated macroeconomic uncertainty during and after the global recession is a pronounced factor behind slow recovery; channels include demand-side "wait-and-see" effects (consumption, investment) and supply-side effects (credit provision, productivity).
- U.S. VAR specification (eight-variable model):
  - Variables: log(S&P500 index), uncertainty measure, Federal funds rate, log(wages), log(consumer price index), hours, log(employment), log(industrial production).
- U.S. VAR findings (Figure 4, impulse responses):
  - VXO or economic policy uncertainty proxies show evidence of overshooting: industrial production and employment increase above initial levels by the middle of the third year after a VXO shock.
  - Jurado et al. (2015) uncertainty shock: both employment and industrial production decline sharply and declines remain persistent for five years.
  - Decomposed CSU effects:
    - Idiosyncratic component (forecast disagreement shock): less significant impacts on industrial production and employment.
    - Common uncertainty shocks: large and long-lived impacts with peak after two to three years.
    - Country-specific uncertainty (combined): responses of industrial production and employment not significant during first nine months after shock; significantly negative impact on industrial production starts around 10 months and remains persistent, keeping industrial production below initial level until the middle of the third year.
  - Interpretation: "wait-and-see" mechanism observed primarily in the common component rather than idiosyncratic component; country-specific measure shows larger persistence than many other proxies (VXO, EPU).
- Global VAR specification (monthly, seven variables):
  - Variables: stock prices, global uncertainty, short term interest rate, oil prices, food prices, unemployment rate, industrial production.
- Global VAR findings (Figure 5, impulse responses):
  - Industrial production (panel A):
    - Global uncertainty shock: immediate decline that dissipates within a few months.
    - Idiosyncratic uncertainty shock: similarly short-lived impact, with overshooting after six months.
    - Common uncertainty shock: long-lived impact with peak after two years.
  - Unemployment (panel B):
    - Global uncertainty shocks lead to sizable and protracted increase in unemployment; significant increase dissipates almost after 30 months.
    - Idiosyncratic shocks: high initial response in unemployment, overshooting after 30 months.
    - Common shocks: more persistent and high unemployment rates peaking around 30 months.
  - Interpretation: employers may "wait-and-see" before laying off after initial shock and then hire later during recovery.
- Joint common and idiosyncratic VAR specification:
  - Variables: log(stock price), common uncertainty, idiosyncratic uncertainty, monetary policy rate, log(consumer price index), log(industrial production).
  - Findings (Figure 6):
    - Common and idiosyncratic components have different effects on industrial production.
    - Common uncertainty shocks: large and persistent impacts.
    - Idiosyncratic uncertainty shocks: short-lived and negligible effects on industrial production.
    - Pattern holds for France, Italy, Japan, United Kingdom, United States.
    - Canada and Germany: both types of uncertainty shocks show very limited and short-lived effects.

*Source: 4. Properties of Economic Uncertainty, wp17219 - 4. Properties of Economic Uncertainty*

### 6. Conclusion

### 6. Conclusion

### Main contributions
- Decompose the uncertainty of a typical forecaster into common and idiosyncratic uncertainty.
- Propose a new monthly index of uncertainty that has both components: perceived variability of future aggregate shocks and the disagreement among forecasters.
- Use actual forecasts of market analysts instead of using hindsight to specify econometric forecasts, thereby incorporating a rich information set and capturing perceived uncertainty for market participants.

### Properties of the proposed uncertainty measure
- Country-specific uncertainty measures have fewer peaks compared to other leading measures, but when peaks occur they are larger and more persistent.
- The measure captures economic uncertainty along different dimensions by including both perceived variability of aggregate shocks and forecaster disagreement.

### Empirical findings (VAR analysis)
- Shocks to country-specific uncertainty are associated with a sizable and protracted decline in real activity.
- Global uncertainty shocks have long-lived effects on industrial production and unemployment.
- The two components of economic uncertainty behave differently:
  - Common uncertainty shocks account for a large fraction of fluctuations in economic activity at business cycle frequencies.
  - Idiosyncratic uncertainty shocks play a small role.

### Implications and directions for future research
- The persistence and size of uncertainty peaks are particularly relevant for theories where uncertainty drives recessions.
- Future research is warranted to quantify the economic effects of different types of uncertainty and analyze the transmission of uncertainty shocks across countries.

### Acknowledgements
- The paper was presented at the 2nd Society for Economic Measurement Conference, 8th International Conference on Computational and Financial Econometrics, 21st Federal Forecasters Conference, 24th Symposium of the Society for Nonlinear Dynamics and Econometrics, 35th International Symposium on Forecasting, London workshop on “Impact of Uncertainty Shocks on the Global Economy,” Stanford workshop on “The Macroeconomics of Uncertainty and Volatility,” Banque de France, George Washington University and Texas Tech University.
- The authors thank conference and seminar participants and internal reviewers from various departments of the IMF for helpful comments.
- The views expressed in this paper are those of the authors and do not necessarily represent views and policies of the IMF.

*Source: wp17219 - 6. Conclusion*

### References

### References (wp17219)

### Empirical correlations: Variable-specific uncertainty measures (United States)
- Table 1.A — Correlation between variable-specific uncertainty measures (Output, Inflation, Consumption, Investment, Industrial production, Unemployment rate, Short-term interest rate, Long-term interest rate). Selected entries (preserve matrix values exactly):
  - Output–Output: 1.00
  - Output–Inflation: 0.57
  - Output–Consumption: 0.79
  - Output–Investment: 0.77
  - Output–Industrial production: 0.82
  - Output–Unemployment rate: 0.77
  - Output–Short-term interest rate: 0.43
  - Output–Long-term interest rate: 0.41
- Table 1.B — Correlation between variable-specific idiosyncratic uncertainty measures. Selected entries:
  - Output–Output: 1.00
  - Output–Inflation: 0.38
  - Output–Consumption: 0.60
  - Output–Investment: 0.54
  - Output–Industrial production: 0.56
  - Output–Unemployment rate: 0.49
  - Output–Short-term interest rate: 0.19
  - Output–Long-term interest rate: 0.35
- Table 1.C — Correlation between variable-specific common uncertainty measures. Selected entries:
  - Output–Output: 1.00
  - Output–Inflation: 0.53
  - Output–Consumption: 0.83
  - Output–Investment: 0.75
  - Output–Industrial production: 0.87
  - Output–Unemployment rate: 0.71
  - Output–Short-term interest rate: 0.54
  - Output–Long-term interest rate: 0.20
- Note: Output, consumption, investment, and industrial production stand for the growth rates of these indicators. The sample is between 1989M11-2014M7 for all estimates.

### R-squared: Variable-specific uncertainty on country-specific uncertainty (Table 2; selected country aggregates)
- United States (Full sample): Output 0.822, Consumption 0.678, Investment 0.763, Industrial production 0.734, Unemployment rate 0.690, Inflation 0.506, Short-term interest rate 0.348, Long-term interest rate 0.303, Average 0.606
- United Kingdom (Full sample): Output 0.777, Consumption 0.805, Investment 0.564, Industrial production 0.659, Unemployment rate 0.479, Inflation 0.627, Short-term interest rate 0.512, Long-term interest rate 0.511, Average 0.617
- France (Full sample): Output 0.696, Consumption 0.498, Investment 0.612, Industrial production 0.640, Unemployment rate 0.429, Inflation 0.224, Short-term interest rate 0.342, Long-term interest rate 0.422, Average 0.483
- Germany (Full sample): Output 0.698, Consumption 0.528, Investment 0.399, Industrial production 0.568, Unemployment rate 0.251, Inflation 0.342, Short-term interest rate 0.540, Long-term interest rate 0.317, Average 0.455
- Euro Zone (Full sample): Output 0.857, Consumption 0.710, Investment 0.887, Industrial production 0.860, Unemployment rate 0.680, Inflation 0.568, Long-term interest rate N/A, Average 0.760
- Note: Each cell presents the R-square of regressions of respective variable-specific uncertainty on country-specific uncertainty measures. Recession episodes are from Claessens, Kose, Ozturk, Terrones (2016, forthcoming). The last column presents the average of the R-square in each economy.

### R-squared: Country-specific uncertainty on global uncertainty (Table 3 — selected economies)
- Sorted by estimated coefficients in uncertainty (total). Each result from bivariate regressions of country-specific uncertainty on global uncertainty. *** indicates significance at 1 percent level.
- Top coefficients (total uncertainty):
  - Estonia: β 1.597***, R2 0.880
  - Bulgaria: β 1.396***, R2 0.886
  - Lithuania: β 1.351***, R2 0.925
  - Latvia: β 1.346***, R2 0.904
  - Taiwan: β 1.320***, R2 0.891
  - Peru: β 1.296***, R2 0.694
  - Russia: β 1.268***, R2 0.836
  - Philippines: β 1.255***, R2 0.857
  - United States: β 1.212***, R2 0.782
  - Canada: β 1.211***, R2 0.677
  - (table continues through Venezuela)
- Bottom coefficients:
  - Indonesia: β 0.208***, R2 0.017
  - Venezuela: β 0.039, R2 0.001
- Note: Columns reported for Uncertainty (total), Idiosyncratic Uncertainty, Common Uncertainty.

### Correlations of uncertainty measures (United States; Table 4)
- Correlation matrix entries (preserve values exactly):
  - Country-specific uncertainty — Country-specific uncertainty: 1.00
  - Country-specific uncertainty — Common uncertainty: 0.94
  - Country-specific uncertainty — Idiosyncratic uncertainty: 0.80
  - Country-specific uncertainty — Economic policy uncertainty: 0.18
  - Country-specific uncertainty — News-based policy uncertainty: 0.19
  - Country-specific uncertainty — Jurado et al. (2015): 0.79
  - Country-specific uncertainty — Scotti (2016): 0.38
  - Country-specific uncertainty — VXO: 0.54
  - Common uncertainty — Idiosyncratic uncertainty: 0.54
  - Common uncertainty — Economic policy uncertainty: 0.05
  - Common uncertainty — News-based policy uncertainty: 0.07
  - Idiosyncratic uncertainty — Economic policy uncertainty: 0.36
  - Idiosyncratic uncertainty — News-based policy uncertainty: 0.35
  - Economic policy uncertainty — News-based policy uncertainty: 0.90
  - Jurado et al. (2015) correlations: 0.79 with country-specific, 0.75 with common, 0.59 with idiosyncratic
  - Scotti (2016) correlations: 0.38 with country-specific, 0.30 with common, 0.44 with idiosyncratic
  - VXO correlations: 0.54 with country-specific, 0.48 with common, 0.49 with idiosyncratic, 0.40 with EPU, 0.49 with NBU, 0.60 with Jurado, 0.51 with Scotti, 1.00 with VXO (diagonal)
- Note: News-based policy uncertainty and economic policy uncertainty measures are from the policy uncertainty website of Baker, Bloom, and Davis (2016). All correlations except those with Scotti (2016) cover 1989M11-2014M7 period and correlations with Scotti (2016) cover 2003M5-2014M7 period.

### Time-series patterns: Country-specific and global uncertainty (Figures 1–3)
- Country-specific uncertainty:
  - Figure 1 presents country-specific uncertainty (sum of idiosyncratic and common uncertainty) for multiple economies. Gray bars indicate recessions identified in Claessens, et. al. (2016, forthcoming).
  - Time axis labels preserved in figures: e.g., 1989M10, 1992M11, 1994M4, 1996M7, 1998M10, 2001M1, 2003M4, 2005M7, 2007M10, 2010M1, 2012M4, 2014M7 (varies by panel).
- Global uncertainty:
  - Figure 2: Each line presents the PPP-weighted average of the respective measure for 46 economies. Gray bars present global recession episodes identified by Kose and Terrones (2015).
  - Series plotted: Global Uncertainty, Common Uncertainty, Idiosyncratic Uncertainty; time axis labels include 1989M10 up to 2013M10.
- Comparison of uncertainty measures:
  - Figure 3: Each uncertainty measure is standardized by subtracting the mean and dividing by the standard deviation. Abbreviations: CSU = country-specific uncertainty, JLN = Jurado et al. (2015), EPU = economic policy uncertainty (Baker et al., 2016), NBU = news-based uncertainty (Baker et al., 2016), Scotti = Monthly average of ex-ante daily uncertainty (Scotti, 2016), VXO = VXO index.
  - Panels include country-specific comparisons (United States, France, Canada, Italy, United Kingdom, Germany, Japan, Spain, Netherlands, China, South Korea, Russia) with recession episodes marked and 1.65 Standard deviation lines shown in some panels.

### Responses to uncertainty shocks (Figures 4–6)
- Responses to uncertainty shocks (Figure 4):
  - Panel A plots responses of industrial production to uncertainty shocks identified recursively in an eight-variable VAR system estimated separately for each uncertainty measure. CSU, JLN, EPU, VXO shown. Dotted (dashed) lines are responses to forecast disagreement shocks where CSU is replaced with idiosyncratic (common) uncertainty component.
  - Panel B plots responses of employment to the same set of uncertainty shocks.
  - Shaded regions present 64 percent confidence intervals using Kilian (1998) bias-corrected bootstrap.
- Responses to global uncertainty shocks (Figure 5):
  - Panel A plots responses of industrial production to uncertainty shocks identified recursively in a seven-variable VAR system estimated separately for total uncertainty and its common and idiosyncratic components.
  - Panel B plots responses of the unemployment rate to the same shocks.
  - Horizontal axes run 0 to 60 periods; panels show differing signs and magnitudes across Total, Common, and Idiosyncratic components.
- Response of industrial production to common and idiosyncratic uncertainty shocks (Figure 6):
  - Country panels (A–G) show responses for United States, United Kingdom, Canada, Japan, Germany, France, Italy.
  - Dashed lines present 64 percent confidence intervals using Kilian (1998) bias-corrected bootstrap.
  - Panels distinguish responses to Common uncertainty vs Idiosyncratic uncertainty.

*References and empirical tables/figures as provided in the source PDF (wp17219 - References).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17219.pdf_