## wp17241

## Source details

**Canonical URL:** [wp17241](https://www.imf.org/-/media/files/publications/wp/2017/wp17241.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17241.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17241.pdf.json)

---

### Introduction and motivation
- Research question: quantify transmission of domestic government spending shocks across 10 euro area countries and identify determinants and channels of fiscal spillovers.
- Policy context: debate on fiscal expansion within the euro area given slow growth and ultra-low interest rates; need to assess cross-border spillovers and country heterogeneity.

### Data, sample, and model
- Sample and frequency:
  - Ten euro area countries: Germany, France, the Netherlands, Belgium, Austria, Italy, Spain, Ireland, Finland, and Portugal.
  - Balanced quarterly panel covering 1999-2016.
- Variables included:
  - Endogenous: government expenditures (sum of consumption and investment), government revenues net of transfers, government debt, output, current account balance, real effective exchange rate (REER), benchmark interest rate (yield on 10-year government bonds).
  - Exogenous controls: Federal fund rate, Eonia, world output, oil prices.
  - Transformations: interest rate in levels; remaining variables in real per capita terms and natural logarithms; current account rescaled prior to log transformation.
- Model specification and dimensionality reduction:
  - Heterogeneous panel VAR: 푦ᵢₜ = 퐷ᵢ(𝐿)푌ₜ₋₁ + 퐹ᵢ(𝐿)푊ₜ + 푒ᵢₜ with cross-unit lagged interdependencies allowed.
  - Error covariance structured as 퐸ₜ ~ N(0, P ⨂ Ω); prior for P centered at bilateral trade weights.
  - Coefficients factorized via 훿 = Ξ휃 + 푢 with Ξ휃 decomposed into mutually orthogonal factors to transform the panel VAR into a parsimonious SUR.
- Identification:
  - Two sign-restriction strategies:
    - Canova and Pappa (2007) type: on-impact restrictions that government expenditure increases, deficit increases, and domestic output increases.
    - Mountford and Uhlig (2009) type: separate business-cycle shock (output and net taxes increase for four quarters) and spending shocks (government spending increases for four consecutive periods and orthogonal to business-cycle shock).
  - For each country, 10,000 orthonormal matrices drawn rotating the contemporaneous variance-covariance matrix; median rotation applied to each posterior draw.

### Estimation and priors
- Estimation approach: Bayesian estimation with Gibbs sampling; Metropolis step for 휎2.
- Prior and posterior diagnostics:
  - Priors are diffuse and their support is much wider than that of the posterior estimates; priors centered around OLS to increase efficiency.
  - Posterior draws: appear well behaved with low serial correlation.
  - Sampling details: results based on the last draw of 500 chains of length 1000 all starting in a small random interval of the last draw of a single (burn-in) chain of 100000 draws.
- Appendix-level prior forms (preserved exactly in source):
  - 휃 ~ 푁(휃0, Θ)
  - 푃 ~ 퐼푊(푟, 푄)
  - Ω ~ 퐼푊(표, 푅)
  - 휎2 ~ 퐼퐺(0.5, 0.5푠2)

### Key empirical findings — multipliers and spillovers (1999-2016)
- Domestic short-run (one year) fiscal multipliers (1999-2016; expressed in percent):
  - Fiscal multipliers are positive and below unity in many cases but country-specific.
  - Magnitude above 0.5 in: Germany, France, the Netherlands, Ireland, and Italy.
  - Magnitude below 0.5 in: Belgium, Austria, Spain, and Portugal.
  - Euro-to-euro (Keynesian) multipliers:
    - Below unity in: Portugal, Spain, Austria, and Belgium.
    - Above one in Germany (median value is 1.5) and in Ireland.
    - Close to one in other countries.
- Selected cross-country year-one spillovers (cumulative responses):
  - Spillovers to Germany from: Netherlands (0.21), Belgium (0.14), Italy (0.11), France (0.22), Austria (0.15).
  - Germany → Netherlands: 0.17; Germany → Austria: 0.16.
  - Germany → Italy and Belgium: 0.13 each.
  - Belgium ↔ Netherlands: 0.11 and 0.13 respectively.
- Small but well-integrated economies:
  - Finland receives from Germany, France, and the Netherlands: 0.19, 0.15, and 0.16 respectively.
  - Portugal receives from Germany and France: 0.12 and 0.15 respectively.
  - Ireland receives large inward spillovers from core EA countries; financial linkages and multinational presence important beyond trade flows.
- Outward impact of small-country expansions:
  - Ireland, Finland, and Portugal generate small spillovers on other member countries; impacts often small and frequently not different from zero.

### Time variation and subsample dynamics
- Rolling subsamples ending each year 2009–2016:
  - Multipliers larger for sample windows ending in 2011-2014 (period corresponding to euro area sovereign debt crisis, LTRO, and ECB quantitative easing near the effective lower bound).
  - Multipliers revert to lower, pre-crisis values in the last two sample windows including normalization periods.
  - Spain and Portugal exhibited very high multipliers in the first sample (1999-2009); potential drivers include exposure to 2007 crisis, boom-bust cycles, and short-sample bias.
  - Germany→spillover profile follows a hump-shaped pattern; countries most exposed to German spillovers remain Netherlands, Finland, Austria, Belgium, Italy across sub-samples.
  - Exposure of Spain and Portugal to German spillovers increased during the euro area crisis.

### Domestic transmission channels — IRF patterns
- Responses to a 1 percent increase in government spending (1999-2016 sample):
  - Government spending: persistent, peaks approximately two years after the shock.
  - Output: expansionary effect that dies out after two years once spending impulse ends.
    - Germany: output jumps on impact (0.7 percent) and expands steadily, plateauing above 0.9 percent.
    - Portugal, Spain, Belgium: output elasticity to spending shock remains below 0.4.
  - Current account:
    - Government spending increase stimulates domestic demand and imports; current account turns negative in almost every country.
    - Germany: current account falls by around 3 percent on impact and stabilizes around 2 percent thereafter.
    - France: current account falls by 3.5 percent on impact and deteriorates further to almost 4 percent.
    - Spain: current account falls by approximately 1 percent (confidence bands overlap zero).
  - REER:
    - Baseline sign restrictions yield mixed REER responses across countries.
    - Under Mountford and Uhlig (2009) restrictions, REER dynamics point more consistently toward appreciation.
    - Largest positive REER response observed for Germany; Netherlands and Ireland show negative REER responses around -0.1 percent.
  - Government debt: increases in most countries following higher government spending; Germany only country where public debt consistently falls across all sub-samples.

### International transmission channels and role of current account
- Foreign output always expands in response to higher domestic spending with heterogeneous strength.
- Current account identified as the single most important series for transmission.
- Examples of cumulative current-account bilateral effects:
  - Netherlands or Belgium shock expands Germany’s current account by 0.4 percent; Austria by 0.2 percent; Italy by 0.1 percent.
  - Austria shock expands Germany’s current account by 0.3 percent and Italy’s by approximately 0.1 percent.
  - Germany shock causes domestic current account to fall and turns negative in the Netherlands and Belgium (approximately 1 percent); smaller and not significant negative responses in France, Austria, Italy and Spain.
- Second-round effects and high import content of exports (GVCs) can reverse or weaken positive net effects of fiscal spillovers on net exports.
- Other variables:
  - Long-term interest rates tend to increase following spending shocks, but effect is quantitatively small.
  - No systematic evidence that markets systematically raise government financing costs due to less conservative fiscal management, though spikes are possible under stressed conditions.

### Determinants of fiscal spillovers — regression evidence
- Regression specification: S_{i,j,t} = d_r + d_s + d_t + β x_{i,j} + ε_{i,j,t} with S_{i,j,t} cumulative fiscal spillover from i to j; bilateral characteristics x_{i,j} include bilateral trade and financial flows.
- Key regression estimates (Table 2 examples preserved as in source):
  - Bilateral trade 0.014***; 0.012*** (Sample Average)
  - Bilateral financial links 0.013***; 0.012*** (Sample Average)
  - Observations 1,800; R-squared 0.825 (sample average row)
  - Alternate panels report Observations 1,720; R-squared 0.83; Observations 1,800; R-squared 0.81
- Interpretation:
  - Stronger bilateral trade links increase fiscal spillovers.
  - Bilateral financial flows are positive and statistically significant predictors of higher spillovers.
  - Importance of the trade channel has increased over time; financial-link impact stronger before the global financial crisis.
  - Coefficients stable when controlling for sending- and receiving-country fixed effects and time effects.

### Robustness and alternative identification
- Robustness checks implemented:
  - Alternative sample sizes, different endogenous and exogenous series, alternative prior specifications, excluding Ireland.
  - Replacing REER with CPI and alternative REER definition (relative to rest of world) — results broadly unchanged.
  - Sensitivity tests on prior distributions had no measurable impact given diffuse priors.
- Alternative identification (Mountford and Uhlig, 2009):
  - Median estimates generally in the ballpark of baseline and fall within baseline confidence bands in many cases.
  - Multipliers larger under Mountford and Uhlig for Belgium, Austria, Italy, Spain and Ireland; smaller for remaining countries.
  - Examples: Italy 0.54 (Mountford & Uhlig) vs 0.5 (baseline); Spain 0.27 vs 0.31; Ireland 0.71 vs 0.68.
  - Under Mountford and Uhlig restrictions, REER dynamics point more consistently to appreciation; government bond yields remain unchanged or turn negative and small.

### Price effects and additional outcomes
- Replacing REER with CPI among endogenous series:
  - Domestic CPI responses:
    - France: CPI increases by 0.1 percent on impact, peaking at 0.15 percent in the third year.
    - Italy and Austria: positive and statistically significant CPI changes, smaller than France.
    - Germany and other countries: muted and insignificant CPI responses.
    - Spain: negative response of the price level (related to prolonged deflationary period post-2007).
  - International prices rise two to three quarters after the shock with magnitude abroad typically not exceeding 0.05 percent.
  - Interpretation: foreign price propagation likely works mostly through the demand channel rather than immediate expectations.

### Conclusions and policy implications
- Empirical summary for 10 euro area countries (post-euro period):
  - Domestic fiscal multipliers are positive and heterogeneous across countries; average multipliers can be misleading.
  - Fiscal spillovers can be significant and are larger when:
    - the source economy is large;
    - countries are highly integrated through trade or financial linkages;
    - the receiving economy is small and has a narrow export base.
  - The current account is the main transmission channel domestically and internationally, but strong trade integration and spillback effects tend to zero-out net impact in the post-euro sample.
  - No systematic evidence for a general "twin deficit" effect; decoupling between output gains and negative current account responses observed in some cases.
  - REER and interest-rate channels show limited systematic effects; Germany is an exception with some evidence of REER appreciation following fiscal expansion.
  - Sub-sample patterns: larger multipliers and spillovers detected between 2011 and 2014; idiosyncratic country factors can drive substantial differences (e.g., Ireland, Spain).
- Suggested avenues for future research:
  - Distinguish spillovers from alternative fiscal instruments (public consumption vs public investment).
  - Examine tax shocks separately from government spending shocks.

*Source: wp17241 (excerpt from PDF chapter).*

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

### wp17241 - References

### Introduction and motivation
- Active fiscal policy, in coordination with other policies, has been proposed as a way out from a global environment characterized by slow growth and ultra-low interest rates (see Gaspar and others, 2016; Furman, 2016; OECD, 2016; and G20, 2016).
- Within the euro area, debate centers on fiscal expansion in countries with fiscal space while most other countries face a combination of high debt, negative output gaps, and high unemployment (EC, 2016; EPSC, 2016).
- Proponents argue stimulus could contribute to the region’s recovery through fiscal spillovers given growing interconnectedness; opponents argue cross-border spillovers are small.
- The paper aims to provide new empirical evidence on transmission of fiscal policy shocks in the euro area, motivated by:
  - Policy need to understand magnitude and impact of domestic fiscal shocks on output and key macroeconomic variables of other member countries.
  - Theoretical ambiguity: competing models predict differing size and sign of fiscal spillovers in a monetary union where trade links are primary transmission channels.
  - Mixed and relatively scarce empirical evidence on cross-country fiscal spillovers compared to the literature on domestic fiscal multipliers.

### Theoretical context and prior empirical evidence
- In a monetary union with a common currency and monetary policy rate, trade links between countries are the main transmission channels for fiscal shocks; Rose and van Wincoop (2001) note higher price elasticity of trade within currency unions.
- Theoretical predictions vary:
  - Some models imply endogenous domestic interest rate increases offset positive domestic demand effects, yielding negligible spillovers under fixed exchange rates (Cwik and Wieland, 2010; Kollmann and others, 2015).
  - Alternative assumptions (shock process, expectations, import content of government spending, financial market imperfections, liquidity trap) can reverse conclusions and yield large or even negative fiscal spillovers (Farhi and Werning, 2012; Clancy and others, 2015; Veld, 2016; Blanchard and others, 2015; Corsetti and others, 2010).
- Existing empirical evidence on fiscal multipliers suggests a broad range of possible values, with values ranging from 0.2 to 1.5 depending on country sample and empirical model considered (Kilponen and others, 2015).
- Empirical studies on spillovers find economically significant effects but mixed evidence on magnitude and sign (Beetsma and others, 2006; Auerbach and Gorodnichenko, 2013; Hebous and Zimmerman, 2013; Canova and others, 2013).

### Data, model, and identification
- Model and data:
  - Large-scale panel VAR estimated for 10 euro area countries using quarterly data for the period 1999-2016.
  - Model includes variables capturing economic and financial links.
- Identification:
  - Relies on sign restrictions to identify fiscal policy shocks, following Canova and Pappa (2007) in a multi-country setting, and on Mountford and Uhlig (2009) using US data.
  - Sign restrictions impose minimal constraints on the shape of impulse response functions and are consistent with a large class of macroeconomic models.
- Methodological advances relative to prior work:
  - Allows heterogeneous shock sizes and propagation across countries.
  - Captures interlinkages across series and countries and dynamic feedback loops from fiscal shocks to other variables.
  - Estimates both domestic fiscal multipliers and cross-country spillovers to shocks in individual countries, capturing country-specific heterogeneous effects rather than applying average multipliers multiplied by trade weights.
  - Enables differentiation of spillovers originating from different countries and exploration of transmission channels (trade, financial, exchange rate) including over time.

### Key empirical findings
- Fiscal spillovers are positive and non-negligible, but vary across countries originating the fiscal shock.
- Domestic fiscal multipliers are positive and country-specific; average estimates of fiscal multipliers can be misleading.
- Determinants of the magnitude of fiscal spillovers include:
  - Relative size of the country originating the fiscal shock.
  - Integration through trade and financial channels.
- Offset and spillback effects:
  - Strong trade integration among euro area countries and spillback effects can zero-out net impacts in some cases.
  - In small and open economies, inward spillovers raising domestic income can increase imports, undoing expansionary effects on the current account.
  - In large economies, strong international trade interlinkages throughout the manufacturing value chain can reduce the positive net effect of spillovers on a country’s external position.
- Robustness:
  - Results are broadly robust to alternative sign restriction approaches and other robustness checks.
- Time variation:
  - Subsample analysis shows effects of fiscal policy changed over time, reflecting different business cycle states and policy framework changes.
  - Typically larger domestic multipliers and spillovers are detected during the euro area crisis, though idiosyncrasies exist across some countries.

*Source: wp17241 - References (excerpt).*

### Section IV summarizes the dataset. Section V presents the results. Finally, section VI concludes.

### wp17241 - Section IV summarizes the dataset. Section V presents the results. Finally, section VI concludes.

### II. LITERATURE REVIEW
- Empirical approaches to fiscal spillovers:
  - VAR-based approaches
    - Beetsma and Giuliodori (2004): standard VAR in pre-EMU period; focus on change in imports from other countries; find statistically and economically significant trade spillovers from fiscal impulses in Germany, Italy, and France.
    - Beetsma and others (2006): homogeneous coefficients panel VAR with fiscal and trade block for eleven EU countries over the period 1965-2002; identify fiscal shocks using structural restrictions as in Blanchard and Perotti (2002); find spillovers from a 1 percent increase in German government spending lead to output responses between 0.05 percent of GDP in Greece and 0.4 percent of GDP in Belgium.
    - Shortcomings of recursive VAR ordering with government spending ordered first: fiscal shocks may be anticipated; difficult to distinguish genuine fiscal policy shocks from business cycle-driven movements; less accurate than multilateral models in terms of bias and mean squared error.
  - Local projections method (LPM)
    - Auerbach and Gorodnichenko (2013): LPM measuring fiscal spillovers from a weighted average of fiscal shocks where weights reflect bilateral trade flows; fiscal shocks defined as forecast errors of government spending in OECD projections.
      - Find output impact (over a 6-year period) of a 1 percent of GDP fiscal consolidation in all trading partners ranging between 1.6 and 2.0 percent, and reaching 3 percent during periods of economic slack.
      - Homogeneity assumption yields only average spillover impact; estimates for slack periods sensitive to assumed length of recession episodes.
    - Goujard (2017): LPM with narrative fiscal consolidations from Devries and others (2011); finds sizeable spillovers, stronger within currency unions or with limited exchange rate adjustment.
  - Narrative (external instrument) approaches
    - Narrative approach may include consolidations implemented in response to bad news about future growth, complicating causal interpretation (Ramey, 2011).
  - Global VAR (GVAR)
    - Studies (Hebous and Zimmerman, 2013; Ricci-Risquete and Ramajo-Hernandez, 2015; Belke and Osowski, 2016; Georgiadis and Hollmayr, 2016) estimate GVAR with fiscal shocks identified using orthogonalized impulse response functions to capture heterogeneous impacts.
      - Example: Hebous and Zimmermann (2013) find spillovers from a 1 percent of GDP fiscal shock in Germany ranging from -0.2 percent of GDP in Italy to 0.13 percent of GDP in Luxembourg.
      - GVAR captures trade and other channels (interest rate, exchange rate) but relies on exogenously set weights and orthogonalized responses that are not structural.
- Identification choice in this paper:
  - Use sign restrictions on responses of endogenous variables.
  - Two types of sign restrictions:
    - Canova and Pappa (2007) style: identify government spending shocks by imposing that domestic government spending, the budget deficit, and output increase on impact.
      - Restrictions imposed only on impact; output may increase or decrease after the first period; deficit may fall in the longer run.
    - Mountford and Uhlig (2009) style: first identify a business cycle shock (output and net taxes increase for four quarters, with net taxes increasing more than government spending), then define government spending shocks as shocks where government spending rises for four quarters and is orthogonal to the business cycle shock.
      - No restriction on output in this approach; responses restricted for a year following the shock.
  - Advantages: minimal assumptions on impulse response shape; allows heterogeneity and dynamic interlinkages across countries.
  - Extensions relative to related work: extends Mountford and Uhlig–type identification to a model with cross-country and dynamic interlinkages; differs from Bicu and Lieb (2015) by developing cross-sectional dimension and sub-sample analysis rather than nesting panel and factor-augmented VAR structures.

### III. MODEL
- Model specification (heterogeneous panel VAR):
  - Equation: 푦ᵢₜ = 퐷ᵢ(𝐿)푌ₜ₋₁ + 퐹ᵢ(𝐿)푊ₜ + 푒ᵢₜ
    - i = 1, ..., N (countries); t = 1, ..., T (time); L is lag operator.
    - 푦ᵢₜ is G x 1 vector of endogenous series for country i; 푌ₜ = (푦₁ₜ′,...,푦_Nₜ′).
    - 퐷ᵢ,푗 are G x NG matrices for lags j = 1, ..., p; 퐹ᵢ are G x M matrices for lags j = 1, ..., q.
    - 푊ₜ is M x 1 vector of common exogenous variables.
    - 푒ᵢₜ ~ N(0, Σᵢ) is G x 1 vector of disturbances.
    - No constant included because each variable is deseasonalized, demeaned and standardized.
- Model features and motivations:
  - Instantaneous and lagged dynamics are unit-specific → shock sizes and propagation are potentially heterogeneous.
  - Cross-unit lagged interdependencies allowed when D(L) is not block diagonal → dynamic feedbacks across countries possible.
  - Multi-country heterogeneous models preferred over bilateral VARs to better capture multilateral and higher-order spillovers and avoid underestimating spillovers.
- Dimensionality challenge:
  - Regression form: 푌ₜ = 푍ₜ훿 + 퐸ₜ with 푍ₜ = 퐼_NG ⨂ 푋ₜ′ and 훿 stacked coefficient vector.
  - Example: 10 countries, 5 variables per country, 2 lags → dimensionality of δ would be 5000 x 1; an unrestricted variance-covariance matrix would have 1125 free parameters.
  - Two assumptions to reduce dimensionality:
    - Kronecker structure for covariance matrix:
      - 퐸ₜ ~ N(0, P ⨂ Ω)
      - P is N x N capturing correlation across countries; Ω is G x G capturing correlation across variables.
      - Prior for P centered at bilateral trade weights; off-diagonal elements left unrestricted.
    - Factorization of coefficient vector:
      - 훿 = Ξ휃 + 푢 where Ξ is a selection matrix of zeros and ones; u ~ N(0, (P ⨂ Ω) ⨂ V).
      - Ξθ decomposed into Ξ₁휃₁ + Ξ₂휃₂ + Ξ₃휃₃ + Ξ₄휃₄ with loading matrices Ξⱼ and mutually orthogonal factors 휃ⱼ capturing country, endogenous series, exogenous series, and lag information respectively.
      - Transforms over-parametrized panel VAR into a parsimonious SUR model; random pooling improves parameter accuracy and reduces standard errors.

### B. Identification and Estimation
- Identification via sign restrictions (two approaches summarized above):
  - Canova and Pappa (2007) type:
    - Instantaneous sign restrictions: i) government expenditure increases; ii) the deficit increases; iii) domestic output increases.
    - Restrictions imposed on impact only.
  - Mountford and Uhlig (2009) type:
    - Business cycle shock: output and net taxes increase for four quarters; net taxes increase more than government spending.
    - Government spending shock: government spending increases for four consecutive periods, increase larger than taxes, and orthogonal to business cycle shock.
- Implementation details:
  - Economic restrictions derived from endogenous series identifying domestic block; rotations satisfying restrictions imposed on matrix Ω (average-country relationship among endogenous variables).
  - Impulse responses computed as difference between two conditional forecasts:
    - IR_y(t, τ) = E(y_{t+τ} | ℱ_t^1) − E(y_{t+τ} | ℱ_t^2) for τ = 1,2,...,20 quarters.
    - ℱ_t^j include initial conditions drawn from posterior distribution and a value for the structural shock.
  - Data standardized before estimation; impulse responses rescaled by each series’ standard deviation when computing dynamic responses; figures in paper show cumulated responses.
- Estimation approach:
  - Bayesian estimation due to relatively small sample.
  - Appendix 1 contains estimation algorithm details; Appendix 2 describes priors and posterior draws used for inference.
  - Use of priors includes centering P at bilateral trade weights and imposing specific priors to reduce the parameter constellation given granularity.

*Source: wp17241 - Section IV summarizes the dataset. Section V presents the results. Finally, section VI concludes.*

### Appendix 2, the prior distributions are diffuse, and their support is much wider than that of the

### Appendix 2, the prior distributions are diffuse, and their support is much wider than that of the

### IV. Estimation setup and priors
- Priors are diffuse and their support is much wider than that of the posterior estimates.
- To increase estimation efficiency, priors are centered around the OLS estimated coefficients.
- Posterior draws:
  - Appear well behaved, with no clear trends or patterns.
  - Exhibit low serial correlation.
- Robustness checks:
  - A number of checks were conducted on the parameters controlling for the properties of the priors’ distributions, without appreciable effects on the empirical results.
- Estimation windows:
  - Model estimated over the full sample (1999-2016).
  - Also estimated on recursive windows to accommodate structural breaks (e.g., Global Financial Crisis, ultra-low interest rates).
  - Non-parametric setup to time variation chosen for computational tractability.

### IV. DATASET (sample and variables)
- Countries (ten euro area countries): Germany, France, the Netherlands, Belgium, Austria, Italy, Spain, Ireland, Finland, and Portugal.
- Sample coverage: balanced panel covering the period from 1999 to 2016 at quarterly frequency.
- Rationale for 1999 start:
  - Minimizes biases arising from the introduction of the euro and other potential structural breaks (e.g., unification of Germany and the currency crisis in the run-up to the monetary union).
- Sub-sample windows:
  - First sample window covers 1999-2009.
  - Subsequent windows are extended by one year until 2016 to assess changes in fiscal spillovers following the global financial crisis and with monetary policy at the effective lower bound.
  - Note: An alternative split before and after 2008 yields subsamples too short and subject to estimation biases.
- Variables included:
  - Government expenditures (sum of consumption and investment).
  - Government revenues net of transfers.
  - Government debt (checked ex-post to ensure path is not explosive).
  - Output.
  - Current account balance.
  - Real effective exchange rate (REER).
  - Benchmark interest rate (yield on 10-year government bonds) — proxies for monetary policy stance and financial conditions.
- Exogenous controls:
  - Federal fund rate and Eonia (US and EA monetary policy).
  - World output (global business cycles).
  - Oil prices (commodity market developments).
- Data transformations:
  - Interest rate variable enters the model in levels.
  - Remaining variables expressed in real per capita terms and transformed into natural logarithms.
  - Current account series rescaled by adding a constant value prior to log transformation.
- Identification and uncertainty:
  - For each country, 10,000 orthonormal matrices drawn rotating the contemporaneous variance-covariance matrix of shocks; median rotation applied to each posterior draw of covariance matrix.
  - Approach captures coefficient uncertainty and, to some extent, identification uncertainty.

### V. RESULTS — overview of methods
- Impulse response functions (IRFs) and fiscal multipliers used to present results.
- Multiplier definition adopted (Owyang and others (2013)):
  - Fiscal multipliers and spillovers = cumulative response of output (in percent) divided by cumulative change in government spending (in percent).
  - Interpretation: multipliers/spillovers can also be interpreted as an elasticity in this VAR-type application.
  - Cumulative/integral approach controls for persistence in the shock dynamics and overall policy impact.
- Graphical convention:
  - Statistically significant IRF results are marked using triangles to avoid clutter.

### V.A Multipliers and spillovers in the euro area — key findings
- Short-run (one year) domestic fiscal multipliers (1999-2016 sample), expressed in percent:
  - Fiscal multipliers are positive and below unity, but size varies across countries.
  - Magnitude above 0.5 in: Germany, France, the Netherlands, Ireland, and Italy.
  - Magnitude below 0.5 in: Belgium, Austria, Spain, and Portugal.
- Euro-to-euro (Keynesian) multipliers:
  - Below unity in: Portugal, Spain, Austria, and Belgium.
  - Above one in Germany (median value is 1.5) and in Ireland.
  - Close to one in other countries.
- Cross-country spillovers:
  - Generally positive, can be large and significant.
  - Larger economies and more integrated countries generate more sizeable spillovers.
  - Year-one spillovers to Germany from: Netherlands (0.21), Belgium (0.14), Italy (0.11), France (0.22), Austria (0.15).
  - Spillovers from Germany to the Netherlands: 0.17; to Austria: 0.16.
  - Germany → Italy and Belgium: 0.13 each.
  - Belgium ↔ Netherlands spillovers: 0.11 and 0.13 respectively.
- Smaller economies:
  - Finland, Portugal, and Ireland are relatively small but well integrated, and receive sizable spillovers:
    - Finland receives from Germany, France, and the Netherlands: 0.19, 0.15, and 0.16 respectively.
    - Portugal receives from Germany and France: 0.12 and 0.15 respectively.
  - Ireland receives large spillovers from core EA countries; trade flows alone do not account for this — financial linkages and multinational presence are important factors.
- Financial interconnectedness:
  - BIS consolidated foreign claims used as proxy for cross-country financial linkages.
  - Average (1999-2016) bilateral consolidated foreign claims and total foreign claims indicate Ireland’s financial links to the EA block were significantly higher than trade flows during the sample.
- Impact of small-country expansions:
  - Ireland, Finland, and Portugal generate small spillovers on other member countries; impacts often small and frequently not different from zero.

### V.B Evolution of spillovers over time
- Time variation of multipliers and spillovers examined across rolling subsamples ending each year from 2009 to 2016.
- Domestic fiscal multipliers:
  - Time pattern tends to be similar across countries.
  - Larger multipliers for sample windows ending in 2011-2014 (period corresponding to euro area sovereign debt crisis, launch of Long-Term Refinancing Operations, and ECB quantitative easing amid zero-lower bound concerns).
  - Multipliers revert to lower, pre-crisis values in the last two sample windows that include normalization periods.
  - Spain, Portugal, and Ireland stand out:
    - Multipliers in Spain and Portugal evolve similarly across samples, possibly reflecting exposure to common cycle and shocks.
    - Multipliers in Spain and Portugal very high in first sample (1999-2009), possibly due to greater exposure to the 2007 crisis in economies with profound structural imbalances.
    - A short sample combined with a boom-bust cycle may further explain large multiplier in Spain.
  - Main results unaffected if Ireland is dropped from the sample.
- Spillovers from Germany:
  - Time profile follows a hump-shaped pattern, similar to domestic multipliers.
  - List of countries with largest Germany→spillovers (Netherlands, Finland, Austria, Belgium, Italy) remains relatively stable across sub-samples.
  - Spain and Portugal were less exposed to German spillovers in earlier samples; exposure increased during the euro area crisis.

### V.C Domestic transmission channels — key responses
- Responses to a one percent increase in government spending (1999-2016 sample):
  - Government spending:
    - Shows persistence after initial increase.
    - Peaks approximately two years after the shock.
  - Output:
    - Effect always expansionary; dies out after two years once spending impulse ends.
    - Germany: output jumps on impact (0.7 percent) and expands steadily, plateauing above 0.9 percent.
    - Portugal, Spain, Belgium: elasticity of output to spending shock remains below 0.4.
  - Current account:
    - Standard theory prediction confirmed: government spending increase stimulates domestic demand and imports, leading to a trade deficit relative to the rest of the world.
    - Current account turns negative in almost every country; quantitatively large in some cases.
    - Germany: current account falls by around 3 percent on impact and stabilizes around 2 percent thereafter.
    - France: current account falls by (text truncated in source before exact value).

*Source: wp17241 - Appendix 2, the prior distributions are diffuse, and their support is much wider than that of the (PDF chapter).*

### 3.5 percent on impact and deteriorates further to almost 4 percent. In Spain, it falls by

### wp17241 - 3.5 percent on impact and deteriorates further to almost 4 percent. In Spain, it falls by

### Current account responses and "twin deficits"
- In Spain, the current account falls by approximately 1 percent, however the confidence bands overlap the zero line.
- In a number of other countries, the response of the current account is either muted or slightly positive.
- Increased integration and Global Value Chains (GVC) have heightened trade interlinkages and the import content of exports, complicating the net impact of fiscal policy on the current account.
- Models that omit lagged interdependencies and spillback effects can fail to characterize the true overall impact on the current account.

### Real Effective Exchange Rate (REER) and competitiveness
- Theoretical predictions for the REER response to government spending shocks are ambiguous; empirical evidence is mixed.
- Under the baseline identifying restrictions, the REER response is either positive or negative, depending on the country.
- Using Mountford and Uhlig (2009) identification restrictions, results point more consistently toward a real exchange rate appreciation.
- Among IRFs in Figure 10:
  - The largest positive REER response is observed for Germany.
  - Countries with negative REER responses include the Netherlands and Ireland (around -0.1 percent).
- A positive response of the REER indicates an appreciation.

### Government debt dynamics
- Government debt increases in most countries following higher government spending.
- Efforts to reduce debt buildup lead to a negative response in some sample windows and countries.
- Germany is the only case where public debt consistently falls across all sub-samples.

### International transmission channels and spillovers
- Foreign output always expands in response to higher domestic spending, with heterogeneous strength across countries.
- The current account is identified as the single most important series to capture transmission of shocks.
- Examples of bilateral spillovers (cumulative responses described):
  - Fiscal shock in the Netherlands or Belgium expands the current account in Germany by 0.4 percent, in Austria by 0.2 percent, and in Italy by 0.1 percent.
  - Fiscal shock in Austria expands the current account in Germany by 0.3 percent and in Italy by approximately 0.1 percent.
  - Fiscal shock in Germany: domestic current account falls and turns negative in the Netherlands and Belgium (approximately 1 percent); smaller and not significant negative responses also in France, Austria, Italy and Spain.
  - Fiscal shock in France: the current account in Belgium falls by 1 percent, while it expands or remains broadly stable in the remaining countries.
- Second-round effects and strong trade integration (including high import content of exports) can reverse or weaken the positive net effect of fiscal spillovers on net exports.
- Beyond the current account, impulse responses for other variables are typically quantitatively small or negligible:
  - Long-term interest rates tend to increase in response to government spending shocks, but the size of this effect is small.
  - No systematic evidence that less conservative fiscal management leads markets to increase the cost of government financing, though spikes remain possible under strained public finances or market uncertainty.

### Regression analysis: determinants of fiscal spillovers
- Regression specification:
  - S_{i,j,t} = d_r + d_s + d_t + β x_{i,j} + ε_{i,j,t}, where S_{i,j,t} denotes cumulative fiscal spillover from country i to country j in quarter t (ranging between 1 and 20); d_r, d_s, d_t are receiving-country, sending-country, and quarter dummies; x_{i,j} contains bilateral country characteristics (either bilateral trade or bilateral financial flows).
- Bilateral trade and financial linkages:
  - Stronger bilateral trade links increase fiscal spillovers.
  - The importance of the trade channel has increased over time.
  - Bilateral financial flows are positive and statistically significant predictors of higher fiscal spillovers.
  - Subsample evidence indicates the impact of financial links was stronger before the global financial crisis.
  - Coefficients remain stable and significant when controlling for country and time effects.
- Definitions:
  - Bilateral trade flows = sum of exports and imports between two countries, rescaled by total trade among all countries in the sample.
  - Bilateral financial flows = defined similarly for consolidated foreign claims.

### Extensions: price effects and alternative identification
- Replacing REER with CPI among endogenous series:
  - Both domestic and foreign price levels generally rise in response to fiscal shocks (Spain is a notable exception).
  - Domestic CPI responses:
    - France: CPI increases by 0.1 percent on impact, peaking at 0.15 percent in the third year.
    - Italy and Austria: positive and statistically significant CPI changes, smaller than France.
    - Germany and other countries: muted and insignificant responses.
    - Spain: negative response of the price level, likely related to prolonged deflationary period after 2007 crisis.
  - International (foreign) prices start rising two to three quarters after the shock; magnitude abroad typically does not exceed 0.05 percent.
  - Interpretation: propagation to foreign prices likely works mostly through the demand channel rather than immediate expectations.
- Alternative identification (Mountford and Uhlig, 2009):
  - Fiscal spillovers do not change substantially; median estimates are generally in the ballpark of baseline and fall within baseline confidence bands in many cases.
  - Fiscal multipliers larger under Mountford and Uhlig (2009) for Belgium, Austria, Italy, Spain and Ireland; smaller for remaining countries.
  - Examples of multiplier comparisons:
    - Italy: 0.54 (Mountford & Uhlig) vs 0.5 (baseline).
    - Spain: 0.27 vs 0.31.
    - Ireland: 0.71 vs 0.68.
  - Under Mountford and Uhlig (2009) sign restrictions, the REER dynamic consistently points to an appreciation.
  - Government bond yields remain either unchanged, or turn negative and small.

### Robustness checks
- Battery of tests including alternative sample size, different endogenous and exogenous series, and changes in prior specifications; broad thrust and magnitude of spillovers largely unchanged.
- Excluding Ireland did not produce significant differences in spillover estimates for other countries.
- Alternative REER definition (relative to rest of the world using CPI) and replacing REER with CPI while using Eonia for the benchmark interest rate left results broadly unchanged.
- Sensitivity tests on prior distributions had no measurable impact given loose priors.

### Conclusions and policy implications
- Empirical findings for 10 euro area countries over the post-euro period using a large-scale panel-VAR with leading and lagged interdependencies:
  - Domestic fiscal multipliers are positive and heterogeneous across countries.
  - Fiscal spillovers can be significant in some cases.
  - Heterogeneity implies average fiscal multiplier estimates can be misleading.
  - An inverse relationship observed between size of domestic fiscal multipliers and trade openness.
  - Fiscal spillovers are larger when:
    - (i) the source economy is large;
    - (ii) countries are highly integrated through trade or financial linkages;
    - (iii) the receiving economy is small and has a narrow export base.
- The current account is the main channel of transmission domestically and abroad, but strong trade integration and spillback effects tend to zero-out net impact when using only the post-euro sample.
- No systematic evidence supporting the "twin deficit hypothesis"; instances of decoupling between positive output dynamics and negative current account responses are observed.
- Other transmission channels (interest rates, exchange rate) show limited systematic effects:
  - REER moves little in response to government spending shocks; prices and wages stickiness and centralized monetary policy in the euro area likely contribute.
  - Germany is a notable exception: some evidence of a positive impact of fiscal policy on the real exchange rate.
  - Domestic and foreign price levels generally rise in response to fiscal shocks, though effects are small.
- Sub-sample analysis:
  - Effects of fiscal policy change over time with business cycle and policy framework; larger domestic multipliers and spillovers observed between 2011 and 2014.
  - Idiosyncratic factors (e.g., rebasing of GDP in Ireland, boom-bust in Spain) can drive sizeable changes.
  - Ireland: very high inward spillovers but low outward spillovers; legislative framework (taxation of corporate income, repatriation of profits, establishment of a company, etc.) may shape transmission abroad.
- Suggested avenues for future research:
  - Investigate spillovers from alternative fiscal instruments (public consumption vs public investment).
  - Examine tax shocks as distinct from government spending shocks.

*Source: wp17241 (excerpts from PDF chapter).*

### REFERENCES

### REFERENCES

### Bibliographic sources
- Comprehensive list of cited works covering fiscal policy, fiscal multipliers, spillovers, Bayesian macroeconometrics, GVAR and VAR methodologies, and empirical studies in the Euro Area and other economies. Key authors and works included (selection from the list):
  - Amador, J.; Cappariello, R.; Stehrer, R., (2015), “Global Value Chains: A View from the Euro Area”, Asian Economic Journal, 29, No. 2, pp. 99–120
  - Attinasi, M.G.; Lalik, M.; Vetlov, I., (2017), “Fiscal Spillovers in the Euro Area: A Model-Based Analysis”, ECB Working Paper No. 2040
  - Auerbach, A.J.; Gorodnichenko, Y., (2013), “Output Spillovers from Fiscal Policy”, American Economic Review, 103, pp. 141-146
  - Ball, L.; Furceri, D.; Leigh, D.; Loungani, P., (2013), “The Distributional Effects of Fiscal Consolidation”, IMF Working Paper No. 151
  - Blanchard, O.; Perotti, R., (2002), “An Empirical Characterization of the Dynamic Effects of Changes in Government Spending and Taxes on Output”, The Quarterly Journal of Economics, 117, No. 4, pp. 1329-1368
  - Mountford, A.; Uhlig, H., (2009), “What are the effects of fiscal policy shocks?”, Journal of Applied Econometrics, Volume 24, Issue 6, pp. 960–992
  - Del Negro, M.; Schorfheide, F., (2011), “Bayesian Macroeconometrics”, in The Oxford Handbook of Bayesian Econometrics
  - Devries, P.; Guajardo, J.; Leigh, D.; Pescatori, A., (2011), “A New Action-based Dataset of Fiscal Consolidation”, IMF Working Paper WP/11/128
  - Farhi, E.; Werning, I., (2012), “Fiscal Multipliers: Liquidity Traps and Currency Unions”, NBER Working Paper No. 18381
  - Pesaran, M.H.; Smith, R.P.; Im, K.S., (1996), “Dynamic Linear Models for Heterogeneous Panels”, in The Econometrics of Panel Data
  - Primiceri, G., (2005), “Time Varying Structural Vector Autoregressions and Monetary Policy”, The Review of Economic Studies, 72, pp. 821-852
  - Poghosyan, T. (2017), “Cross-Country Spillovers of Fiscal Consolidations in the Euro Area”, IMF Working Paper WP/17/140
  - Ramey, V.A., (2011), “Can Government Purchases Stimulate the Economy?”, Journal of Economic Literature, 49, No. 3, pp. 673-85
  - OECD, (2016), “Stronger Growth Remains Elusive: Urgent Policy Response is Needed”, OECD Interim Economic Outlook, 18 February 2016
- Institutional and policy documents cited include EC, 2016; EPSC, 2016; G20 (2016); Gaspar, Obstfeld, and Sahay (2016); Furman (2016).

### Tables and figures included or referenced
- Table 1. Fiscal Multipliers and Spillovers (1 year)
  - Multipliers (diagonal) and spillovers (off-diagonal) at year one using sample 1999-2016; numbers in parenthesis indicate 68 percent confidence intervals.
- Table 2. Determinants of Fiscal Spillovers
  - Regression results reported for: Sample Average, 2006, and 2016.
  - Coefficients (examples preserved exactly as in table headers and cells):
    - Bilateral trade 0.014***; 0.012*** (Sample Average)
    - Bilateral financial links 0.013***; 0.012*** (Sample Average)
    - Observations 1,800; R-squared 0.825; and alternate panels: Observations 1,720; R-squared 0.83; Observations 1,800; R-squared 0.81
  - Dummies for receiving countries: Yes; Dummies for quarters: No/Yes as reported.
  - Note: Dependent variable is bilateral fiscal spillovers measured by IRFs. Confidence intervals: ***<0.01, **<0.5, and *<0.1.
- Table 3. Fiscal Multipliers and Spillovers Under Alternative Identification Restrictions (1 year)
  - Reproduces multipliers and spillovers using identification restrictions in Mountford and Uhlig (2009); sample 1999-2016; numbers in parenthesis indicate 68 percent confidence intervals.
- Figures (titles and notes preserved):
  - Figure 1. Openness to Trade and Fiscal Multipliers — Trade openness defined as sum of exports and imports over GDP; domestic fiscal multipliers correspond to Table 1 diagonal; bubble diameters proportional to country GDP.
  - Figure 2. Bilateral Trade Flows — Sample average 1999-2016; country codes: Austria (AT), Belgium (BE), Finland (FI), France (FR), Germany (DE), Italy (IT), Ireland (IE), the Netherlands (NL), Portugal (PT), Spain (ES).
  - Figure 3. Bilateral Financial Flows — Consolidated foreign claims, immediate borrower basis; sample average 1999-2016.
  - Figure 4. Foreign Claims by Country in 2006 (US $, billions) — Consolidated foreign claims, immediate borrower basis.
  - Figure 5. Foreign Claims by Country in 2006 (in percent of GDP) — Consolidated foreign claims, immediate borrower basis, expressed in percent of domestic GDP.
  - Figures 6–11, 12–13 — Evolution of exports/imports, fiscal multipliers across sub-samples, fiscal spillovers from Germany across sub-samples, domestic and international responses (government expenditure, output, current account, REER, CPI) to a 1 percent government spending shock; units, horizons, and statistical significance markers described in notes.

### Appendix 1 — Estimation Approach (Bayesian model specification and sampling)
- Model specification and notation (preserved exactly where shown):
  - Model in regression format: 푌푡 = ∑풵푗푡푝푗=1휃푗 + 푣푡
  - 풵푗푡 = 푍푡 Ξ푗 and 푣푡 = 퐸푡 + 푍푡 푢푡.
  - Assumption 푉 = 휎2퐼, leading to 푣푡 ~ 푁(0, (퐼+휎2푍푡′푍푡′)(푃⊗Ω)).
  - Unknowns: vector of factors 휃, scale factor 휎2, blocks P and Ω of VAR shock covariance.
- Prior distributions (exact forms):
  - 휃 ~ 푁(휃0, Θ)
  - 푃 ~ 퐼푊(푟, 푄)
  - Ω ~ 퐼푊(표, 푅)
  - 휎2 ~ 퐼퐺(0.5, 0.5푠2)
- Hyperparameter choices and computation:
  - 휃0 computed as average of cross-sectional data; Θ fixed constant.
  - Q estimated using residuals of country-specific models; R using residuals of variable-specific models.
  - 푠2 obtained using average of variance of residuals of AR(p) regressions of the NG endogenous variables.
- Likelihood function (preserved structure):
  - 퐿 ∝ (∏ |(퐼+휎2푍푡′푍푡′)(푃⊗Ω)|−1/2) exp[−1/2 ∑(푌푡−풵푡휃)′((퐼+휎2푍푡′푍푡′)(푃⊗Ω))−1(푡푌푡−풵푡휃)]
- Estimation algorithm:
  - Gibbs sampling cycles from conditional posteriors of each block of unknowns.
  - Metropolis step for 휎2 due to non-standard conditional posterior with Jacobian term.
  - Candidate draws in Metropolis: 휎2푖 = 휎2푖−1 + ℎ, with h ~ Normal(0, l); l chosen to target acceptance rate approximately 33 percent. Footnote 32: "The value of l is selected to have an acceptance rate of the order of approximately 33 percent."
  - Sampling details: results based on the last draw of 500 chains of length 1000 all starting in a small random interval of the last draw of a single (burn-in) chain of 100000 draws.
- Model features and departures from literature:
  - Model has heteroskedastic error term where time variation derived from lags of endogenous variables rather than an exogenous process.
  - Time-invariant parameters and VAR errors with fixed volatilities — noted as a departure from models with both time-varying coefficients and volatilities (Primiceri, 2005).

### Appendix 2 — Gibbs sampler output (posterior diagnostics and figures)
- Appendix Figure 1. Posterior Draws for Vector of VAR Coefficients θ
  - Note: Red line indicates initial value; posterior draws in blue; x-axis = total number of posterior draws used for inference, in this case 500.
- Appendix Figure 2. Posterior Draws for Country Covariance Matrix P
  - Same plotting conventions and x-axis = 500.
- Appendix Figure 3. Posterior Draws for Series Covariance Matrix Ω
  - Same plotting conventions and x-axis = 500.
- Appendix Figures 4–5. Posterior density distributions for θ, P, Ω, and 흈2
  - Dashed lines denote prior densities; solid lines denote posterior densities.
- Appendix Figures 6–8. Autocorrelation of posterior draws for θ, P, and Ω
  - Red lines denote 1.5 SD interval; dots indicate serial correlation at given lag (in quarters).

### Appendix 3 — Data sources (series names, series codes, and sources)
- Series list (selection preserving exact series codes and sources):
  - Population — XPOPT.P — Oxford Economics
  - GDP deflator — XPGDP.E / XPGDP.F — Oxford Economics
  - CPI — XCPI..E / XCPI..F — Oxford Economics
  - 10-year government bond yield — GBOND. — Datastream
  - Short-term interest rate — XRSHR.R — Oxford Economics
  - REER — Q..RECE — IMF - International Financial Statistics
  - Nominal GDP — XGDP..B / XGDP..E — Oxford Economics
  - Government revenue — XGREV.A / XGREV.B — Oxford Economics
  - Government expenditure — XGEXB.A / XGEXB.B — Oxford Economics
  - Government consumption — XGCN..B — Oxford Economics
  - Government interest — XGDPI.A / XGDPI.B — Oxford Economics
  - Government debt — XGGDB.A / XGGDB.B — Oxford Economics
  - Government transfers — XGCGP.A / XGCGP.B — Oxford Economics
  - Current AB — XBCU..A / XBCU..B — Oxford Economics
  - Government investments — XGINV.C / XGINV.D — Oxford Economics
  - Exports — Q7D0EXA — IMF - Direction of Trade Statistics
  - Imports — Q7D1EXA — IMF - Direction of Trade Statistics
  - US FFR — USPRATE. — Reuters
  - EU short-term interest rate, next 6 months — EXIFIRSTR — World Economic Survey, IFO
  - OPEC Oil Basket Price U$/Bbl — OILOPEC — OPEC
  - Crude Oil Average Spot Price — HWWICGE — HWWI
  - Nominal world GDP — WDXGDP..A — Oxford Economics
  - World GDP, PPP — WDXGPP..A — Oxford Economics
  - EUR to USD exchange rate — EUDOLLR — Reuters

*Italic: Content unit: wp17241 - REFERENCES (source PDF: wp17241 - REFERENCES).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17241.pdf_