Professional Documents
Culture Documents
Seminar TSLSM
Seminar TSLSM
Seminar TSLSM
Seminar 7
Econometric analysis - procedure
Formulating of an assumption or a hypothesis
Eg. Keynes model of income: what the economic theory says about the influence of income on consumption
Specification of an economic model
C = B1 + B2Y B2 < 0
Y=C+I
Formulation of the econometric model
y1t = γ11x1t + β12y2t + u1t y1= consumption y2 = income
y2t = β21y1t + γ12x2t + u2t x1=constant x2= investment
Identification of variables in the model
1. equation: 1=1
2. equation: 1=1 model is exactly identified
Data collection
Time series.
Cross sectional data
Panel data
Estimation of parameters
β12t β21 is estimated through indirect least square method.
Simultaneous model overidentified are estimated with the use of the 2SLS or other method.
Testing of the hypothesis of the parameter significance
Is our assumption on B2≠ 0 adequately supported by the data?
Use of model for prognosis
What will be the consumption trend in next 3 years if the income declines by 1% per year?
Methods of structural parameters
estimation
y1t= γ11 x1t + γ16 y1t-1 + γ12 X2 + γ13 X3t + γ14 x4t + u1t
y2t = β21y1t + γ21 x1t + + γ25 x5t + u2t
RECURSIVE MODELS
y1=variation in prices
Second equation represents one-
y2= variation in wages
directional causal dependence (y1t affects
x1= constant.
y2t but not wise versa).
X2 = variation in capital prices
X3= variation in import prices Although y1t is an endogenous variable,
X4= variation in labour productivity its presence in the second equation has
X5= unemployment rate an exogenous character.
X6= variation in wages in t-1
Procedure:
• We transform the equations into the REDUCED form (on right hand side
there are only exogenous variables)
y1t = m11x1t + m12x2t + v1t
y2t =m21x1t + m22x2t + v2t
• We apply OLS we get values of parameters m
• Return from REDUCED to STRUCTURAL form (by solving 4 equations with
4 variables)
Overidentified simultaneous models
We estimate with TWO STAGE LEAST SQUARE METHOD
Procedure:
• Overidentified models have a problem of transforming from reduced to structural
form (one equation in reduced form can correspond to more equations in simultaneous
forms)
•We must start from the structural form
• The trick is to replace endogenous variables on the right hand side of the equation by
an artificial (instrumental) variable – removing the simultaneous component
• We perform a regression of that particular endogenous variable on all exogenous
variables in the whole model
•After obtaining the instrumental variables, we substitute them into equation and apply
OLS
Review of the estimation methods
y∆ = y1 Y2 X* n x kv
X=
nxk
Xn n x kn
n x gv n x1 n x (gv –1)
Input matrices
• Example for this equation :
y2t = ß24y4t + γ21x1t + γ23x3t + u2t
Input matrices:
y1= y2t
Y2= y4t
Xv= x1, x3
X= x1, x2, x3
Principle of the TSLSM
1) Replacing matrix Y2 by theoretical Y^2 –
regression on all X
Y^2 = X ( XTX)-1 XT Y 2
ß2 T
= (PT P)-1 PT y1 P = [Y^2, Xv ]
Γ 1v T
Principle of the TSLSM
PT = Y^2T x P = [Y^2, Xv ]
Xv T
-1
Y^2T Y^2 Y2T Xv Y^2T
ß2 T
y1
=
XvT Y2 XvT Xv XvT
Γ 1v T
Matrices in the TSLSM
-1
Y^2T Y^2 Y^2T Xv Y^2T
ß2 T
y1
= Matrix kT
Xv Y2
T
Xv Xv XvT
Γ 1v T
-1
Y^2T Y^2 Y^2T Xv Y^2T
ß2 T
= Matrix c y1
XvT Y2 XvT Xv XvT
Γ 1v T