Week 4 - Testing Solow Swan; convergence Flashcards
What are proximate determinants of growth?
There is a direct link from these variables to the growth of income per capita
True or false: Solow Swan’s basic predictions are seen in the data?
True
What is the steady state output per worker along the balanced growth path in the solow-swan model?
y*= A (s/n+g+δ)^a/1-a
How do you take the steady state equation and turn it to something testifiable (a regression equation)? Whats the outcome?
Take logs to make it linear:
lny = lnA + (a/1-a)lns - (a/1-a)ln(n+g+δ)
Add subscript i’s to the variables which are the specific country in the regression.
Which variables are constant for all the countries in the regression equation?
lnA, g, δ
What do you have to remember to add to the end of a regression model?
An error term - everything that affects lny that is not in our model
Putting it all together, what regression does the SS model suggest to run?
lny = B0 + B1lnsi + B2ln(ni+0.05) + ei
assuming g + δ = 0.05
In the regression, why might i estimate b1 = 0.5 and b2 = -0.5?
because a is estimated to be 1/3rd
What are Mankir, Romar and Weil’s esimates of alpha when they carry out their regression
a = 2/3rds not 1/3rd
What does Mankiw, Romer and Weil (1992) conclude about the ss model?
An augmented ss model that incorporates human capital does a very goof job in explaining differences in output per worker.
Do most economists agree or disagree with MR and W?
disagree
What is the consensus about the variations in income per capita?
differences in physical and human capital do not explain most of the variation in income per capita, so there is a large role for TFP
What is the steady state income per capita with skilled human capital H in the cobb douglas PF instead of L?
y = hA(s/n+g+δ)^(a/1-a)
How can we get the gap of income per capita between 2 countries?
Take the ratio: Y(Zambia)/Y(US)
How do I carry out a simple level/development accounting exercise?
Take the ratio: Y(Z)/Y(US)=…
rearrange for A(Z)/A(US)=…
*Try and put everything that we can measure on the RHS
What do we calculate A as?
A residual for TFP
Human capital per capita can’t be measured directly, what can we do?
measure average years of education in each country
What is the equation for human capital Ht? Change into ht
Ht=e^φu(L)T
Divide both sides by L
ht = e^φu
What is u?
average years of education in a country
What is φ?
100 x φ tells us: if u increases by 1 unit, by what percentage does Ht rise:
dlnHt/du = φ
How can we estimate φ?
Assuming that a persons increase in productivity is reflected in higher wages, we can estimate φ from so-called mincerian earnings function
What is the intuition behind this assumption?
A more educated person is more skilled/productive and will produce more output and get paid more as a result
What is the regression for this mincerian earnings function?
lnwage = b0 +b1edu + ei
What is b1 in that function? What can we use this to estimate?
How much more productive is a person with an extra year of education i.e. how much more human capital does this person have. So we can use this as an estimate for φ
How do you get the empirical model for the regression-based approached i.e. MRW with human capital?
Recall the human capital augmented ss steady state for y. Take the logs
In this empirical model, what is lnh and the coefficient for lnh
lnh = φu, therefore lnh = u and the coefficient = φ
Since MRW didn’t have data on ui, what did they do?
used a proxy for the investment in human capital (not the level of human capital)
What does this do to the human capital variable in the empirical model
changes the variable to b4lnschooli
- schooli is something akin to the secondary school enrolment ratio
What does this change do to our empirical model?
Changes the predictions for all the coefficients
What does this do to the coeffcients for population growth and investment?
Now the coefficient for investment will be bigger than 0.5 and the coefficient for pop growth will be smaller than 0.5
What is the main problem with this regression
lns and lnschool are likely to be endogenous i.e. correlated with the error term. This means we get biased OLS estimtators for the coefficients of interest. May lead to MRW to overstate the explanatory power of physical and especially human capital.
What are some other problems with this regression? (3)
- Reverse causality: richer countries investment more in human capital
- Omitted variables that affect both school and y
- Our large and significant coefficients could just be picking up the effect of some other omitted variables that happen to be correlated with the investment in physical and human capital.
What is the broad methodology of the level accounting approach with respect to regression accounting
regression accounting attempts to pin down the contribution of physical and human capital/education, and in particular parameters a and φ, from some additional information
Whats an advantage of this level accounting approach
less chance of endogeneity exaggerating the contribution of physical and human capital
What is a disadvantage?
Need to make extra assumptions to pin down these parameters
What assumption do you have to make for the equation for lny in the ss model without human capital?
Assume that each country is in equilibrium (in its steady state)
True or false: It is likely most countries have reached their balanced growth path?
False
What is the end resulting difference between regression (MRW) and level accounting?
Development accounting suggests MRW overstates the contribution of human capital
How can we measure out of steady state behaviour (conditional convergence)?
Change the dependant variable to growth of income per capita.
Compare growth when your close to the steady state compared to if your further away from it
The growth rate of y is faster when you’re further away from your steady state. This is a condition of convergence
How can you show this convergence with the empirical growth model?
you add another x variable: lnyt-1
Using the empirical growth model, if 2 countries are starting at the same initial income and 1 has a higher investment, which one will grow faster?
The one with the high investment level because they will have a higher steady state.
Baumol found that if you have a higher GDP per capita, you have a lower growth rate, supporting convergence. What 2 problems of this paper are there?
1.Selection bias - only included countries that ended up rich post 1980 due to data availability and purposeful choice.
2. measurement error for income per capita, leading to a spurious negative correlation between the initial income and subsequent growth.
What did we see when De long dropped japan and added 7 richer countries?
Still get a negative correlation between growth and GDP, but the correlation was much weaker
How could we solve this selection bias?
Choose a sample of countries that at the start of the had the necessary ‘social capability’ to industrialise and catch up with world leader.
Further work shows convergence clubs. What are these?
Groups of countries with similar features (same level and growth of tech, same pop growth, same investment etc) that start out at different points and then converge.
What happens when we take the measurement error into account?
Convergence disappears completely
What type of convergence does the ss model predict
Countries converge only if they share the same steady state (conditional convergence)