Fertility, Longevity, and Capital Flows

Transcription

1 Fertility, Longevity, and Capital Flows Zsófia Bárány (SciencesPo) Nicolas Coeurdacier (SciencesPo & CEPR) Stéphane Guibaud (SciencesPo) International Macro Finance Conference Chicago Booth, December / 50

2 Aging across the world The world is aging. Drop in fertility. Falling mortality rates, rising life expectancy. But there is large variation in aging across countries. Fertility started to fall decades ago in developed countries, while the drop happened more recently in other countries. In some developing countries, fertility (adjusted for infant mortality) actually increased. Mortality rates fell (almost) everywhere but at very different paces across countries. 2 / 50

3 Fertility across the world in 1970 Fertility rate adjusted for infant mortality 7 Africa Latin America Emerging Asia Central & Eastern Europe Old Europe & Japan Anglo-saxon countries USA GBR AUS JPN SWE FRA DEU POL LVA CZE TUR THA KOR IDN CHN URY PER HND BRA ARG MAR ZAF ZMB MLI MDG BDI 3 / 50

4 Fertility across the world in 1990 Fertility rate adjusted for infant mortality 7 Africa Latin America Emerging Asia Central & Eastern Europe Old Europe & Japan Anglo-saxon countries USA GBR AUS JPN SWE FRA DEU POL LVA CZE TUR THA KOR IDN CHN URY PER HND BRA ARG MAR ZAF ZMB MLI MDG BDI 4 / 50

5 Fertility across the world in 2010 Fertility rate adjusted for infant mortality 7 Africa Latin America Emerging Asia Central & Eastern Europe Old Europe & Japan Anglo-saxon countries USA GBR AUS JPN SWE FRA DEU POL LVA CZE TUR THA KOR IDN CHN URY PER HND BRA ARG MAR ZAF ZMB MLI MDG BDI 5 / 50

6 Mortality across the world in 1970 Probability of dying before 65 conditional on being 25 60% Africa Latin America Emerging Asia Central & Eastern Europe Old Europe & Japan Anglo-saxon countries 50% 40% 30% 20% 10% 0% USA GBR AUS JPN SWE FRA DEU POL LVA CZE TUR THA KOR IDN CHN URY PER HND BRA ARG MAR ZAF ZMB MLI MDG BDI 6 / 50

7 Mortality across the world in 1990 Probability of dying before 65 conditional on being 25 60% Africa Latin America Emerging Asia Central & Eastern Europe Old Europe & Japan Anglo-saxon countries 50% 40% 30% 20% 10% 0% USA GBR AUS JPN SWE FRA DEU POL LVA CZE TUR THA KOR IDN CHN URY PER HND BRA ARG MAR ZAF ZMB MLI MDG BDI 7 / 50

8 Mortality across the world in 2010 Probability of dying before 65 conditional on being 25 60% Africa Latin America Emerging Asia Central & Eastern Europe Old Europe & Japan Anglo-saxon countries 50% 40% 30% 20% 10% 0% USA GBR AUS JPN SWE FRA DEU POL LVA CZE TUR THA KOR IDN CHN URY PER HND BRA ARG MAR ZAF ZMB MLI MDG BDI 8 / 50

9 Capital flows across across the world: Average current account over GDP 12% 9% Africa Latin America Emerging Asia Central & Eastern Europe Old Europe & Japan Anglo-saxon countries 6% 3% 0% -3% -6% -9% -12% -15% -18% USA GBR AUS JPN SWE FRA DEU POL LVA CZE TUR THA KOR IDN CHN URY PER HND BRA ARG MAR ZAF ZMB MLI MDG BDI 9 / 50

10 Capital flows across across the world: Average current account over GDP 12% 9% Africa Latin America Emerging Asia Central & Eastern Europe Old Europe & Japan Anglo-saxon countries 6% 3% 0% -3% -6% -9% -12% -15% -18% USA GBR AUS JPN SWE FRA DEU POL LVA CZE TUR THA KOR IDN CHN URY PER HND BRA ARG MAR ZAF ZMB MLI MDG BDI 10 / 50

11 World aging and capital flows World aging depresses the world interest rate. If countries age the same way and are otherwise identical, world aging should leave capital flows unaffected. But if countries are heterogenous and respond differently to the fall in interest rate, world aging can affect capital flows. The response of savings in a given country depends on the relative strength of income vs. substitution & wealth effects. The strength of wealth effect depends on access to credit and social security (which affects shape of age-income profile). 11 / 50

12 Idiosyncratic aging and capital flows Differences in aging patterns across countries can generate international capital flows. Countries with faster aging prospects should turn into capital exporters. A drop in fertility lowers the marginal productivity of capital and investment; it also increases aggregate savings, by increasing the share of high propensity-to-save households. A rise in life expectancy raises aggregate savings. The strength of the impact of aging on aggregate savings depends on the saving profile by age (composition effect). The shape of the profile is affected by the young s ability to borrow and the extent of social security. 12 / 50

13 Aging and the puzzling patterns of capital flows Uphill capital flows. In response to the fall in interest rate caused by world aging, the response of savings depends on the level of development. The savings of less developed countries should increase more. The allocation puzzle among emerging markets. Those that are growing fast are also the ones aging faster. 13 / 50

14 Outline 1. Theory: three-period OLG model with multiple countries and integrated capital markets, incorporating heterogeneity across countries in growth and aging prospects; other dimensions of heterogeneity driving cross-country differences in savings response to world/idiosyncratic components of growth and demographic evolutions. Useful framework to elucidate the main mechanisms. Derive (approximate) closed-form expressions for capital flows that can be tested against the data. 14 / 50

15 Outline (cont d) 2. Empirical test of the model s cross-sectional implications. Data on capital flows, demographics & growth over for large cross section of countries, developed and developing. Finding: idiosyncratic expected aging is a strong predictor of capital flows. The effect is quantitatively significant. 3. Quantitative version of multi-country OLG model. Calibrated to cross-country demographics, growth, household debt (and social security) data. Simulate the world economy and compare outcomes to data (cross-country/time series); run counterfactual experiments. 15 / 50

16 Related literature Demographics and capital flows. Obstfeld and Rogoff (1996), Brooks (2003), Domeij and Floden (2006), Ferrero (2010), Backus, Cooley and Henriksen (2014) Attanasio, Kitao, Violante (2007), Krueger and Ludwig (2007), Borsch-Supan, Ludwig, Winter (2006) Global imbalances and the allocation puzzle. Bacchetta and Benhima (2013), Benhima (2013), Caballero, Farhi and Gourinchas (2008), Carroll and Jeanne (2009), Gourinchas and Jeanne (2011), Gourinchas and Rey (2013), Mendoza, Quadrini and Rios-Rull (2009), Sandri (2010), Song, Storesletten and Zilibotti (2011), Coeurdacier, Guibaud and Jin (2014) Empirical literature on determinants of current account (or aggregate savings) across countries. Higgins (1998), Chinn and Prasad (2003). See meta-analysis by Cazorzi et al. (2012) for references Bloom, Canning, Mansfield and Moore (2003) 16 / 50

17 Basic model Key ingredients Multi-country OLG model with log utility full depreciation of capital. Agents live for at most three periods (y, m, o). Young: do not work, face credit constraints. Middle aged: work and save for retirement. Old: consume out of accumulated assets, no bequest. Mortality: a middle-aged individual in period t in country i reaches retirement with probability p i t. Fertility: L i y,t+1 = ni tl i y,t. 17 / 50

18 Production Output where productivity evolves as Y i t = ( K i t ) α (A i t L i m,t) 1 α, A i t+1 = (1 + γ i A,t+1 )Ai t. Wage rate wt i = (1 α)a i t(kt) i α, with kt i Kt i /(A i tl i m,t). Rate of return between periods t and t + 1 Rt+1 i = α(kt+1) i α / 50

19 The household s problem An agent born in country i in period t maximizes U i t = log(c i y,t) + β log(c i m,t+1) + p i t+1β 2 log(c i o,t+2), subject to budget constraints c i y,t + a i y,t = 0, cm,t+1 i + am,t+1 i = wt+1 i + Rt+1a i y,t, i ( and credit constraint c i o,t+2 = R i t+2 am,t+1 i + 1 pi t+1 pt+1 i am,t+1 i ), ay,t i θ i w t+1 i Rt+1 i. 19 / 50

20 Savings decisions Assume that the credit constraint is always binding ay,t i = θ i w t+1 i Rt+1 i. The optimal savings of the middle aged is then a i m,t = βpi t 1 + βp i t ( 1 θ i ) w i t. Role of θ i and p i t. Role of n i t 1 (composition effect). 20 / 50

21 Capital market equilibrium Under financial autarky K i t+1 = L i y,ta i y,t + L i m,ta i m,t. Under financial integration ( L i y,t ay,t i + L i m,tam,t i ). i K i t+1 = i Financial integration in period t implies R i t+1 = R t+1 for all i. 21 / 50

22 Autarky interest rate Steady state interest rate under autarky R i = n i (1 + γa i ) 1 + βp i ( ) α βp i (1 θ i ) 1 α + θi. Role of credit constraints (θ i ). Impact of aging: n i and/or p i R i. 22 / 50

23 NFA under integration Country s net foreign assets NFA i t = L i y,ta i y,t + L i m,ta i m,t K i t+1. Under financial integration NFA i t Y i t = βpi t (1 α) ( 1 θ i) 1 + βp i t ni t 1 (1 + γi A,t+1 ) R t+1 [ α + (1 α)θ i ] ( kt+1 k t ) α. 23 / 50

24 NFA under integration Special case: n i t = n t, p i t = p t, γ i A,t = γ A,t. Suppose all countries have identical demographic and growth evolutions, and only differ in terms of θ. Then NFA i t Y i t = βp t (1 α) ( 1 θ ) [ 1 θ i 1 + βp t 1 θ ] α + (1 α)θi α + (1 α) θ where θ i ω i θ i, ω i A i,tl i m,t j A j,tl j. m,t Corollary 1: NFA > 0 if and only if θ i < θ, i.e., high-θ countries are debtors, while low-θ countries are creditors. Corollary 2: An increase in life expectancy common across all countries causes further divergence in NFA/GDP. 24 / 50

25 NFA under integration Impact of idiosyncratic aging: comparative statics for small open economy NFA i t Y i t = βpi t (1 α) ( 1 θ i) 1 + βp i t ni t 1 (1 + γi A,t+1 ) R t+1 [ α + (1 α)θ i ] ( kt+1 k t ) α. Partial equilibrium effects of an increase in p i t a fall in n i t 1 fast growth γ i A,t / 50

26 From NFA to CA By definition, CA i t = NFA i t NFA i t 1. Hence CA i t/y i t depends on the change in survival prob of middle-aged p i t p i t 1 the change in fertility n i t 1 nt 2 i growth acceleration through period t, γ i A,t+1 = γa,t+1 i γi A,t. Define the old dependency ratio in period t o i t Li o,t L i m,t = pi t 1 Li m,t 1 L i y,t 1 = pi t 1 nt 2 i. The demographic evolutions that matter to determine CA i t/yt i are captured by one variable: the growth rate γo,t+1 i of the old dependency ratio between periods t and t / 50

27 CA/GDP - approximate formulas Suppose all countries have similar levels of financial development, i.e., θ i = θ for all i. Then CA i t Y i t µ t + λ t ( γ i o,t+1 γ i A,t+1 ) + ν i NFA i t 1 t Yt 1 i. Otherwise CA i t Y i t µ t (θ i ) + λ t (θ i ) ( γ i o,t+1 γ i A,t+1 ) + ν i NFA i t 1 t Yt 1 i, with λ t increasing in θ i. 27 / 50

28 Data Sample of 146 countries, over the period Capital flows: current account over GDP (World Bank, IMF). Average CA/GDP for each country over 20y (or 5y) period. Demographics: use UN data on demographic composition by age (incl projection until 2040) to construct aging prospects. γ i o,t+1 defined as the expected change in old dependency ratio between current period t and 25y forward, annualized. Growth: use data on GDP per worker in constant USD (PPP adjusted) from PWT to construct growth acceleration. γ i A,t+1 defined as the difference in the annual growth rate of GDP/worker over current period t and past 20y period. 28 / 50

29 Data Access to credit: cannot observe θ i. Graph Proxy with household debt over GDP for OECD countries. For developing countries, proxy with predicted household debt over GDP in 2010, using share of workers having a mortgage, and share of workers having a credit card (World Bank data). Standard controls: Oil dummies Regional (and time) dummies Pre-1990 conditions: NFA/GDP (Lane and Milesi-Feretti), initial GDP per worker Trade openness Capital account openness from Chinn and Ito. 29 / 50

30 Cross-sectional regression Start with average current account over : CA i Y i = µ + λ o γ i o + λ A γ i A + νz i + ɛ i. Model predicts λ o > 0 and λ A < / 50

31 Current account: cross-sectional regression ( ) (1) (2) (3) (4) (5) γo i (.296) (.278) γ i A (.165) γ i o γ i A HDebt GDP (.143) (.192) (.206) (.019) (.045) ( NFA GDP ) past (.009) (.010) Sample All All All All Emerging Controls No No No Yes Yes Obs R Notes: Data source: PWT, World Bank, IFS, Chinn and Ito. Oil dummy always included. Additional Controls include, initial GDP per worker, capital account and trade openness. *** p<0.01, ** p<0.05, * p< / 50

32 Expected aging and capital flows ExpectedAgeing CurrentAccount_GDP Fitted values Notes: variable γo i is demeaned. Sample mean 1.3%. 32 / 50

33 Expected aging and capital flows Adjusting for growth acceleration ExpectedAgeing_Adjusted CurrentAccount_GDP Fitted values 33 / 50

34 Panel regressions Using average CA/GDP for each country over 5-year periods: CA i t Y i t = µ t + λ ( γ i o,t+1 γ i A,t+1) + νz i t + ɛ i t. 34 / 50

35 Panel regressions γ i,t o γ i,t A γ i,t o (1) (2) (3) (4) (5) (6) (.364) (.341) γ i,t A (.224) (.292) (.202) (.268) (.084) HDebt GDP (.024) (.020) ( NFA GDP ) (.009) (.008) Est. Pool Pool Pool BE BE Within Controls No No Yes No Yes Yes Obs R Notes: Data source: PWT, World Bank, IFS, Chinn and Ito. Oil and time dummies always included. Additional Controls include, initial NFA/GDP and GDP per worker, capital account and trade openness. *** p<0.01, ** p<0.05, * p<0.1. For pool regressions, obs. clustered by countries. 35 / 50

36 The role of household debt Use average CA/GDP over 5-year periods. Classify countries into three groups according to level of household debt over GDP, h = 1,..., 3. ( ) CA ( = µ t + µ h + λ h γ i Y o,t+1 γa,t+1) i + νz i t + ɛ i t. i,h,t 36 / 50

37 The role of household debt γ i,t o γ i,t o γ i,t o (1) (2) (3) (4) γ i,t A (h = 1) (.169) (.146) (.143) (.140) γ i,t A (h = 2) (.368) (.372) (.300) (.391) γ i,t A (h = 3) (.314) (.315) (.242) (.232) ( NFA GDP ) (.009) (.009) Sample All All Emerging Emerging Controls No Yes No Yes Obs R Notes: Data source: PWT, World Bank, IFS, Chinn and Ito. Oil and time dummies always included. Additional Controls include, initial NFA/GDP and GDP per worker, capital account and trade openness. *** p<0.01, ** p<0.05, * p<0.1. Obs. clustered by countries. 37 / 50

38 Quantitative OLG model Agents live for at most J + 1 periods. Age j = 0,..., J. Conditional survival probability pj,t i. Lifetime utility of agent born in period t in country i U i t = J j=0 with isoelastic preferences ( j 1 pl,t+l i l=0 ) u(c) = c1 1 σ σ β j u(c i j,t+j), 38 / 50

39 Quantitative model Output Y i t = (K i t ) α J A i t ej,tl i i j,t j=0 1 α. Labor income w i j,t = e i j,t(1 α)a i t(k i t) α, k i t K i t A i t J j=0 ei j,t Li j,t. Gross rate of return R i t+1 = 1 δ + α(k i t) α / 50

40 Quantitative model Credit constraints aj,t+j i θ i pi j,t+j Hi j+1,t+j+1 Rt+j+1 i, with H i j,t w i ( τ 1 J j s=0 pi j+s,t+s j,t + τ τ=1 s=1 Ri t+s ) w i j+τ,t+τ. 40 / 50

41 Quantitative model Unintentional bequests J 1 Qt i (1 pj,t)l i i j,taj,t. i j=0 Redistributed to the surviving agents as lump sum transfers. 41 / 50

42 Illustrative experiments: Calibration Adult life lasts for at most 7 periods (one period = 10 years). Standard preferences/technology parameters. β = 0.98 (annual), 1/σ = 2, α = 0.3, δ = (annual). Calibrate demographics & growth on 4 exemplary economies: US, China, Germany/Japan, Zambia. Calibrate θs on household debt/gdp. θ US = 0, θ CH = 0.025, θ GJ = 0.10, θ ZB = Shape of age-income profiles (ej,t i ) calibrated on the US. 42 / 50

43 Illustrative experiments Two countries, heterogenous θs, integration at t = Old dependency ratio US Asia Rate of return S/Y 0.1 CA/Y Log productivity 0.8 GDP shares 0.3 I/Y NFA/Y / 50

44 Illustrative experiments Two countries, integration, common aging Old dependency ratio US Asia Rate of return S/Y 0.1 CA/Y Log productivity 0.8 GDP shares 0.3 I/Y NFA/Y / 50

45 Illustrative experiments Two countries, integration, differences in aging Old dependency ratio US Asia Rate of return S/Y 0.1 CA/Y Log productivity 0.8 GDP shares 0.3 I/Y NFA/Y / 50

46 Illustrative experiments Two countries, integration, differences in aging & growth Old dependency ratio US Asia Rate of return S/Y 0.1 CA/Y Log productivity 0.8 GDP shares 0.3 I/Y NFA/Y / 50

47 Illustrative experiments Third country: fast aging, decelerating growth, intermediate θ Old dependency ratio US Asia Ger/Jap Rate of return S/Y CA/Y Log productivity 0.5 GDP shares I/Y NFA/Y / 50

48 Illustrative experiments Fourth country: no aging, low growth, low θ Old dependency ratio US Asia Ger/Jap Afr Rate of return S/Y CA/Y Log productivity 0.5 GDP shares I/Y NFA/Y / 50

49 Conclusion Large variations across countries in timing and pace of aging. Lifecycle model predicts capital flows depend on expected aging and growth acceleration. Strong empirical support in large cross section of countries. Quantitative performance: to be assessed. Interactions between demographics, growth and level of development seem to have the potential to account for patterns of capital flows. Next step: incorporating social security. 49 / 50

50 THANK YOU. 50 / 50

51 Household debt over GDP back 0% 20% 40% 60% 80% 100% 120% 140% 160% SEN BFA BDI MLI MDG IDN SLE YEM UGA MDA LSO ARM CMR LAO ZMB TGO RWA BGD HND SLV VNM MNG TUN PER PHL LKA DOM NPL COL MAR SYR DZA ZAF CRI KAZ CHN MUS MKD LBN TUR TZA CIV BLR PNG SVK SAU BGR LTU CHL POL LVA DMA HUN BLZ HRV HKG SGP MDV FRA KWT DEU GRC SWE NOR USA PRT NZL AUS IRL CYP 51 / 50

52 Household debt over GDP back PredictedHHdebt hhdebtgdp Fitted values 52 / 50