## _wp15214

## Source details

**Canonical URL:** [_wp15214](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15214.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15214.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15214.pdf.json)

---

### I. Introduction — scope and dataset
- Common policy practice assumes indirect tax changes are fully and exactly passed through to consumer prices; theory permits undershifting, overshifting, and price declines (Edgeworth paradox).
- This paper exploits a large Eurozone panel (1999 to 2013) to estimate VAT pass through:
  - Monthly price and tax data for 67 consumption items.
  - Pooled data from 17 integrated economies sharing a common currency.
  - Dataset covers 1,231 VAT changes.
- Research objectives:
  - Estimate pass through for standard rate changes, reduced rates, and reclassifications.
  - Test heterogeneity by scope (share of consumption affected), asymmetries between increases and decreases, dynamics (anticipation and sluggish response), and durables versus non-durables.
- Robustness checks address measurement error and endogeneity.

### II. Theory and empirical specification
- Pass-through definition (equation (1)):
  - σ_i ≡ (Δp_i / p_i) / (Δτ_i / (1+τ_i)).
  - Full pass through: σ_i = 1; zero pass through: σ_i = 0.
- Partial- and general-equilibrium expressions highlight roles of own-price supply elasticity, own-price demand elasticity, and cross-price elasticities; compact expression summarized in equation (5) involves elasticity of substitution ε, income elasticity η_1, and budget share s_1.
- Empirical reduced-form (equation (6)):
  - Δln(p_{i,c,t}) = sum_{j=-12}^{12} β_j (1 + Δln τ_{c,t+j}) + X_{c,t} γ + α_c + α_i + δ_t + ε_{i,c,t}.
  - Sum of β_j over ±12 months interpreted as long-run pass through elasticity.
  - Controls X include monthly unemployment rate and real quarterly GDP growth.
  - Fixed effects: country α_c, consumption category α_i, month δ_t.
  - Regressions weighted by consumption category share; standard errors clustered at panel level.

### III. Data and sample statistics (Appendix Table A1 summary)
- Panel size and coverage:
  - Total unbalanced panel: 155,385 monthly observations.
  - Longest individual panels: January 1999 to September 2013.
  - 65 distinct VAT reform episodes generate 1,231 individual VAT rate changes.
- VAT level statistics (Number of obs, Mean, St. Dev., Min, Max):
  - Standard: 107,089; 19.24; 2.14; 15.00; 23.00
  - Reduced: 41,667; 7.92; 3.40; 2.10; 17.00
  - Zero-rated: 3,341; 0.00; 0.00; 0.00; 0.00
- VAT changes (log changes in the tax factor):
  - Standard changes: 972 observations; Mean 1.05; St. Dev. 0.96; Min -0.83; Max 2.55
  - Reduced changes: 191 observations; Mean 0.42; St. Dev. 1.98; Min -6.45; Max 6.45
  - Reclassification: 68 observations; Mean -0.32; St. Dev. 8.58; Min -13.38; Max 14.87
  - VAT increase: 1,009 observations; Mean 1.67; St. Dev. 1.69; Min 0.41; Max 14.87
  - VAT decrease: 222 observations; Mean -3.02; St. Dev. 3.49; Min -13.38; Max -0.41
- Price changes and controls (Number of obs, Mean, St. Dev., Min, Max):
  - HICP Index: 155,385; 0.01; 1.34; -117.96; 142.82
  - Unemployment rate: 155,385; 8.22; 4.15; 1.70; 27.60
  - Real Quarterly GDP growth: 155,385; 0.38; 1.20; -8.50; 7.90
  - Consumption weights: 155,385; 3.38; 3.23; 0.00; 17.56
- Dataset notes:
  - Approximately 70 percent of consumption items are taxed at the standard VAT rate, accounting for slightly less than 65 percent of the value of total consumption.
  - Standard VAT rates range between 15 and 23 percent, with an average rate of close to 20 percent. Reduced rates vary between 2.1 and 17 percent, averaging 8 percent.
- Pre-estimation processing: price indices de-seasonalized and de-trended.

### IV. Average pass-through estimates (Table 2)
- Estimation variants reported with 155,385 observations:
  - Column (1): no controls/fixed effects.
  - Column (2): country-month-category fixed effects.
  - Column (3): fixed effects plus unemployment rate and real quarterly GDP growth (preferred).
- Cumulative pass through estimates (sum over ±12 months; standard errors in parentheses):
  - Column (1): Pre-Reform 0.12* (0.06); Contemporaneous 0.17*** (0.03); Post-Reform 0.11* (0.06); Total 0.40*** (0.09); R-squared 0.00
  - Column (2): Pre-Reform 0.07 (0.07); Contemporaneous 0.15*** (0.03); Post-Reform 0.07 (0.06); Total 0.29*** (0.10); Fixed Effects Yes; R-squared 0.02
  - Column (3): Pre-Reform 0.08 (0.07); Contemporaneous 0.15*** (0.03); Post-Reform 0.09 (0.06); Total 0.32*** (0.09); Controls: Unemployment Rate -0.01*** (0.00); Real quarterly GDP growth -0.01 (0.01); Fixed Effects Yes; R-squared 0.02
- Key interpretations:
  - Preferred specification (column (3)) yields Total = 0.32*** (0.09).
  - Point estimates imply roughly one-third of a VAT change is passed to consumer prices on average.
  - Significant anticipation and lag effects: roughly one-third to one-half of total effects occur before or after the reform month.
  - Null of full pass through (σ = 1) is rejected in pooled estimates.

### V. Pass through by type of VAT change (Table 3, Figure 2; OLS and IV)
- Sample counts: Standard 972, Reduced 191, Reclassification 68 (155,385 observations).
- OLS cumulative pass through by type (Pre-Reform, Contemporaneous, Post-Reform, Total; standard errors in parentheses):
  - Standard: 0.71*** (0.22); 0.30*** (0.06); 0.38** (0.17); Total 1.39*** (0.36)
  - Reduced: -0.13 (0.25); 0.36*** (0.13); 0.07 (0.18); Total 0.30 (0.23)
  - Reclassification: -0.06 (0.06); 0.10*** (0.03); 0.04 (0.06); Total 0.08 (0.07)
- Main OLS interpretations:
  - Standard rate changes: large anticipation effects (cumulative lead 0.71***), contemporaneous 0.30***, post-reform 0.38**; Total 1.39*** indicates overshifting (~39 percent) though standard error 0.36 means long-run full pass through cannot be rejected in some specifications.
  - Reduced rate changes: no significant anticipation; contemporaneous 0.36***; Total 0.30 (0.23) — substantially below full pass through.
  - Reclassifications: mostly contemporaneous small effect 0.10***; Total 0.08 (0.07) — not different from zero.
- Heterogeneity tests:
  - Wald tests reject equality of pass through across types for pre-reform, contemporaneous, and total cumulative effects (p-values reported as 0.00).
- IV strategy using HICP-CT-derived VAT changes as instruments (Tables 5–6 summary):
  - Correlations between HICP-CT series and own VAT measure:
    - Standard: 0.81; Reduced: 0.64; Reclassifications: 0.66.
  - OLS on HICP-CT-coincident sample (104,212 observations) yields similar qualitative patterns; standard total 1.30*** (0.42).
  - IV second-stage sums (VAT changes instrumented with HICP-CT; 104,212 observations) — panel-clustered SEs:
    - Standard: Pre-Reform 1.10** (0.50); Contemporaneous 0.37*** (0.13); Post-Reform 0.51* (0.30); Total 1.98** (0.82). First-stage Angrist-Pischke F-Stat 194.7.
    - Reduced: Pre-Reform -0.50** (0.24); Contemporaneous 0.59*** (0.23); Post-Reform 0.08 (0.39); Total 0.16 (0.35). First-stage F-Stat 278.7.
    - Reclassification: Pre-Reform -0.38* (0.22); Contemporaneous 0.21** (0.09); Post-Reform 0.39 (0.41); Total 0.23 (0.22). First-stage F-Stat 29.1.
  - OLS=IV Wald test p-values: Standard 0.17; Reduced 0.61; Reclassification 0.29 (cannot reject equality of OLS and IV for each type).
- Robustness conclusion: measurement error/attenuation bias unlikely to drive main results; IV results broadly corroborate OLS patterns though with larger standard errors.

### VI. Durables versus non-durables
- Durables exhibit:
  - Significant anticipation effects: pass through of 35 percent achieved before VAT change.
  - Long-run overall pass through around 50 percent (larger than non-durables), but still significantly different from full pass through.
  - More than 95 percent of durables taxed at the standard rate.
- Non-durables exhibit:
  - No anticipation effects.
  - Smaller long-run pass through than durables.
  - Close to 45 percent of non-durables taxed at reduced or zero-rate.

### VII. Scope of reform (consumption share affected) — inverted U relationship (Figure 5)
- Heterogeneity strategies:
  - Interact VAT change with quantile bins of total consumption share affected: 0-10%, 10-20%, 20-26%, 26-40%, 40-50%, 50-60%, 60-74%.
  - Interact VAT change with a cubic polynomial in the consumption share affected.
- Main empirical patterns:
  - Inverted U-shaped relationship between total pass through and consumption share affected:
    - Pass through small when reforms affect <10% of consumption.
    - Pass through highest—and close to full—for reforms affecting around 50% of consumption.
    - Pass through declines for reforms affecting >65%, though declining part not statistically significant given large SEs.
  - Precise statements:
    - Pass through not significantly different from zero when VAT applies to consumption share <10% or >65%.
    - Given few reforms affecting >60% of consumption, null of full pass through cannot be rejected for reforms affecting >30% of consumption.
- Implication: pass through does not increase monotonically with breadth; highest pass through occurs at intermediate coverage.

### VIII. Asymmetry between VAT increases and decreases (Table 4)
- Sample split: VAT increases 1,009 changes; VAT decreases 222 changes; observations 155,385.
- Cumulative estimates (Pre-Reform, Contemporaneous, Post-Reform, Total; clustered SEs in parentheses):
  - Pooled: 0.08 (0.07); 0.15*** (0.03); 0.09 (0.06); Total 0.32*** (0.09)
  - Increase: 0.13 (0.11); 0.16*** (0.04); 0.10 (0.09); Total 0.40*** (0.15)
  - Decrease: 0.01 (0.07); 0.14*** (0.04); 0.09 (0.06); Total 0.24*** (0.08)
- Wald test p-values for equality across increases and decreases:
  - Pre-Reform: 0.30
  - Contemporaneous: 0.74
  - Post-Reform: 0.88
  - Total: 0.33
- Key findings:
  - Estimated pass through for increases and decreases are close; Wald tests cannot reject equality.
  - Timing asymmetry: point estimate of pre-reform pass through larger for increases (0.13) than decreases (0.01), but not statistically different.
  - Results contrast with some prior single-country findings of asymmetry.

### IX. Endogeneity and fiscal consolidation episodes (Table 7)
- Identification strategy: use IMF WEO narrative-based fiscal reform database to classify reforms as "identified exogenous" (occurring in fiscal consolidation episodes) versus "others".
- Sample: 16 advanced economies for period 1980-2009; 98,287 observations.
- Sum of coefficients (Pre-Reform, Contemporaneous, Post-Reform, Total):
  - Identified Exogenous (198 VAT changes): 0.27 (0.20); 0.22*** (0.07); 0.27 (0.42); Total 0.76* (0.44)
  - Others (636 VAT changes): 0.50* (0.28); 0.47*** (0.09); 0.41 (0.27); Total 1.39*** (0.33)
- Wald test p-values for equality across groups:
  - Pre-Reform: 0.50
  - Contemporaneous: 0.01
  - Post-Reform: 0.77
  - Total: 0.20
- Interpretation:
  - Contemporaneous pass through lower for identified exogenous reforms (difference significant at 1 percent).
  - Total pass through lower for exogenous reforms (0.76* vs 1.39***), but equality of totals cannot be rejected (p-value 0.20).
  - Endogeneity may inflate contemporaneous OLS pass-through estimates, but main long-run conclusions are robust.

### X. Robustness: measurement error and HICP-CT instrument construction
- Measurement error concern: imperfect matching of VAT reforms to COICOP categories; reduced rates and reclassifications often apply only to subsets of a category.
- HICP-CT characteristics:
  - Calculated holding indirect taxes constant within the calendar year and assuming full and immediate pass through; difference in growth rates between HICP and HICP-CT reflects tax changes (VAT and other indirect taxes).
  - Available for aggregated consumption groups from 2002 to 2013.
- Instrumental variables approach:
  - Under classical errors-in-variables (CEV), instrument own VAT change measure with HICP-CT-derived measure.
  - Correlations between HICP-CT and own VAT measure: Standard 0.81; Reduced 0.64; Reclassification 0.66.
  - First-stage Angrist-Pischke F-Stats reported: Standard 194.7; Reduced 278.7; Reclassification 29.1.
  - OLS=IV Wald test p-values: Standard 0.17; Reduced 0.61; Reclassification 0.29 (cannot reject equality).
- HICP-CT construction relies on national statistical agencies’ identification of indirect tax changes and the Laspeyres-type index algebra described in annex.

### XI. Main policy-relevant conclusions
- Heterogeneity by type:
  - Standard rate changes: conventional treatment of 100 percent pass through is "a reasonable starting point"; OLS total 1.30*** (0.42), IV total 1.98** (0.82) — evidence of large pass through and sometimes overshifting for standard changes.
  - Reduced rate changes: pass through substantially below one, "perhaps around 30 percent"; OLS total 0.30 (0.28), IV total 0.16 (0.35).
  - Reclassifications: pass through close to zero; OLS total 0.05 (0.08), IV total 0.23 (0.22).
- Breadth of VAT reform:
  - Pass through tends to increase with breadth up to intermediate coverage (around 50 percent of consumption), then flattens or declines at very high coverage — inverted U-shape.
- Goods type and dynamics:
  - Pass through greater for durables (anticipation important) than non-durables.
  - Significant anticipation and lagged responses imply timing matters for incidence analysis.
- Asymmetry:
  - No systematic tendency for pass through to be greater for tax increases than for tax cuts.
- Practical implications for policymakers:
  - Treating standard-rate VAT changes as fully passed through is broadly justified, but heterogeneity across reform types cautions against blanket assumptions.
  - Benefits of reduced rates are not fully passed on to consumers; distributional analyses assuming full pass through of reduced rates are likely to overstate incidence on households.
  - Case-by-case assessment needed given unaddressed determinants: announcement dates, market structure, trade intensity, compliance, business cycle, exchange rates, and monetary policy (especially outside monetary unions).

*Source: _wp15214 (PDF chapter/section; Tables and Figures Inventory, Sections I–II.C, Appendix Table A1, Tables 2–7, robustness and annex text as provided).*

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

### _wp15214 - References

### Tables and Figures Inventory
- Tables
  - Table 1: Summary Statistics
  - Table 2: Consumer Price Elasticities
  - Table 3: Regression Results of Pass Through by Type of VAT Rate Change
  - Table 4: Pass Through for VAT Increases and Decreases
  - Table 5: Pass Through by Type of VAT Change – OLS Estimates
  - Table 6: Pass Through by Type of VAT Change – IV Estimates
  - Table 7: Pass Through During Fiscal Consolidation Episodes
  - Table A1. Country List and VAT Reforms
  - Table A2. Consumption Categories

- Figures
  - Figure 1: Elasticity of Price Responses in Two-Year Interval around the VAT Change
  - Figure 2: Cumulative Pass Through by Type of VAT Rate Change
  - Figure 3: Cumulative Pass Through for Durables versus Non-Durables
  - Figure 4: Number of VAT Rate Changes by Type of Reform and Total Consumption Share Targeted
  - Figure 5: Pass Through by Share of Consumption Affected
  - Figure A1. Comparison of VAT Changes by Source

- Appendices
  - Appendix A: Constructing HICP-CT-based VAT Rate Changes
  - Appendix B: The Elasticity of Substitution and Derivation of Equation (5)

### I. INTRODUCTION — Core points from the text
- Common policy practice assumes indirect tax changes are fully and exactly passed through to consumer prices; references include Bird and Gendron (2007) and Lustig et al. (2013).
- Theory permits a range of pass through outcomes: undershifting, overshifting, and even price declines (Edgeworth paradox).
- Timing of pass through may include anticipation effects and sluggish response.
- Existing empirical literature finds heterogeneous results: many studies find less than full pass through, while others find full pass through or overshifting.
- Limitations of many prior studies:
  - Often single-country, narrow-set commodity focus.
  - Use domestic untaxed goods as controls, raising identification concerns when cross-price effects exist.
  - Limited ability to answer the core policy question for standard VAT rate changes that affect a broad consumption base.
- This paper exploits a large panel dataset for the Eurozone (1999 to 2013):
  - Monthly price and tax data for 67 consumption items.
  - Covers 1,231 VAT changes.
  - Pooled data from 17 integrated economies sharing a common currency used to provide more credible counterfactuals (same commodity in other countries as controls).
- Research scope enabled by dataset:
  - Estimate pass through for standard rate changes, reduced rates, and reclassifications.
  - Test heterogeneity by scope (share of consumption affected) and asymmetries between VAT increases and decreases.
  - Explore dynamics, including anticipation and sluggish response, and differences for durables versus non-durables.
- Robustness assessment includes measurement error and endogeneity — issues not previously tested in incidence studies.
- Paper organization:
  - Section II: theoretical framework, methodology, data.
  - Section III: main empirical results and dynamics.
  - Section IV: robustness checks.
  - Section V: conclusion.

### II.A Theory — Key equations and intuition
- Definition: For commodity i with consumer price p_i and producer price w_i, taxed at ad valorem tax-exclusive rate τ_i (so p_i = (1+τ_i) w_i), degree of pass through σ_i defined as
  - σ_i ≡ (Δp_i / p_i) / (Δτ_i / (1+τ_i))  (equation (1) in text).
  - Full pass through corresponds to σ_i = 1; zero pass through corresponds to σ_i = 0.
- Partial-equilibrium formula (two-commodity example) yields familiar expression:
  - σ_1 = θ^0_11 / (θ^0_11 + θ^m_11)  (equation (2)), where θ^0_ii is own-price supply elasticity and θ^m_ii denotes minus own-price demand elasticity.
- General-equilibrium extension (both goods considered) gives
  - σ_1 = θ^0_11 / (θ^0_11 + θ^m_11 + correction term involving cross-price elasticities)  (equation (3)).
  - If cross-price elasticities θ^m_12 and θ^m_21 have same sign, pass through is unambiguously higher than partial-equilibrium formula suggests.
- Pass through of untaxed good 2 given by σ_2 formula in equation (4); movements in p_2 shift demand for taxed good and affect σ_1.
- Appendix B leads to a compact expression (equation (5)):
  - σ_1 = θ^0_1 / (θ^0_1 + σ{ 1 + (η_1 / ε) (s_1) } ), where symbols represent elasticity of substitution ε (ε > 0), income elasticity η_1, and budget share s_1 — exact form preserved as in source.
- Implications:
  - Effect of elasticity of substitution on pass through is ambiguous — opposing forces via demand elasticity and general equilibrium impacts on alternative good prices.
  - Wider tax base may increase pass through in some cases (e.g., homothetic CES with ε = 1) but not guaranteed in general; relationship between budget share and pass through may be non-monotonic.
  - Imperfect competition and product differentiation complicate pass through; curvatures of demand and market behavior matter (Weyl and Fabinger (2013) literature).
  - Timing effects: menu costs can produce lags; pre-announcement can induce lead effects via changed consumer timing and firm pricing.

### II.B Empirical Model — Specification
- Reduced-form monthly model for log changes of consumer price index Δln(p_{i,c,t}):
  - Δln(p_{i,c,t}) = sum_{j=-12}^{12} β_j (1 + Δln τ_{c,t+j}) + X_{c,t} γ + α_c + α_i + δ_t + ε_{i,c,t}  (equation (6) in text).
  - β_j measures impact on consumer price at time t of a VAT change in month t+j; j ∈ {−12,...,12}.
  - Sum of β_j over two-year window interpreted as long-run pass through elasticity (consistent with equation (1)).
- Controls X include current monthly unemployment rate and real quarterly GDP growth.
- Fixed effects: country fixed effects α_c, consumption category fixed effects α_i, and month fixed effects δ_t capturing Eurozone-wide trends and seasonality.
- Using only Eurozone countries removes need to control for exchange rate movements or monetary policy variations.
- Note: Reduced-form cannot distinguish between different supply models nor separately identify underlying supply and demand elasticities without further restrictions.
- Regressions with lead and lags up to 24 months suggest no significant effects outside the ±12 month window.

### II.C Data and Estimation — Data sources and construction
- Price data: Harmonized Index of Consumer Prices (HICP) from Eurostat, monthly, by COICOP classification.
  - Sample comprises 67 COICOP consumption categories for which prices were market driven and not VAT-exempt in most countries.
- Expenditure shares: country-specific expenditure shares from Eurostat; shares are constant within calendar years and updated annually.
- VAT rates dataset: novel monthly dataset constructed for the paper matching VAT rates by month to the 67 consumption categories in 17 Eurozone countries (1999 to 2013).
  - Primary source: "VAT Rates Applied in the Member States of the European Union" (European Commission).
  - Complemented, where necessary, with information from the International Bureau for Fiscal Documentation (IBFD).
  - When annual publications lacked precise month detail, January 1 was assigned as date of reform.
- Dataset covers 1,231 VAT changes across the 17 countries over 1999 to 2013.
- The full list of countries and number of VAT reforms by type is provided in Table A1 (appendix).

*Source: _wp15214 - References, PDF (pages and tables/figures list and Sections I–II.C as provided).*

### Appendix Table A1.

### Appendix Table A1

### Summary statistics (VAT rates, price indices, controls, panel)
- Sample size: total unbalanced panel consists of 155,385 monthly observations. Longest individual panels: January 1999 to September 2013.
- VAT level statistics (Number of obs, Mean, St. Dev., Min, Max):
  - Standard: 107,089; 19.24; 2.14; 15.00; 23.00
  - Reduced: 41,667; 7.92; 3.40; 2.10; 17.00
  - Zero-rated: 3,341; 0.00; 0.00; 0.00; 0.00
- VAT changes (expressed as log changes in the tax factor):
  - Standard changes: 972 observations; Mean 1.05; St. Dev. 0.96; Min -0.83; Max 2.55
  - Reduced changes: 191 observations; Mean 0.42; St. Dev. 1.98; Min -6.45; Max 6.45
  - Reclassification: 68 observations; Mean -0.32; St. Dev. 8.58; Min -13.38; Max 14.87
  - VAT increase: 1,009 observations; Mean 1.67; St. Dev. 1.69; Min 0.41; Max 14.87
  - VAT decrease: 222 observations; Mean -3.02; St. Dev. 3.49; Min -13.38; Max -0.41
- Price changes and controls (Number of obs, Mean, St. Dev., Min, Max):
  - HICP Index: 155,385; 0.01; 1.34; -117.96; 142.82
  - Unemployment rate: 155,385; 8.22; 4.15; 1.70; 27.60
  - Real Quarterly GDP growth: 155,385; 0.38; 1.20; -8.50; 7.90
  - Consumption weights: 155,385; 3.38; 3.23; 0.00; 17.56
- Other dataset notes:
  - Approximately 70 percent of consumption items are taxed at the standard VAT rate, accounting for slightly less than 65 percent of the value of total consumption.
  - Standard VAT rates range between 15 and 23 percent, with an average rate of close to 20 percent. Reduced rates vary between 2.1 and 17 percent, averaging 8 percent.
  - There are 65 distinct VAT reform episodes generating 1,231 individual VAT rate changes (examples and counting conventions clarified in text).
  - Prior to estimating (6), all price indices are de-seasonalized and de-trended. Regressions are weighted by the share of each consumption category in the national consumption basket. Standard errors in (6) are clustered at the panel level.

### Results — Average VAT pass through (Table 2)
- Estimation variants: (1) no controls/fixed effects, (2) country-month-category fixed effects, (3) fixed effects plus unemployment rate and real quarterly GDP growth. Observations: 155,385 in all columns.
- Cumulative pass through estimates (sum of coefficients over 2-year window; standard errors in parentheses):
  - Column (1):
    - Pre-Reform: 0.12* (0.06)
    - Contemporaneous: 0.17*** (0.03)
    - Post-Reform: 0.11* (0.06)
    - Total: 0.40*** (0.09)
    - R-squared: 0.00
  - Column (2):
    - Pre-Reform: 0.07 (0.07)
    - Contemporaneous: 0.15*** (0.03)
    - Post-Reform: 0.07 (0.06)
    - Total: 0.29*** (0.10)
    - Fixed Effects: Yes; R-squared: 0.02
  - Column (3):
    - Pre-Reform: 0.08 (0.07)
    - Contemporaneous: 0.15*** (0.03)
    - Post-Reform: 0.09 (0.06)
    - Total: 0.32*** (0.09)
    - Controls: Unemployment Rate coefficient -0.01*** (0.00); Real quarterly GDP growth coefficient -0.01 (0.01)
    - Fixed Effects: Yes; R-squared: 0.02
- Key interpretations:
  - The cumulative final pass through ranges from 0.29 to 0.40 across specifications; column (3) (preferred) shows Total = 0.32*** (0.09).
  - Slightly less than half of the total effect occurs contemporaneously; meaningful lead and lag effects exist (anticipation and delayed pass through).
  - The null of full pass through (elasticity = 1) is rejected; point estimates imply only around one-third of a VAT change is passed to consumer prices.
  - Non-contemporaneous effects matter: roughly one-third to one-half of full effects occur before or after reform.

### Pass through by type of VAT change (Table 3 and Figure 2)
- Estimation separates types: Standard rate changes, Reduced rate changes, Reclassifications. Observations: 155,385; Number of VAT changes: Standard 972, Reduced 191, Reclassification 68.
- Cumulative pass through by type (Pre-Reform, Contemporaneous, Post-Reform, Total) with standard errors in parentheses and significance:
  - Standard:
    - Pre-Reform: 0.71*** (0.22)
    - Contemporaneous: 0.30*** (0.06)
    - Post-Reform: 0.38** (0.17)
    - Total: 1.39*** (0.36)
  - Reduced:
    - Pre-Reform: -0.13 (0.25)
    - Contemporaneous: 0.36*** (0.13)
    - Post-Reform: 0.07 (0.18)
    - Total: 0.30 (0.23)
  - Reclassification:
    - Pre-Reform: -0.06 (0.06)
    - Contemporaneous: 0.10*** (0.03)
    - Post-Reform: 0.04 (0.06)
    - Total: 0.08 (0.07)
  - Column (4) reports Wald test p-values for equality across the three types by period; tests reject equality for pre-reform, contemporaneous, and total cumulative effects (p-values shown as 0.00 in table).
- Key findings and interpretations:
  - Standard rate changes:
    - Large positive anticipation effects starting about nine months before the VAT change; cumulative lead effects add up to 0.71***.
    - Contemporaneous effect 0.30***; post-reform adds 0.38**.
    - Total estimated pass through 1.39***, implying overshifting of standard VAT changes by some 39 percent; with standard error 0.36, cannot reject full cumulative pass through in long run.
  - Reduced rate changes:
    - No significant anticipation effects (Pre-Reform = -0.13).
    - Contemporaneous effect 0.36***; post-reform adds 0.07.
    - Total pass through 0.30 (0.23), statistically different from zero but statistically different from unity (full pass through rejected).
  - Reclassifications:
    - Virtually no effect over two-year window aside from contemporaneous pass through of 0.10***.
    - Total pass through 0.08 (0.07); null of zero total pass through cannot be rejected.
  - Wald tests: pass through for standard VAT changes is significantly larger than for reduced changes and for reclassifications (pre-reform, contemporaneous, total).

### Pass through and durables versus non-durables
- Method: split consumption categories into durable and non-durable; run regression with fixed effects and controls.
- Key patterns (Figure 3 summary):
  - Durable items:
    - Significant anticipation effects: pass through of 35 percent is achieved before the VAT change is enacted.
    - Long-run overall pass through around 50 percent (larger than for non-durables), still significantly different from full pass through.
  - Non-durable items:
    - No signs of anticipation effects.
    - Contemporaneous and lagged patterns similar to durables, but overall long-run pass through smaller than for durables.
- Related sample features:
  - Close to 45 percent of non-durables are taxed at either a reduced or zero-rate.
  - More than 95 percent of durables are taxed at the standard rate.

### Pass through and scope of VAT reform (consumption share affected)
- Empirical strategies for heterogeneity by consumption share:
  - Interaction of VAT rate change with indicator bins for quantiles of total consumption share affected. Bins: 0-10%, 10-20%, 20-26%, 26-40%, 40-50%, 50-60%, 60-74%.
  - Interaction of VAT change with a cubic polynomial in the total consumption share affected.
- Main empirical patterns (Figure 5 summary):
  - Both specifications imply an inverted U-shaped relationship between total pass through and the consumption share affected:
    - Pass through is relatively small for reforms affecting a small consumption share (<10%).
    - Pass through is highest—and close to full—for reforms affecting around half of all consumption.
    - Pass through declines again for reforms affecting a very large share (>65%), but this declining part is not statistically significant given large standard errors.
  - More precise statements:
    - Pass through is not significantly different from zero when the VAT applies to a consumption share less than 10 percent or more than 65 percent.
    - Given few reforms affecting more than around 60 percent of consumption, the null of full pass through cannot be rejected for VAT reforms affecting more than 30 percent of consumption.
  - Implication: the presumption that pass through increases monotonically with the breadth of the VAT change is not supported; instead, an inverted U-shape with highest pass through at intermediate coverage is observed.

*Source: Appendix Table A1 and accompanying text in the provided content unit.*

### Section II warns that this is not necessarily the case: the wider the set of commodities to which a

### _wp15214 - Section II warns that this is not necessarily the case: the wider the set of commodities to which a

### E. Asymmetric Responses? Pass Through for Rate Increases and Decreases
- Question addressed: whether pass through of VAT reform is asymmetric between rate increases and rate decreases, especially if downward nominal rigidities matter.
- Method: sample split into VAT rate increases and VAT rate decreases; specification with fixed effects and controls for unemployment rate and real GDP growth; observations weighted by consumption shares; regressions include 12 leads and lags of VAT rate changes.
- Benchmark (column (1), pooled changes):
  - Cumulative pass through = 0.32 (same as column (3) of Table 2).
- Estimates by sign (Table 4):
  - Pre-Reform:
    - Pooled: 0.08 (standard error (0.07))
    - Increase: 0.13 (0.11)
    - Decrease: 0.01 (0.07)
    - Wald test p-value for equality (column (4)): 0.30
  - Contemporaneous:
    - Pooled: 0.15*** (0.03)
    - Increase: 0.16*** (0.04)
    - Decrease: 0.14*** (0.04)
    - Wald test p-value: 0.74
  - Post-Reform:
    - Pooled: 0.09 (0.06)
    - Increase: 0.10 (0.09)
    - Decrease: 0.09 (0.06)
    - Wald test p-value: 0.88
  - Total:
    - Pooled: 0.32*** (0.09)
    - Increase: 0.40*** (0.15)
    - Decrease: 0.24*** (0.08)
    - Wald test p-value: 0.33
  - Number of VAT changes:
    - Pooled: 1231
    - Increase: 1009
    - Decrease: 222
- Statistical notes:
  - Number of observations = 155,385.
  - Additional controls: unemployment rates, real quarterly GDP growth, fixed effects for country, consumption category and month.
  - Panel-clustered standard errors in parentheses.
  - Significance notation: * p<10%, ** p<5%, *** p<1%.
  - Column (4) reports p-values from Wald tests of equality of coefficients in columns (2) and (3).
- Key findings:
  - Estimated pass through for VAT increases and decreases are close to each other; Wald tests cannot reject equivalence.
  - Main asymmetry is timing: pre-reform pass through point estimate larger for rate increases (0.13) than for decreases (0.01), but Wald test cannot reject equivalence.
  - Results contrast with previous findings (e.g., Batista Politi and Mattos (2011) for Brazil) that found greater pass through for increases than decreases.
- Robustness checks noted: main conclusions unchanged when including standard rate changes only, or when excluding reclassifications.

### IV. Robustness — A. Measurement Error
- Potential problem: measurement error in the VAT reform variable due to imperfect matching of VAT and COICOP categories; reduced VAT rates or reclassifications may apply to subsets of a COICOP category.
- Implication: mismatches and incorrect weighting by consumption share could be especially important for narrowly-applicable VAT reforms (reduced rates, reclassifications) and could drive differences in estimated pass through by type.
- Alternative data source used to address measurement error: Eurostat’s Harmonized Index of Consumer Prices at Constant Tax rates (HICP-CT) series.
  - HICP-CT characteristics:
    - Compiled by member states’ statistical agencies.
    - Calculated holding indirect taxes constant throughout the current calendar year and assuming full and immediate pass through of indirect tax changes into consumer prices.
    - Provides an independently-derived measure of VAT changes via the difference in monthly growth rates between the constant-tax price index and the regular HICP price index.
    - Drawback: available only for more aggregated consumption categories (e.g., ‘processed food’, ‘unprocessed food’, ‘non-energy industrial goods’, ‘energy’, ‘services’) and only from 2002 to 2013.
- Instrumental variables approach under the classical errors-in-variables (CEV) assumption:
  - Observed VAT rate change = true VAT rate change + measurement error uncorrelated with true value.
  - Instrument own VAT change measure with HICP-CT-based measure; assume both are sum of true VAT change and uncorrelated measurement error terms.
  - Purpose: recover unbiased estimate of pass through (Ashenfelter and Krueger 1994; Wooldridge 2010).
  - Applied separately for standard VAT rate changes, reduced VAT rate changes, and reclassifications to assess attenuation bias.
  - Interpretation: statistically significant difference between OLS and IV estimates indicates systematic measurement error in the VAT change variable.
- Correlations between HICP-CT-derived series and own VAT rate change measure (by type of change):
  - Standard VAT rates: 0.81
  - Reduced rates: 0.64
  - Reclassifications: 0.66
  - Interpretation: ratio of information to noise likely highest for standard rate changes and lowest for reduced rate changes and reclassifications.
- OLS estimates using HICP-CT-coincident sample (Table 5 summary):
  - Most results qualitatively similar to Table 3.
  - Main difference: timing of pass through for reduced VAT rate changes:
    - Pre-reform negative pass through is much larger in absolute value.
    - Contemporaneous and lagged pass through are also higher.
    - Total pass through for reduced rate changes remains at 0.30.
  - Total estimated pass through remains statistically different from one at the 5 percent level.
  - Estimated pass through for reclassifications remains insignificant.
  - Estimated pass through for standard VAT rate changes decreases from 1.39 to 1.30 in this smaller sample.

*Italic: Source — content unit _wp15214 - Section II warns that this is not necessarily the case: the wider the set of commodities to which a*

### Annex A describes the algebra underlying the construction of the HICP-CT series and how to recover the

### _wp15214 - Annex A describes the algebra underlying the construction of the HICP-CT series and how to recover the

### Pass-through estimates by type of VAT change (OLS)
- Table 5 presents OLS estimates of price elasticity with respect to VAT rate changes, summarized as sums of coefficients over periods (Pre-Reform, Contemporaneous, Post-Reform) with 12 leads and lags and 104,212 observations. Panel-clustered standard errors in parentheses.
- Standard rate changes (Column (1), Number of VAT Changes: 666)
  - Pre-Reform: 0.74*** (0.24)
  - Contemporaneous: 0.27*** (0.07)
  - Post-Reform: 0.29 (0.21)
  - Total: 1.30*** (0.42)
- Reduced rate changes (Column (2), Number of VAT Changes: 167)
  - Pre-Reform: -0.45 (0.29)
  - Contemporaneous: 0.59*** (0.15)
  - Post-Reform: 0.17 (0.30)
  - Total: 0.30 (0.28)
- Reclassification changes (Column (3), Number of VAT Changes: 51)
  - Pre-Reform: -0.10* (0.06)
  - Contemporaneous: 0.13*** (0.03)
  - Post-Reform: 0.02 (0.08)
  - Total: 0.05 (0.08)
- Column (4) reports p-value from Wald test of equality of coefficients across columns (1)-(3): 0.00 (for pre-reform row) and 0.00 (for contemporaneous row) as shown in table (note: table formatting shows zeros in Column (4) for some rows).
- Notes: regressions include unemployment rates, real quarterly GDP growth and fixed effects for type of rate change, country, consumption category and month. Individual observations weighted by share of total national consumption.

### Measurement error and IV robustness
- Approach: Compare OLS estimates with IV estimates where VAT changes are instrumented using VAT changes from HICP-CT series (Eurostat).
- Instrument strength: F statistics for excluded instrument in the contemporaneous VAT change equation "well exceeding 10" (reported in Table 6).
- Wald tests reported at bottom of Table 6 cannot reject equality of pass through between OLS and IV estimates for each VAT change type — implying measurement error/attenuation bias is unlikely to drive main results.

### Pass-through estimates by type of VAT change (IV)
- Table 6 presents IV second-stage sums of coefficients; VAT rate changes instrumented with HICP-CT-derived VAT changes. Number of observations is 104,212. Panel-clustered standard errors in parentheses.
- Standard rate changes (Column (1), Number of VAT Changes: 666)
  - Pre-Reform: 1.10** (0.50)
  - Contemporaneous: 0.37*** (0.13)
  - Post-Reform: 0.51* (0.30)
  - Total: 1.98** (0.82)
- Reduced rate changes (Column (2), Number of VAT Changes: 167)
  - Pre-Reform: -0.50** (0.24)
  - Contemporaneous: 0.59*** (0.23)
  - Post-Reform: 0.08 (0.39)
  - Total: 0.16 (0.35)
- Reclassification changes (Column (3), Number of VAT Changes: 51)
  - Pre-Reform: -0.38* (0.22)
  - Contemporaneous: 0.21** (0.09)
  - Post-Reform: 0.39 (0.41)
  - Total: 0.23 (0.22)
- First-stage diagnostics (Angrist-Pischke F-Stat for excluded instrument in contemporaneous VAT change regressions):
  - Standard: 194.7
  - Reduced: 278.7
  - Reclassification: 29.1
- OLS=IV (p-value) reported:
  - Standard: 0.17
  - Reduced: 0.61
  - Reclassification: 0.29
- Interpretation: IV point estimates broadly similar to OLS; standard rate total pass through larger under IV (1.98**), but standard errors larger so Wald test cannot reject full pass through. Reduced and reclassification totals remain statistically different from one at the 5 percent level in both OLS and IV results.

### Endogeneity: identification and results
- Concern: VAT reforms may be endogenous—undertaken for fiscal consolidation or stimulus, potentially correlated with unobserved consumer price shocks and biasing OLS (e.g., stimulus-motivated VAT cuts could coincide with softer prices, leading to overestimate of pass through).
- Identification strategy: Use IMF WEO action-based fiscal reform database (Romer and Romer (2010) narrative approach) to classify VAT reforms as:
  - Identified exogenous (consolidation-driven): reforms occurring in same year as broader fiscal consolidation in WEO (2010) dataset.
  - Others: remaining reforms.
- Sample for this exercise: 16 advanced economies for period 1980-2009 (WEO dataset), focusing on standard and reduced rate changes.
- Table 7: Pass through during fiscal consolidation episodes (Number of observations: 98,287). Column (3) shows Wald test p-value for equality of coefficients across the two groups.

Table 7 estimates (sum of coefficients):
- Identified Exogenous (Column (1), Number of VAT Changes: 198)
  - Pre-Reform: 0.27 (0.20)
  - Contemporaneous: 0.22*** (0.07)
  - Post-Reform: 0.27 (0.42)
  - Total: 0.76* (0.44)
- Others (Column (2), Number of VAT Changes: 636)
  - Pre-Reform: 0.50* (0.28)
  - Contemporaneous: 0.47*** (0.09)
  - Post-Reform: 0.41 (0.27)
  - Total: 1.39*** (0.33)
- Column (3) Wald test p-values for equality across columns (1) and (2):
  - Pre-Reform: 0.50
  - Contemporaneous: 0.01
  - Post-Reform: 0.77
  - Total: 0.20
- Interpretation:
  - Pre-reform estimates lower for exogenous reforms (consistent with expectation that business-cycle–driven reforms raise OLS pass-through).
  - Contemporaneous effect for exogenous reforms is about half of that for others, difference statistically significant at 1 percent (p-value 0.01).
  - Overall, final pass-through lower for exogenous reforms (0.76* vs 1.39***), but Wald test cannot reject equality of totals (p-value 0.20).
  - Conclusion: endogeneity unlikely to fundamentally change long-run cumulative pass-through conclusions, though it may imply some overestimation of contemporaneous pass-through in earlier results.

### Main conclusions and policy-relevant findings
- Heterogeneity by type of VAT change:
  - Standard rate changes: assumption of 100 percent pass through is a reasonable starting point; OLS total 1.30*** (0.42), IV total 1.98** (0.82).
  - Reduced rate changes: pass through significantly less than one, "perhaps around 30 percent"; OLS total 0.30 (0.28), IV total 0.16 (0.35).
  - Reclassifications: pass through seems close to zero; OLS total 0.05 (0.08), IV total 0.23 (0.22).
- Breadth of consumption base: pass through appears to increase with breadth of coverage over most of the practically relevant range; but relationship may flatten or decrease at highest levels of coverage, consistent with theoretical expectations.
- Goods type: pass through seemingly greater for durables than non-durables.
- Asymmetry: no systematic tendency for pass through to be greater for tax increases than for tax cuts.
- Dynamics and anticipation:
  - For standard rate changes, majority of pass through occurs in months before the VAT change (significant anticipation effects), particularly for durables.
  - For reduced VAT rates, anticipation effects appear weaker.
- Implications for policy makers and analysts:
  - Conventional treatment of standard rate changes (100 percent pass through) is broadly validated.
  - Conventional assumption that benefits of reduced rates are fully passed on to poorer households is cautioned against, since reduced-rate pass through is substantially below one.
  - Practical application requires case-by-case consideration; several potentially important determinants are not fully addressed due to data limitations:
    - Announcement dates (could further unpack dynamics)
    - Market structure
    - Trade intensity
    - Degree of compliance
    - Business cycle, exchange rate and monetary policies (in non-monetary-union countries)
- Overall message: marked heterogeneities in VAT pass through exist; assuming uniform 100 percent pass through is called into question by the empirical evidence presented.

*Source: _wp15214 - Annex A describes the algebra underlying the construction of the HICP-CT series and how to recover the implied VAT rate changes.*

### References

### _wp15214 - References

### HICP-CT construction and use as an instrument
- HICP-CT series are constructed to provide consumer price indexes that reflect underlying economic activity and are not affected by changes in tax rates.
- National statistical agencies identify changes in national-level indirect taxes (including VAT, excise duties, consumption and other specific and special taxes); under the assumption of full and instantaneous pass-through, hypothetical underlying producer prices are recovered and then grossed up by taxes that applied in a reference period to yield the HICP-CT index.
- The difference in growth rates between the HICP and HICP-CT series in any time period reflects only changes in taxes; these implied tax-change values encompass VAT changes and other ad valorem and specific tax rate changes as well.
- The authors restrict the HICP-CT-derived tax rate change series to months in which their own series records a VAT rate change to reduce contamination from non-VAT tax changes, while noting that perfect correlation is not expected because VAT reforms can coincide with other indirect tax changes or due to matching errors.
- The HICP-CT is calculated as a Laspeyres-type index (formula provided in the source) where:
  - ݍ௜௧߬(ݍ) denotes the hypothetical producer price of sub-item ݅ in period ݐ, recovered by assuming full and instantaneous pass-through;
  - ߬௜଴ is the VAT rate applicable during a reference period 0;
  - ߛ௜଴ is an ad valorem sales tax rate;
  - ߙ௜଴ is any specific tax;
  - ݓ௜ is the consumption weight of sub-items belonging to a COICOP category.
- Under simplifying assumptions (uniform VAT and other taxes within a COICOP category), HICP-CT and HICP can be written compactly as:
  - HICP-CT୲ = ݍ෤௧/ݍ෤଴
  - HICP୲ = (1+߬௧)ݍ෤௧ (1+߬଴)ݍ෤଴
- The paper concludes that, provided changes in VAT rates are not systematically correlated with deviations from the uniform-tax assumption within COICOP categories, the HICP-CT series constitutes a suitable instrument for the authors’ VAT reform measure.

### Appendix B — elasticity of substitution and derivation of equation (5)
- Appendix B defines the elasticity of substitution between two goods using compensated demands ܪ(݌,݊) and derives relationships between compensated elasticities and budget shares.
- Key derivation steps (equation labels preserved as in source):
  - Definition: ≡ߪ ܪ(݈߲݊)ܪ/ଶ ݌(݈߲݊)  (A.1)
  - Differentiation and manipulation using symmetry of compensated demand effects and homogeneity of degree zero lead through (A.2)–(A.5).
  - Expressed in terms of compensated elasticities ߝ and budget shares yields (A.6) and, using the Slutsky equation, (A.7).
  - Substituting (A.7) into (3) and assuming ߝ௦ଵଵ = ߝ௦ଶଶ leads to (A.8) and, after algebraic simplification, to (A.9):
    - ߛଵ = ߝ௦ ߝ௦ ߪ൅ [ߝ௦െߪ൅ (ߟെߪଶ)߱ଶ ߝ௦െߪ൅ (ߟെߪଵ)߱ଵെ(ߪെηଶ)߱ଶ ]
  - The denominator simplifies to ߝ1൅ଶ and the numerator can be written as ߝ௦൅1൅(ߟെߪଵ)߱ଵ, using the adding-up implication ߱ଵߟଵ݅߱ଶߟଶ = 1.
- The appendix provides the formal microeconomic foundation for the paper’s equation (5), linking substitution elasticities, compensated demand effects, and tax pass-through expressions under the stated symmetry and homogeneity assumptions.

### Data tables and figures (content and purpose)
- Table A1: Country list and VAT reforms — presents, for each country, the first year in sample, number of VAT reforms, and number of consumption categories affected by Standard, Reduced, and Reclassification changes (data sources: EC (various years) and IBFD). Notes explicitly state: “Data from EC (various years) and IBFD.”
- Table A2: Consumption Categories — lists COICOP codes, consumption items, Average Expenditure Share, Average VAT Rate, Number of Durable, Standard Rate Change, Reduced Rate Change, and Reclassification. Notes: “Data on VAT rates taken from EC (various years) and IBFD. Price, unemployment rates and consumption weights are taken from Eurostat.”
- Figure A1: Comparison of VAT Changes by Source — three panels:
  - a) Standard Rate Changes
  - b) Reduced Rate Changes
  - c) Reclassification
  - Notes: scatter plots of VAT changes weighted by consumption share as measured in the sample against VAT changes derived from the HICP-CT data and the linear relation estimated via OLS by type of VAT reform separately.

*Source: _wp15214 - References (appendices, tables and figures) — content as provided in the PDF chapter/section.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15214.pdf_