Download Econ 302. Bilkent University Taskin Econometrics Department of

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Linear regression wikipedia , lookup

Transcript
Econ 302.
Econometrics
Bilkent University
Department of Economics
Taskin
Lab Exercise 7
INSTRUMENTAL VARIABLES
In growth equation, the pace of output growth is perceived as the result of
economic and noneconomic factors in the country. One such exercise examines the
role of “institutions” on output growth.
The paper written by Daron Acemoglu, Simon Johnson and James Robinson “The
Colonial Origins of Comparative Development: An Empirical Investigation”,
published in the American Economic Review, December 2001, vol. 91,
pp1369-1401, examines the role of ‘property rights’ in the output growth. The
presence of secure property rights will create an advantage in creating faster capital
accumulation and output growth.
However, a simple regression of GDP per capita on a measure of property rights may
suffer from endogeneity problem or reverse causation of higher income also
‘causing’ better and stronger property rights in the economy. Hence simple OLS will
create a biased and inconsistent estimates.
The paper makes use of the Instrumental Variable estimation technique and found
one variable strongly correlated with property rights (institutions) but independent
of the error in the equation because of no intertemporal correlation. (Not related in
the same time period) . This variable is “settler mortality” for the countries which
has colonization period in their history.
Settler mortality is seen as a predictor of good institutions, hence property rights
because high death rates in the first settlers forced the colonizers to establish good
institution in the countries from which they want to extract resources. The good
institutions then imply good institutions now.
Daron Acemoglu’s web site provides data for 64 ex-colonies:
Variables:
Counryn: countryname
Shortnam: country abbreviaion
Lgdp: log of GDP per capita in 1995
Logmort: log of settler mortality
Latitude: absolute latitude of capital
Prot: measure of protection of property rights
Euro: proportion of population of European descent in 1975
The estimating equation is;
Logyi = m + a Ri + X'i g + ei
Where the variable of interest is the R and the coefficient is
a.
OLS estimations
1. Conduct an OLS estimate. What are the properties of the estimate? How do you
interpret the coefficients? You may also use control variables such as Latitude.
( estimate using the file colony64_2.dta in stata or XXX in eviews, conduct robust
estimation, you can reproduce the 2nd and 5th column of Table 1)
2. Plot Log y and R.
How to deal with the possible endogeneity problem,
Output growth has an impact on institutional development. What variables might be an
instrument for R, protection of property rights.
Theoretically it is expected that present institutions are functions of past institutions and
hence,
Ri = lR + bRCi + X'i g R + uR where C is the past institutions.
Furthermore past institutions depend on measure of early settlements S with the following
function:
Ci = lC + bCSi + X'i g C + uC ,
Early settlements are functions of mortality rate of the early colonies, log of M, with the
following function;
Si = lS + bS log Mi + X'i g S + uS
Hence, the log of mortality rates in the early colonies can be used as an instrument for R,
protection of property rights today.
3. To see the instrument relevance let’s plot the relationship between R and log M.
4. Estimate the relationship and see what variation of R does the log M explains.
Compute the fitted value of Rhat in this equation. (First stage of the 2SLS)
5. Estimate the relationship between growth rate and fitted value of R, (Rhat). This is the
second stage of the 2SLS
These are not reported in the article, since most researchers will conduct 2SLS by using
the given software, which reports the corrected std errors, hence reliable t stats)
6. Estimate the IV directly.
You can reproduce the same results in Table 4 in columns (1) and (2).
If you use the 2SLS command the EVIEWS should produce the same results, with some
probability of minor difference resulting from difference in algorithms.
7. Which of the three methods would you prefer and why hint: explain why standard
errors will be different in a part c and d.
Overidentification tests
8. Check for overidentification.
Stata Program commands
regress logpgp95 avexpr, robust
ivreg logpgp95 (avexpr=logem4), first
ivreg logpgp95 (avexpr=cons00a), first
ivreg logpgp95 (avexpr=cons00a logem4), first
estimates store efficient
ivreg logpgp95 (avexpr=cons00a), first
estimates store consistent
hausman consistent efficient
….for overidentification test (Hausman)
ivreg logpgp95 (avexpr=cons00a) logem4 ……….
for another overidentification test (with mortality as an independent variable directly in the
equation rather than an instrument.)