## 1espea2026002

## Source details

**Canonical URL:** [1espea2026002](https://www.imf.org/-/media/files/publications/cr/2026/english/1espea2026002.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/cr/2026/english/1espea2026002.pdf.md)
- [Structured JSON version](/-/media/files/publications/cr/2026/english/1espea2026002.pdf.json)

---

### Primary balance, debt, and institutional context of autonomous communities
- Institutional features:
  - “Common regime” (15 communities): receive resources from a selection of taxes with partial rate-setting power and full autonomy over other taxes; revenues partly pooled into redistributive funds.
  - “Foral regime” (Navarra and País Vasco): pay a negotiated contribution to the general government’s budget but retain the entirety of their tax collections.
  - Some communities issue bonds, receive bank loans, and several finance part or all of their debt through Fondos de Financiación Autonómica.
  - Subnational fiscal rule: Organic Law on Budget Stability and Financial Sustainability 2/2012 (LOEPSF) sets yearly targets for deficits, primary spending growth, and debt levels.
- Key statistics and historical trajectory:
  - Regional spending ≈ 18 percent of GDP (2021); regions’ share of total public sector expenditure ≈ 27 percent (OECD, 2024).
  - Pre-GFC: consolidated regional budgets near balanced; regional debt ≈ 5 percent of GDP.
  - 2008–2019: primary (overall) balance mostly negative; peak deficits of 4.6 (primary) and 5.1 (overall) percent of GDP in 2011.
  - Regional debt rose to close to 25 percent of GDP in 2016; declined thereafter except in 2020.
  - By end-2025 total regional debt (including government-sponsored funds) remained above 20 percent of GDP.
  - Regional debt (20 percent of GDP) equals one-fifth of Spain’s total public sector debt and exceeds the LOEPSF long-term debt anchor of 13 percent of GDP.
  - Autonomous communities account for ~50 percent of all capital spending, amounting to 1.6 percent of GDP.

### Main descriptive and econometric findings on cyclicality and debt
- LOEPSF performance:
  - Historically produced partial compliance and inconsistencies across communities due to multiplicity of fiscal targets.
  - Methodology for yearly spending-growth targets under LOEPSF not aligned with new EU economic governance framework (operational in 2025).
- Cyclicality (econometric results 2004–2024):
  - Regional public spending positively correlated with the business cycle; spending fell in downturns (notably post-GFC).
  - Strong procyclicality of health and education spending in Spain’s autonomous communities, contrasting with peer euro area countries where health and education spending is mostly acyclical.
  - Consolidated regional governments mostly failed to stabilize debt dynamics over the period.
- Quantitative estimates — consolidated and panel results:
  - Consolidated sector: for total spending and public consumption, estimated β ≈ 0.5 (a 1 percent increase in the output gap associated with a 0.5 percent increase in spending/public consumption); primary spending coefficient about three times as large for the consolidated regional sector.
  - Panel of 17 communities (2004–2024, IV): Regional GDP - Detrended coefficients 0.366*** (Total Expenditure), 0.437*** (Primary Expenditure), 0.349*** (Public Consumption); Observations = 340; Number of CAs' = 17; First-Stage F-stat = 264; P-val = 0.00; R2 - Between: 0.32, 0.31, 0.39; R2 - Within: 0.03, 0.04, 0.07.
  - Robustness: results hold excluding Navarra and País Vasco, using single instruments, and controlling for liquidaciones two years prior.
  - GFC (2009–2012): procyclicality 4 to 6 times baseline value.
- Debt-stabilization evidence:
  - Panel fiscal reaction coefficient ≈ 0.02 — implying a 10-percentage point increase in debt associated with a 0.2-percentage point improvement in the primary balance.
  - Many individual communities show coefficients close to 0; most estimates statistically insignificant at 5 percent.
  - Stationarity condition considered with g̅ = 1.7 percent and r̅ = 1.2 (staff analyses): with g̅ = 1.7 percent, most communities stationary at pre-pandemic borrowing rates, but 6 communities become non-stationary with a 2 percentage point rise in financing costs; with r̅ = 1.2, a 2 percentage point fall in growth would render 6 regions non-stationary.

### Historical episode: GFC adjustment and consequences
- Revenue and transfer dynamics:
  - Consolidated regional revenues slowed from 2007; net transfers initially stable then rose, but contracted sharply in 2010–2011 (liquidaciones), producing a decade-long revenue slump.
  - Spending rose 2007–2009, then retrenched in 2008 and more decisively in 2010, including health and education.
- Composition of adjustment:
  - Capital expenditure fell by more than 50 percent in real terms over 2009–2014 and accounted for 54 percent (€16 billion) of the €30 billion contraction (2019 prices) in total regional spending over that period.
  - Soaring debt servicing costs 2009–2014 eased after Fondos de Financiación Autonómica reduced effective interest rates; debt plateaued but did not return to pre-GFC levels.

### Policy recommendations on subnational fiscal rules and spending
- Core recommendation:
  - Center a revised subnational fiscal rule on an expenditure growth target — consistent with the EU framework — to keep regional spending (especially health and education) mostly acyclical and achieve debt-stabilizing regional fiscal policy while avoiding procyclical cuts.
- Rule design options:
  - Region-specific spending growth limits; or
  - A common spending growth limit with tighter adjustment requirements for high-debt regions (e.g., regions with debt above 13 percent of GDP).
- Complementary design and enforcement features:
  - Ensure central government transfers to autonomous communities do not fall during downturns to reduce incentives for procyclical spending cuts.
  - Include a clear and applicable corrective arm for non-compliance, enforced by the Ministry of Finance.
  - Use multi-year spending trajectories and spending-efficiency reviews.
- Rationale against deficit-based regional rules:
  - Deficit targets can induce procyclical spending cuts, harming education and health and lengthening recessions (hysteresis).
  - Autonomous communities have limited debt-carrying capacity; central government is better placed for macro stabilization.

---

### Detailed recommendations and reform considerations for LOEPSF and rule implementation
- LOEPSF structure and compliance record:
  - LOEPSF (2013) objectives: structural deficit, spending growth (homogeneous nominal medium-term potential growth), and debt (13 percent of GDP for autonomous communities).
  - AIReF (2025): over 2013–2019 many communities annually did not comply with either the deficit or expenditure target; consolidated sector rarely met all three LOEPSF objectives simultaneously.
  - 2024–2025: fiscal rule paused during COVID-19; indicative targets 2021–2024; in 2024 a majority of communities exceeded then-binding expenditure rule with average growth about twice the target.
  - Multiplicity of yearly targets creates inconsistencies: identical deficit targets can imply different expenditure growth targets across regions.
- Design tradeoffs for a spending rule:
  - Individual targets: tailored but data-intensive and may complicate aggregate consistency and political acceptance.
  - Common target: simpler and transparent; can include differentiation (stricter adjustments for high-debt entities, escape clauses).
  - Methodologies: parsimonious statistical trends vs. aggregate/national approaches using projected spending needs and tax-raising capacity.
  - Time horizon: year-on-year targets offer flexibility but risk volatility; multi-year horizons provide commitment and protection during downturns, with escape clauses for large shocks.
- Compliance and enforcement upgrades:
  - Make coercive measures mandatory rather than optional; align spending limits with the maximum allowed spending rather than previous-year realized spending.
  - Improve timing and information content of Economic and Financial Plans (PEFs) to enable Ministry of Finance action.
  - Revisit access rules to Fondos de Financiación to ensure non-compliance triggers appropriate conditionality.

### Proposals recently advanced (AIReF 2025; Martínez‑López and others 2026)
- AIReF (2025):
  - Translate Spain’s Medium-Term Fiscal Structural Plan (MTFSP) primary-spending path into differentiated spending targets across levels of government, ensuring consistency with EU rule.
  - Apply net primary spending growth objective to each regional government, possibly with stronger adjustment for high-debt entities.
- Martínez‑López and others (2026):
  - Econometric, community-specific expenditure limits using Holt‑Winters trend revenue estimates over horizon h; set yearly growth rates to reach structural balance by t+h.
  - Consequence: communities with trend revenues above current expenditure face looser limits; those with trend revenues below face stricter limits.
- Additional considerations:
  - Ensure revenue cyclicality addressed and transfers predictable; some taxes (VAT, PIT) are strongly procyclical.
  - Account for demographic uncertainty; allow rule to adapt to evolving spending needs.
  - Clarify sequencing with any debt-reduction operation and consider central government fiscal space losses.

---

### Rationale and empirical support for maintaining acyclical health and education spending (chapter "10")
- Health and education:
  - Crucial for long-term growth; lower short-run multipliers; inelastic demand for essential services.
  - Cuts in downturns can lengthen recessions via hysteresis and produce lasting human capital scars (Jackson and others, 2021).
- Preconditions for acyclical spending:
  - Debt-issuance capacity to smooth essential spending when revenues fall (noting limited regional debt-carrying capacity relative to national government).
  - Stable and predictable revenue sources; central transfers should not fall in downturns.
  - During upswings, revenue surpluses should be used for debt repayment rather than discretionary expansion.
- Institutional actions:
  - Predictable multi-year spending trajectories, spending-efficiency reviews, and stability/transparency of transfer and liquidaciones mechanisms.

---

### Mortgage market: borrower-based measures (BBMs), loan-level evidence, and stress-test results
- BBMs overview and international context:
  - BBMs: LTV, LTI/LSTI, maturity and amortization limits; used by 18 of 21 banking-union countries as of 2025 (ECB 2025).
  - Spain: BBMs not yet activated by Bank of Spain; capital measures included CCyB = 1 percent (effective October 2026), G-SII/O-SII buffers 0.25–1.25 percent (2025 Bank of Spain); SyRB and Sectoral CCyB exist but not activated (2025 ESRB).
- Empirical loan-level strategy:
  - Dataset: EDW securitized loan records for Spanish residential mortgages 1999–2021 merged with Banco de España bank capital ratios and province-level data (50 provinces, excluding Ceuta and Melilla).
  - Regression baseline: reduced-form linear probability model for default (miss ≥3 consecutive monthly payments) with lending-standard dummies, loan/borrower/collateral controls, bank-by-quarter and province-by-quarter fixed effects.
- Loan-level findings:
  - Single-measure: LTV>80% → default probability +1.2 percentage points; LTI>6 or LSTI>40% → +0.2–0.4 percentage points.
  - Joint specifications: LTV effect remains ≈ 1.2 percentage points; LTI becomes insignificant; LSTI smaller but significant.
  - Interaction: LTV × LSTI ≈ +0.5 percentage points and significant.
  - Bank capital interaction: LTV × capital ratio ≈ −0.4 percentage points (attenuation); high-LTV loans retain significant default premium even with well-capitalized lenders.
- Annex I regression highlights:
  - Column (1) LTV>80% = 0.0117*** (0.0019); Observations = 1,015,509.
  - Column (3) LSTI>40% = 0.0037*** (0.0010); Observations = 1,012,602.
  - Annex I Table 2: LTV>80% # RWA Capital Ratio = -0.0039*** (0.0012); Observations = 463,933.
- Stress-test structure and calibration:
  - Portfolio coverage: CIRBE confidential credit registry covering Spain's €505.7 billion mortgage portfolio as of 2025Q2 (99.5 percent of outstanding mortgages).
  - Vintage-risk segmentation: 660 buckets (55 vintages × 12 LTV categories); borrower liquid financial assets = 1.15 percent of average property value (2023 HFCS).
  - Scenarios: Baseline follows January 2026 WEO; Adverse scenario (three-year, 2027Q2–2030Q2) per 2025 EU-wide stress test assumptions.
  - Adverse scenario key shocks (cumulative/changes):
    - House prices fall 17.2 percent cumulatively (Adverse N [-17.2%, standard deviation = 15%]).
    - Unemployment rises 5.6 percentage points.
    - Mortgage rates widen by average 160 basis points.
    - Household disposable income Adverse YoY growth = 0% (Baseline YoY growth = 12.9%).
    - Benchmark rate Adverse = 1.62 pps (Baseline = 0.19 pps).
- Monte Carlo PD/LGD simulation:
  - For each bucket draw N = 10,000 house prices; repeat simulation 1,000 times per bucket; compute PD_{i,t} = (∑ ED / N) × Pr(FD).
  - LGD calculation uses foreclosure discount δ, time-to-sell n = 6.17 years, discount rate on collateral = 2.7%, spread = 1.0%; foreclosure discount = 27.26%.
- Stress-test results (representative, point-in-time projections):
  - Outstanding portfolio (2027Q2):
    - Baseline: PD 0.77 ; LGD 10.92 ; Loss rate 0.29
    - Adverse: PD 2.63 ; LGD 32.2 ; Loss rate 0.99
  - New vintages (2025Q3–2027Q2):
    - Baseline: PD 3.75 ; LGD 14.65 ; Loss rate 1.53
    - Adverse: PD 12.76 ; LGD 36.13 ; Loss rate 5.21
  - Annualized mortgage loss rates: baseline 30 basis points → adverse 150 basis points.
  - Cumulative mortgage losses ≈ 4.6 percent of total mortgage portfolio ≈ €23 billion over three years.
  - Mortgages originated 2025Q3–2027Q2 face loss rates of 520 basis points in adverse scenario.
  - The 1 percent CCyB (scheduled Oct 2026) estimated ≈ EUR 16 billion and would mainly absorb non‑mortgage losses.
  - Under stress, credit losses concentrated more in non‑collateralized retail and corporate loans than in residential mortgages.
- BBMs impact from scenario analysis:
  - Regulatory limits assumed introduced 8 quarters before stress materializes.
  - Well-calibrated BBMs could reduce portfolio losses under severe stress by up to 25 percent after two years — ≈ €6 billion.
  - Most effective: LTV cap of 80 percent (hard limit or combined with LTI limit of 5 under speed-limit framework).
  - DSTI caps (30 or 40 percent) yield minimal benefits given system-average DSTI well below those thresholds.
  - An 85 percent LTV cap reduces losses by ≈ 6 percent; a 90 percent LTV cap is almost non-binding.
  - Behavioral assumptions: bunching below limits; speed limits adjust shares proportionally.
  - Limitations: analysis uses LTV rather than LTP; CIRBE covers main-residence mortgages only.

### Policy conclusions on BBMs and macroprudential mix
- Main diagnostics:
  - Spain currently lacks acute mortgage imbalances; household leverage relatively low and banks broadly solid.
  - Recent vintages show weakening lending standards (higher high‑LTV shares, rising LTI/DSTI among newer borrowers), increasing vulnerability.
- Recommendations:
  - Anchor BBM framework on a collateral‑based cap (LTV or LTP), possibly complemented by an LSTI cap targeted at loans that are both highly leveraged and have stretched debt-service burdens.
  - Preferred calibration: LTV cap ≈ 80 percent (hard or speed-limit with LTI ≈ 5) to achieve largest adverse-loss reduction.
  - Implement BBMs pre‑emptively, gradually, with clear communication, speed/exemption quotas (5–20 percent), and borrower-type or geographic differentiation to mitigate distributional impacts.
  - View BBMs and capital-based measures as complements; capital buffers alone do not address origination risks.

---

### Firm-level innovation, frictions, and policy simulations (chapters 13 and 26)
- Stylized facts and barriers:
  - Spain’s labor productivity ≈ 74 percent of US level.
  - R&D tax credit statutory design: 25 percent up to average of previous two years; 42 percent on additional expenditure; AIReF (2020) finds take-up ≈ 30 percent.
  - Patenting per million much lower than peers; within-sector margins ≈ 70 percent of patenting gap with US.
- Empirical two-step strategy:
  - Step 1: regress patenting indicator on friction measures (size-dependent regulations, financial constraints MRPK proxy, regulatory burdens).
  - Step 2: regress firm employment growth on interaction between patenting and friction to assess post-innovation growth dampening.
- Key empirical results:
  - Size-dependent threshold (45≤Emp<50): negative coefficients on patenting and citations (e.g., 45≤Emp<50 = -0.141*** for Any Patents).
  - Financially constrained (MRPK above sector-year median): Constrained = -0.087*** for Any Patents.
  - Regulatory intensity (∆ log # regulations): -0.022*** for Any Patents.
  - Post-innovation growth interactions: Innovation × 45≤Emp<50 = -0.191***; Innovation × Constrained = -0.035**; Innovation × Regulated = -0.046**.
  - Aggregate interpretation: three constraints together might account for ~88 percent of innovation gap relative to Sweden at employment 45.
- Structural model and calibration (extension of Klette and Kortum (2004)):
  - Frictions introduced: fixed cost φ when employment > 50; financial friction τ(n) = −τ log(n); entry cost δ; innovation types λ_H and λ_L; R&D cost C(x;n) with convexity η = 2.
  - Calibration targets include R&D/GDP = 1.49 percent, average TFP growth 2015–2025 = 0.67 percent, share of firms emp>50 ≈ 0.015.
  - Parameters: φ calibrated to match patenting decline at 45–50 employees; τ calibrated to MRPK slope; δ calibrated to entry rate.
- Quantitative counterfactuals and policy simulations:
  - Halving each friction one at a time:
    - Cutting financial frictions and regulatory burdens yields larger productivity gains than only addressing size thresholds.
    - Reducing all three frictions raises long-term productivity growth by 0.18 percentage points.
  - R&D tax-credit reforms:
    - Increasing take-up from 30% to 60% (effective subsidy from 10% to 20%) raises TFP growth by additional 0.06 percentage points.
    - Targeting increased take-up to young firms (<5 years) yields further 0.03 percentage points; combined targeted increase adds 0.09 percentage points.
  - Combined package (reduce three frictions + increase/target R&D subsidy) raises long-term productivity growth by 0.27 percentage points.
- Policy recommendations:
  - Pursue and amplify Regime 20; encourage autonomous communities to adopt open-market laws.
  - Support EU-level initiatives (EU 28th Regime, Savings and Investment Union).
  - Orient España Crece toward young, innovative firms.
  - Smooth size-dependent thresholds (particularly the 50-employee threshold).
  - Simplify and strengthen legal certainty of R&D tax credit (IMV coordination, ease refund-option restrictions: 20 percent discount, one-year wait, post-refund R&D staffing requirement).

*Source: 1espea2026002.*

### 1. Primary Balance and Debt of Autonomous Communities ______________________________ 5

### 1. Primary Balance and Debt of Autonomous Communities

### Overview and institutional context
- Autonomous communities deliver several essential public services including education, health, long-term care, and other social assistance and can finance spending by issuing debt.
- Under the “common regime” (15 communities) regions receive resources from a selection of taxes with partial rate-setting power and full autonomy over other taxes; revenues are partly pooled into redistributive funds.
- The “foral regime” (Navarra and País Vasco) pay a negotiated contribution to the general government’s budget but retain the entirety of their tax collections.
- Some communities issue bonds and receive bank loans; several finance part or all of their debt through central-government sponsored vehicles collectively denominated Fondos de Financiación Autonómica.
- The subnational fiscal rule is set by the Organic Law on Budget Stability and Financial Sustainability 2/2012 (LOEPSF) and sets yearly targets for deficits, primary spending growth, and debt levels.

### Key statistics and historical fiscal trajectory
- In 2021 Spain ranked third among OECD and EU economies for the size of public spending carried out by regional governments as a share of GDP—approximately 18 percent—and fifth as regards regions’ share of total public sector expenditure—approximately 27 percent (OECD, 2024).
- Prior to the Global Financial Crisis (GFC), consolidated regional budgets were close to balanced and total regional debt remained stable at around 5 percent of GDP.
- From 2008 to 2019 the primary (overall) balance of autonomous communities was mostly negative, with a peak deficit of 4.6 (5.1) percent of GDP in 2011.
- Regional debt rose pronouncedly after the GFC, edging close to 25 percent of GDP in 2016.
- Since 2016 consolidated deficits have decreased in size with several regions reaching surpluses; with the exception of 2020, regional debt has declined every year since its peak.
- By end-2025 total debt—including that issued via the government-sponsored funds—remained above 20 percent of GDP.
- The 20 percent of GDP in regional debt amounts to a fifth of Spain’s total public sector debt and exceeds the long-term debt anchor of 13 percent of GDP envisaged in Spanish law (Organic Law 2/2012).
- Autonomous communities account for approximately half of all capital spending (amounting to 1.6 percent of GDP), concentrated under economic affairs and general public service programs.

### Main findings from the analysis (descriptive and econometric)
- The current subnational framework under LOEPSF has historically produced partial compliance and inconsistencies across communities due to multiplicity of fiscal targets.
- Econometric analysis for 2004–2024 finds:
  - Regional public spending was positively correlated with the business cycle—i.e., spending fell in downturns, notably in the aftermath of the GFC.
  - Most regions failed to stabilize their debt dynamics over this period.
  - Spain shows strong procyclicality of spending on health and education (the communities’ main responsibilities), in contrast with peer euro area countries where health and education spending is mostly acyclical.
- The methodology for setting yearly targets on spending growth under LOEPSF is not aligned with the new EU economic governance framework (operational in 2025), creating potential conflicts in fulfilling both rules.

### Policy recommendations and rule design implications
- Center a revised subnational fiscal rule on an expenditure growth target—consistent with the EU framework—to:
  - Keep regional spending, particularly on health and education, mostly acyclical.
  - Achieve debt-stabilizing regional fiscal policy while avoiding procyclical cuts in recessions.
- Design options for an expenditure-based rule:
  - Region-specific spending growth limits, or
  - A common spending growth limit with tighter adjustment requirements for high-debt regions (for example, regions with debt above the 13 percent of GDP threshold).
- Complementary design and enforcement features:
  - Ensure central government transfers to autonomous communities do not fall during downturns to reduce the incentive for procyclical spending cuts.
  - Include a clear and applicable corrective arm for non-compliance, enforced by the Ministry of Finance.
- Rationale against deficit-based regional rules:
  - Deficit targets can lead to procyclical spending cuts that exacerbate recessions and undermine long-term growth via reductions in education and health.
  - Autonomous communities have limited debt-carrying capacity and are not well placed to conduct active countercyclical fiscal policy; the central government is better positioned for macroeconomic stabilization.

### Complementary evidence and literature context
- The paper relates to a substantial literature on regional fiscal performance and the effectiveness of fiscal rules:
  - Historical analyses of regional debt and fiscal stress following the GFC and euro area crisis.
  - Studies highlighting drivers of non-compliance with fiscal rules: ambitious/volatile targets, exogenous shocks, persistence of non-compliance, soft budget constraints, degree of fiscal autonomy, and political-economy factors.
  - Recent assessments that the expenditure rule has helped contain growth in current and primary expenditure to some extent despite partial compliance.

*Source: 1. Primary Balance and Debt of Autonomous Communities (excerpt), IMF staff paper, May 4, 2026.*

### 10.      Autonomous communities’ spending should not fluctuate significantly over the

### 10.      Autonomous communities’ spending should not fluctuate significantly over the business cycle.

### Rationale for maintaining acyclical spending in health and education
- Health and education are crucial for fostering long-term economic growth but do not have strong macroeconomic stabilization functions, as they generally have lower short-run multipliers (Atolia and others, 2017) and are mostly determined by inelastic demand for essential services.
- These areas do not lend themselves well to countercyclical spending and procyclical spending should be avoided because:
  - Health and education services provision should not be diminished or discontinued due to macroeconomic circumstances.
  - Evidence shows that cuts in health and education during downturns can lengthen the duration of the recession—a phenomenon called hysteresis—and inflict long-lasting scars on human capital and potential output (Jackson and others, 2021).
- Such spending programs should remain acyclical while contributing to debt sustainability and fiscal consolidation by following predictable multi-year trajectories and through spending efficiency reviews.

### Preconditions to sustain stable (acyclical) regional spending
- Autonomous communities require a combination of:
  - Debt-issuance capacity that allows maintenance of stable spending on essential services when revenues decline due to cyclical shocks.
  - Stability in revenue sources to reduce the need for debt issuance.
- Debt capacity constraints:
  - Autonomous communities can issue debt in the form of bonds and loans, but their overall debt-carrying capacity is more limited than that of the national government.
- Fiscal behavior during upswings:
  - During economic upswings, revenue surpluses should be devoted to debt repayments rather than increased provision of services above initial plans.
- Role of transfers:
  - Transfers from the central government should remain stable and predictable, if not actively offsetting fluctuations in communities’ own tax revenues, whose bases fluctuate with the business cycle.

### Historical experience: Global Financial Crisis (GFC) and euro area crisis in Spain
- Timeline and dynamics:
  - Consolidated revenues of autonomous communities began to slow down in 2007, driven primarily by lower tax revenues, while net transfers from the central government remained stable at first and eventually rose.
  - Spending continued to rise over 2007–2009, outgrowing revenues and leading to an initial rise in debt.
  - In 2010 and 2011, transfers from the central government, partly linked to the system of reconciliation of past tax revenues (liquidaciones), contracted sharply and persistently, producing a long-lasting slump in communities’ revenues that recovered their 2007 level in real terms only 10 years later.
- Adjustment path and composition:
  - The slump was accommodated through both a steep rise in debt and eventually a durable reduction in spending.
  - In 2008 and more decisively in 2010, autonomous communities retrenched spending in all areas, including health and education.
  - The adjustment was mostly carried out via cuts to capital expenditure, which fell by more than 50 percent in real terms over 2009–2014 and accounted for 54 percent (€16 billion) of the €30 billion contraction, in 2019 prices, in total spending by regional governments over the same period.
- Debt servicing and eventual relief:
  - Soaring debt servicing costs for most communities started in 2009 until 2014, when establishment of the Fondos de Financiación Autonómica reduced effective interest rates and allowed regions to regain fiscal space.
  - In the following decade, despite recovery in revenues and lower borrowing costs, debt plateaued but did not decline to pre-GFC levels.

### Empirical evidence on cyclicality of regional spending
- Definitions and methodology:
  - Three definitions of spending analyzed: total spending; primary spending (excludes interest costs); public consumption (excludes interest, investment, and transfers).
  - Series expressed in constant 2019 prices. Cyclical components extracted from log series using Hodrick-Prescott filter with smoothing parameter 100.
  - Baseline regression: Cycle(Log Expenditure_t) = α + β Cycle(Log GDP_t) + γ X_t + ε_t. Coefficient β is the focus: negative → countercyclical; positive → procyclical.
  - Instrumental variables used in panel regressions: (i) average output gap of bordering autonomous communities; (ii) export-weighted output gap of Spain’s trading partners.
- Visual and regression findings:
  - Autonomous communities’ spending has been strongly procyclical throughout the last three decades, with alignment to the business cycle (and its lag) and larger amplitude of fluctuations than the central government.
  - The COVID-19 pandemic is an exception: spending by autonomous communities rose markedly in response to the health emergency despite a sharp economic contraction.
  - Primary expenditure appears most procyclical because it excludes interest costs (which can be countercyclical) but includes capital expenditure, historically strongly positively correlated with the cycle.
- Quantitative estimates — consolidated levels of government:
  - For autonomous communities, estimated β coefficients are positive, statistically different from zero, and larger than for the central government.
  - For total spending and public consumption, the estimated coefficient is close to 0.5: a one percent increase in the output gap has historically been associated with a 0.5 percent increase in total spending and public consumption.
  - The coefficient for primary spending is about three times as large for the consolidated regional government sector.
  - Estimated coefficients for the central government are not statistically different from zero (acyclical overall), though primary spending coefficient appears slightly more positive.
- Quantitative estimates — panel of 17 individual autonomous communities:
  - Panel regression (community-specific fixed effects, 2004–2024) with instruments yields estimated coefficients in the range of 0.3-0.4 for total expenditure, primary expenditure, and public consumption.
  - Table results (IV regression summary):
    - Regional GDP - Detrended: 0.366*** (Total Expenditure); 0.437*** (Primary Expenditure); 0.349*** (Public Consumption).
    - Constant: 0.003*** for all three specifications.
    - Observations: 340; Number of CAs': 17.
    - IV 1: Neighbors' GDP = YES; IV 2: Trading Partners GDP = YES.
    - First-Stage F-stat: 264; P-val: 0.00.
    - R2 - Between: 0.32 (Total), 0.31 (Primary), 0.39 (Public Consumption).
    - R2 - Within: 0.03 (Total), 0.04 (Primary), 0.07 (Public Consumption).
    - Robust standard errors in parentheses; significance: *** p<0.01.
- Robustness and sensitivity:
  - Results robust to excluding Navarra and País Vasco (foral regime), using one instrumental variable at a time, and controlling for liquidaciones of tax revenues from two years prior.
  - GFC period (2009–2012) featured very strong procyclicality (4 to 6 times the baseline value), but the rest of the sample also shows statistically significant procyclicality.
  - Alternative specifications: first-differences regressions yield smaller but still positive and significant coefficients; System-GMM (Arellano-Bond) shows spending persistence with lagged dependent variable coefficient around 0.7 for all three expenditure definitions, and output gap coefficient about half as large but still statistically significant.

### Policy implications and recommendations
- Maintain acyclical provision of health and education by:
  - Ensuring predictable multi-year spending trajectories and conducting spending efficiency reviews.
  - Strengthening autonomous communities’ access to stable revenue sources and predictable central government transfers to reduce forced procyclical adjustments.
  - Using upcycle revenue surpluses for debt repayment rather than discretionary service expansions that raise vulnerability to downturns.
  - Preserving some debt-issuance capacity or smoothing mechanisms so that communities can maintain essential spending when revenues decline.
- Institutional reforms and monitoring:
  - Stability in the central-to-regional transfer system and transparency around liquidaciones and reconciliation mechanisms to avoid persistent revenue slumps.
  - Consider comprehensive reviews of health and education spending (as recommended in the 2025 Article IV) to improve efficiency and long-term sustainability.

*Source: IMF staff analysis as presented in the chapter "10.      Autonomous communities’ spending should not fluctuate significantly over the business cycle."*

### 22.       Spending on health and

### 22.       Spending on health and

### Procyclicality of health and education spending
- Regression specification (1) is adapted to separately examine business cycle behavior of spending on health and education for:
  - Spain’s general government;
  - Spain’s autonomous communities;
  - Germany, France, and Italy;
  - all EU countries excluding Spain.
- For both health and education, the estimated coefficient for Spain’s total general government spending and for spending by regions is above that obtained for other European countries.
- In contrast to Spain, procyclicality of spending in health and education is not the norm among the peer economies studied.
- Estimation details:
  - System GMM is used to handle the lagged dependent variable in a fixed-effects framework, using up to the tenth lag of the dependent variable and the “collapse” option to reduce instruments.
  - The output gap remains instrumented using neighbor regions’ and trading partners’ GDP.
  - Baseline specification uses the contemporaneous output gap; results are robust to using the one-year or two-year lag of the output gap.

### Debt stabilization of autonomous communities
- Historical descriptive evidence:
  - Since 2000, autonomous communities’ fiscal policy has not been strongly responsive to debt levels.
  - Figure 9 plots communities’ debt-to-GDP ratios against fiscal impulse (yearly change in the primary balance) and shows a flat relationship for the full period, with a positive association only during 2008–2014.
  - As a result, debt ratios remained generally high after 2010.
- Fiscal reaction function (Bohn-style) specification for an individual community:
  - PBt/GDPt = α + β (Dt−1/GDPt) + γ Cycle(Log GDPt) + εt
  - Coefficient of interest: β — higher β implies stronger primary balance response to debt (greater debt stabilization).
- Estimated coefficients:
  - Many individual communities show coefficients close to 0; sample sizes yield large confidence intervals and mostly statistically insignificant coefficients at the 5 percent level.
  - The panel of communities’ coefficient is approximately 0.02 — implying a 10-percentage point increase in debt is associated with a 0.2-percentage point improvement in the primary balance.
  - A few communities show coefficients comparable to Bohn (1998) for the US, but most do not.
- Stationarity condition and sensitivity to r̅ and g̅:
  - Stationarity requires (1+ r̅ − g̅)(1− β) < 1.
  - Staff sets g̅ = 1.7 percent (medium-term projection for Spain) in one analysis and r̅ = 1.2 in another.
  - Historical reference: Spain’s 10-year bond yields averaged 4.4 percent and inflation averaged 3.2 percent in 2000–GFC, approximating r̅ ≈ 1.2 after adjustment.
- Scenarios for non-stationary debt dynamics:
  - With g̅ = 1.7 percent, most communities remain stationary at real borrowing rates in line with pre-pandemic Spain central government rates, but the debt dynamics of 6 communities would become non-stationary with an increase in financing costs of 2 percentage points.
  - With r̅ = 1.2, a reduction in growth of 2 percentage points would result in non-stationary debt dynamics for 6 regions; such a durable fall in real growth into negative territory is deemed unlikely relative to financing-cost shocks.

### Historical compliance with the subnational fiscal rule (LOEPSF and predecessors)
- Evolution of rules:
  - From 1992: yearly fiscal deficit targets agreed between regions and central administration.
  - 2001 Budget Stability Law: from 2002 regions required balanced budgets or surplus.
  - 2006 Budget Stability Law (in force 2007): fiscal deficit limits in percent of regional GDP, with corrective/coercive mechanisms.
  - LOEPSF (introduced 2013) articulates objectives based on three fiscal measures differentiated by government level: structural deficit, spending growth, and debt.
- LOEPSF specifics:
  - Structural deficit: all administrations should achieve a structural balance; Ministry of Finance sets a deficit limit for consolidated general government and breaks it down; since LOEPSF inception, with the exception of 2013, all autonomous communities have been subject to the same deficit target.
  - Spending growth: limit computed based on Spain’s nominal medium-term potential growth rate, set homogenously for central, regional, and local governments.
  - Debt limits compatible with EU target of 60 percent of GDP are split: 44 percent for central government, 13 percent for autonomous communities, and 3 percent for municipalities. Entities above their limits must publish multi-year debt reduction plans with yearly debt targets compatible with deficit and spending growth targets.
- Compliance record (summary of empirical findings):
  - AIReF (2025) analysis: over 2013–2019, a substantial share of communities each year did not comply with either the deficit or expenditure target, or both.
  - Consolidated regional sector: some years met either the deficit target (2017, 2018) or the expenditure target (2013, 2014, 2016), but never both simultaneously.
  - Debt target: higher compliance by individual communities (often above 80 percent); consolidated regional sector collectively met the debt target each year except 2015.
  - Simultaneous compliance with all three LOEPSF objectives was rare at the individual level and never occurred at the consolidated level.
- Recent developments and 2024–2025 context:
  - Over COVID-19 pandemic the fiscal rule was paused; indicative non-binding targets were set over 2021–2024, and communities’ deficits moved gradually closer to targets with the post-pandemic recovery.
  - In 2024, when the expenditure rule was the only legally binding limit, a large majority of communities exceeded it, with an average growth rate about twice as high as the target.
  - Although a consolidated regional-sector limit for 2024 and 2025 was prepared and approved by the Council of Ministers, these and individual limits for communities were not approved by Congress and by the CPFF; hence the national rule was in place but not operationally binding.
- Inconsistencies from multiplicity of yearly targets:
  - A given deficit target can imply different expenditure growth targets across communities depending on revenues and macro growth, producing situations where one target is met but the other is not.
  - Example year 2016: Extremadura, Cantabria, and Comunidad Valenciana met the expenditure limit (growth ≈ 0) but had deficits ≈ 1.5 percent of GDP (above 0.7 limit), while Castilla y León and Comunidad de Madrid had similar expenditure growth but complied with the deficit rule; País Vasco, Andalucía, and Galicia had comparable deficits but much higher expenditure growth.

### Considerations for reforming the national fiscal rule
- Core recommendation:
  - Given the case for acyclical spending and autonomous communities’ limited debt issuance capacity, a rule focused on spending growth would be preferable to a deficit rule.
  - A spending rule, possibly based on a multi-year framework, would ensure stable and predictable spending on health, social care, and education, independent of business-cycle surprises — preventing expenditure cuts during downturns and limiting procyclicality.
- Design tradeoffs and key choices:
  - Individual vs common targets:
    - Individual targets: account for community-specific structural and conjunctural factors and could better achieve debt sustainability, but require granular data (potential growth, output gap, structural revenues), may be hard to make compatible with aggregate targets, and could create political-economy issues.
    - Common target: less tailored but easier to implement, more transparent, and can be made ex-ante consistent with EU fiscal rule via top-down allocation; differentiation can be added through additional features (e.g., stricter adjustments for high-debt communities or escape clauses).
  - Methodology for setting the target:
    - Parsimonious statistical approaches (e.g., trend growth in revenues) require only historical fiscal data but may not capture structural differences (demographics, service delivery costs, climate exposure) or planned revenue measures.
    - Aggregate/national approaches can use projected spending on health, long-term care, and education and communities’ tax-raising capacity.
  - Time horizon:
    - Year-on-year targets (current subnational rule) provide flexibility but may yield volatile spending if sensitive to cyclical measures.
    - Multi-year horizons provide stronger commitment to stable spending, protect spending during downturns, and allow better assessment of whether revenue surprises are temporary or structural; escape clauses can preserve flexibility for large shocks.
  - Additional features:
    - Escape clauses for large exogenous shocks (natural disasters, deep recessions).
    - Stricter adjustments for communities with debt above chosen thresholds.
    - Debt limits and escape clauses are embedded in both the LOEPSF and the EU fiscal rule.
- Implementation considerations:
  - Choices should be grounded in economic principles and the broader context of the budget cycle, political economy, and technical feasibility.
  - Data availability, transparency, and perceived fairness are important constraints when choosing between granularity and feasibility.

*Sources: Eurostat, IGAE, INE, Bank of Spain, Ministry of Finance, AIReF, and IMF staff calculations.*

### 35.      Two recent proposals for reforming the regional fiscal framework by AIReF (2025) and

### Two recent proposals for reforming the regional fiscal framework by AIReF (2025) and Martínez-López and others (2026)

### Proposal summaries and methodological differences
- AIReF (2025) approach:
  - Starts from Spain’s Medium-Term Fiscal Structural Plan (MTFSP) path under the EU rule as the national starting point to ensure consistency between national and EU targets.
  - Spain’s target for primary spending growth net of revenue measures under the MTFSP in a given year is translated into differentiated targets across levels of government.
  - The translation into regional spending limits entails a policy decision over which areas of spending will carry the efforts of fiscal adjustment.
  - Multiple decision-making processes and methodologies could be used to set spending growth limits.
  - The resulting net primary spending growth objective for the collective of autonomous communities would then be applied to each individual regional government, possibly with stronger adjustment required for entities with high debt.

- Martínez-López and others (2026) econometric approach:
  - Compute community-specific expenditure limits using an econometric method.
  - For each community, estimate the trend of revenues based on a (Holt-Winters) filter for h years ahead.
  - From the community’s current expenditure level at time t, set yearly growth rates until t+h such that by t+h the community is in structural balance (expenditures equal trend revenues).
  - Consequence: communities with trend revenues currently higher than expenditure or with revenues expected to rise faster will face looser expenditure growth limits; communities where expenditures currently exceed trend revenues or where revenues are projected to grow slowly will face stricter expenditure growth limits.

### Additional considerations for an expenditure-based fiscal rule
- Revenue cyclicality and debt capacity:
  - Regions have limited debt-carrying capacity; acyclical spending requires revenues to be acyclical to the extent possible.
  - Some taxes assigned to communities, like the VAT and the PIT, are strongly procyclical; it is essential that other revenue components remain at a minimum acyclical.
  - Central government transfers could, if necessary, be explicitly countercyclical in case of shocks disproportionately affecting regional spending (example: COVID-19 pandemic).

- Demographic uncertainty and adaptability:
  - High uncertainty related to demographic dynamics and aging-related pressures requires the fiscal framework to adjust over time to evolving spending needs.
  - Methodologies should reflect developments such as slower demographic aging via higher migration inflows into revised spending and revenue projections without excessive lags.

### Compliance, corrective and coercive mechanisms
- Need for clear, enforceable mechanisms:
  - An effective rule requires a clear and enforceable set of preemptive, corrective, and coercive mechanisms to ensure compliance.
- Weaknesses in the current LOEPSF framework (as identified by Martínez-López, 2020):
  - LOEPSF is on paper one of the strictest subnational rules across the EU, but its potential strictness is not matched by a strong compliance track record.
  - Several coercive measures to be taken by supervisory entities are specified as optional rather than mandatory.
  - For the spending growth rule, non-compliance does not preclude access to the borrowing instruments of the Fondos de Financiación that entail lower conditionality, while non-compliance of the deficit rule does preclude such access.
  - Yearly limits are set based on the previous year’s realized spending rather than the (lower) maximum allowed spending, so absence of correction for past deviations provides an indirect incentive for non-compliance.
  - Economic and financial plans (PEFs) required in case of non-compliance often lack the information necessary for the Ministry of Finance to decide on potential coercive actions due to misaligned timing with the regular budget cycle.
- Recommendation:
  - Revisit corrective and coercive mechanisms to make them compatible with reformed targets and time frames, emphasizing graduality and political feasibility so that provisions on paper can be strictly enforced in practice.

### Interaction with other reforms and sequencing
- Complementarity with multi-year targets and revenue predictability:
  - Multi-year targets under the revised rule could be more easily set if revenues were predictable over several years ahead.
- Debt reduction operation and debt limit clause:
  - If the rule maintained a debt limit clause (example given: current threshold of 13 percent of GDP or a revised one), the exact design of a one-time debt absorption operation would determine whether high-debt autonomous communities still need strict adjustments in the first few years under the revised rule.
  - If the amount of relief received is not large enough for some communities’ total debt level to fall below the debt threshold of the revised rule, those communities would still face significant initial adjustment needs.
- Loss of central government fiscal space:
  - The design of the rule should recognize and address the loss of central government fiscal space resulting from any proposed debt reduction and revision to the “common regime” framework.
  - During the adjustment period of the MTFSP, different formulations of the rule would shift the shares of the envisioned aggregate consolidation borne by the central government and the autonomous communities.

### Political economy, ownership, and enforceability
- Ownership and political commitment:
  - An enhanced subnational fiscal rule should foster ownership and political commitment by both regional and central governments.
  - Effectiveness depends on technical design and acceptability of the process through which the rule is developed and enforced.
  - The fiscal rule reflects societal choices on redistribution and risk-sharing across regions (e.g., region-specific versus common targets) and preferences about prioritization of spending across public entities.
- Importance of dialogue and collaboration:
  - Dialogue among stakeholders, especially between central and regional governments with diverse positions, is essential to reach politically viable compromises.
  - A predominantly top-down framework, even if economically well grounded, risks limited buy-in from autonomous communities and could undermine compliance and credibility.
  - A collaborative approach is especially important to ensure acceptability and strong enforcement in the case of differentiated targets across autonomous communities.

*Source: IMF staff summary of section on regional fiscal framework reform proposals in the provided content unit.*

### 3.      Despite relatively low household leverage, persistently strong house price growth and

### 3.      Despite relatively low household leverage, persistently strong house price growth and emerging signs of easing in lending standards raise the question of whether pre-emptive borrower-based measures (BBMs) are warranted

### Overview of BBMs and complementary macroprudential measures
- BBMs are regulatory limits on loan underwriting criteria most often applied to residential real estate lending, directly constraining borrowers’ capacity to borrow and containing build-up of vulnerabilities (ECB 2025).
- Common BBMs and their roles:
  - LTV ratios: cap the loan amount at a fixed share of the property's value at origination; borrower equity acts as a first-loss buffer and lowers banks' loss given default.
  - LSTI or LTI ratios: restrict borrowers’ total debt service or debt stock relative to gross income; lower probability of default by strengthening households’ ability to withstand interest rate or income shocks.
- Non-BBM complements frequently applied alongside BBMs:
  - Loan maturity limits (e.g., 30 years for mortgages) to prevent excessively long loans and accelerate amortization (2025 France FSSA).
  - Amortization requirements and interest-only restrictions to impose repayment structure and reduce default risks.
- Current Spanish capital-based measures (complements to BBMs):
  - Countercyclical capital buffer (CCyB) set at 1 percent, effective October 2026.
  - Capital buffers for systemically important institutions (G-SII and O-SII) with rates ranging from 0.25 to 1.25 percent (2025 Bank of Spain).
  - Systemic Risk Buffer (SyRB) and Sectoral CCyB exist in framework but are not currently activated (2025 ESRB).
- Adoption across euro area:
  - As of 2025, 18 out of 21 banking union countries have at least one active BBM (ECB 2025).
  - Use of combined collateral-, income- and maturity-based limits has risen markedly; evidence indicates BBMs have helped curb risky mortgage lending and strengthen household and bank resilience (e.g., Ahuja and Nabar, 2011; Kuttner and Shim, 2016; Valderrama, 2023).
- Scope:
  - BBMs are primarily designed for residential real estate lending; some countries extend them to consumer credit, while commercial real estate is generally addressed through capital-based tools.
- Spanish context:
  - To date, BBMs have not been activated as part of the Bank of Spain’s macroprudential toolkit, while they are in place in most other euro area countries.

### Design and calibration considerations
- Overarching objective: balance financial-stability benefits (limiting excessive leverage and debt burdens) against costs in reduced credit access and potential distributive consequences.
- Key design principles:
  1. Timing: most effective when introduced early in the financial cycle, before systemic risks accumulate (Bianchi, 2011; Valderrama, 2023; Biljanovska and others, 2023). Preemptive activation is less distortionary than late tightening.
  2. Calibration: should be data-driven and grounded in country-specific risk analysis using granular, loan-level data and model-based stress testing (e.g., Gornicka and Valderrama, 2020).
     - International benchmarks: LTV limits typically set between 80 and 100 percent for first-time buyers purchasing a primary residence, and 70-90 percent for repeat buyers.
     - Income-based (LSTI/LTI) caps: range from 30 to 80 percent of borrower’s monthly income across countries.
  3. Multi-pronged approach: LTV and LSTI caps address different dimensions of risk—loss given default and likelihood of default—and are mutually reinforcing; maturity caps and amortization requirements help prevent circumvention.
  4. Flexibility: use of exemption quotas or “speed limits” typically allowing 5-20 percent of new loans to exceed caps preserves banks’ discretion and credit access for atypical but creditworthy borrowers.
  5. Targeted calibration: geographic or borrower-type differentiation can focus tightening where risks concentrate (examples provided from Norway, New Zealand, Denmark, Hungary, Slovenia).
     - Norway: stricter LTV and lower speed limit in Oslo for secondary dwellings through end-2022; since January 2023, uniform national LTV with Oslo speed limit 8 percent of new lending per quarter vs 10 percent elsewhere.
     - New Zealand: tighter LTV limits and lower speed limits for Auckland investment properties introduced November 2015; investor LTVs tightened nationwide to 60 percent in October 2016; LTI restrictions added from July 2024.
     - Denmark: Growth Area Guidelines for Copenhagen and Aarhus from 2016, with additional 2018 restrictions for borrowers with LTI above 4 and LTV above 60 percent.
     - Hungary: differentiates LSTI limits by income level, with a higher cap applying above a monthly net income threshold (about EUR 2,000).
     - Slovenia: income-tiered LSTI caps in 2018 with 50 percent for borrowers earning up to twice the gross minimum wage and 67 percent for the portion above that threshold; since July 2023, uniform 50 percent LSTI cap.
  6. Implementation: adopt gradual transition periods or supervisory guidance before binding limits; communicate clearly that BBMs are ceilings (permanent structural safeguards) not targets to avoid loosening of standards.
- Distributional trade-offs: tighter BBMs can be more binding for lower-income households and may affect access to home ownership; exemptions and targeted measures can mitigate adverse distributional impacts.

### Empirical approach and data for Spain
- Analytical building blocks of the Selected Issues paper:
  1. Review of international experience on BBMs: objectives, benefits/costs, design features, calibration practices across euro area peers.
  2. Loan-level econometric analysis: assessment of how LTV, LTI, and LSTI ratios at origination affect probability of mortgage default in Spain and interaction with lenders’ capitalization.
  3. Scenario-based stress analysis: quantify how alternative BBM calibrations could lower bank losses under an adverse macroeconomic scenario.
- Loan-level dataset and linkage:
  - Source: European DataWarehouse (EDW) securitized loan records for residential mortgages originated in Spain between 1999 and 2021.
  - EDW reports LTV at origination plus loan, borrower, and collateral characteristics and performance indicators.
  - EDW records are merged with bank-level capital ratios from Banco de España (matched by originating lender) and province-level data for the 50 Spanish provinces (excluding Ceuta and Melilla), including GDP and real house prices (deflated by province-level CPI) from Sociedad de Tasación.
  - Caveat: EDW captures securitized loans rather than the universe of Spanish mortgage originations; potential representativeness bias. However, Galán and Lamas (2025) show securitized Spanish mortgages are broadly representative along key risk dimensions for the period they analyze.

### Key findings, implications, and policy-relevant takeaways
- Rationale for preemptive BBMs:
  - Sustained increases in valuations may create pressure to ease bank lending standards or encourage riskier lending behavior, consistent with historical evidence (Martín and others, 2021; Duca and others, 2011; Justiniano and others, 2019).
  - Implementing BBMs before they become strongly binding can minimize economic, social, and political costs because they are unlikely to materially constrain credit if introduced early.
- Complementarity with capital measures:
  - BBMs limit leverage and ensure debt serviceability at issuance, strengthening households' balance sheets and improving banks' asset quality ex ante, while capital-based measures act on the outstanding stock of exposures and improve resilience ex post.
- International experience and effectiveness:
  - Evidence indicates BBMs have helped curb high-LTV and high-LSTI lending, strengthen household and bank resilience, and have been associated with milder housing corrections and smaller increases in default rates during recent monetary tightening episodes (2024 IMF WEO).
- Design implications for Spain:
  - Authorities should consider whether to introduce BBMs given persistently strong house price growth and emerging easing in lending standards.
  - If considered, design and calibration should be:
    - Data-driven using granular loan-level information and scenario-based stress testing.
    - Multi-instrument (LTV + LSTI/LTI ± maturity/amortization limits) and possibly targeted (geographic, borrower-type, or income-tiered) to minimize unintended distributional impacts.
    - Implemented early, gradually, and with clear communication and transition arrangements.
  - Use of exemption/speed limits (typically 5-20 percent in peer countries) and borrower-type differentiation (e.g., first-time buyers) can preserve access for creditworthy borrowers while containing systemic risk.

*Source: IMF Selected Issues paper excerpt.*

### 19.      The empirical strategy follows a standard reduced-form linear probability model

### 19.      The empirical strategy follows a standard reduced-form linear probability model

### Empirical strategy and baseline specification
- Baseline model: reduced-form linear probability model relating an indicator of default (equal to one if the borrower misses at least three consecutive monthly payments) to:
  - dummies for loans with risky lending standards at origination,
  - controls for loan, borrower, and collateral characteristics,
  - high-dimensional fixed effects.
- Baseline regression equation (variables and indexing preserved as in source):
  - Defaultribt = α + β1 LS1 irbt + β2 LSj irbt + β3 (LS1 irbt × LSj irbt) + γ X irbt + δ Y irbt + θ Z irbt + λ tb + μ rt + ε it
  - where i, r, b, t index loans, provinces, banks, and origination quarters.
- Lending-standard dummies:
  - LS1: dummy = 1 for loans with LTV above 80 percent.
  - LSj: income-based lending-standard dummy (LTI>6 or LSTI>40 percent), used on its own or to form combinations with the collateral-based LTV dummy.
- Controls:
  - X (borrower characteristics): log income, job status, nationality, multiple loans.
  - Y (loan terms): maturity, fixed versus variable rate, loan purpose, interest rate.
  - Z (collateral features): property type and recourse.
- Fixed effects:
  - Bank-by-origination-quarter fixed effects (λ tb) to absorb time-varying lender heterogeneity.
  - Province-by-origination-quarter fixed effects (μ rt) to absorb local macroeconomic and housing-market conditions at origination.
- Three nested estimations:
  1. Single-measure specification: each lending-standard dummy entered alone.
  2. Joint specification: LTV together with either LTI or LSTI.
  3. Interaction specification: augmented with LTV × (LTI or LSTI) interaction to isolate marginal risk of loans that breach both collateral- and income-based thresholds.

### Main empirical findings on lending standards and default risk
- Single-measure specification:
  - Loans with LTV above 80 percent: default probability 1.2 percentage points higher than otherwise comparable loans.
  - Loans with LTI above 6 or LSTI above 40 percent: default probability 0.2–0.4 percentage points higher.
- Joint specifications (LTV with LTI or LSTI):
  - LTV effect remains about 1.2 percentage points and significant at the 1 percent level.
  - LTI coefficient becomes statistically insignificant.
  - LSTI coefficient gets smaller but remains significant (results not shown).
- Interaction (LTV × LSTI):
  - Standalone LSTI coefficient loses significance.
  - Interaction term (LTV × LSTI) is positive, about 0.5 percentage points, and statistically significant.
- Interpretation:
  - Ordering of risk: collateral-based looseness (high LTV) is the primary driver of default.
  - Income-based looseness matters mainly through interaction with high-LTV lending.
  - Among income-based measures, LSTI appears to matter more than LTI.

### Interaction with originating bank capital ratio
- Augmented model: baseline specification plus interaction between lending-standard dummies and originating bank's capital ratio at issuance.
- Key findings:
  - High-LTV loans continue to be linked to higher default probability: about 1.1 percentage points.
  - Interaction term LTV × capital ratio: negative and significant at the 1 percent level, around -0.4 percentage points.
  - Similar, though weaker, attenuation effect for high-LTI loans.
  - No attenuation effect found for high-LSTI loans.
- Interpretation and caveats:
  - Better-capitalized banks appear to originate loans with lower realized default risk for given loan characteristics, consistent with better screening, monitoring, and workout capacity.
  - Caveat 1: estimates capture average relationship and mask heterogeneity across banks; lenders with capital ratios below average would see a much smaller mitigating effect.
  - Caveat 2: attenuation is partial—high-LTV loans retain an economically and statistically significant default premium even among well-capitalized lenders.

### Scenario-based calibration of borrower-based measures (BBMs) and portfolio coverage
- Data and coverage:
  - Confidential credit registry data from CIRBE covering Spain's €505.7 billion mortgage portfolio as of 2025Q2, representing 99.5 percent of outstanding mortgages.
- Purpose:
  - Assess resilience of portfolio to an adverse macroeconomic scenario with versus without BBMs.
  - Evaluate effectiveness of alternative tools: LTV, DSTI, LSTI, LTI, and combinations.
- Structural model:
  - Builds on framework by Górnicka and Valderrama (2020), adapted to Spanish mortgage and housing markets.

### Portfolio segmentation and characteristics used in stress analysis
- Segmentation:
  - Portfolio segmented into 660 loan vintage-risk buckets: 55 loan vintages (quarterly from 1990Q1 to 2025Q2, with pre-2012Q1 loans grouped together) crossed with 12 LTV categories.
  - Each bucket characterized by risk metrics (LTI and DSTI ratios), repricing schedule (six repricing buckets), and average effective maturity.
- Borrower liquid financial assets:
  - Estimated at 1.15 percent of average property value based on the 2023 HFCS; incorporated into default trigger to assess ability to draw down savings.
- Macroeconomic paths:
  - Baseline scenario variables (unemployment, household income, GDP) drawn from January 2026 WEO.
  - Adverse scenario deviations and house price projections follow 2025 EU-wide stress test assumptions.

### Recent vintage composition and trends in lending standards (key statistics)
- Vintage composition:
  - 20 percent of outstanding loans were originated between 2004–2008 (pre-crisis boom).
  - About half of outstanding loans were issued after 2020.
- Rate type:
  - Mortgages predominantly issued at floating rates; share of fixed-rate loans increased following the 2022 ECB monetary hiking cycle.
- LTI trends:
  - Average LTI ratio peaked at 4.9 in 2020Q2 and declined to 4.2 by 2025Q2.
  - Share of loans with LTI ratios above 4 fell from 58.6 percent in 2021Q4 to 43.7 percent in 2024Q1, then increased to over 50 percent in 2025Q2.
  - In 2025Q2, one-third of new issuances had LTI ratio above 5.
- LTV and DSTI trends:
  - Share of high‑LTV mortgages (LTV>80 percent) rose from 32.2 percent in 2018 to 34.7 percent in 2025Q2.
  - Average DSTI ratios across all mortgages rose from 30 percent to 33 percent between 2018 and 2025Q2.
  - Average LTV ratio fell from about 80 percent in 2012 to 64 percent in 2023, then rose to 68 percent by 2025Q2.
  - High‑LTV originations declined from over 40 percent to 9 percent over the previous decade, but rose to 16 percent by 2025Q2.

### Risks, model structure, and calibration
- Risk amplification:
  - Deterioration in lending standards over past two years increases vulnerability of recent vintages; recent borrowers have repaid less principal and hold smaller equity buffers, making them more vulnerable to rising interest rates and falling house prices.
- Default concept: “double trigger” approach
  - Default occurs when both:
    1. Financial distress (inability to service the loan; liquidity shock), and
    2. Economic default (negative home equity: property value below outstanding loan value).
- Financial distress probability (equation preserved from source):
  - Pr(FD i,t) = Φ(DSTI i,t) ⋅ D i + β1 ⋅ ΔDSTI i,t γ + Φ(DSTI i,t) ⋅ (β2 ⋅ U t + β3 ⋅ ΔU t α)
  - where Φ(∙) is a non-linear function sharply increasing in DSTI i,t; D i, α, β1, β2, β3, γ are parameters set to match observed financial distress probabilities; U t denotes national unemployment rate.
  - Mitigating actions (loan restructuring, family borrowing, consumption cuts) implicitly captured via parameter calibration.
- Economic default condition (equation preserved):
  - ED i,t = { 1 if P̃ i,t − C + FinWealth i,t < NPV(L i,t, r t type,M, r t f , T t,s | FD i,t) ; 0 otherwise }
  - where P̃ i,t is market property value, C is transaction cost (assumed constant), FinWealth i,t is liquid financial assets (1.15 percent of average property value), and NPV includes outstanding loan amount L i,t plus estimated prepayment penalty (discounted foregone interest payments), depending on mortgage rate locked-in r t type,M, remaining maturity (T t,s − s), and discount rate r t f (average mortgage rate as of 2025Q2).
- Prepayment penalty and assumptions:
  - Prepayment penalties are estimated as discounted foregone interest payments and vary by mortgage type and time elapsed since signing.
  - Model assumes a positive penalty for early re-payment even if market rates rise, calculated as net present value of foregone interest for specified months.
- Calibration and validation:
  - Calibration draws on historical crises:
    - Spanish 2013–14 real estate crisis to capture unemployment and interest rate sensitivities in the financial distress equation.
    - 1990s UK real estate crisis to estimate sensitivity of financial distress to initial DSTI ratio.
    - Spanish IRB banks' 2024Q4 reported default rates for demographic factors.
  - Model validated by successfully replicating Spanish IRB banks' reported default risk under baseline conditions and during the 2014 real estate crisis. 

*Source: IMF staff summary of empirical strategy, findings, and scenario-based calibration (text as provided).*

### 32.      Before conducting stress-test

### 32.      Before conducting stress-test

### Updating vintage-risk buckets to point-in-time (as of 2025Q2)
- For each vintage-risk bucket the model calculates:
  - (i) the outstanding loan amount after amortization;
  - (ii) the current property value using the average cumulative house price growth in Spain;
  - (iii) borrower income updated with cumulative wage growth proxied by the average household disposable income; and
  - (iv) the current interest rate reflecting the bucket's repricing schedule and mortgage type composition.
- These inputs generate point-in-time LTV, LTI, and LSTI ratios for each bucket.
- The stress scenario is applied to these point-in-time estimates, ensuring the analysis reflects the portfolio's actual risk profile as of 2025Q2.

### Stress-test scenarios and key macrofinancial assumptions
- The model simulations assume a baseline and an adverse scenario.
  - Baseline: follows the IMF's January 2026 WEO projections.
  - Adverse: a three-year adverse scenario (2027Q2-2030Q2) combining an inflationary shock with a domestic recession, based on the 2025 EU-wide stress test exercise for Spain.
- Key adverse scenario assumptions:
  - House prices falling 17.2 percent cumulatively (with stochastic variation allowing fluctuations of ±15 percent to capture idiosyncratic property-level risk).
  - Unemployment rising 5.6 percentage points.
  - Mortgage rates widening by an average of 160 basis points (varying by mortgage type and repricing schedule).
  - Household disposable income remaining broadly flat in nominal terms.
- Table of scenario shocks (reported as cumulative or changes):
  - Household disposable income: Baseline YoY growth = 12.9% ; Adverse YoY growth = 0%
  - Real estate price: Baseline N [0, standard deviation = 15%] ; Adverse N [-17.2%, standard deviation = 15%]
  - Unemployment: Baseline -0.1 ppts ; Adverse 5.6 ppts
  - Benchmark rate: Baseline 0.19 ppts ; Adverse 1.62 ppts

### Model calibration and validation
- Parameters:
  - 훼 is set to 1 (following Harrison and Mathew (2008)).
  - Sensitivity parameters (훽1, 훽2, 훽3, 휸) are calibrated using the Spanish 2013-14 crisis.
  - Demographic parameters (훽2, 퐷) are jointly calibrated to match Spanish IRB banks' reported default rates in 2024Q4.
  - See Annex II. Table 1 and Table 2 for detailed parameter values.
- Validation:
  - The average PD over the 3-year baseline scenario is compared against the PD reported by IRB Spanish banks in 2024Q4 (1-year ahead PD).
  - The average loss rate over the 3-year adverse scenario is compared against the peak loss rate observed during the 2014 Spanish real estate crisis.

### Monte Carlo simulation procedure for PD and LGD (660 buckets)
- Simulation scope:
  - 660 vintage-risk buckets (12 LTV categories crossed with 55 quarterly vintages).
- Three-step simulation for each bucket:
  1. Compute the probability of financial distress for each bucket using equation (1), yielding a value between 0 and 1 that represents the share of borrowers in financial distress under the scenario's macroeconomic conditions.
  2. Conditional on financial distress, simulate economic default outcomes:
     - All loan characteristics in equation (2)—outstanding loan amount, interest rate, remaining maturity, transaction costs, and liquid financial assets—are LTI-LTV bucket-level averages.
     - Draw N = 10,000 house prices for each LTV-vintage bucket from a distribution centered on the scenario's house price path to capture property-level variation.
     - For each draw, equation (2) produces a binary outcome: 1 if that house price results in economic default (sale proceeds cannot cover the loan), 0 otherwise.
  3. Calculate the bucket's PD using the N binary outcomes:
     - PD_{i,t} = (∑_{n=1}^{N} ED_{i,t} / N) × Pr(FD_{i,t})
     - Pr(FD_{i,t}) is the (scalar, continuous) probability of financial distress from equation (1).
     - The bucket-specific LGD is computed as the average of LGDs among defaulted loans shown in equation (4).
- Repetition for stability:
  - The process is repeated 1,000 times for each LTV-vintage bucket.
  - Monte Carlo draws account for loan-level heterogeneity; the error of a Monte Carlo simulation decreases in proportion to 1/√N.

### LGD calculation and recovery assumptions
- LGD formula:
  - LGD_{i,t} = NPV(L_{i,t}, r^{type}_t, M, r^{f}_t, T_{t,s}) − (1 − δ) × P̃_{i,t+n} / (1 + r^{f}_t + spread)^n
  - First term: outstanding debt net present value.
  - Second term: discounted recovery value from selling the repossessed property.
- Key parameters and assumptions:
  - δ denotes the foreclosure discount at which the bank sells the property at time t+n.
  - n is the average time needed to sell off the collateral: 6.17 years in Spain (based on estimates from the Bank of Spain for 2022-2024).
  - spread is the risk-adjusted spread used to discount the value of the risky asset.
- Aggregation:
  - Portfolio-wide PD, LGD, and loss rates are calculated by aggregating bucket-specific results weighted by each bucket's outstanding share of the total portfolio in 2025Q2.

### Stress test results: PD, LGD, loss rates, and concentration
- Representative results (percent, point-in-time projections):
  - Outstanding portfolio (2027Q2):
    - Baseline: PD 0.77 ; LGD 10.92 ; Loss rate 0.29
    - Adverse: PD 2.63 ; LGD 32.2 ; Loss rate 0.99
    - (Chart annotations suggest baseline loss rates around 30 bps and adverse around 150 bps; see findings below.)
  - New vintages (2025Q3-2027Q2):
    - Baseline: PD 3.75 ; LGD 14.65 ; Loss rate 1.53
    - Adverse: PD 12.76 ; LGD 36.13 ; Loss rate 5.21
- Key quantified findings:
  - Annualized mortgage loss rates for the overall portfolio rise from 30 basis points in the baseline to 150 basis points in the adverse scenario.
  - Cumulative losses of roughly 4.6 percent of the total mortgage portfolio—about €23 billion over the three-year horizon.
  - Mortgages originated during 2025Q3-2027Q2 would face loss rates of 520 basis points—nearly 3.5 times the portfolio average.
  - The 1 percent countercyclical capital buffer (CCyB), scheduled for October 2026 and estimated at around EUR 16 billion, would mainly help absorb non‑mortgage losses.
  - Under stress, banks’ credit losses are primarily concentrated in non‑collateralized retail and corporate loans rather than residential mortgages.
- Systemic implications:
  - The projected EUR 23 billion in gross mortgage losses (i.e. before netting out credit provisions) is significant—particularly if losses were to be unevenly distributed across banks, which could prompt affected banks to curtail lending and amplify stress through negative feedback loops.

### Impact of borrower-based measures (BBMs)
- Timing for interventions:
  - Regulatory limits on new vintage flows are assumed to be introduced 8 quarters before the 3-year stress materializes.
- Effectiveness:
  - Well-calibrated borrower-based measures could reduce portfolio losses under severe stress by up to 25 percent after two years of implementation—equivalent to roughly €6 billion.
  - An LTV cap of 80 percent—either as a hard limit or combined with an LTI limit of 5 under a speed-limit framework—is found to be most effective.
  - DSTI caps calibrated at standard ratios (30 or 40 percent) yield minimal benefits because system-wide average DSTI ratios already fall well below tested thresholds.
  - An 85 percent LTV cap would reduce adverse-scenario losses by only about 6 percent relative to the no-BBM baseline.
  - A 90 percent LTV cap would be almost non-binding.
- Modeling assumptions on behavioral responses:
  - New lending exhibits “bunching” just below regulatory limits; when “speed limits” are applied, the share of loans exceeding the limits is adjusted proportionally to the initial distribution of originations in the absence of BBMs.
- Limitations noted:
  - Benefits from collateral-based measures are based on LTV ratios rather than loan-to-price (LTP) ratios; calibration using LTP ratios would likely yield larger resilience gains.
  - CIRBE data cover only mortgages secured on households' main residences; extending analysis to second homes and buy-to-let mortgages could further strengthen estimated benefits of BBMs.

### Conclusion and policy recommendations
- Main diagnostic conclusions:
  - Spain's mortgage market does not currently display acute imbalances; average household leverage remains low; and bank balance sheets are solid.
  - However, recent vintages show a gradual accumulation of risk—rising high-LTV shares, higher LTI ratios, and smaller equity buffers among post-2020 borrowers—that would translate into material portfolio losses under a severe housing correction, particularly among the most recent cohorts.
- Policy implications and recommendations:
  - Evidence supports anchoring a BBM framework on a collateral-based cap (such as LTV or LTP), possibly complemented by an LSTI cap.
    - Collateral-based looseness is identified as the primary driver of mortgage default in Spain.
    - Income-based looseness matters mainly through interaction with high-LTV lending.
    - An LTV cap of around 80 percent—whether as a hard limit or combined with an LTI limit under a speed-limit framework—delivers the largest reduction in adverse-scenario losses.
    - A standalone LSTI cap calibrated at standard international thresholds would add little given average LSTI ratios in Spain already fall well below those levels; LSTI can act as a complementary tool targeting particularly risky loans that are both highly leveraged and have stretched debt-service burdens.
  - Three further considerations for implementation:
    1. BBMs and capital-based measures should be viewed as complements rather than alternatives; capital buffers alone cannot substitute for addressing risk at origination.
    2. Introducing BBMs pre-emptively—before they become binding—would minimize economic and social costs and likely face less public resistance.
    3. Flexibility features such as speed limits and differentiation by borrower type, together with a gradual, well-communicated implementation path, could mitigate distributive concerns and ease the transition.

*International Monetary Fund — Spain: stress-test chapter (text as provided).*

### Annex I. Full Regression Results for the Loan-Level Analysis

### Annex I. Full Regression Results for the Loan-Level Analysis

### Effects of High-Risk Loans on Mortgage Defaults (Loan-Level Regressions)
- Dependent variable: mortgage delinquency.
- Lending-standard indicator thresholds: LTV > 80%, LTI > 6, LSTI > 40.
- All specifications control for loan, borrower risk and property characteristics. Regressions include bank-by-time and province-by-time fixed effects. Robust standard errors in parentheses. ***, **, * denote statistical significance at 1%, 5% and 10% levels.

- Annex I. Table 1 — Selected coefficient estimates and statistics:
  - Column (1): LTV>80% = 0.0117*** (0.0019). Observations = 1,015,509. R-squared = 0.0453. Time-Bank FE = Yes. Time-Province FE = Yes.
  - Column (2): LTI>6 = 0.0025*** (0.0009). Observations = 1,030,201. R-squared = 0.0448.
  - Column (3): LSTI>40% = 0.0037*** (0.0010). Observations = 1,012,602. R-squared = 0.0435.
  - Column (4) (joint LTV and LTI): LTV>80% = 0.0116*** (0.0019); LTI>6 = 0.0012 (0.0008). Observations = 1,015,509. R-squared = 0.0454.
  - Column (5) (joint LTV and LSTI): LTV>80% = 0.0115*** (0.0020); LSTI>40% = 0.0026** (0.0010). Observations = 993,698. R-squared = 0.0441.
  - Column (6) (LTV × LTI interaction): LTV>80% = 0.0100*** (0.0016); LTV>80% # LTI>6 = 0.0030 (0.0019). Observations = 1,015,509. R-squared = 0.0454.
  - Column (7) (LTV × LSTI interaction): LTV>80% = 0.0093*** (0.0017); LTV>80% # LSTI>40% = 0.0054*** (0.0020). Observations = 993,698. R-squared = 0.0441.

- Notes: Columns (1)-(3) show single-measure specifications; (4)-(5) joint specifications entering LTV with either LTI or LSTI; (6)-(7) augmented with interaction between the LTV dummy and the income-based dummy.

### Effects Contingent on Lenders’ Capital Ratio (Bank Interaction Regressions)
- Risk-weighted assets (RWA) capital ratio is winsorized at the 1st and 99th percentiles and standardized.
- Annex I. Table 2 — Selected coefficient estimates and statistics:
  - Column (1): LTV>80% = 0.0106*** (0.0018). LTV>80% # RWA Capital Ratio = -0.0039*** (0.0012). Observations = 463,933. R-squared = 0.0451. Time-Bank FE = Yes. Time-Province FE = Yes.
  - Column (2): LTI>6 = 0.0023* (0.0013). LTI>6 # RWA Capital Ratio = -0.0018** (0.0007). Observations = 470,862. R-squared = 0.0445.
  - Column (3): LSTI>40% = 0.0042*** (0.0014). LSTI>40% # RWA Capital Ratio = 0.0004 (0.0011). Observations = 462,178. R-squared = 0.0433.

- Notes: Regressions include bank-by-time and province-by-time fixed effects. Robust standard errors in parentheses. ***, **, * denote coefficients statistically different from 0 at 1%, 5% and 10% levels.

### Interpretation of Loan-Level Results (from provided notes and table structure)
- LTV>80% consistently associated with higher mortgage delinquency across specifications, with magnitudes between 0.0093 and 0.0117 and statistically significant at 1%.
- LTI>6 and LSTI>40% show positive associations with delinquency in single-measure specifications; joint and interaction specifications yield smaller or statistically weaker point estimates.
- Higher bank RWA capital ratio attenuates the delinquency effect of high LTV (LTV>80% # RWA Capital Ratio = -0.0039***).

---

### Annex II. Scenario-Based Default Model Calibration

### Model calibration basis and purpose
- Calibration and parameterization based on loan loss experience of the burst of the real estate bubble in Spain in 2012-2014.
- Risk characteristics of vintages adjusted to reflect current conditions as of 2025Q2.
- Sources: Bank of Spain and IMF staff calculations.

### Key calibrated parameters (selected from Annex II tables and text)
- Calibrate sensitivity of FD to shifts in DSTI:
  - 2
  - 0.06
  - β = 1
- Calibrate sensitivity of FD to initial DSTI by bucket:
  - 0.0016
  - β = 2.5
- Calibrate sensitivity of FD to unemployment:
  - γ = 0.02
- Allocate financial distress to shocks during the Spanish real estate crisis 2013-14:
  - D = , 17.6% (as presented)
- Estimate sensitivity to demographic factors and foreclosure transaction costs: values listed below.
- Additional parameter values explicitly reported:
  - Foreclosure discount = 27.26%
  - Time to sell collateral after foreclosure = 6.17 year
  - Discount rate of real estate collateral = 2.7%
  - Spread on discount rate on foreclosed collateral = 1.0%
- Other parameter entries noted in table (as rendered): 3, 0.727, β = 1, α = 3.65%, C = 1.15%, P · ~ , 14.8%, FD i FD i ξ ∂ == ∂ (presented in source table format).

---

### Annex III. Macrofinancial Variables Path Assumptions

### Scenario construction and sources
- Baseline paths drawn from the January 2026 WEO (except house price projections).
- House price projections and adverse paths follow the European Banking Authority's (EBA) 2025 EU-wide stress test exercise.
- Adverse paths generated by applying deviations from the baseline assumed in the EBA 2026 EU-wide stress test scenario to the January 2026 WEO projections.
- These trajectories feed into model simulations underlying mortgage portfolio loss estimates reported in Section E.

### Annex III. Table 1 — Baseline and Adverse scenario time paths (selected series)
- Baseline (January 2026 WEO) values by year:
  - Unemployment: 2025 = 10.8, 2026 = 10.7, 2027 = 10.7, 2028 = 10.7, 2029 = 10.7, 2030 = 10.7
  - Interest rates: 2025 = 2.6, 2026 = 2.5, 2027 = 2.5, 2028 = 2.5, 2029 = 2.5, 2030 = 2.5
  - Income growth (year-on-year): 2025 = 6.0, 2026 = 5.1, 2027 = 4.2, 2028 = 4.4, 2029 = 4.0, 2030 = 4.0
  - House prices (EBA 2025 stress test projections applied to baseline): 2025 = 8.5, 2026 = 7.6, 2027 = 6.7, 2028 = 4.9, 2029 = 4.9, 2030 = 4.9

- Adverse (EBA 2025 stress test deviations applied) values by year:
  - Unemployment: 2025 = 10.8, 2026 = 10.7, 2027 = 10.7, 2028 = 14.0, 2029 = 17.2, 2030 = 16.4
  - Interest rates: 2025 = 2.6, 2026 = 2.5, 2027 = 2.5, 2028 = 5.1, 2029 = 4.5, 2030 = 4.1
  - Income growth: 2025 = 6.0, 2026 = 5.1, 2027 = 4.2, 2028 = 4.4, 2029 = 4.0, 2030 = 4.0
  - House prices: 2025 = 8.5, 2026 = 7.6, 2027 = 6.7, 2028 = -4.3, 2029 = -11.0, 2030 = -5.6

---

### Firm-Level Innovation Findings (excerpts from following annex text relevant to model and policy implications)

### Stylized facts on Spain’s innovation gap
- Spain’s labor productivity now at about 74 percent of the US level (statement in text).
- Spain’s R&D tax credit statutory design:
  - 25 percent tax credit up to the average expenditure level of the previous two years.
  - 42 percent on additional expenditure above that level.
  - AIReF (2020) finds actual take-up rate of the R&D tax credit is only around 30 percent.
- Patenting and R&D patterns:
  - R&D intensity stagnated since late 2000s; patenting per million population substantially lower than peers.
  - Within-sector margins account for about 70 percent of patenting gap between Spain and the US; across-sector (composition) margin accounts for about 30 percent.

### Firm-level patenting patterns and heterogeneity
- Matching rates for merged Orbis–PATSTAT:
  - Spain matching rate = 87 percent.
  - France matching rate = 78 percent.
  - Sweden matching rate = 85 percent.
- Patenting rates and quality:
  - Spanish firms file fewer patents than French and Swedish firms, gap larger for younger firms and for USPTO (high-quality) patents.
  - Patent quality (forward citations) increases with firm size and decreases with firm age.
- Innovation polarization statistics:
  - Only 0.17 percent of firms file a patent when young (age five or below).
  - Among firms that innovate when young, 36 percent also innovate at older ages.
  - Among firms that do not innovate when young, only 0.3 percent file a patent later in life.

### Policy-relevant model and counterfactual findings (from the paper’s introduction and calibration summary)
- Endogenous growth model calibrated to micro patterns:
  - Counterfactuals indicate addressing frictions (size-dependent regulations, financial constraints, regulatory burdens) could increase long-run TFP growth rate by 0.18 percentage points.
  - Streamlining the R&D tax credit and better targeting it toward young firms could increase TFP growth rate by a further 0.1 percentage point.
- Rationale for R&D tax-credit reform:
  - Complexity, rejection risk, lack of immediate monetization, discounting of monetized credits, and administrative burden reduce take-up and effective subsidy for small and young firms.

*Source: 1espea2026002 - Annex I. Full Regression Results for the Loan-Level Analysis.*

### 13.      Firm-level analysis is carried out to assess the impact of frictions on firm-level

### 13.      Firm-level analysis is carried out to assess the impact of frictions on firm-level innovation

### Empirical approach
- Two-step empirical strategy:
  - Step 1: Regress a patenting indicator on friction measures to assess whether the friction is negatively correlated with innovation.
  - Step 2: Regress firm employment growth on the interaction between the patenting indicator and the friction measure to examine whether firm growth after innovation is dampened by the presence of the friction.
- Three frictions examined: size-dependent labor regulations, financial constraints, and regulatory burdens (region–sector regulations).

### Size-dependent labor regulations (50-employee threshold)
- Context:
  - Spanish firms face more stringent labor regulations once they exceed 50 employees (higher layoff costs, requirement to establish a works council).
  - Literature finds size-dependent regulations can discourage firm growth and innovation.
- Empirical specification:
  - Dependent variables (three alternatives):
    1. Dummy: firm files any patent.
    2. Number of patents filed at Spanish, European, and US patent offices.
    3. Inverse hyperbolic sine of average forward citations.
  - Controls include (4-digit NACE) sector and year fixed effects.
- Key regression results (Table 1):
  - log(emp): 0.181*** (Any Patents), 0.039*** (Patents filed in ES, EU, US), 0.086*** (Average Citations).
    - Standard errors: (0.001), (0.001), (0.008).
  - 45≤ Emp < 50: -0.141*** (Any Patents), -0.073*** (Patents filed in ES, EU, US), -0.163* (Average Citations).
    - Standard errors: (0.019), (0.008), (0.089).
  - Observations: 5,049,483; 5,049,483; 6,496.
  - Mean of dependent variables: 0.130; 0.022; 1.099.
- Interpretation:
  - Coefficients on the 45–49 employment dummy are negative across all dependent variables, suggesting firm-level innovation is negatively correlated with size-dependent regulations.
  - Footnote: Another threshold at operating revenue €6 million is noted, but no statistically significant effect on innovation is found.

### Financial constraints (MRPK proxy)
- Measurement:
  - Marginal revenue product of capital (MRPK) is used as a proxy for severity of financial constraints; calculated under Cobb-Douglas assumptions:
    - y_i = a_i l_i^α k_i^(1−α)
    - MRPK_i = (1−α) p_i y_i / k_i
  - Firms classified as financially constrained if MRPK exceeds the sector–year median.
- Empirical findings:
  - Figure 7: negative correlation between log(MRPK) and log(firm age) conditional on firm sales and sector–year fixed effects (younger firms more likely financially constrained).
  - Regression specification: Y_it = β·1(Constrained) + γ·log(emp_it) + Γ·X_it + ε_it.
- Key regression results (Table 2):
  - log(emp): 0.180***; 0.039***; 0.084*** (standard errors (0.001), (0.001), (0.008)).
  - Constrained: -0.087***; -0.021***; -0.144*** (standard errors (0.003), (0.001), (0.030)).
  - Observations: 5,049,483; 5,049,483; 6,496.
  - Mean of dependent variables: 0.130; 0.022; 1.099.
- Interpretation:
  - Innovation outcomes—both probability of patenting and patent quality—are negatively associated with financial constraints.

### Regulatory burdens (region–sector regulatory intensity)
- Measurement:
  - Intensity proxied by number of newly introduced regulations at the region–sector–year level (Mora-Sanguinetti and others).
  - Regression: Y_it = β·∆Log(# of regulations) + γ·log(emp_it) + Γ·X_it + ε_it, with X_it including sector, year, and region (two-digit NUTS) fixed effects.
- Key regression results (Table 3):
  - log(emp): 0.130***; 0.028***; 0.077*** (standard errors (0.001), (0.001), (0.010)).
  - ∆ Log # of regulations: -0.022***; -0.006*; -0.027 (standard errors (0.008), (0.003), (0.078)).
  - Observations: 4,227,020; 4,227,020; 3,935.
  - Mean of dependent variables: 0.094; 0.016; 1.102.
- Interpretation:
  - Increases in the number of regulations are negatively associated with firm-level innovation activity. The coefficient on average citations is negative but not statistically significant.

### Aggregate statement on empirical estimates
- While non-causal interpretation cautioned, the analysis indicates material impacts of the three frictions on innovation.
- Figure 8 compares innovation probabilities for high-friction (top decile) versus low-friction (bottom decile) firms.
- For size-dependent regulations, comparison is firms with 45–49 employees versus all other firms.
- Taken at face value, the three constraints together might account for about 88 percent of the innovation gap relative to Swedish firms at employment 45 (33 out of 37 percentage points).

### Post-innovation firm growth and frictions
- Regression to test conditional growth:
  - log l_{i,t+5} − log l_{i,t} = β1·1(Innovate) + β2·1(Friction) + β3·1(Innovate)×1(Friction) + β4·log l_{i,t} + ε_{i,t}.
  - 1(Innovate) = dummy if firm files a patent. 1(Friction) = dummy if firm’s friction measure is above median (for employment threshold, employment between 45 and 50).
  - β3<0 implies conditional on innovating, firms grow less when facing larger frictions.
- Key interaction results (Table 4):
  - Innovation: 0.344***; 0.352***; 0.362*** (standard errors (0.008), (0.010), (0.014)).
  - Innovation × 45≤ Emp < 50: -0.191*** (standard error (0.052)).
  - Innovation × Constrained: -0.035** (standard error (0.017)).
  - Innovation × Regulated: -0.046** (standard error (0.022)).
  - Observations: 4,705,084; 4,705,084; 3,940,540.
- Interpretation:
  - In all three cases the interaction coefficient is negative and statistically significant, indicating post-innovation firm growth is negatively associated with each of the three friction measures.
  - The 50-employee regulatory threshold appears quantitatively most strongly associated with reduced post-innovation firm growth (Figure 9).

### Model: Extension of Klette and Kortum (2004) to include frictions
- Model features:
  - Continuous-time general equilibrium endogenous growth model with firm dynamics driven by innovation and creative destruction.
  - Firms have multiple product lines; R&D generates innovations that improve product-level productivity by a fixed step.
  - R&D costs with Poisson arrival rate x_i: C(x_i; n_i) = x_i^η n_i^(1−η) Y/α (base form).
- Friction extensions:
  - Size-dependent regulation: fixed cost φ>0 when employment exceeds 50.
  - Financial frictions: additional R&D cost τ(n_i) decreasing with firm-level employment; in model τ(n) = −τ log(n) parameterization later used in calibration.
  - Regulatory burdens: incorporated as entry cost δ>0 paid by entrants upon successful innovation.
  - Heterogeneous innovation ability: step size λ ∈ {λ_H, λ_L} with λ_H > λ_L; all firms start with λ_H and transition to λ_L with Poisson rate κ.
- R&D cost with subsidy s and financial friction τ(n_i):
  - C(x_i; n_i) = ((1−s) (1+τ(n_i)) x_i^η n_i^(1−η) Y)/α.
- Mechanisms:
  - Frictions reduce net return to innovation by lowering private returns or increasing costs, thus lowering innovation rates and aggregate productivity growth.
  - Optimal R&D subsidy rate is higher for high-type firms because the wedge between private and social returns increases with step size.

### Quantitative calibration
- Macro targets and parameter calibration (selected):
  - R&D / GDP (Spain): 1.49 percent targeted via α (innovation cost shifter).
  - Average TFP growth 2015–2025: 0.67 percent used to calibrate λ_L; λ_H assumed twice λ_L.
  - Labor supply L calibrated to match share of firms with more than 50 employees.
- Micro moment calibrations:
  - φ calibrated to match marginal decline in probability of patenting as firm size increases from 45 to 50 employees.
  - τ(n) set to −τ log(n), τ calibrated to match slope of log MRPK with respect to log employment.
  - δ calibrated to match firm entry rate in Spain (4.66 percent).
  - κ set to match patent quality difference between young and old firms.
  - Baseline R&D subsidy rate s = 10 percent (effective subsidy rate 33 percent × take-up rate 30 percent per AIReF (2020)).
  - R&D cost convexity η = 2.
- Calibration results (Table 5 excerpts):
  - L: Main target Share of firm with emp>50 = 0.015; Model = 0.017.
  - α: R&D / GDP target 0.158; Model 0.149.
  - λ_L: TFP growth rate target 0.71; Model 0.67.
  - φ: Patenting decline at 45<emp<50 target -0.44; Model -0.41.
  - τ: Slope of log(MRPK) over log(emp) target -0.08; Model -0.08.
  - δ: Entry rate target 0.05; Model 0.047.
  - κ: Pr(Low type | High type) target 0.225; Model 0.22.
  - S (R&D subsidy rate): Effective subsidy rate x take up rate target 0.1; Model 0.1.
  - η = 2 (Acemoglu and others (2018)).

### Quantitative counterfactuals and policy simulations
- Friction reduction simulations:
  - Reduce the cost parameter associated with each friction by half, one at a time (size-dependent regulations, financial frictions, regulatory burdens).
  - Long-term productivity growth gains (Figure 10 summary):
    - Addressing size-dependent regulations alone yields relatively small productivity growth increase (limited share of firms near threshold).
    - Cutting financial frictions and regulatory burdens yields larger productivity gains as they affect a broader set of firms.
    - Reducing all three frictions increases long-term productivity growth rate by 0.18 percentage points.
- R&D subsidy / tax credit counterfactuals:
  - Streamline R&D tax credit to increase take-up rate from 30 percent to 60 percent:
    - Effective R&D subsidy rate rises from 10 to 20 percent.
    - TFP growth rate increases by an additional 0.06 percentage points on top of gains from cutting the three frictions.
  - Targeting the increased take-up to young firms (<5 years) while keeping government spending/GDP constant:
    - Relative to broad take-up increase, targeting raises growth by an additional 0.03 percentage points.
    - Total additional gain from targeting versus baseline take-up increase: 0.09 percentage points.
- Combined reforms:
  - All reforms considered (cutting three frictions + increasing and targeting R&D subsidy) raise long-term productivity growth rate by 0.27 percentage points in total.

*Source: Chapter 13, Spain section (firm-level analysis, model, and quantitative exercises) from the provided IMF content unit.*

### 26.      Addressing key frictions is key to boosting Spain’s innovation and long-term

### 26.      Addressing key frictions is key to boosting Spain’s innovation and long-term productivity growth

### Major regulatory and financial frictions
- Ongoing “Regime 20” initiative: a common regulatory framework aimed at reducing administrative barriers across regions; recommended to be pursued and amplified.
- Complementary bottom-up actions: broader adoption of open-market laws by individual autonomous communities could deliver immediate benefits to participating regions and create incentives for others to follow.
- EU-level reforms: EU 28th Regime initiative could further ease regulatory burdens and incentivize firms to scale up and innovate.
- Financial frictions: progress toward the EU’s Savings and Investment Union would benefit comparatively more financially-constrained Spanish firms.
- Public investment fund España Crece: could be helpful if it focuses on stimulating innovation and targets young and innovative firms rather than smaller, older, less innovative firms.
- Size-dependent regulations: streamline by smoothing existing regulatory thresholds—particularly the 50-employee threshold, which concentrates multiple regulations and creates discontinuities.

### R&D tax credit: obstacles and suggested adjustments
- Main obstacles to take-up:
  - Costly application process and limited legal certainty (AIReF (2020) identifies these as key obstacles).
  - Ex post review by tax authorities can lead to repayment if expenditures are later judged not R&D-related.
- Current mitigant: Ministry of Science issues an ex ante Binding Motivated Report (Informe Motivado Vinculante, IMV) as certification for R&D projects, but tax authorities still review expenditures ex post.
- Suggested improvements:
  - Greater coordination between the Ministry of Science and the tax authorities to strengthen legal certainty provided by the IMV.
  - Simplify access to the refund option and reduce practical complexity and costs.
- Specific refund-option restrictions to ease:
  - Refunds are currently subject to a 20 percent discount.
  - Firms must wait at least one year after the expenditure is incurred to obtain a refund.
  - Firms must maintain at least the same number of R&D workers after receiving the refund.
- Rationale: While these conditions likely limit abuse, they also weaken innovation; streamlining them could increase take-up, particularly among young and innovative firms.

### Empirical data and matching performance (Annex I)
- Main sample period: 2000 to 2019.
- ORBIS coverage: Spain accounts for, on average, 82 percent of gross output in the manufacturing sector.
- Comparable coverage rates: France 87 percent; Sweden 77 percent.
- Matching procedure: PATSTAT merged with ORBIS using standardized firm names and exact matching at subsidiary level.
- Matching rates for patents:
  - Spain: 87 percent.
  - France: 78 percent.
  - Sweden: 85 percent.
- Entrants and subsidiaries: example that Airbus France and Airbus Spain are treated as distinct firms.

### Model structure and mechanisms (Annex II)
- Environment:
  - Continuous time, unit continuum of products j ∈ [0,1], Cobb–Douglas aggregator Y = ∫ y_j dj.
  - Production function: y_j = a_j l_j.
  - Profits per product: π_i = (1−1/λ_i) Y.
  - Size-dependent regulation: deadweight cost φY incurred if employment exceeds 50 workers (equivalently firms with more than n̅ product lines are subject to this regulation).
- Innovation:
  - Incumbents and entrants can innovate; innovation draws a product line and increases its productivity.
  - When average innovation rate is m, a firm with n lines loses a product at hazard rate n m.
  - Firms choose R&D expenditure to determine Poisson arrival rate x of innovation.
  - R&D flow cost: C(x_i; n_i) = (1−s)(1+τ(n_i)) x_i^η n_i^{1−η} Y/α, where s is R&D subsidy rate, τ(n_i) is financial friction costs (decreases with n_i), and α is a cost shifter.
  - Entrants pay fixed entry cost δ Y and face same R&D cost as incumbents with one product line.
  - Innovation step sizes: λ_i ∈ {λ_H, λ_L} with λ_H > λ_L > 1. Entrants born high-type and transition to low type at Poisson rate κ > 0.
- Equilibrium and policy drivers:
  - Normalized incumbent value function v(n,q) satisfies:
    r v(n,q) − v̇(n,q) = max_x [ x n π(q) − φ⋅1(l≥50) − c(x;n) + x(v(n+1,q)−v(n,q)) − m(v(n−1,q)−v(n,q)) + κ(v(n,L)−v(n,q)) ].
  - Incumbent innovation rate:
    x(n,q) = [ (v(n+1,q) − v(n,q)) / (η (1−s) (1+τ(n))) ]^{η−1} n.
    - Implication: absent frictions, innovation scales linearly with firm size; subsidies raise innovation; financial frictions reduce it.
  - Entrant value and innovation:
    r v_E − v̇_E = max_x [ x v(1,H) − c(x;1) − δ ], 
    with entrant innovation rate:
    x_E = [ v(1,H) / (η (1−s) (1+τ(1))) ]^{η−1}.
  - Aggregate growth rate:
    g = ∫ log(λ_H) x(n,H) dμ(n,H) + ∫ log(λ_L) x(n,L) dμ(n,L) + log(λ_H) x_E.

### Policy implications emphasized by the analysis
- Pursue and amplify Regime 20 while encouraging autonomous communities to adopt open-market laws to create immediate regional gains and peer incentives.
- Support EU-level regulatory initiatives (EU 28th Regime) and financial integration (Savings and Investment Union) to ease scaling and financing constraints.
- Orient España Crece toward stimulating innovation, targeting young and innovative firms rather than smaller, older, less innovative firms.
- Smooth size-dependent regulatory thresholds, particularly the 50-employee threshold, to reduce discontinuities that discourage firm growth.
- Strengthen legal certainty for the R&D tax credit via better coordination between the Ministry of Science and tax authorities and simplify the refund option by reconsidering the 20 percent discount, the one-year waiting requirement, and the post-refund R&D staffing maintenance rule.

*Source: 1espea2026002*

---


_Source: https://www.imf.org/-/media/files/publications/cr/2026/english/1espea2026002.pdf_