Master 2 Macro I. Lecture 8 : Empirical studies of convergence

Transcription

1 Master 2 Macro I Lecture 8 : Empirical studies of convergence Franck Portier (based on Gilles Saint-Paul lecture notes) franck.portier@tse-fr.eu Toulouse School of Economics Version /10/ / 38

2 Disclaimer These are the slides I am using in class. They are not self-contained, do not always constitute original material and do contain some cut and paste pieces from various sources that I am not always explicitly referring to (not on purpose but because it takes time). Therefore, they are not intended to be used outside of the course or to be distributed. Thank you for signalling me typos or mistakes at franck.portier@tse-fr.eu. 2 / 38

3 1. Facts The gap between rich and poor countries is large It is rather persistent Standard neo-classical models would predict rather rapid convergence 3 / 38

4 1. Facts Productivity growth seems to accelerate over time GDP Per Capita (1990 Int'l $) Western Europe Western Offshoots Asia Latin America Africa Eastern Europe Figure 1: The Evolution of Regional Income Per Capita, CE (Source: Maddison, 2003) The forces that generated the remarkable escape from the Malthusian epoch and 4 / 38

5 1. Facts The gap between rich and poor countries is huge 5 / 38

6 1. Facts The world s countries shaped with area in proportion to the gross domestic product adjusted for purchasing power parity (2002 data) 6 / 38

7 1. Facts Can we explain it? Let s take the Solow model (constant and exogenous saving rate s) Y t = A t Kt α K t = sy t δk t Along a BGP : Kt g+δ ( ) 1 K t = Ats 1 α g+δ Y t = A Y Y = α s 1 α s Y t = s 1 1 α t ( ) α s 1 α g+δ 7 / 38

8 1. Facts It does not add up Common estimates suggest α = 1/3 Therefore, to have a 10-fold difference in GDP, we need a 20-fold difference in savings rate (2 % vs. 40 %) More leeway if technology differed across countries, but unlikely if technology is transferable So what do we do? 8 / 38

9 1. Facts If α were greater? Growth would take more time to fall to zero Convergence would be slower (The speed of convergence is the coefficient of gdp growth on (local) initial log gdp) From Lecture 7, we know that the speed of convergence is v = (g + δ)(1 α) Income differences between countries would be magnified 9 / 38

10 1. Facts An extreme case : α = 1, g = 0 This is the AK model The speed of convergence goes to zero The convergence path becomes a balanced growth path at a constant rate MPK is no longer falling. Therefore, capital accumulation can sustain long-run growth The growth rate is now endogenous and depends on preferences (in a Ramsey-type model) 10 / 38

11 2. Convergence Convergence in neo-classical models Neo-Classical models : each country converges to its own steady state All own steady states grow at the same rate But the level depend on policies, savings rates, etc Therefore, similar countries converge to same GDP per capita 11 / 38

12 2. Convergence Convergence in endogenous growth models A laggard never closes the gap Therefore, no convergence in income levels This because MPK is no higher for the laggard Furthermore, differences in policies affect the long-run growth rate 12 / 38

13 2. Convergence Looking at convergence allows us to : Test the relevance of endogenous growth models Assess the magnitude of the returns to accumulable factors Recall : v = (g + δ)(1 α) 13 / 38

14 2. Convergence Two approaches Barro and Sala-i-Martin : take a data set of similar economic units and look at convergence between them in per capita GDP Mankiw-Romer-Weil : take a cross-country regression of growth rates on initial income controlling for own long-run steady state 14 / 38

15 3. Barro and Sala-i-Martin Estimated equation They use a data-base of U.S. states over a long-run period They estimate the equivalent of our local speed of convergence regression ( ) ( ) 1 T log yi,t0 +T 1 e βt = B log(y i,t0 )+u i,t0,t y i,t0 T 0 +T 15 / 38

16 3. Barro and Sala-i-Martin Results TABLE 1 CROSS-STATE REGRESSIONS FOR PERSONAL INCOME Sectoral Composition Sample (Sit) R v (.0046) (.0040) (.0063) (.0074) (.0051) (.0053) (.0052) (.0043) (.0069) (.0114) 11. Nine periods, ,b restricted* (.0019) 16 / 38

17 (.0043) 2. Convergence (.0069) Results 10. (continued) (.0114) 11. Nine periods, ,b restricted* (.0019) (.0048) (.0079) (.0075) (.0094) (.0112) (.0175) (.0048) (.81) (.0055) (.57) (.0055) (.09) (.0052) (.25) (.0076) (.37) (.0106) (.22) 21. Nine periods,.0249 individual...,b restricted* (.0021) (.0067) NOTE.-Standard errors of coefficients are shown in parentheses. Regression 22 has 29 observations, regressions 1 and 2 have 47 observations (excluding Oklahoma), and regression 12 has 46 observations (excluding Oklahoma 17 / 38

18 3. Barro and Sala-i-Martin Results CONVERGENCE (continued) FL VA Ui _ ~~~~~~~~A~ 0.02 WV ~~~~~~~~~~~MD C!,~~~~~~~~~~~~~~~~~~~~~~C O 0.01 CL _AZ\ LOG(1880 PER CAPITA PERSONAL INCOME) FIG. 1.-Growth rate from 1880 to 1988 vs per capita income 18 / 38

19 3. Barro and Sala-i-Martin The BSM Universal Law of Convergence The speed of convergence is 2 % per year 19 / 38

20 3. Barro and Sala-i-Martin Solow model prediction What do we get in a Solow model? The Solow model predicts a speed of convergence v equal to (δ + g)(1 α) A reasonable calibration is δ = 0.06, g = 0.02, α = 0.3 This gives v = 5.6 % per year 20 / 38

21 3. Barro and Sala-i-Martin How universal is the law? TABLE 3 COMPARISON OF REGRESSIONS ACROSS COUNTRIES AND U.S. STATES Additional Sample Variables R countries, no (.0018) countries,.0184 yes (.0045) OECD countries,.0095 no (.0028) OECD countries,.0203 yes (.0068) U.S. states,.0218 no (.0053) U.S. states,.0236 yes (.0013) NOTE.-The dependent variable in regressions 1-4 is the growth rate of real per capita GDP from 1960 to 1985; in regressions 5 and 6 it is the growth rate of real per capita GSP (the variable used in table 2) from 1963 to The coefficient P applies in regressions 1-4 to the logarithm of real per capita GDP in 1960, and in regressions 5 and 6 to the logarithm of real per capita GSP in Each regression also includes a constant. The additional variables included in regressions 2 and 4 are the primary and secondary school enrollment rates in 1960, the average ratio of government consumption expenditure (standard figures less spending on defense and education) 21 / 38

22 3. Barro and Sala-i-Martin How universal is the law? (continued) LO 33 OD CD 0) I ( O 0.0~~~~~~~ W ~51 70,,, ~~~~~~~~~~~ 7 7#459 8H32 < (!) 8196 a < 0~~~~~~~~~~~58 C, l1l9 W LOG(1960 PER CAPITA GDP) FIG. 4.-Growth rate from 1960 to 1985 vs per capita GDP, sample of 98 countries (listed in App. B) / 38

23 3. Barro and Sala-i-Martin Findings The more similar the countries, the more it holds unconditionally The less similar the countries, the more likely we find divergence But the law is restored if controls are added, controlling for own steady state 23 / 38

24 3. Barro and Sala-i-Martin Implication for how to eradicate poverty? 1. Adopt the policies and institutions of advanced countries 2. Wait! How long? Suppose I am 10 times poorer than the US. How long does it take to be 2( times ) poorer? ( ) We need to solve d dt log Y (t) Y US (t) = β log Y (t) Y US (t) which implies log Y (t) log Y US (t) = (log(y ( (0) ) = log(y US (0)) e βt and therefore t = 1 β log log ρ1 log ρ 0 24 / 38

25 3. Barro and Sala-i-Martin Wait... For how long? With v = 0.02, ρ 0 = 0.1, ρ 1 = 0.5, we obtain t = 60 years! With v = 0.056,get instead t = 21 years. We want to understand why the speed of convergence is so low Can policy increase the speed of convergence? In principle, the speed of convergence only depends on the deep technological parameters That it is low tells us that the technology is not what we thought it was But it does not tell us we can increase v A gloomy perspective. 25 / 38

26 4. Mankiw, Romer and Weil Approach National accounts suggest that the elasticity to capital α is 0.3 Observed speeds of convergence suggest α = 1 v/(g + δ) = /0.08 = 0.75 To reconcile these two facts, MRW introduce another form of capital : Human capital 26 / 38

27 4. Mankiw, Romer and Weil The Augmented Solow model Y (t) = K(t) α H(t) β (A(t)L(t)) 1 α β K(t) = sk Y (t) δk(t) Ḣ(t) = s h Y (t) δh(t) Denoting x(t) = X (t) A(t)L(t), we have k(t) = sk y(t) (n + g + δ)k(t) ḣ(t) = sh y(t) (n + g + δ)k(t) 27 / 38

28 4. Mankiw, Romer and Weil BGP Along a BGP, deflated variables are at a Steady State : ( ) k s = 1 β k s β 1 1 α β h ( h = n+g+δ s α k s1 α h n+g+δ ) 1 1 α β 28 / 38

29 4. Mankiw, Romer and Weil Explaining cross-country differenced in per capita GDP : The preceding equations define own steady state MRW use it to see if it explains cross-country income differences : ( ) Y (t) log L(t) = log A(0) + g t α + β log(n + g + δ) 1 α β + α 1 α β log s β k + 1 α β log s h 29 / 38

30 4. Mankiw, Romer and Weil Measuring s h 30 / 38

31 4. Mankiw, Romer and Weil Results 31 / 38

32 4. Mankiw, Romer and Weil What have we learned? We have seen previously that with α = 0.3, it is difficult to explain country income differences But now what matters is α + β, which acts as α. So with α + β large enough we can explain cross-country differences. A natural question is : can we also expect slow convergence? 32 / 38

33 4. Mankiw, Romer and Weil Recomputing the speed of convergence Let x = X X LR One must have s k Y LR K LR = n + g + δ Define ŷ = log y, then v = ŷ ŷ. ẏ y = α k k + β ḣ h k k = K K (n + g) = s k Y K (δ + n + g) Y = s LR k K LR (1 ŷ k) (δ + n + g) = (δ + n + g)(ŷ k) Similarly, ḣ h = (δ + n + g)(ŷ ĥ) Therefore, ẏ y = α(δ + n + g)(ŷ k) + β(δ + n + g)(ŷ ĥ) = (δ + n + g)(1 α β)ŷ 33 / 38

34 4. Mankiw, Romer and Weil Empirical strategy Investment rates and schooling are kept to proxy for own steady state Initial output is added Coefficient in initial output related to SOV as in BSM No other control variable is added in strict interpretation of Solow model 34 / 38

35 4. Mankiw, Romer and Weil Old Solow does not work / 38

36 4. Mankiw, Romer and Weil... but new does. 36 / 38

37 4. Mankiw, Romer and Weil Does it add up? α =.3 β =.3 n + g + δ =.06 Therefore v = / 38

38 5. Summary The Solow model predicts too low income disparities and too quick convergence The AK model predicts zero convergence and widening disparities The Augmented Solow model does well to predict both the disparities and the speed of convergence 38 / 38