## cr17351

## Source details

**Canonical URL:** [cr17351](https://www.imf.org/-/media/files/publications/cr/2017/cr17351.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/cr/2017/cr17351.pdf.md)
- [Structured JSON version](/-/media/files/publications/cr/2017/cr17351.pdf.json)

---

### High Household Savings — Key Facts and Overview
- Gross household saving rate averaged some 17 percent since the global financial crisis, increasing by 7.5 percentage points over its pre-crisis level.
- Gross savings defined as gross disposable income minus final consumption expenditure plus the change in the net equity position of pension fund reserves.
- Household sector includes non-profit institutions serving households (NPISHs).
- Household sector’s share in total savings grew from one-eighth (3½ percent of GDP) to one-third (over 10 percent of GDP) in recent years.
- Sweden’s gross investment is 25 percent of GDP; comparable economies (Denmark, Germany and Netherlands) have gross investment around 20 percent of GDP.
- Implication: High household savings, rather than low investment, is the predominant driver of Sweden’s significant current account surplus.

### How Household Saving Evolved Since the Global Financial Crisis
- Pre-crisis saving rates displayed cyclical behavior aligned with macro developments (credit deregulation in 1985, banking crisis in 1992, 1994 pension reform).
- 1994 pension reform created a transition to higher mandatory retirement savings, stabilizing at an average of 3¾ percent of GDP per year.
- Abolition of the wealth tax in 2007 and the global financial crisis coincided with a considerable ramp-up in household savings, with no significant decline subsequently.
- Excess of savings over mandatory pension contributions:
  - Averaged 1.2 percent in the five years to 2007.
  - Averaged 6 percent of GDP in the last 5 years (post-crisis period referenced).
- Micro-distribution (Household Budget Survey 2012):
  - Households in the upper half of the income distribution do virtually all the saving.
  - Some 80 percent of all household savings (or 13.7 percentage points of the entire household sector’s 17 percent of income) were by households in the highest quartile of income.
  - Households in the lowest quartile dis-saved by an estimated 27 percent of their income in 2012.
- Post-crisis compositional change:
  - Second and third income quartiles increased saving rates by 10–14 percentage points between 2006 and 2012.
  - Upper quartile increased by 5 percentage points over the same period.
  - Middle-income households’ contribution to total household savings rose to 31 percent, up from -4 percent before the crisis.
- Caveat on measurement: Approximating saving units with consumption units can bias estimates (children do not save), possibly overestimating savings for households with children.

### Evaluating Household Savings — Determinants and Econometric Analysis
- Typical explanatory factors considered: economic uncertainty, demographic factors, and balance sheets.
- Variables and expected signs used in regression work:
  - Unemployment rate: Positive
  - Expected income growth: Negative
  - Young-age dependency: Negative
  - Old-age dependency: Negative
  - Net wealth levels and growth: Negative
  - Assets level and growth: Negative
- Country fixed effects panel model (1995–onward; Denmark, Finland, France, Germany, Netherlands, Norway, Sweden, United Kingdom) — statistically significant coefficients:
  - ur: 0.42***; Standard Error: 0.11; Statistic: 3.71
  - 1-year lead of gdi.g (gdi.g t + 1Y): -0.20*; Standard Error: 0.09; Statistic: -2.27
  - oad: -0.27**; Standard Error: 0.09; Statistic: -2.90
  - emp65: 0.47**; Standard Error: 0.17; Statistic: 2.73
  - assets: -0.03***; Standard Error: 0.01; Statistic: -3.36
  - assets.g: -0.20***; Standard Error: 0.04; Statistic: -5.26
  - nwealth: 0.05***; Standard Error: 0.01; Statistic: 5.14
- Panel model statistics:
  - R2 = 0.41
  - K = 8 countries
  - T = 20-21 years
  - N = 167 observations
  - model F-statistic: 15.1 on 7 and 152 degrees of freedom, p-value < 0.001
- Interpretation:
  - Signs and magnitudes align with theoretical expectations except for net wealth (nwealth) which is positive (0.05***), contrary to the expected negative wealth effect; hypothesis: desired net wealth may have increased over time.
  - Model-based residual indicates Swedish households save 2.6 percentage points per year more than in other similar economies, controlling for typical explanatory variables.

### Country-Specific Error Correction Models (Appendix II) — Excluding Pension Savings
- Dependent variable: srxpen = Gross household saving rate excluding pension savings, percent.
- Frequency: quarterly from 1980 (variables seasonally adjusted or interpolated as needed).
- Additional proxy introduced: el.care = Deflator of residential care and social work activities (2016 prices) / deflator of final consumption expenditure of households (2016 prices) / (population of 65+ people / population) — used to capture cost of elderly care per elderly person.
- Long-run Basic Model (selected coefficients):
  - Intercept: 11.73; Standard Error: 8.16; Statistic: 1.44
  - ur: 0.66***; Standard Error: 0.13; Statistic: 5.23
  - gdi.g t + 4Q: -1.03***; Standard Error: 0.11; Statistic: -9.28
  - yad: -0.56*; Standard Error: 0.27; Statistic: -2.09
  - nwealth: 0.01***; Standard Error: 0.00; Statistic: 5.23
  - nwealth.g: -0.19***; Standard Error: 0.03; Statistic: -6.60
  - Adjusted R2 = 0.64; model F-statistic: 50.2 on 5 and 136 degrees of freedom, p-value < 0.001
- Long-run Expanded Model (introducing el.care and interaction el.care * oad):
  - Intercept: 235.94***; Standard Error: 22.30; Statistic: 10.58
  - ur: 0.86***; Standard Error: 0.10; Statistic: 8.26
  - gdi.g t + 4Q: -0.66***; Standard Error: 0.08; Statistic: -8.43
  - yad: -2.50***; Standard Error: 0.27; Statistic: -9.13
  - nwealth: 0.04*; Standard Error: 0.01; Statistic: 2.53
  - nwealth.g: -0.35***; Standard Error: 0.06; Statistic: -5.81
  - assets t – 4Q: -0.05***; Standard Error: 0.01; Statistic: -3.53
  - el.care: -44.44***; Standard Error: 3.89; Statistic: -11.43
  - oad: -5.92***; Standard Error: 0.63; Statistic: -9.38
  - el.care * oad: 1.68***; Standard Error: 0.15; Statistic: 10.97
  - Adjusted R2 = 0.85; model F-statistic: 92.7 on 9 and 132 degrees of freedom, p-value < 0.001
- Cointegration and short-run dynamics:
  - Unit root tests show residuals of the expanded model are stationary → evidence of cointegration.
  - Error Correction Model (∆4 srxpen) key coefficients:
    - Residual of expanded model t – 4Q: -0.51***; Standard Error: 0.08; Statistic: -6.77
    - ∆4 gdi.g: 0.5***; Standard Error: 0.05; Statistic: 9.92
    - oad t – 4Q: 1.19***; Standard Error: 0.2; Statistic: 6.09
    - ∆4 oad: -3.63***; Standard Error: 0.51; Statistic: -7.1
    - yad t – 4Q: -1.03***; Standard Error: 0.11; Statistic: -9.23
    - ∆4 yad: -3.53***; Standard Error: 0.68; Statistic: -5.16
    - ∆4 assets: -0.03***; Standard Error: 0; Statistic: -8.65
    - assets.g t – 4Q: -0.19***; Standard Error: 0.02; Statistic: -9.38
  - Model statistics:
    - Adjusted R2 = 0.73
    - model F-statistic: 48.2 on 8 and 131 degrees of freedom, p-value < 0.001
  - Interpretation: the error correction coefficient (-0.51***) implies about half of the remaining gap from the long-run equation is closed each year.
  - Expanded specification substantially improves fit and captures the higher post-crisis saving rates; the positive net wealth coefficient persists across specifications.

### Hypotheses and Explanations Considered for High Savings
- Potential drivers explored:
  - Structural decline in consumer confidence since the global financial crisis.
  - Households raising savings to reach higher wealth targets due to slowed productivity growth and reduced expectations for trend income growth.
  - Crisis “scarring effects” increasing uncertainty around future asset values.
  - Abolition of the wealth tax and reduction in nominal property taxes altering preferences for wealth holdings.
  - High and rising house prices leading younger households to rely on parental transfers — inducing older/wealthier generations to save more to finance intergenerational transfers while remaining in their dwelling.
  - Concerns about viability of the Swedish model due to migrant inflows with lower employment rates possibly eroding confidence.
- Demographic nuance:
  - Sweden has both a large share approaching retirement and a booming young population (increase in fertility since 2000 and recent refugee arrivals).
  - Old- and young-age dependency ratios are rising rapidly.
  - Elderly employment rate has increased as more people postpone retirement, which may dampen the decline in saving typically associated with ageing.
- Role of elderly care:
  - The “orange envelope” conjecture (annual pension statement showing projected pension income and effect of postponing retirement) and rising cost of elderly care proxied by el.care are consistent with higher precautionary saving for elderly care costs.

### Conclusions and Policy Implications on Household Savings
- Main empirical conclusion: Preliminary evidence suggests the large increase in savings after the financial crisis may be due to rising costs and concerns about elderly care; econometric analysis supports anecdotal explanations that extended life expectancies and preference for higher quality residential care have contributed to higher savings.
- Distributional implication: explains higher savings among middle-income groups since the crisis.
- Policy implications suggested:
  - If validated, policies to reduce public concerns around elderly care costs may be appropriate.
  - If parental assistance for young homebuyers is confirmed as a driver, policies to improve housing affordability could have beneficial effects.
- Research gaps and recommendations:
  - Further analysis using more granular household-level data is required to test alternative hypotheses (parental transfers, bequest motives, housing down-payment needs).
  - More definitive conclusions depend on household-level microdata and further research.

### Wage Formation and Recent Wage Moderation — Background and Stylized Facts
- Since the Industrial Agreement of 1997, wage formation led by bargaining in the industrial sector; collective agreements cover over 90 percent of employees.
- Firm/local level negotiations increased from 10 percent in 2013 to over 20 percent by 2015; National Mediation Office reports "more than 20 percent as of now."
- Trend labor productivity growth declined to about 1¼ percent in recent years from over 2 percent before the crisis.
- Inflation expectations remained below the 2 percent target for several years until 2017.
- Swedish wages have moved very closely with agreed wages in Germany in recent years, reflecting employers’ emphasis on safeguarding cost competitiveness and export market share.

### Analytical Approach to Wage Dynamics
- Core model: wage curve (Blanchard 1997) extended with trend labor productivity (TLP) and error-correction specification (ECM).
- Real wages negatively related to aggregate unemployment in wage-curve tradition and positively related to trend productivity.
- Data and key variable definitions (selected):
  - Unemployment Gap (UnemGap): Percent deviation from HP-filtered equilibrium unemployment rate (lambda=12000), source LFS.
  - Trend productivity (TLP): HP-filtered trend real labor productivity per hour (lambda=1600), source Eurostat.
  - Real Compensation/wage (RW): Compensation or wage / GDP deflator, source Statistics Sweden.
  - Expected inflation: 2 year ahead expected inflation, source TNS Prospera.

### Estimation Results — Cointegration and Key Determinants
- Sample period: 1995:Q1 to 2017:Q1.
- Cointegration evidence: coefficients on lagged level of real wage statistically significant in all variants; error correction rates range from -0.2 to -0.4.
- Domestic determinants significant in models:
  - Unemployment gap, changes in involuntary part-time employment, and labor productivity growth.
- International spillovers:
  - German compensation/wage growth and euro area labor market slack significantly affect Swedish wage dynamics.
  - In fuller models, expected inflation and euro area unemployment changes become significant alongside domestic unemployment gap.
- Interpretation: wage formation influenced by domestic fundamentals and international spillovers; deviations of real wages from trend productivity drive subsequent nominal wage growth via ECM dynamics.

### Model Fit, Scenario Analysis and Comparative Results
- Models using total labor compensation fit better than models using wages and salaries excluding employer social security contributions.
- Error correction model (combined model, Table 1, model 9) explains close to 80 percent of total variation in actual wage development (R-squared = 0.78); domestic model explains about 60 percent.
- Spillovers raise model fit by about 20 percentage points (about 10 percent each from euro area labor market and German wage growth).
- Scenario analysis (based on model 9), baseline = 2017Q1 levels:
  - Domestic shocks:
    - Increase in productivity growth from 1.25 to 1.75 percent → raises nominal labor compensation growth by 0.5 percent in short run.
    - Decline in unemployment rate by 1 percentage point → raises wage growth by 0.2 percent.
    - Decline in involuntary PT employment share by 1 percentage point → raises wage growth by 0.1 percent.
    - Combined domestic effect ≈ 0.8 percent y/y increase in nominal labor compensation growth.
  - Foreign shocks:
    - Increase in German wage growth from 2 to 3 percent → spillover ~0.6 percent.
    - Fall in euro area unemployment from 9.5 to 8.5 percent → impact ~0.1 percent.
    - Combined foreign effect ≈ 0.7 percent y/y increase.
  - Combined domestic and foreign shocks → increase nominal compensation growth by about 1.5 percent.
- Panel regression covering 22 EU countries:
  - Sweden’s spillover from German labor compensation growth: 0.6 (Sweden) versus about 0.2 (EU average).
  - Long-run unemployment gap responsiveness: Sweden: -0.13; average EU country: -0.11.

### Analysis of Recent Wage Moderation and Conclusions
- Decomposition attributes recent wage moderation to:
  - Sluggish productivity growth.
  - Low expected inflation.
  - Error-correction adjustment after earlier real wage outperformance (correction toward equilibrium).
- Rolling regressions:
  - Wage responsiveness to unemployment broadly stable after accounting for foreign spillovers.
  - Impact of expected inflation fluctuated with no clear trend.
  - Rolling coefficient on German wages rose since early 2000s and continued rising.
- Other structural contributors noted (not fully captured econometrically):
  - Decline in profitability: net return on capital in Sweden declined since the crisis to be the lowest among peer economies.
  - Structural shift toward less capital-intensive sectors, globalization, increased use of temporary and part-time workers, outsourcing, rising labor supply not captured in standard surveys.
- Longer-run perspective:
  - Swedish real wages grew slightly over 1 percent on average as real productivity increased by 1 percent.
  - Real wages rose about 0.1 percent as the unemployment rate fell by 1 percentage point.
  - Nominal wages react to expected inflation and trend productivity with magnitudes comparable to other EU countries.
- Suggested directions for further research:
  - Assess structural factors underlying the rise in the impact of external labor market conditions.
  - Investigate whether omission of broader indicators of domestic labor market slack, foreign labor market conditions, and deviation from long-run equilibrium wage levels drives findings of a flattened Phillips curve.
  - Explore role of rising European labor market integration (e.g., Schengen expansion from 2003) as a contributor to rising external influence.

*International Monetary Fund — SWEDEN, November 1, 2017*

### References _____________________________________________________________________________ 11

### References

### High Household Savings in Sweden — Introduction
- Gross household saving rate averaged some 17 percent since the global financial crisis, increasing by 7.5 percentage points over its pre-crisis level.
- Gross savings are defined as gross disposable income minus final consumption expenditure plus the change in the net equity position of pension fund reserves.
- The household sector includes non-profit institutions serving households (NPISHs).
- The household sector’s share in total savings grew from one-eighth (3½ percent of GDP) to one-third (over 10 percent of GDP) in recent years.
- Sweden’s gross investment is 25 percent of GDP; comparable northern European economies (Denmark, Germany and Netherlands) have gross investment around 20 percent of GDP.
- Implication: High household savings, rather than low investment, is the predominant driver of Sweden’s significant current account surplus.

### How Household Saving Evolved Since the Global Financial Crisis
- Before the crisis, saving rates showed cyclical behavior aligned with macro developments (credit deregulation in 1985, banking crisis in 1992, 1994 pension reform).
- The 1994 pension reform created a transition to higher mandatory retirement savings, stabilizing at an average of 3¾ percent of GDP per year.
- The abolition of the wealth tax in 2007 and the global financial crisis coincided with a considerable ramp-up in household savings, with no significant decline subsequently.
- Excess of savings over mandatory pension contributions:
  - Averaged 1.2 percent in the five years to 2007.
  - Averaged 6 percent of GDP in the last 5 years (post-crisis period referenced).
- Household Budget Surveys (Statistics Sweden) are used to allocate national-account gross savings across household cohorts using:
  - Proportional scaling of gross disposable income and final consumption expenditure by household-type shares.
  - Adjustment for change in pension net equity position multiplied by population size and average household persons divided by average number of consumer units.
- Caveat: Approximating saving units with consumption units can bias estimates (children do not save), possibly overestimating savings for households with children.
- Distributional findings (Household Budget Survey 2012):
  - Households in the upper half of the income distribution do virtually all the saving.
  - Some 80 percent of all household savings (or 13.7 percentage points of the entire household sector’s 17 percent of income) were by households in the highest quartile of income.
  - Households in the lowest quartile dis-saved by an estimated 27 percent of their income in 2012.
- Post-crisis compositional change:
  - Second and third income quartiles increased saving rates by 10–14 percentage points between 2006 and 2012.
  - Upper quartile increased by 5 percentage points over the same period.
  - Middle-income households’ contribution to total household savings rose to 31 percent, up from -4 percent before the crisis.

### Evaluating Household Savings — Determinants and Econometric Analysis
- Typical explanatory factors referenced: economic uncertainty, demographic factors, and balance sheets.
- Typical variables and expected signs listed in the text:
  - Unemployment rate: Positive (economic/income uncertainty → higher savings)
  - Expected income growth: Negative
  - Young-age dependency: Negative
  - Old-age dependency: Negative
  - Net wealth levels and growth: Negative (wealth effect)
  - Assets level and growth: Negative (wealth effect)
- Fixed effects panel model:
  - Estimated for 1995–onward across 4 Nordic economies (Sweden, Denmark, Finland, Norway) and peers (Germany, France, Netherlands, United Kingdom).
  - Statistically significant explanatory variables: unemployment rate; 1-year forward looking gross disposable income growth; old-age dependency ratio; elderly employment ratio; household assets; household asset growth; net wealth of households.
  - Elderly employment ratio improves model fit, especially post-crisis.
  - Unexpected result: coefficient on net wealth is positive.
- Model-based residual for Sweden:
  - Swedish households save 2.6 percentage points per year more than in other similar economies, controlling for typical explanatory variables.
- Country-specific error correction models (Appendix II):
  - Models regress gross household saving rate excluding pension savings to isolate non-mandatory savings.
  - Basic model fits reasonably but fails to capture post-crisis increase.
  - Expanded model introduces a proxy for the cost of elderly care interacted with the old-age dependency ratio; this improves fit and captures the higher post-crisis saving rates well.
  - Short-term specification shows cointegration and that almost half of the gap is closed from year to year.
  - Net wealth coefficient remains incorrectly positive in extended specifications.

### Hypotheses and Explanations Considered
- Possible drivers explored in text:
  - Structural decline in consumer confidence since the global financial crisis.
  - Households raising savings to reach higher wealth targets due to slowed productivity growth and reduced expectations for trend income growth.
  - Crisis “scarring effects” increasing uncertainty around future asset values.
  - Abolition of the wealth tax and reduction in nominal property taxes altering preferences for wealth holdings.
  - High and rising house prices leading younger households to rely on parental transfers — inducing older/wealthier generations to save more to finance intergenerational transfers while remaining in their dwelling.
  - Concerns about viability of the Swedish model due to migrant inflows with lower employment rates possibly eroding confidence.
- Demographic nuance:
  - Sweden has both a large share of people approaching retirement age and a booming young population (enduring increase in fertility since 2000 and recent refugee arrivals).
  - Both old- and young-age dependency ratios are rising rapidly.
  - Elderly employment rate has increased as more people postpone retirement, which may dampen the decline in household saving typically associated with ageing.
  - A proxy variable for the cost of elderly care per elderly person is introduced to capture the “orange envelope” conjecture (the annual pension statement in an orange envelope that shows projected pension income and the effect of postponing retirement).

### Conclusions and Policy Implications
- Main finding: Preliminary evidence suggests the large increase in savings after the financial crisis may be due to rising costs and concerns about elderly care.
  - Econometric analysis supports anecdotal explanations that extended life expectancies and preference for higher quality residential care have contributed to higher savings.
  - This may explain higher savings among middle-income groups since the crisis.
- Policy implications suggested in the text:
  - If validated, policies to reduce public concerns around elderly care costs may be appropriate.
  - If parental assistance for young homebuyers is confirmed as a driver, policies to improve housing affordability could have beneficial effects.
- Research gaps and recommendations:
  - Further analysis using more granular household-level data is required to test alternative hypotheses (e.g., parental transfers, bequest motives, housing down-payment needs).
  - More definitive conclusions depend on household-level microdata and further research.

*International Monetary Fund — SWEDEN, November 1, 2017*

### Appendix I. Fixed Effects Panel Model for Gross Saving Rates

### Appendix I. Fixed Effects Panel Model for Gross Saving Rates

### Data and sample
- Countries: Denmark, Finland, France, Germany, Netherlands, Norway, Sweden and the United Kingdom.
- Frequency and sample period: annual frequency from 1995 (1996 for Norway) to 2016.
- Variables (as labeled in regression):
  - sr: Gross household saving rate, percent (dependent). Source: Eurostat (nasa_10_ki).
  - ur: Unemployment rate, percent. Source: Eurostat (une_rt_a).
  - gdi.g: Real disposable income y/y growth, percent. Source: OECD (table 14A) and IMF WEO.
  - oad: Old-age dependency ratio (population of 65+ years old to population of 15–64 years old), percent. Source: Eurostat (demo_pjanind).
  - emp65: Elderly employment rate (employed 65+ years old to population of 65+ years old), percent. Source: Eurostat (ifsa_egais and demo_pjanbroad).
  - assets: Level of total household assets divided by gross disposable income, percent. Source: OECD (tables 720R, 9B and 14A).
  - assets.g: Total household asset y/y growth, percent. Source: OECD (tables 720R and 9B).
  - nwealth: Level of household net wealth (assets – liabilities) divided by gross disposable income, percent. Source: OECD (tables 720R, 9B and 14A).

### Estimation approach
- Country fixed effects panel regression estimated using the 7 explanatory variables with the 1-year forward growth of gross disposable income (gdi.g t + 1Y) instead of the coincident indicator (IMF WEO forecast for 2017 used for the lead).
- Rationale: control for second-order effect of unemployment on saving rates (higher unemployment → higher savings via labor income risk, but also reduces expected income).

### Regression estimates (Country Fixed Effects Panel Estimates)
- ur: 0.42***; Standard Error: 0.11; Statistic: 3.71
- 1-year lead of gdi.g (gdi.g t + 1Y): -0.20*; Standard Error: 0.09; Statistic: -2.27
- oad: -0.27**; Standard Error: 0.09; Statistic: -2.90
- emp65: 0.47**; Standard Error: 0.17; Statistic: 2.73
- assets: -0.03***; Standard Error: 0.01; Statistic: -3.36
- assets.g: -0.20***; Standard Error: 0.04; Statistic: -5.26
- nwealth: 0.05***; Standard Error: 0.01; Statistic: 5.14

- Model statistics:
  - R2 = 0.41
  - K = 8 countries
  - T = 20-21 years
  - N = 167 observations
  - model F-statistic: 15.1 on 7 and 152 degrees of freedom, p-value < 0.001
  - Significance levels: *** < 0.001 < ** < 0.01 < * < 0.05

### Key findings and interpretation
- Signs and magnitudes of estimated coefficients are consistent with literature and theoretical considerations (e.g., Mody et al., 2012), with the exception of the net wealth variable.
- The positive coefficient on nwealth (0.05***) is at odds with expectation that higher net wealth reduces the need for additional savings; hypothesis offered: desired net wealth may have increased over time.
- Model fit is reported by country in the panel chart (model fits and data).

*Source: Fund staff calculations.*

### Appendix II. Error Correction Model for the Saving Rate Excluding Pensions

### Data, frequency and variable definitions
- Frequency and sample period: quarterly frequency starting from 1980 (Statistics Sweden series used for longer quarterly series).
- Seasonality and temporal interpolation:
  - Variables seasonally-adjusted either by source or via X-13ARIMA-SEATS.
  - Variables available only annually (household balance sheet and demographics) converted to quarterly frequency via the Denton-Cholette method.
- Additional variables (new labels and sources):
  - srxpen: Gross household saving rate excluding pension savings, percent (dependent). Source: Statistics Sweden (NR0103DT).
  - yad: Young-age dependency ratio (population of 0–14 years old to population of 15–64 years old), percent. Source: Eurostat (demo_pjanind).
  - nwealth.g: Household wealth y/y growth, percent. Source: Statistics Sweden (000000KI).
  - el.care: Deflator of residential care and social work activities (2016 prices) / deflator of final consumption expenditure of households (2016 prices) / (population of 65+ people / population). Source: Statistics Sweden (NR0103DA, NR0103BV and BE0101C£).

### Dependent variable construction
- Gross household saving rate excluding pension savings is calculated as:
  - gross saving rate ex pension savings = (gross savings – change in net equity of pension fund reserves) / gross disposable income.
- Rationale: preferable for identifying drivers of non-mandatory savings since the share of mandatory savings in GDP has remained relatively stable since the mid-1990s.

### Additional proxy
- el.care introduced as a proxy for the cost of elderly care; estimated as relative cost of elderly residential care to overall household consumption, scaled by the share of elderly in population.

### Long-run regression specifications and estimates
- Two long-run models estimated: Basic Model and Expanded Model (expanded model introduces el.care interacting with oad).

Basic Model (Long-Run)
- Intercept: 11.73; Standard Error: 8.16; Statistic: 1.44
- ur: 0.66***; Standard Error: 0.13; Statistic: 5.23
- gdi.g t + 4Q: -1.03***; Standard Error: 0.11; Statistic: -9.28
- yad: -0.56*; Standard Error: 0.27; Statistic: -2.09
- nwealth: 0.01***; Standard Error: 0.00; Statistic: 5.23
- nwealth.g: -0.19***; Standard Error: 0.03; Statistic: -6.60

- Adjusted R2 = 0.64; model F-statistic: 50.2 on 5 and 136 degrees of freedom, p-value < 0.001.

Expanded Model (Long-Run)
- Intercept: 235.94***; Standard Error: 22.30; Statistic: 10.58
- ur: 0.86***; Standard Error: 0.10; Statistic: 8.26
- gdi.g t + 4Q: -0.66***; Standard Error: 0.08; Statistic: -8.43
- yad: -2.50***; Standard Error: 0.27; Statistic: -9.13
- nwealth: 0.04*; Standard Error: 0.01; Statistic: 2.53
- nwealth.g: -0.35***; Standard Error: 0.06; Statistic: -5.81
- assets t – 4Q: -0.05***; Standard Error: 0.01; Statistic: -3.53
- el.care: -44.44***; Standard Error: 3.89; Statistic: -11.43
- oad: -5.92***; Standard Error: 0.63; Statistic: -9.38
- el.care * oad: 1.68***; Standard Error: 0.15; Statistic: 10.97

- Adjusted R2 = 0.85; model F-statistic: 92.7 on 9 and 132 degrees of freedom, p-value < 0.001.

- Significance levels: *** < 0.001 < ** < 0.01 < * < 0.05.

### Cointegration and stationarity
- Unit root tests show residuals of the expanded model are stationary.
- Combined with inspection of orders of integration, evidence suggests a cointegrating relationship between the variables.

### Error Correction Model (short-run dynamics in 4-quarter increments)
- Dependent variable: ∆4 saving rate excluding pension savings (∆4 denotes the difference operator over 4 periods).
- Coefficients (Error Correction Model for the Gross Household Saving ex-Pension Savings Rate):
  - Intercept: -2.49; Standard Error: 6.38; Statistic: -0.39
  - Residual of expanded model t – 4Q: -0.51***; Standard Error: 0.08; Statistic: -6.77
  - ∆4 gdi.g: 0.5***; Standard Error: 0.05; Statistic: 9.92
  - oad t – 4Q: 1.19***; Standard Error: 0.2; Statistic: 6.09
  - ∆4 oad: -3.63***; Standard Error: 0.51; Statistic: -7.1
  - yad t – 4Q: -1.03***; Standard Error: 0.11; Statistic: -9.23
  - ∆4 yad: -3.53***; Standard Error: 0.68; Statistic: -5.16
  - ∆4 assets: -0.03***; Standard Error: 0; Statistic: -8.65
  - assets.g t – 4Q: -0.19***; Standard Error: 0.02; Statistic: -9.38

- Model statistics:
  - Adjusted R2 = 0.73
  - model F-statistic: 48.2 on 8 and 131 degrees of freedom, p-value < 0.001
  - Note: The coefficient of the error correction term (-0.51***) implies about half of the remaining gap from the long-run equation is closed each year.

### Interpretation and implications
- The expanded long-run specification (including el.care and its interaction with oad) substantially improves fit, particularly in the post-crisis period.
- The non-significance of differences and lags of el.care and nwealth in the final differences equation may support the hypothesis that concerns about increased cost of elderly care and an increase in target net wealth are relatively recent phenomena.
- Short-run dynamics operate in whole-year (4-quarter) increments; specifications with 1-, 2-, and 3-quarter increments did not provide meaningful or parsimonious estimates, and long-lag effects (8 quarters and longer) lacked consistently statistically significant estimates.

*Source: Fund staff calculations.*

### 7.      The following sections investigate the drivers of recent wage moderation in Sweden.

### 7.      The following sections investigate the drivers of recent wage moderation in Sweden

### Background on Swedish wage formation and stylized facts
- Since the Industrial Agreement of 1997, wage formation has been led by bargaining in the industrial sector, which sets a benchmark expected to be adopted by other sectors.
- Collective agreements cover over 90 percent of the employees.
- Firm/local level negotiations have increased from 10 percent in 2013 to over 20 percent by 2015; National Mediation Office reports "more than 20 percent as of now."
- Real labor compensation per hour and real GDP per hour worked historically track closely; there was a period in 2011–13 where real wage growth exceeded trend productivity.
- Trend labor productivity growth:
  - declined to about 1¼ percent in recent years from over 2 percent before the crisis.
- Inflation expectations remained below the 2 percent target for several years until 2017.
- Labor market slack indicators:
  - Domestic unemployment and other slack measures point to improving/tightening labor market conditions.
  - Euro area labor market slack has declined, though the unemployment gap has not closed and involuntary part-time employment remains high.
- Recent alignment with external wages:
  - Swedish wages have moved very closely with agreed wages in Germany in recent years, reflecting employers’ emphasis on safeguarding cost competitiveness and export market share.

### Analytical approach and model specification
- Core model: wage curve based on Blanchard (1997) extended with trend labor productivity (TLP) and an error-correction specification (ECM).
- Long-run relationship: loglogRW = α − β logTLP (real wages negatively related to aggregate unemployment in wage-curve tradition, augmented with productivity).
- Error-correction model (ECM) features:
  - Error correction term depends on lagged levels of real wages and trend labor productivity (4 quarters earlier).
  - Difference operator covers 4 quarters, reflecting annual wage adjustment practice.
  - Specification controls for lagged dependent variable (4-quarter lag) to reflect inertia or base effects.
  - A statistically significant negative coefficient on lagged real wages (4t RW−4) indicates cointegration between real wages and trend productivity.
  - Parameter φ1 expected to be approximately equal to φ2 so real wages grow broadly at the pace of trend productivity in equilibrium.

### Data and variable definitions
- Nominal wages: two measures used
  - (1) wages and salaries from the national accounts as a ratio to hours worked;
  - (2) total labor compensation from the national accounts as a ratio to hours worked (includes employers’ social security contributions).
- Rationale: national accounts measures reflect structural changes in composition of the labor force and align with labor productivity measures; LCI covers business sector and is based on a more stable job basket.
- Real wages: nominal wage indicator deflated by the GDP deflator (not a consumer price measure).
- Inflation expectations: TNS-Prospera survey; 2-year ahead measure used.
- Key variable definitions (as stated):
  - Unemployment Gap (UnemGap): Percent deviation from HP-filtered equilibrium unemployment rate (lambda=12000), source LFS.
  - Involuntary PT employment: Share of involuntary PT workers in share of total employment, source Eurostat.
  - Labor compensation per hour (W): total labor compensation per hour worked, source Eurostat, National Accounts.
  - Wages and salaries per hour (W): wages and salaries per hour worked, source Eurostat, National Accounts.
  - Trend productivity (TLP): HP-filtered trend real labor productivity per hour (lambda=1600), source Eurostat.
  - Real Compensation/wage (RW): Compensation or wage / GDP deflator, source GDP deflator, Nominal GDP/Real GDP, Statistics Sweden.
  - Expected inflation: 2 year ahead expected inflation, source TNS Prospera.

### Estimation results and main findings
- Sample period: 1995:Q1 to 2017:Q1.
- Dependent variables: labor compensation per hour (national accounts) and wages and salaries per hour (national accounts) in separate estimations.
- Cointegration evidence:
  - Coefficients on the lagged level of the real wage are statistically significant in all variants, indicating cointegration between real wages and productivity.
  - Rate of error correction ranges from -0.2 to -0.4 across specifications.
- Model variants and key significant determinants:
  - Models 1–3 (domestic variables only):
    - ECM term significant.
    - Significant variables: unemployment gap, changes in involuntary part-time employment, and labor productivity growth.
  - Models 4–6 (add euro area labor market slack):
    - Changes in euro area involuntary part-time employment are statistically significant.
    - Domestic labor slack indicators remain statistically significant.
  - Models 7–9 (add German compensation/wage growth):
    - German compensation/wage growth has a significant impact on Swedish compensation growth.
    - In these more complete models, coefficients on expected inflation and euro area unemployment changes become significant, while the domestic unemployment gap remains statistically significant.
- Interpretation:
  - Wage formation has been influenced by both domestic fundamentals (productivity, domestic labor slack, inflation expectations) and international spillovers (especially German wage developments and euro area labor slack).
  - The ECM dynamics imply that deviations of real wages from trend productivity (e.g., the 2011–13 overhang) influence subsequent nominal wage growth as wages adjust back toward equilibrium.

*Source: cr17351 - 7.      The following sections investigate the drivers of recent wage moderation in Sweden.*

### 18.      While results are robust to using data on wages and salaries excluding employer’s

### 18.      While results are robust to using data on wages and salaries excluding employer’s

### Model fit and data distinctions
- Models using total labor compensation fit better than models using wages and salaries excluding employer’s social security contributions.
- Hourly compensation growth is more closely related to trend productivity, labor market conditions, and German labor compensation growth (larger and more significant coefficients) than is hourly wages and salaries growth.
- Expected inflation has more direct and larger impact on wages and salaries than on total compensation.
- Further analysis focuses on Table 1, model 9.

### Long-run drivers of real labor compensation
- Estimation results confirm that labor productivity is the main driver of real labor compensation in the long run.
- Estimated coefficients on lagged real labor compensation and trend productivity are approximately equal in absolute values (Table 1).
- The long-run solution of Table 1, model 9 shows:
  - a coefficient just over unity on labor productivity.
  - a long-run coefficient on the unemployment gap consistent with wage curve literature.
- Reported long-run equation fragment: log4.00  1.06*log0.13*RWLTPUnemGap   
- Footnote: The coefficient on trend productivity is significantly higher than 1 at a 99 percent confidence level.

### Error correction model (ECM) performance
- The error correction model explains close to 80 percent of the total variation of the actual wage development (combined model, Table 1, model 9: R-squared = 0.78).
- The domestic wage curve model (includes unemployment, productivity, and expected inflation) has an R-square of only 60 percent.
- Spillover effects from the euro area labor market condition and German wage growth each contribute an additional 10 percent to the model fit (together raising fit by about 20 percentage points relative to the domestic model).

### Scenario analysis (based on model 9)
- Baseline: underlying variables remain unchanged at the 2017Q1 level.
- Domestic shocks:
  - An increase in productivity growth from the current 1.25 to 1.75 percent would raise nominal labor compensation growth by 0.5 percent in the short run; the immediate impact dissipates over time via the error correction mechanism.
  - A decline in unemployment rate by 1 percentage point has positive impact on wage growth of 0.2 percent.
  - A decline in involuntary part-time (PT) employment share by 1 percentage point has positive impact on wage growth of 0.1 percent.
  - The combined effect from the domestic shocks could lift nominal labor compensation growth by about 0.8 percent y/y.
- Foreign shocks:
  - An increase in German wage growth from 2 to 3 percent yields a spillover effect close to 0.6 percent.
  - A fall in euro area unemployment rate from 9.5 to 8.5 percent has an impact of about 0.1 percent.
  - The combined foreign effect could raise nominal compensation growth by about 0.7 percent y/y.
- Combined domestic and foreign shocks:
  - A combination of domestic and foreign factors could increase nominal compensation growth by about 1.5 percent.

### Comparison with other EU countries (panel regression covering 22 EU countries)
- Domestic and foreign variables in the panel regression are statistically significant with expected signs.
- Impacts of productivity growth, expected inflation, and euro area labor market slack on labor compensation growth are broadly comparable between Sweden and the EU average.
- Spillover of German wages:
  - In Sweden, a 1 percentage point increase in German labor compensation growth leads to a 0.6 percentage point increase in Swedish labor compensation growth.
  - The average impact on EU labor compensation growth is about 0.2 percentage point (one third of the Swedish magnitude).
- Long-run responsiveness to unemployment gap:
  - Sweden: -0.13 on unemployment gap.
  - Average EU country: -0.11 on unemployment gap.

### Analysis of recent wage moderation
- Decomposition (using short-run regression model 9) indicates recent wage moderation is associated with:
  - Sluggish productivity growth.
  - Low expected inflation.
  - An adjustment from higher wage growth in earlier years (negative contribution from the error correction term reflecting correction toward equilibrium after real labor compensation rose faster than trend productivity).
- Rolling regressions (10-year windows) find:
  - Wage responsiveness to unemployment remains broadly stable after accounting for foreign spillovers.
  - Impact of expected inflation has fluctuated with no clear trend.
- Increasing external linkages:
  - Rolling coefficient on German wages rose notably since the early 2000s and continued rising; recent German labor compensation growth is around 2 percent (noted as somewhat higher than pre-crisis averages close to 1 percent).
  - Rolling coefficients on euro area unemployment changes increased significantly since the early 2000s and stabilized more recently, consistent with higher trade integration and freer labor flows across EU countries.
- Other possible structural contributors to wage moderation (not included in econometric analysis):
  - Decline in profitability: net return on capital in Sweden declined since the crisis to be the lowest among peer economies (Net Return on Capital series shown for Sweden, Germany, Finland, Denmark, Netherlands).
  - Structural change toward less capital-intensive sectors (e.g., services).
  - Globalization-induced competition, increased use of temporary and part-time workers, outsourcing.
  - Rising labor supply not covered by labor force survey data (e.g., posted workers in construction).

### Conclusions and future research
- Longer-run perspective:
  - Swedish real wages grew slightly over 1 percent on average as real productivity increased by 1 percent.
  - Real wages also rose about 0.1 percent as the unemployment rate fell by 1 percentage point.
  - Nominal wages react to expected inflation and trend productivity with magnitudes comparable to other EU countries.
  - Domestic fundamentals explain about 60 percent of total variation of the wage development.
- Recent disconnect between nominal wage growth and unemployment:
  - Reflects closer external linkages in Swedish wage formation and a process of correction after unusually rapid real wage rises relative to productivity post-global financial crisis.
  - Including foreign factors increases model fit by about 20 percentage points.
  - Rolling regressions confirm increased wage responsiveness to euro area labor market conditions and German wage growth, especially after 2000.
- Suggested directions for further research:
  - Assess structural factors underlying the rise in the impact of external labor market conditions.
  - Investigate whether omission of broader indicators of domestic labor market slack, foreign labor market conditions, and deviation from long-run equilibrium wage levels drives findings of a flattened Phillips curve.
  - Explore the role of rising European labor market integration (e.g., Schengen expansion from 2003) as a contributor to rising external influence.

*International Monetary Fund — Sweden chapter content (extracted from the supplied PDF section).*

---


_Source: https://www.imf.org/-/media/files/publications/cr/2017/cr17351.pdf_