Growth Regressions 2014 Jean-Bernard CHATELAIN Université Paris I Panthéon Sorbonne Paris School...
-
Upload
bartholomew-horton -
Category
Documents
-
view
218 -
download
0
Transcript of Growth Regressions 2014 Jean-Bernard CHATELAIN Université Paris I Panthéon Sorbonne Paris School...
Growth Regressions2014
Jean-Bernard CHATELAIN
Université Paris I Panthéon Sorbonne
Paris School of Economics
Growth regressions Plan Courses 1 - 3
A1. Data and convergence
A2. Reinhart and Rogoff example
A2. Burnside and Dollar: panel data, outliers
A3. Other statistical issues: inference, instrumental variables, and so on.
A1. Growth regressions
1. The dependent variable: cross country growth of GDP per capita: data issues.
2. Descriptive statistics
3. Bivariate between regressions
4. List of regressors
5. Multi-factor explanations, endogeneity including reverse causality, outliers, non-linearity with poverty trap.
6. Reinhart and Rogoff: Growth GDP, public debt
A2. Burnside and Dollar (2000) replication
1. Burnside and Dollar (2000) paper.
2. Data and specification
3. Spurious regressions and outliers
4. Panel data estimators
5. Instrumental variables estimators
A3. Statistical issues
1. Inference: statistical versus substantive significance
2. Publication bias
3. Multiple testing
4. Power: minimal number N of observations
5. Maximal number k of regressors, contributions to R2.
A3-bis. Statistical issues
1. Omitted variable bias / Spurious regressions and near multicolinearity: Outliers detection, Robust estimates, Graphs; overfitting. Quadratic and interaction terms, spurious and/or unstable effects.
2. Panel data: Within versus Between: time trends versus endogeneity, time invariant variables in panel data.
3. Instrumental variables
4. Instrumental variables with GMM using panel data.
A1. Growth regressions: General case
1. Motivation
2. Measurement issues
3. GDP/head descriptive statistics
4. Growth of GDP/head descriptive statistics
1. Motivation: Convergence?
Solow, Ramsey-Cass-Koopmans
Predict convergence with decreasing returns to scale aggregate macro production function:
Low Y/L imply high growth of Y/L
Convergence, growth econometrics
are no longer
what they used to be,
as in this 2003 textbook.
Motivation: Catch up and convergence: recent trends
(Subramanian Kessler: hyperglobalization):
1960-2000: only 29.2% of developing countries Y/L grew more than the USA (+1.53% a year on average).
2000-2011: 73% to 90% did it (with +3% a year on average)
Most impressive: China India (43% of world population) Brazil Russia : (BRIC) growth.
The past of convergence:2000 versus 1960 (in textbooks)Convergence: no evidence:
A group of poor countries below the 45 degree line did not grew more than the USA, large country leader of GDP/head
(excluding Luxembourg offshore financial center with highest GDP/head).
Measurement Problems of GDP data
2013
The Wealth of Nations GDP/Land its Growth
GDP/Head: Alternative measures: Happiness? Consumption/head? Health indicator/head? green economy?
Measurement error: Hidden economy.
Inequality of income inside a large, a small country: still many poor people in wealthy economies.
Inequality between around 190 countries of various population size: China, Iceland, St Kitts and Nevis.
PPP: purchasing power parity in US dollar given year.
Penn World tables (Website, PWT8.0 new version august 2013), WDI, IMF, OECD
http://www.rug.nl/research/ggdc/data/penn-world-table
Historical cross country data sources before 1960: Angus Maddison project (Website, link), break on measurement errors which increases before 1960.
Measurement errors: hidden economy.
World Development Indicators (WDI) 2012 CD-ROM Penn World Tables 7.1 (PWT)
and International Comparison Program (ICP)
“Huge differences are found between the two sources for numerous countries in both the current and the last versions. The number of countries for which WDI and the ICP benchmark numbers show huge differences is small, but there are many countries for which PWT and the ICP benchmark numbers show large differences.”
Ram and Ural, Social Indicators Research, march 2013.
GDP recent boundaries changes: US, Australia, Canada
Kuznets report (1934), Stone SNA (1947)
System of National Accounts 2008:
Investment: « intellectual property products » measured by firms and government innovation related costs and expected royalties on original artwork
by the US Bureau of Economic Analysis (BEA) [+3.6% of US GDP in august 2013].
Other countries should join by 2014.
Unchanged boundary: Services consumed at home
Cleaning a home, caring for a relative.
Market price for these activities do exist.
Remark: National accounts are revised up to T-3 years by statisticians. Latest data not stable.
Other measures than Y/L:Cross section simple correlationConsumption
Life expectancy
Happiness
Green sustainable ressources wealth
Growth versus CyclesAveraging the Dependent variable
Averages over arbitrary 5, 6,…, 10 years.
Trend versus cycles using filters (example Hodrick Prescott).
Interaction between cycles of GDP/head and the growth trend, long term effects of crisis?
Researchers endogenous sample selection: data availibility (1960s) varies for regressors.
3. Descriptive statistics on GDP/Head
Google: Gapminder software
Cross section
Size of the country per population
Continent.
Time series:
Pre-industrial
1st and 2nd industrial revolution
1960’s to 2000s.
Time series of GDP/head (Oded Galor)
History: Malthusian (population growth and fluctuations) then modern regime
1. Neolithic revolution (G. Childe): -7000 to -3000
2. Empires and nations
3. Going backwards: early middle ages, black plague (trade) 1350, fall of population.
4. 1750 First industrial revolution, UK
5. 1880 2nd industrial revolution and first trade and financial globalization
6. 1930-40 going backwards
7. Stability then 2nd globalization 1970s.
Complementary Ingredients
Population: health, diseases, culture, knowledge transmission, slavery.
Natural resources (nature’s capital stock) and their fluctuations with climate changes.
Technology: innovations, blocked or not, unintended consequences.
Coordination: predation, wars, empires (pax romana and trade), colonies, law and property rights, trust, trade, cities and agglomerations, institutions, religion, culture.
2 regimes
Malthusian
Demographic transition to current period.
« unified growth theory » (Galor).
Genetic diversity and output per head?
http://www.nature.com/news/economics-and-genetics-meet-in-uneasy-union-1.11565
Ashraf and Galor (2012), « The out of africa hypothesis, human genetic diversity and comparative economic development. » American Economic Review.
Distribution of cross sections of GDP/head (given year) [Kernel estimates]:
GDP/head: skewed.
Log(GPD/head): less skewed.
Weighted by population (China, India) of Log(GDP/head).
GDP/worker (labour force, less young and retired): productivity.
Log-normal: income distributionGDP/head
Countries observations weigthed by population
China and India, two points (countries) corresponds to 43% of the weights. The histogram is smoother than with equal weights for each countries.
Reminder: Within these large countries inequality of regional or personal income is not taken into account.
3. Descriptive statistics on the growth of GDP/Head:
growth of GDPminus growth of population
peaked triangular distributionper group of countries
The dependent variable:Growth of GDP/head ppp ajusted
Growth of real output
(growth of nominal output less growth of GDP deflator) less growth of population
1. Cross section triangular or Laplace distribution
Cf. micro level growth of firms output (Bottazi-Secchi), of individuals wealth, of animals size,…
2. Skewed, Twin peaks? Mixture of distributions.
Laplace double – exponentialdistribution
When the dependent variable is not following a normal distribution
Residuals of the multiple regression are likely not to follow a normal distribution.
Estimated standard errors of the estimated parameters may be biased.
Explaining growth
Many causal factors: up to 500 indicators for 50 effects explaining growth (some of the indicators intend to measure the same effect).
Reverse causality: endogeneity, except for geography and far in the past.
Outliers.
Poverty traps: thresholds, non linear effects.
From the country monograph (Bostwana) to general effects and policy?
« Between » simple correlations« average over time of cross sections »
i=country; t=year
Average over time of variables x(it)
denoted x(i.):
Corr ( y(i.), x(i.) )
If one variable is time invariant z(i) (GDP/head in 1960):
Corr ( y(i.), z(i) )
« Between » simple correlations« average over time of cross sections »
We will see later that between correlations:
Corr ( y(i.), x(i.) )
may be biased because of correlations of x(i.) with country specific (time invariant) random country effects
Perhaps 3000 published papers on growth regressions: anything robust at all?
Multiple testing on regressors
Panel data econometrics: within versus between
Endogeneity, weak instrumental variables
Multiple testing on instruments
Outliers and spurious regressions: Aid and Growth.
Meta-analysis
Multiple Regression in statistics
Well designed for multinormal variables.
Up to 4 or 5 explanatory variables leading to R2>0.5.
At least simple correlations with dependent at least over 0.1.
Orthogonal regressors: simple correlation close to zero between regressors (for interpretation of ceteris paribus effects)
Any data set which differs may lead to odd results
Controversies upon the statistical inference of determinants of growth
Growth (not per head) and Public Debt: 90% threshold (Reinhart and Rogoff) (2010)
Growth and Foreign Aid: Burnside Dollar,
Doucouliagos Paldam meta-analysis, Roodman, Chatelain Ralf.
Growth and Finance? Arcand; Beck; Levine Zervos versus Pollin.
Genetic diversity and GDP/head?