120 likes | 257 Views
ENDOGENEITY. Development Workshop. What is endogeneity and why we do not like it. Three causes: X influences Y, but Y reinforces X too Z causes both X and Y fairly contemporaneusly X causes Y, but we cannot observe X and Z (which we observe) is influenced by X but also by Y Consequences:
E N D
ENDOGENEITY Development Workshop
What is endogeneity and why we do not like it • Three causes: • X influences Y, but Y reinforces X too • Z causes both X and Y fairly contemporaneusly • X causes Y, but we cannot observe X and Z (which we observe) is influenced by X but also by Y • Consequences: • No matter how many observations – estimators biased (this is called: inconsistent) • Ergo: whatever point estimates we find, we can’t even tell if they are positive/negative/significant, because we do not know the size of bias + no way to estimate the size of bias
Solutions • If X/Y are not very sticky, using past values is enough • Most economic variables are sticky=>instruments: • Easy: use lag of X instead of X as its instrument (only if not very sticky) • Hard: find an adequate instrument, i.e. variable uninfluenced by Y that correlates well with X • Few words about instruments and their quality (paper by Frank Kleibergen, Econometrica, 2002) • Method: find instruments, run 2SLS, trust your results • Problem: t-test and F-test may loose power under 2SLS • In other words: estimators are fine, standard errors are fine, but the tests we use for them behave far from normal
So what about IV • Really hard to find instruments • Good instruments do not need a good theory • But to be good, correlation needs to be reliable • Once we find good instrument – run 2SLS • But remember that p-values may be out of the moon and there is little we can do to avoid this problem • If really want to work hard, use alternative tests (nobody puts them on statistical packages)
IV method • Variables X and Y expected to be endogenous • Z correlates well with X (purely statistical property) and by definition cannot depend on Y • Z is called an instrument: • Regress Z on X => get fitted values X* • Regress X* on Y => these are your final results • Interpretation problems • Instruments and theories – Acemoglu, Johnson and Robinson 2001 (AER)
IV method • One needs as many equations as endogenous variables • For each endogenous variable one needs one (separate for each) instrument • For each first stage regression one can use the same instruments as well.
IV in STATA • Syntax in STATA • ivregressy (x1x2x3= z1z2z3z4) x4 x5 x6 We run a regression in which • y is explained • x1, x2, x3 are endogenous (instrumented for) • z1, z2, z3 are exogenous (instruments) • x3, x4 x5 are exogenous too (do not need to be instrumented) Difference between automatic and manual 2SLS?
Papers • Ann Harrison, Opennes and growth: A time-series, cross-country analysis of developing countries, Journal of Development Economics, 1996 • Jeffrey Frankel i David Romer, Does Trade Cause Growth?, AER 1999 • Jeffrey Frankel i Andrew K. Rose, One Money, One Currency,