## wpiea2022091-print-pdf - 0.21 percentage points.

## Source details

**Canonical URL:** [wpiea2022091-print-pdf - 0.21 percentage points.](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022091-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022091-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022091-print-pdf.pdf.json)

---

### Key empirical patterns: downward trend in core non-cyclical inflation
- Core non-cyclical inflation on average declined by 1.1 percentage points between 2012 and 2019.
- Core cyclical inflation increased by 0.26 percentage points between 2012 and 2019.
- The coefficient of the linear trend for 2012-2019 (estimated from a Phillips curve augmented with two separate linear time trends for 2000-11 and 2012-19, with equal weights across countries) is significantly negative for core non-cyclical inflation.
- The downward trend in core non-cyclical inflation:
  - Was not reversed between 2016 and 2019 when oil price started to rise.
  - Is robust to excluding any first-digit COICOP components, and to restricting the exercise to CPI components classified as goods or as services.
  - Was not driven by large countries (estimation uses equal country weights to avoid disproportionate influence).

### Timing and overlaps with other episodes
- The episode of inflation surges overlaps with the missing disinflation period analyzed by Coibion and Gorodnichenko (2015): 2010Q4 – 2011Q2 had the largest positive surprises in short-term professional forecasters’ expectations between 2009 and 2011.
- Fact 1 (inflation surges in 2011):
  - Core cyclical inflation and core non-cyclical inflation increased by 0.87 and 0.88 percentage points between December 2010 and December 2011, when the unemployment rate declined by only 0.2 percentage points.
  - For the increase to be explained by contemporaneous changes in unemployment rate, the slope of the Phillips curve would need to be around 4; the literature estimate for the slope is much smaller than this value.
  - Between September 2008 and December 2009, the average unemployment rate increased by 2.56 percentage points, core cyclical inflation declined by 2.3 percentage points, and core non-cyclical inflation declined by 0.23 percentage points.
  - The quarters with the largest positive deviation from the historical relationship between inflation surprise and unemployment rate were 2009Q2 – 2009Q4 and 2010Q4 – 2011Q2.
  - The 2011 surge reversed between 2012 – 2013, although more gradually than the surge.

### Global cost-push indexes and their (lack of) explanatory power
- Global indexes considered: oil price, Baltic Dry Index (BDI) as shipping costs, and the global supply chain pressure index (GSCPI); food price also included in extended checks.
- Including these global indexes in the Phillips curve has:
  - Insignificant or even negative impact on core non-cyclical inflation.
  - Little impact on the negative linear trend between 2012 and 2019.
- Key results (augmented regressions):
  - The global supply chain pressure index and food prices do not have significant impact on core non-cyclical inflation.
  - The coefficients of oil price and the BDI are insignificant in most cases and sometimes significantly negative.
  - Time trend estimates controlling for global indexes are similar to baseline estimates, leaving the downward trend intact.
- Conditions required for a cost-push explanation (all four must hold):
  1. They affect more than one sector;
  2. They influence multiple countries;
  3. They are not captured by any of the global indexes mentioned above;
  4. Firms responded to these cost-push factors significantly, while they absorbed the impact of oil prices, shipping costs and global supply chain disruption without passing the cost pressure to consumers.
- Interpretation: these patterns suggest cost-push shocks were not the cause of the downward trend in core non-cyclical inflation prior to the pandemic.

### Core cyclical inflation: downward trend after controlling for slack
- Core cyclical inflation also exhibited a downward trend between 2012 and 2019 after controlling for economic slack.
- Identification strategy and assumptions:
  - OLS estimation of a Phillips curve for core cyclical inflation is treated as a lower bound of the true Phillips curve slope (acknowledging endogeneity concerns).
  - Given falling unemployment rates between 2012 and 2019, the lower bound on the slope yields an upper bound for the linear time trend of non-cyclical factors behind core cyclical inflation — identifying a negative trend.
- Implementation and key numeric estimates:
  - Step 1: OLS estimate of slope of Phillips curve for 2000Q1–2019Q4: β_C = −0.297
  - Step 2: Quarterly regression for 2012Q1–2019Q4 of π_{i,t}^C + 0.297 u_{i,t}^C on a linear time trend yields:
    - β̂ = −0.0242
    - standard error = 0.0076
  - Interpretation: Under the assumption that OLS estimation of the Phillips curve slope is downward biased, these estimates imply that cyclical core inflation exhibits a downward trend between 2012 and 2019.

### Interpretation and plausible drivers
- The disconnect between inflation and unemployment rates is attributed to counter-cyclical movements of inflation-driving forces that did not come from:
  - Cost-push shocks,
  - Core goods price inflation,
  - Country- or sector-specific factors.
- Among remaining candidates, counter-cyclical movements of firms’ inflation expectations are suggested as more plausible than measurement errors, given:
  - Broad-based movements across countries and sectors.
  - Consistency with literature finding that firms’ inflation expectations are not fully anchored.

### Relation to literature and contribution
- Distinctive aspects of this paper:
  - Decomposes core inflation dynamics in a cross-country context (28 advanced economies).
  - Explores both the missing disinflation and the missing reflation.
  - Studies the role of competing factors that are not changes in economic slack, rather than focusing solely on the Phillips curve slope.
- Connects to literature constructing alternative inflation measures and to studies on supply-chain disruptions during the pandemic; finds that global cost-push shocks did not cause persistent movements in core non-cyclical inflation before the pandemic.

### Data, definitions, and methodology highlights
- Sample: 28 advanced economies.
- Cyclical sensitivity measure:
  - Defined as the median, across advanced economies, of the correlation between a component’s inflation and the unemployment gap (unemployment rate minus NAIRU) over January 2000–December 2019.
  - Results robust to replacing the unemployment gap with HP-filtered output gap.
- Classification of CPI components:
  - Cyclical CPI components: “restaurants and hotels”, “furnishings, household equipment, and routine maintenance of the house”, “housing excluding gas and electricity”, “food”, and “recreation”.
  - Non-cyclical CPI components: remaining core components (examples include “miscellaneous goods and services”, “health”, “clothing”, “education”, and “communications”); “transportation excluding energy” excluded as an outlier.
- Construction of core cyclical and core non-cyclical inflation:
  - Weighted averages of the designated COICOP components.
  - Three-month trailing averages are used to reduce high-frequency noise.
- Common trends estimation:
  - Regression Z_{i,t} = μ_{t}^Z + δ_{i}^Z + ε_{i,t}^Z; estimation uses GDP based on Purchasing Power Parity (PPPGDP) as weights.
- Precise numeric and temporal markers preserved: 0.21 percentage points; 1.1 percentage points decline (2012–2019); 0.26 percentage points increase (2012–2019); 2010Q4 – 2011Q2; 2009–2011; January 2000 – December 2019; three-month trailing averages; 28 countries.

*IMF Working Paper: Inflation Dynamics in Advanced Economies: A Decomposition into Cyclical and Non-Cyclical Factors (source PDF: wpiea2022091-print-pdf).*

### 0.21 percentage points.

### wpiea2022091-print-pdf - 0.21 percentage points.

### Key empirical patterns: downward trend in core non-cyclical inflation
- Core non-cyclical inflation on average declined by 1.1 percentage points between 2012 and 2019.
- Core cyclical inflation increased by 0.26 percentage points between 2012 and 2019.
- The coefficient of the linear trend for 2012-2019 (estimated from a Phillips curve augmented with two separate linear time trends for 2000-11 and 2012-19, with equal weights across countries) is significantly negative for core non-cyclical inflation.
- The downward trend in core non-cyclical inflation:
  - Was not reversed between 2016 and 2019 when oil price started to rise.
  - Is robust to excluding any first-digit COICOP components, and to restricting the exercise to CPI components classified as goods or as services.
  - Was not driven by large countries (estimation uses equal country weights to avoid disproportionate influence).

### Timing and overlaps with other episodes
- The episode of inflation surges overlaps with the missing disinflation period analyzed by Coibion and Gorodnichenko (2015): 2010Q4 – 2011Q2 had the largest positive surprises in short-term professional forecasters’ expectations between 2009 and 2011.

### Global cost-push indexes and their (lack of) explanatory power
- Global indexes considered: oil price, Baltic Dry Index (BDI) as shipping costs, and the global supply chain pressure index (GSCPI).
- Including these global indexes in the Phillips curve has:
  - Insignificant or even negative impact on core non-cyclical inflation.
  - Little impact on the negative linear trend between 2012 and 2019.
- Interpretation: these patterns suggest cost-push shocks were not the cause of the downward trend in core non-cyclical inflation prior to the pandemic.

### Core cyclical inflation: downward trend after controlling for slack
- Core cyclical inflation also exhibited a downward trend between 2012 and 2019 after controlling for economic slack.
- Identification strategy:
  - OLS estimation of a Phillips curve for core cyclical inflation is treated as a lower bound of the true Phillips curve slope (acknowledging endogeneity concerns).
  - Given falling unemployment rates between 2012 and 2019, the lower bound on the slope yields an upper bound for the linear time trend of non-cyclical factors behind core cyclical inflation — identifying a negative trend.

### Interpretation and plausible drivers
- The disconnect between inflation and unemployment rates is attributed to counter-cyclical movements of inflation-driving forces that did not come from:
  - Cost-push shocks,
  - Core goods price inflation,
  - Country- or sector-specific factors.
- Among remaining candidates, counter-cyclical movements of firms’ inflation expectations are suggested as more plausible than measurement errors, given:
  - Broad-based movements across countries and sectors.
  - Consistency with literature finding that firms’ inflation expectations are not fully anchored (Coibion and Gorodnichenko (2015), Kumar et al. (2015), Jorgenson and Lansing (2019), Candia et al. (2021)).

### Relation to literature and contribution
- Distinctive aspects of this paper:
  - Decomposes core inflation dynamics in a cross-country context (28 advanced economies).
  - Explores both the missing disinflation and the missing reflation.
  - Studies the role of competing factors that are not changes in economic slack, rather than focusing solely on the Phillips curve slope.
- Connects to literature constructing alternative inflation measures and to studies on supply-chain disruptions during the pandemic; finds that global cost-push shocks did not cause persistent movements in core non-cyclical inflation before the pandemic.

### Data, definitions, and methodology highlights
- Sample: 28 advanced economies (list of countries provided in source).
- Cyclical sensitivity measure:
  - Defined as the median, across advanced economies, of the correlation between a component’s inflation and the unemployment gap (unemployment rate minus NAIRU) over January 2000–December 2019.
  - Results robust to replacing the unemployment gap with HP-filtered output gap.
- Classification:
  - Cyclical CPI components: “restaurants and hotels”, “furnishings, household equipment, and routine maintenance of the house”, “housing excluding gas and electricity”, “food”, and “recreation”.
  - Non-cyclical CPI components: remaining core components (examples include “miscellaneous goods and services”, “health”, “clothing”, “education”, and “communications”); “transportation excluding energy” excluded as an outlier.
- Construction of core cyclical and core non-cyclical inflation:
  - Weighted averages of the designated COICOP components.
  - Three-month trailing averages are used to reduce high-frequency noise.
- Common trends estimation:
  - Regression Z_{i,t} = μ_{t}^Z + δ_{i}^Z + ε_{i,t}^Z, where Z_{i,t} denotes core cyclical inflation or core non-cyclical inflation.
  - Estimation uses GDP based on Purchasing Power Parity (PPPGDP) as weights to avoid small-country-driven results.
- Precise numeric and temporal markers preserved: 0.21 percentage points; 1.1 percentage points decline (2012–2019); 0.26 percentage points increase (2012–2019); 2010Q4 – 2011Q2; 2009–2011; January 2000 – December 2019; three-month trailing averages; 28 countries.

*IMF Working Paper: Inflation Dynamics in Advanced Economies: A Decomposition into Cyclical and Non-Cyclical Factors (source PDF: wpiea2022091-print-pdf).*

### 4.1 Fact 1: Inflation surges in 2011

### 4.1 Fact 1: Inflation surges in 2011

### Key findings
- Core cyclical inflation and core non-cyclical inflation increased by 0.87 and 0.88 percentage points between December 2010 and December 2011, when the unemployment rate declined by only 0.2 percentage points.
- For the increase to be explained by contemporaneous changes in unemployment rate, the slope of the Phillips curve would need to be around 4; the literature estimate for the slope is much smaller than this value.
- Co-movement in 2011 was stronger than in 2009:
  - Between September 2008 and December 2009, the average unemployment rate increased by 2.56 percentage points, core cyclical inflation declined by 2.3 percentage points, and core non-cyclical inflation declined by 0.23 percentage points.
- The quarters with the largest positive deviation from the historical relationship between inflation surprise and unemployment rate (per Coibion and Gorodnichenko (2015)) were 2009Q2 – 2009Q4 and 2010Q4 – 2011Q2.
- The coincidence between inflation surges across advanced economies and positive inflation surprises in 2011 suggests factors other than changes in economic slack played a significant role in driving up inflation.
- There was a reversal of the inflation surge in 2011 between 2012 – 2013, although at a more gradual pace compared with the surge.

---

### 4.2 Fact 2: The downward trend of core non-cyclical inflation

### Overall trend
- Core non-cyclical inflation declined by 1.1 percentage points between January 2012 and December 2019; the decline was statistically significant.
- The decline was not reversed between 2016 and 2019, when the oil price increased.
- The cyclical part of core inflation recovered to a level in 2019 that was higher than that in 2011, though the difference is not statistically significant.

### Within goods and services
- Inflation surges in 2011 were broad-based across both goods and services. All four components had a significant increase in 2011, with similar timing and magnitude:
  - Cyclical goods increased by 1.08 percentage points between December 2010 and December 2011.
  - Non-cyclical goods increased by 1.56 percentage points between December 2010 and December 2011.
  - Cyclical services increased by around 0.79 percentage points between December 2010 and December 2011.
  - Non-cyclical services increased by around 0.86 percentage points between December 2010 and December 2011.
- Between January 2012 and December 2019:
  - Cyclical goods and cyclical services increased by 1.20 and 0.59 percentage points, respectively.
  - Non-cyclical goods and non-cyclical services declined by 1.45 and 0.35 percentage points, respectively.

### Across sectors (sensitivity to exclusion of components)
- Excluding the health sector from core non-cyclical inflation:
  - The average increase in non-cyclical core inflation from December 2010 to December 2011 was 1.21 percentage points.
  - Its decline from January 2012 to December 2019 was 1.19 percentage points.
- Results are robust to excluding other first-digit COICOP components (see referenced robustness exercises).

### Across countries (equal-weight panel evidence)
- Estimating a linear time trend between 2012-19 for core non-cyclical inflation (equal weights across countries) yields a negative and statistically significant linear trend coefficient.
- Based on the estimated coefficient in column (1) of Table 1, the implied cumulative decline in core non-cyclical inflation is 0.924 percentage points, broadly consistent with common time trends.
- The negative trend holds whether estimated for 2012–2019 or for the entire period allowing separate linear trends before and after 2012, supporting that the downward trend is not driven by country-specific factors.

---

### 4.3 The downward trend not driven by cost-push factors measured by global indexes

### Empirical approach
- Augment equation (2) with global indexes (equation (3)): oil price, food price, shipping costs (Baltic Dry Index, BDI), and a Global Supply Chain Pressure index.
- Variables defined as changes refer to differences relative to 12 months ago to be consistent with the definition of core cyclical and non-cyclical inflation.

### Key results
- The global supply chain pressure index and food prices do not have significant impact on core non-cyclical inflation.
- The coefficients of oil price and the BDI are insignificant in most cases and sometimes significantly negative.
- The downward trend of non-cyclical core inflation is not significantly different after controlling for global indexes: time trend estimates in columns (2)–(4) are similar to column (1), and columns (6)–(8) similar to column (5).
- These findings make it questionable that cost-push shocks (as proxied by these global indexes) drove the downward trend of core non-cyclical inflation across advanced economies.

### Conditions required for a cost-push explanation
- For cost-push shocks to be the driver, they would need to satisfy all four conditions:
  1. They affect more than one sector;
  2. They influence multiple countries;
  3. They are not captured by any of the global indexes mentioned above;
  4. Firms responded to these cost-push factors significantly, while they absorbed the impact of oil prices, shipping costs and global supply chain disruption without passing the cost pressure to consumers.

*Source: IMF Working Paper — “Inflation Dynamics in Advanced Economies: A Decomposition into Cyclical and Non-Cyclical Factors” (sections 4.1–4.3).*

### 4.4 Fact 4: Downward trend of core cyclical inflation

### 4.4 Fact 4: Downward trend of core cyclical inflation

### Framework and key assumptions
- Definition: Core cyclical inflation π_{i,t}^C satisfies
  - π_{i,t}^C = α_i^C + (β_C + b_C) u_{i,t} + ε_{i,t}^C, (3)
  - where u_{i,t} is the unemployment gap, ε_{i,t}^C captures non-cyclical factors, and β_C + b_C is the true slope of the Phillips curve (written as a sum to highlight potential OLS bias).
- Objective: determine the sign of the underlying trend of ε_{i,t}^C (factors other than changes in economic slack that affect core cyclical inflation).
- Assumptions:
  - (A1) u_{i,t} = α_i^u + β_u φ + v_{i,t}^u, and β_u < 0. (Assumes unemployment rate follows a downward trend for 2012–2019.)
  - (A2) ε_{i,t}^C = α_i^{εC} + β_{εC} φ + v_{i,t}^{εC}.
  - (A3) cov(φ, v_{i,t}^u) = cov(φ, v_{i,t}^{εC}) = 0. (Technical assumptions to ease interpretation.)

### Lemma and identification logic
- Lemma 1: Under assumptions A1–A3, an OLS regression of π_{i,t}^C − β_C u_{i,t} on the linear time trend yields an upper bound of the trend of factors that are not changes in economic slack and that drive core cyclical inflation.
- Proof sketch (as in source): The estimation gives β_{εC} + β_u b_C. Since b_C < 0, β_{εC} + β_u b_C > β_{εC}. Q.E.D.
- Identification strategy implemented:
  - Use the slope of the Phillips curve estimated over 2000Q1–2019Q4 as a lower bound for the slope over 2012Q1–2019Q4. A longer sample (2000Q1-2019Q4) yields a steeper slope, which strengthens the implied downward trend of ε_{i,t}^C.
  - Literature evidence mitigates concerns about slope instability over the post-2000 period (cited finding: slope broadly stable between 2000Q1 and 2019Q4).

### Implementation and empirical results
- Step 1: OLS estimate of slope of Phillips curve for 2000Q1–2019Q4:
  - β_C = −0.297
- Step 2: Quarterly regression for 2012Q1–2019Q4 estimated:
  - Equation: π_{i,t}^C + 0.297 u_{i,t}^C = α_i + β̂ φ + v_{i,t}
  - Country fixed effects α_i included; standard errors clustered at the country level.
  - Estimated trend coefficient:
    - β̂ = −0.0242
    - standard error = 0.0076
- Interpretation: Under the assumption that OLS estimation of the Phillips curve slope is downward biased, these estimates imply that cyclical core inflation exhibits a downward trend between 2012 and 2019.

### Substantive findings and context from concluding remarks
- Documented patterns:
  - The intrinsic slope of the Phillips curve is significantly positive, implying counter-cyclical movements of non-slack factors weakened the inflation–unemployment link.
  - The non-cyclical part of core inflation fell persistently across advanced economies between 2012 and 2019, offsetting a recovery of the cyclical part of core inflation.
  - The downward trend was broad-based across countries, sectors, goods, and services and was not explained by global cost-push measures.
- Interpretation on candidate explanations:
  - Changes in firms’ inflation expectations are a more promising candidate to explain the disconnect between inflation and unemployment than measurement errors.
  - Evidence noted: core non-cyclical inflation did not co-move well with oil prices (the inflation surge concentrated in 2011, while the oil-price rebound started in 2009; the downward trend of core non-cyclical inflation persisted through 2016–2019 despite oil price increases).
  - Including oil prices in Phillips curve estimation does not affect the presence of a downward time trend for the non-cyclical part of core inflation.
- Relation to the literature:
  - Differences across studies about "missing disinflation" after the global financial crisis may stem from the proxies used to capture firms’ inflation expectations and whether studies consider the 2011 inflation rise.

*Source: IMF Working Paper — "Inflation Dynamics in Advanced Economies: A Decomposition into Cyclical and Non-Cyclical Factors" (section 4.4 and concluding remarks).*

### 3. Excluding communication

### 3. Excluding communication

### Series shown
- 1. Excluding health care
  - Vertical axis ticks: -1, -.5, 0, .5, 1, 1.5
  - Time axis ticks: 2010m1, 2012m1, 2014m1, 2016m1, 2018m1, 2020m1
  - Series labels: Cyclical; Cyclical: 90 percent CI; Non-Cyclical; Non-Cyclical: 90 percent CI
- 2. Excluding clothing and footwear
  - Vertical axis ticks: -1, -.5, 0, .5, 1, 1.5
  - Time axis ticks: 2010m1, 2012m1, 2014m1, 2016m1, 2018m1, 2020m1
  - Series labels: Cyclical; Cyclical: 90 percent CI; Non-Cyclical; Non-Cyclical: 90 percent CI
- 3. Excluding communication
  - Vertical axis ticks: -2, -1, 0, 1, 2
  - Time axis ticks: 2010m1, 2012m1, 2014m1, 2016m1, 2018m1, 2020m1
  - Series labels: Cyclical; Cyclical: 90 percent CI; Non-Cyclical; Non-Cyclical: 90 percent CI
- 4. Excluding education
  - Vertical axis ticks: -2, -1, 0, 1, 2
  - Time axis ticks: 2010m1, 2012m1, 2014m1, 2016m1, 2018m1, 2020m1
  - Series labels: Cyclical; Cyclical: 90 percent CI; Non-Cyclical; Non-Cyclical: 90 percent CI
- 5. Excluding miscellaneous goods and services
  - Vertical axis ticks: -2, -1, 0, 1, 2
  - Time axis ticks: 2010m1, 2012m1, 2014m1, 2016m1, 2018m1, 2020m1
  - Series labels: Cyclical; Cyclical: 90 percent CI; Non-Cyclical; Non-Cyclical: 90 percent CI
- 6. Excluding furnishing
  - Vertical axis ticks: -1, 0, 1, 2
  - Time axis ticks: 2010m1, 2012m1, 2014m1, 2016m1, 2018m1, 2020m1
  - Series labels: Cyclical; Cyclical: 90 percent CI; Non-Cyclical; Non-Cyclical: 90 percent CI
- 7. Excluding housing
  - Vertical axis ticks: -1, -.5, 0, .5, 1, 1.5
  - Time axis ticks: 2010m1, 2012m1, 2014m1, 2016m1, 2018m1, 2020m1
  - Series labels: Cyclical; Cyclical: 90 percent CI; Non-Cyclical; Non-Cyclical: 90 percent CI
- 8. Excluding recreation
  - Vertical axis ticks: -1, -.5, 0, .5, 1, 1.5
  - Time axis ticks: 2010m1, 2012m1, 2014m1, 2016m1, 2018m1, 2020m1
  - Series labels: Cyclical; Cyclical: 90 percent CI; Non-Cyclical; Non-Cyclical: 90 percent CI
- 9. Excluding restaurants and hotels
  - (No explicit axis ticks printed in source fragment)
  - Series labels implied: Cyclical; Cyclical: 90 percent CI; Non-Cyclical; Non-Cyclical: 90 percent CI

### Notes and context
- Document title fragment: Inflation Dynamics in Advanced Economies: A Decomposition into Cyclical and Non-Cyclical Factors
- Publication identifier: Working Paper No. WP/2022/091
- Sources listed: Haven Analytics, Euro stat, the OECD stat, and authors’ calculation.

*International Monetary Fund — IMF WORKING PAPERS Inflation Dynamics in Advanced Economies: A Decomposition into Cyclical and Non-Cyclical Factors — Working Paper No. WP/2022/091*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022091-print-pdf.pdf_