## 4.1  Two-Country Extension

## Source details

**Canonical URL:** [4.1  Two-Country Extension](https://www.imf.org/-/media/files/publications/wp/2018/wp18264.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18264.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18264.pdf.json)

---

### Model setup and market-clearing
- Interest rate market clearing condition: total savings across the two regions equals total borrowing across the two regions.
- Notation: starred variables denote one country, non-starred variables the other.
- Demand for borrowing in the starred country: n_t^? d_t^? / (1 + r_t). Supply of savings: −n_{t−1}^? b_{t−1}^?.
- Global interest rate formula: 1 + r_t = [n_t^? d_t^? + n_t d_t][−n_{t−1}^? b_{t−1}^? − n_{t−1} b_{t−1}]^{−1}.
- Capital flows between regions defined as the difference between savings demand and supply at the world equilibrium interest rate.

### Calibrations and comparative statics (Figures 5 and 6)
- Prime-age savers share experiment (Figure 5):
  - Two countries calibrated to same parameters and demographics; country A’s middle-aged cohort size varied.
  - Finding: higher prime-age savers share in country A → greater individual savings in country A → excess savings relative to desired borrowing in country A → capital flows from country A to country B.
  - Empirical-consistent slope: linear plot slope of excess savings as a fraction of the endowment to the prime savers share ≈ 0.21.
- Life expectancy experiment (Figure 6a):
  - Country A life expectancy calibrated to 80 years; country B varied between 75 years and 90 years.
  - No formal/informal intergenerational transfers in either country (only bond-market transfers).
  - Finding: capital flows from the country with higher life expectancy to the country with lower life expectancy, reflecting higher savings by those expecting to live longer.
  - Example: when country A life expectancy = 80 and country B = 75, global interest rate > closed-economy interest rate for A → capital flows from A to B.
- Intergenerational transfer generosity experiment (Figure 6b):
  - Both country A and B have equally generous transfers with λ=λ?= 0.15.
  - Finding: with transfers, within-country savings are less sensitive to changes in life expectancy → cross-country capital flows are less sensitive to life expectancy differences.

### Formal results from the multi-country model
- Result 6 (Capital Flows and Life Expectancy): Capital flows from countries with higher life expectancy and savings rates to those with lower life expectancy and savings rates.
- Result 7 (Capital Flows and Prime-Age Share): Capital flows from countries with a relatively larger share of prime-age savers, as those countries also have higher relative savings rate.
- Result 8 (Capital Flows and Transfer Generosity): The sensitivity of cross-country capital flows to life expectancy decreases with an increasing generosity of the formal and informal transfer system.

### Transition dynamics (Section 4.2; Figures 7 and 8)
- Convergence in life expectancy (Panel 7a):
  - Country A constant life expectancy = 80; country B increases from 75 to 80 over 40 years.
  - Country B initially saves less → current account deficit; as life expectancies converge, gap in savings/borrowing narrows.
- Temporary population size shock (Panel 7b):
  - Country B experiences unexpected temporary 20% population increase; country A population fixed.
  - Immediate effect: high borrowing demand in B → current account deficit financed by borrowing from A.
  - Later when the larger cohort reaches prime-age saving, B runs a current account surplus by importing capital from A; after cohort exits, both converge to steady state.
- Role of unfunded pay-as-you-go (PAYG) transfers (Panels 7c and 7d):
  - Both countries operate PAYG with generosity λ= 0.4.
  - Finding: magnitudes of capital flows are muted with PAYG in place because sensitivity of savings to demographic changes falls.
- Asymmetric introduction of PAYG (Figure 8):
  - Panel 8a: Both countries identical (life expectancy 80, no fertility shocks). After 40 years, country B introduces PAYG with generosity rising gradually to λ= 0.5 over 40 years.
    - Finding: country B becomes a net borrower as aggregate saving declines following the windfall transfer to the initial old generation.
  - Panel 8b: Country B increases PAYG generosity from λ= 0.1 to λ= 0.3, experiences life expectancy increase from 70 to 80, and fertility decline relative to country A.
    - Finding: higher relative savings needs induced by higher life expectancy and lower fertility cause country B to become a net lender despite increased PAYG generosity.
  - Overall: net saving/borrowing imbalance depends on interplay between demographic trends and coverage/generosity of old-age transfers.

### Empirical sample and context used for validation
- Empirical sample: 49 countries (advanced and emerging), encompassing about 90 percent of global GDP (as of 2017).
- Time period for most countries: 1986-2016.
- Data sources used: IMF WEO database (current account) and UN World Population Prospects database, 2017 revision (demographics).

### Key empirical findings linking model to data (Section 5 and subsections)
- Prime-saver share and current account (Section 5.1; Figure 9):
  - Prime-aged savers defined as ages 45-64; working age defined as ages 30-64.
  - Cross-section and pooled correlations show a strong positive correlation between prime-saver share and current account balance.
  - Estimated slope coefficient ≈ 0.2 to 0.3; R-squared between 8 and 20 percent (pooled and cross-section).
  - Economic magnitude: a standard deviation change in prime-saver share of 5.6 percent associated with a 1.1 to 1.7 percent higher current account balance.
- Life expectancy and current account (Section 5.2; Figure 10):
  - Life expectancy at prime age defined between age 45-50.
  - Observed non-linearity: at low life expectancy (mostly low-income and emerging markets), improvements in prime-age life expectancy are associated with lower current accounts; relationship turns positive for countries with higher initial life expectancy (advanced economies).
  - Model mechanisms for non-linearity:
    - Higher life expectancy → disproportionately higher end-of-life medical and long-term care costs → non-linear/quadratic relation between saving and life expectancy.
    - Lower life expectancy countries tend to have weaker institutions/financial development, weakening life-cycle saving link.
  - For low-life-expectancy countries, positive health/productivity effects and precautionary savings can produce negative correlation between life expectancy gains and current accounts.
- Old-age transfers, life expectancy and current account (Section 5.3; Figures 11 and 12):
  - Within OECD subsample, variation in future old-age dependency ratios is negatively correlated with level of public pension generosity (public pension spending per elderly relative GDP per worker); correlation coefficient about −0.2.
  - Expected shifts in population age composition are associated with decreases in public pension generosity over the same horizon.
  - Future old-age dependency ratio constructed as moving average of old-age dependency ratio 15 to 25 years forward (reflecting retirement horizon of a prime-age worker).
  - Empirical correlation: current account positively correlated with future old-age dependency ratio (higher future dependency → higher need for life-cycle saving → higher current account).
  - Sensitivity of current accounts to life expectancy increases with higher future dependency ratios: slope of CA vs life expectancy strongly positive for observations at/above median future old-age dependency ratio and essentially flat below median.

*Source: IMF Working Paper section 4.1 and surrounding text (wp18264 - 4.1 Two-Country Extension).*

### 5.4  Putting it All Together

### Multivariate regression: specification and key coefficients
- Demographic explanatory variables are expressed in deviation from world averages.
- Pooled regressions add demographic variables sequentially to a baseline EBA specification (controls include old-age dependency ratio and population growth).
- Selected coefficient estimates from Table 1 (Dependent Variable: CA balance (% of GDP); N = 1495):
  - Old-age dependency:
    - Column (1): 0.044**
    - Column (2): -0.008
    - Column (3): -0.058**
    - Column (4): -0.074***
    - Column (5): -0.074***
  - Population growth:
    - Column (1): -0.643***
    - Column (2): -1.044***
    - Column (3): -0.824***
    - Column (4): -1.025***
    - Column (5): -1.024***
  - Life expectancy:
    - Column (1): 0.000
    - Column (2): -0.037***
    - Column (3): -0.035***
    - Column (4): -0.007
    - Column (5): -0.006***
  - Life expectancy squared:
    - Column (1): 0.001***
    - Column (2): 0.001***
    - Column (3): 0.000
  - Prime-saver share:
    - Column (3): 0.132***
    - Column (4): 0.196***
    - Column (5): 0.198***
  - Life exp × Future old-age dep:
    - Column (4): 0.023***
    - Column (5): 0.024***
  - Constant (selected):
    - Column (1): -0.001
    - Column (2): -0.004***
    - Column (3): -0.003**
    - Column (4): -0.004***
    - Column (5): -0.004***
- R-squared values (Table 1):
  - Column (1): 0.037
  - Column (2): 0.058
  - Column (3): 0.064
  - Column (4): 0.077
  - Column (5): 0.077
- Notes: Significance levels: *, 10%; **, 5%; ***, 1%. Sample is 1986 to 2016.

### Non-linearities and interaction effects
- Life expectancy does not have a simple linear effect; a strongly significant positive life-expectancy squared term in earlier columns indicates a positive effect on the current account only at relatively high levels of life expectancy.
- The interaction term life expectancy × future old-age dependency ratio enters with a strong positive sign, consistent with model predictions that higher longevity combined with unfavorable future old-age support ratios raises the need to rely on own saving.
- When including the interaction term, the life-expectancy squared term becomes insignificant, suggesting the interaction captures most of the non-linearity (captures negative slope for low life expectancy EMs, turning positive for AEs with unfavorable old-age support ratios).

### Economic significance and explanatory power
- Demographic indicators in the joint specification (column 5 of Table 1) explain almost 13 percent of the cross-country variation in current account balances over the long run.
- The observed magnitude (about 13 percent) is consistent with the literature, where demographic forces in calibrated structural models explain between 13 and 27 percent of current account variation across major advanced economies.

### Robustness checks (Table 2)
- Alternative specifications and robustness exercises (sample 1986 to 2016) produce consistent demographic coefficient signs and significance:
  - Column (1) adds country fixed effects (Country FE: Y) — demographic mechanisms operate within countries.
  - Column (2) adds time fixed effects (Time FE: Y) — common trends removed.
  - Column (3) uses Prais-Winsten regression allowing panel-specific heteroskedasticity and autocorrelation.
  - Column (4) introduces additional EBA explanatory variables ((NFA/Y) t−1, (Y/worker) t−1, Oil×Resource Temp., Public Health Spending/GDP t−1, VIX×Cap. Open. t−1, Own currency’s share in world reserves, Output gap, Commodity ToT gap×Trade Openness, Detrended private credit/GDP, etc.).
- Selected coefficient highlights from Table 2 (N varies; example N = 1495 in columns 1–3, N = 1372 in column 4):
  - Old-age dependency:
    - (1): -0.205***
    - (2): -0.190***
    - (3): -0.050
    - (4): -0.100**
  - Population growth:
    - (1): -2.136***
    - (2): -1.673***
    - (3): -1.205***
    - (4): -0.951**
  - Life expectancy (at prime-age):
    - (1): -0.005***
    - (2): -0.007***
    - (3): -0.006***
    - (4): -0.005***
  - Prime saver share:
    - (1): 0.145***
    - (2): 0.172***
    - (3): 0.138**
    - (4): 0.115**
  - Life exp.*Future old-age dep.:
    - (1): 0.030***
    - (2): 0.027***
    - (3): 0.024***
    - (4): 0.011**
  - R-squared values (Table 2):
    - (1): 0.063
    - (2): 0.102
    - (3): 0.023
    - (4): 0.304
- Conclusion from robustness: demographic coefficients retain similar magnitudes and statistical significance when alternative econometric approaches and additional controls are used, indicating empirical salience and stability of the identified mechanisms.

### Mechanisms, model implications, and caveats
- Model framework: open economy overlapping generations model with mortality risk, population growth, and old-age pay-as-you-go transfer systems.
- Key theoretical predictions supported by the data (panel of 49 countries, sample period of 30 years):
  - Saving and capital flows are affected by differences in life expectancy and the share of savers versus dissavers (prime-saver share).
  - Fertility and longevity trends shape the share of savers and thus external balances.
  - Coverage and generosity of old-age transfer schemes modify capital flow implications of demographic transition stages.
  - Non-linearities emerge when there are limits to elderly tax-transfer mechanisms.
- Policy-relevant caveats:
  - Pension systems differ across many dimensions (private fully-funded vs unfunded PAYG, coverage, generosity, maturity), and parameters have evolved substantially over time, complicating simple mapping to current accounts.
  - Distilling these pension-system features into simple time-varying parameters is challenging.
  - Even if pension coverage/generosity can be related to current accounts, the resulting saving-investment balance may not correspond to a welfare-desirable outcome for each country.
- Suggested area for further research: defining benchmark or desirable old-age transfer policies to form normative views.

### Additional theoretical considerations (appendix summaries)
- Discount factor as proxy for financial development:
  - Introducing a discount factor β attached to old-age utility implies steady-state middle-age savings −b_{t−1}^t = βγ/(1 + βγ) (y − d).
  - Sensitivity of savings to changes in life expectancy γ increases in β: ∂(−b_{t−1}^t)/∂γ = β(1 + βγ)^{−2} (y − d).
  - Interpretation: higher β (proxy for deeper financial development) raises the sensitivity of savings to longevity.
- End-of-life care expenses:
  - Introducing an exogenous end-of-life cost κ that increases in survival probability γ reduces resources for saving and consumption.
  - With budget constraint c_{t+2}^t + γ_{t+2} κ ≤ −(1 + r_{t+1})^{-1} γ_{t+2} b_t^{t+1}, optimal consumption growth remains as baseline, but savings increase with the expected end-of-life expense.
  - Calibrating κ = 0.1 yields a nonlinear increase in savings with life expectancy.
- Borrowing constraints as proxy for financial development:
  - If young can only borrow a fraction θ of future endowment (d = θy) and θ is binding, steady-state savings −b_{t−1}^t = γ/(1 + γ) y(1 − θ).
  - Lower per-capita output y implies tighter borrowing constraints and reduces the sensitivity of savings to life expectancy γ.

### Data and summary statistics used in empirical analysis
- Data sources: UN World Population Prospects (Prime Savers Share; Life Expectancy at Prime Age; Population Growth; Old-age Dependency Ratio), Lane and Milesi-Ferretti (Net Foreign Assets), World Economic Outlook (Output Per Worker, Oil/Natural Gas Trade Balance, GDP growth forecast in 5 years, Output Gap, Commodity ToT, Trade Openness, Cyclically adjusted Fiscal Balance, (∆ Reserves)/GDP×Capital Controls), Quinn Database (Capital Openness), WITS and BP Statistical Review (Oil and Natural Gas), OECD/WDI/CEPAL/FAD/ADB (Public Health Spending/GDP), Haver (VIX), COFER (Own currency’s share in world reserves), BIS (Detrended Private Credit/GDP).
- Selected summary statistics (Table 3; Obs. = 1519 unless noted):
  - Old-age dependency ratio (population 30-64/population 65+): Mean 0.249, Std. Dev. 0.094, Min 0.106, Max 0.579.
  - Population growth (Annual growth rate of total population): Mean 0.010, Std. Dev. 0.008, Min -0.005, Max 0.037.
  - Life expectancy at prime age (Avg. life expectancy at 45 to 50 years old): Mean 31.296, Std. Dev. 3.183, Min 22.299, Max 37.873.
  - Prime-saver share (Population 45-64/population 30-64): Mean 0.482, Std. Dev. 0.059, Min 0.361, Max 0.624.
  - Future old-age dependency ratio: Mean 0.336, Std. Dev. 0.142, Min 0.114, Max 0.782.
  - Pension per elderly (in %): Obs. = 775; Mean 25.600, Std. Dev. 9.633, Min 0, Max 59.809.

*Source: IMF working paper content (section 5.4, tables and appendices).*

### 4.1  Two-Country Extension

### 4.1 Two-Country Extension

### Model setup and market-clearing
- Interest rate market clearing condition: total savings across the two regions equals total borrowing across the two regions.
- Notation: starred variables denote one country, non-starred variables the other.
- Demand for borrowing in the starred country: n_t^? d_t^? / (1 + r_t). Supply of savings: −n_{t−1}^? b_{t−1}^?.
- Global interest rate formula (as given): 1 + r_t = [n_t^? d_t^? + n_t d_t][−n_{t−1}^? b_{t−1}^? − n_{t−1} b_{t−1}]^{−1}.
- Capital flows between regions defined as the difference between savings demand and supply at the world equilibrium interest rate.

### Calibrations and comparative statics (Figures 5 and 6)
- Prime-age savers share experiment (Figure 5):
  - Two countries calibrated to same parameters and demographics; country A’s middle-aged cohort size varied.
  - Finding: higher prime-age savers share in country A → greater individual savings in country A → excess savings relative to desired borrowing in country A → capital flows from country A to country B.
  - Empirical-consistent slope: linear plot slope of excess savings as a fraction of the endowment to the prime savers share ≈ 0.21.
- Life expectancy experiment (Figure 6a):
  - Country A life expectancy calibrated to 80 years; country B varied between 75 years and 90 years.
  - No formal/informal intergenerational transfers in either country (only bond-market transfers).
  - Finding: capital flows from the country with higher life expectancy to the country with lower life expectancy, reflecting higher savings by those expecting to live longer.
  - Example: when country A life expectancy = 80 and country B = 75, global interest rate > closed-economy interest rate for A → capital flows from A to B.
- Intergenerational transfer generosity experiment (Figure 6b):
  - Both country A and B have equally generous transfers with λ=λ?= 0.15.
  - Finding: with transfers, within-country savings are less sensitive to changes in life expectancy → cross-country capital flows are less sensitive to life expectancy differences.

### Formal results from the multi-country model
- Result 6 (Capital Flows and Life Expectancy): Capital flows from countries with higher life expectancy and savings rates to those with lower life expectancy and savings rates.
- Result 7 (Capital Flows and Prime-Age Share): Capital flows from countries with a relatively larger share of prime-age savers, as those countries also have higher relative savings rate.
- Result 8 (Capital Flows and Transfer Generosity): The sensitivity of cross-country capital flows to life expectancy decreases with an increasing generosity of the formal and informal transfer system.

### Transition dynamics (Section 4.2; Figures 7 and 8)
- Convergence in life expectancy (Panel 7a):
  - Country A constant life expectancy = 80; country B increases from 75 to 80 over 40 years.
  - Country B initially saves less → current account deficit; as life expectancies converge, gap in savings/borrowing narrows.
- Temporary population size shock (Panel 7b):
  - Country B experiences unexpected temporary 20% population increase; country A population fixed.
  - Immediate effect: high borrowing demand in B → current account deficit financed by borrowing from A.
  - Later when the larger cohort reaches prime-age saving, B runs a current account surplus by importing capital from A; after cohort exits, both converge to steady state.
- Role of unfunded pay-as-you-go (PAYG) transfers (Panels 7c and 7d):
  - Both countries operate PAYG with generosity λ= 0.4.
  - Finding: magnitudes of capital flows are muted with PAYG in place because sensitivity of savings to demographic changes falls.
- Asymmetric introduction of PAYG (Figure 8):
  - Panel 8a: Both countries identical (life expectancy 80, no fertility shocks). After 40 years, country B introduces PAYG with generosity rising gradually to λ= 0.5 over 40 years.
    - Finding: country B becomes a net borrower as aggregate saving declines following the windfall transfer to the initial old generation.
  - Panel 8b: Country B increases PAYG generosity from λ= 0.1 to λ= 0.3, experiences life expectancy increase from 70 to 80, and fertility decline relative to country A.
    - Finding: higher relative savings needs induced by higher life expectancy and lower fertility cause country B to become a net lender despite increased PAYG generosity.
  - Overall: net saving/borrowing imbalance depends on interplay between demographic trends and coverage/generosity of old-age transfers.

### Empirical sample and context used for validation
- Empirical sample: 49 countries (advanced and emerging), encompassing about 90 percent of global GDP (as of 2017).
- Time period for most countries: 1986-2016.
- Data sources referenced for analysis: IMF WEO database (current account) and UN World Population Prospects database, 2017 revision (demographics).

### Key empirical findings linking model to data (Section 5 and subsections)
- Prime-saver share and current account (Section 5.1; Figure 9):
  - Prime-aged savers defined as ages 45-64; working age defined as ages 30-64.
  - Cross-section and pooled correlations show a strong positive correlation between prime-saver share and current account balance.
  - Estimated slope coefficient ≈ 0.2 to 0.3; R-squared between 8 and 20 percent (pooled and cross-section).
  - Economic magnitude: a standard deviation change in prime-saver share of 5.6 percent associated with a 1.1 to 1.7 percent higher current account balance.
- Life expectancy and current account (Section 5.2; Figure 10):
  - Life expectancy at prime age defined between age 45-50.
  - Observed non-linearity: at low life expectancy (mostly low-income and emerging markets), improvements in prime-age life expectancy are associated with lower current accounts; relationship turns positive for countries with higher initial life expectancy (advanced economies).
  - Model mechanisms for non-linearity:
    - Higher life expectancy → disproportionately higher end-of-life medical and long-term care costs → non-linear/quadratic relation between saving and life expectancy.
    - Lower life expectancy countries tend to have weaker institutions/financial development, weakening life-cycle saving link.
  - For low-life-expectancy countries, positive health/productivity effects and precautionary savings can produce negative correlation between life expectancy gains and current accounts.
- Old-age transfers, life expectancy and current account (Section 5.3; Figures 11 and 12):
  - Within OECD subsample, variation in future old-age dependency ratios is negatively correlated with level of public pension generosity (public pension spending per elderly relative GDP per worker); correlation coefficient about −0.2.
  - Expected shifts in population age composition are associated with decreases in public pension generosity over the same horizon.
  - Future old-age dependency ratio constructed as moving average of old-age dependency ratio 15 to 25 years forward (reflecting retirement horizon of a prime-age worker).
  - Empirical correlation: current account positively correlated with future old-age dependency ratio (higher future dependency → higher need for life-cycle saving → higher current account).
  - Sensitivity of current accounts to life expectancy increases with higher future dependency ratios: slope of CA vs life expectancy strongly positive for observations at/above median future old-age dependency ratio and essentially flat below median.

*Source: IMF Working Paper section 4.1 and surrounding text (wp18264 - 4.1 Two-Country Extension).*

### 5.4  Putting it All Together

### 5.4  Putting it All Together

### Multivariate regression: specification and key coefficients
- Demographic explanatory variables are expressed in deviation from world averages.
- Pooled regressions add demographic variables sequentially to a baseline EBA specification (controls include old-age dependency ratio and population growth).
- Selected coefficient estimates from Table 1 (Dependent Variable: CA balance (% of GDP); N = 1495):
  - Old-age dependency:
    - Column (1): 0.044**
    - Column (2): -0.008
    - Column (3): -0.058**
    - Column (4): -0.074***
    - Column (5): -0.074***
  - Population growth:
    - Column (1): -0.643***
    - Column (2): -1.044***
    - Column (3): -0.824***
    - Column (4): -1.025***
    - Column (5): -1.024***
  - Life expectancy:
    - Column (1): 0.000
    - Column (2): -0.037***
    - Column (3): -0.035***
    - Column (4): -0.007
    - Column (5): -0.006***
  - Life expectancy squared:
    - Column (1): 0.001***
    - Column (2): 0.001***
    - Column (3): 0.000
  - Prime-saver share:
    - Column (3): 0.132***
    - Column (4): 0.196***
    - Column (5): 0.198***
  - Life exp × Future old-age dep:
    - Column (4): 0.023***
    - Column (5): 0.024***
  - Constant (selected):
    - Column (1): -0.001
    - Column (2): -0.004***
    - Column (3): -0.003**
    - Column (4): -0.004***
    - Column (5): -0.004***
- R-squared values (Table 1): 
  - Column (1): 0.037
  - Column (2): 0.058
  - Column (3): 0.064
  - Column (4): 0.077
  - Column (5): 0.077
- Notes: Significance levels: *, 10%; **, 5%; ***, 1%. Sample is 1986 to 2016.

### Non-linearities and interaction effects
- Life expectancy does not have a simple linear effect; a strongly significant positive life-expectancy squared term in earlier columns indicates a positive effect on the current account only at relatively high levels of life expectancy.
- The interaction term life expectancy × future old-age dependency ratio enters with a strong positive sign, consistent with model predictions that higher longevity combined with unfavorable future old-age support ratios raises the need to rely on own saving.
- When including the interaction term, the life-expectancy squared term becomes insignificant, suggesting the interaction captures most of the non-linearity (captures negative slope for low life expectancy EMs, turning positive for AEs with unfavorable old-age support ratios).

### Economic significance and explanatory power
- Demographic indicators in the joint specification (column 5 of Table 1) explain almost 13 percent of the cross-country variation in current account balances over the long run.
- The observed magnitude (about 13 percent) is consistent with the literature, where demographic forces in calibrated structural models explain between 13 and 27 percent of current account variation across major advanced economies.

### Robustness checks (Table 2)
- Alternative specifications and robustness exercises (sample 1986 to 2016) produce consistent demographic coefficient signs and significance:
  - Column (1) adds country fixed effects (Country FE: Y) — demographic mechanisms operate within countries.
  - Column (2) adds time fixed effects (Time FE: Y) — common trends removed.
  - Column (3) uses Prais-Winsten regression allowing panel-specific heteroskedasticity and autocorrelation.
  - Column (4) introduces additional EBA explanatory variables (controls such as (NFA/Y) t−1, (Y/worker) t−1, Oil×Resource Temp., Public Health Spending/GDP t−1, VIX×Cap. Open. t−1, Own currency’s share in world reserves, Output gap, Commodity ToT gap×Trade Openness, Detrended private credit/GDP, etc.).
- Selected coefficient highlights from Table 2 (N varies; example N = 1495 in columns 1–3, N = 1372 in column 4):
  - Old-age dependency:
    - (1): -0.205***
    - (2): -0.190***
    - (3): -0.050
    - (4): -0.100**
  - Population growth:
    - (1): -2.136***
    - (2): -1.673***
    - (3): -1.205***
    - (4): -0.951**
  - Life expectancy (at prime-age):
    - (1): -0.005***
    - (2): -0.007***
    - (3): -0.006***
    - (4): -0.005***
  - Prime saver share:
    - (1): 0.145***
    - (2): 0.172***
    - (3): 0.138**
    - (4): 0.115**
  - Life exp.*Future old-age dep.:
    - (1): 0.030***
    - (2): 0.027***
    - (3): 0.024***
    - (4): 0.011**
  - R-squared values (Table 2):
    - (1): 0.063
    - (2): 0.102
    - (3): 0.023
    - (4): 0.304
- Conclusion from robustness: demographic coefficients retain similar magnitudes and statistical significance when alternative econometric approaches and additional controls are used, indicating empirical salience and stability of the identified mechanisms.

### Mechanisms, model implications, and caveats
- Model framework: open economy overlapping generations model with mortality risk, population growth, and old-age pay-as-you-go transfer systems.
- Key theoretical predictions supported by the data (panel of 49 countries, sample period of 30 years):
  - Saving and capital flows are affected by differences in life expectancy and the share of savers versus dissavers (prime-saver share).
  - Fertility and longevity trends shape the share of savers and thus external balances.
  - Coverage and generosity of old-age transfer schemes modify capital flow implications of demographic transition stages.
  - Non-linearities emerge when there are limits to elderly tax-transfer mechanisms.
- Policy-relevant caveats:
  - Pension systems differ across many dimensions (private fully-funded vs unfunded PAYG, coverage, generosity, maturity), and parameters have evolved substantially over time, complicating simple mapping to current accounts.
  - Distilling these pension-system features into simple time-varying parameters is challenging.
  - Even if pension coverage/generosity can be related to current accounts, the resulting saving-investment balance may not correspond to a welfare-desirable outcome for each country.
- Suggested area for further research: defining benchmark or desirable old-age transfer policies to form normative views.

### Additional theoretical considerations (appendix summaries)
- Discount factor as proxy for financial development:
  - Introducing a discount factor β attached to old-age utility implies steady-state middle-age savings −b_{t−1}^t = βγ/(1 + βγ) (y − d).
  - Sensitivity of savings to changes in life expectancy γ increases in β: ∂(−b_{t−1}^t)/∂γ = β(1 + βγ)^{−2} (y − d).
  - Interpretation: higher β (proxy for deeper financial development) raises the sensitivity of savings to longevity.
- End-of-life care expenses:
  - Introducing an exogenous end-of-life cost κ that increases in survival probability γ reduces resources for saving and consumption.
  - With budget constraint c_{t+2}^t + γ_{t+2} κ ≤ −(1 + r_{t+1})^{-1} γ_{t+2} b_t^{t+1}, optimal consumption growth remains as baseline, but savings increase with the expected end-of-life expense.
  - Calibrating κ = 0.1 yields a nonlinear increase in savings with life expectancy.
- Borrowing constraints as proxy for financial development:
  - If young can only borrow a fraction θ of future endowment (d = θy) and θ is binding, steady-state savings −b_{t−1}^t = γ/(1 + γ) y(1 − θ).
  - Lower per-capita output y implies tighter borrowing constraints and reduces the sensitivity of savings to life expectancy γ.

### Data and summary statistics used in empirical analysis
- Data sources include: UN World Population Prospects (Prime Savers Share; Life Expectancy at Prime Age; Population Growth; Old-age Dependency Ratio), Lane and Milesi-Ferretti (Net Foreign Assets), World Economic Outlook (Output Per Worker, Oil/Natural Gas Trade Balance, GDP growth forecast in 5 years, Output Gap, Commodity ToT, Trade Openness, Cyclically adjusted Fiscal Balance, (∆ Reserves)/GDP×Capital Controls), Quinn Database (Capital Openness), WITS and BP Statistical Review (Oil and Natural Gas), OECD/WDI/CEPAL/FAD/ADB (Public Health Spending/GDP), Haver (VIX), COFER (Own currency’s share in world reserves), BIS (Detrended Private Credit/GDP).
- Selected summary statistics (Table 3; Obs. = 1519 unless noted):
  - Old-age dependency ratio (population 30-64/population 65+): Mean 0.249, Std. Dev. 0.094, Min 0.106, Max 0.579.
  - Population growth (Annual growth rate of total population): Mean 0.010, Std. Dev. 0.008, Min -0.005, Max 0.037.
  - Life expectancy at prime age (Avg. life expectancy at 45 to 50 years old): Mean 31.296, Std. Dev. 3.183, Min 22.299, Max 37.873.
  - Prime-saver share (Population 45-64/population 30-64): Mean 0.482, Std. Dev. 0.059, Min 0.361, Max 0.624.
  - Future old-age dependency ratio: Mean 0.336, Std. Dev. 0.142, Min 0.114, Max 0.782.
  - Pension per elderly (in %): Obs. = 775; Mean 25.600, Std. Dev. 9.633, Min 0, Max 59.809.

*Source: IMF working paper content (section 5.4, tables and appendices) as provided in the source PDF.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18264.pdf_