Demonstration of the Gauss-Markov theorem

Gauss was undoubtedly one of the greatest mathematicians.

from history, made enormous contributions in the field of
statistics such as the normal distribution, likewise, was
the first in introduce the method of the minimums squares
ordinary MCO1 in 1821 before the variance estimator
minimum of Markov, these two greats united creating the
Gauss-Markov theorem where the properties of the
estimators called MELI, Best Linear Estimators
Unbiased or what in its English acronym BLUE also
considered as ELIO.

The parameters are MELI if the following conditions are met.

properties listed below, the reader is advised
revise some concepts of matrix algebra such as,
Inverse of the matrix, transposes and their properties, calculation
from the determinant, classes of matrices, scalar matrices,
symmetrical, singular among others; However, for the
reader what no it find familiarized him will expose step a
I present the resolution of these statements without leaving rigor.
mathematician that is implicitly present in such a way that the
resolution should be easy to follow even for those who do not
they manage matrix algebra.

It is worth emphasizing that demonstrations can also be

carry out through algebraic manipulation, which I do not
will be done in this work since I consider it a better way
simple and less complicated than matrix algebra which
I am sure it complicates things more for the student.

On the other hand, it should be considered that even being the

parameters MELI how good lo propose the theorem no take in
consideration of consistency properties, which for
many econometricians is the most important within the
econometric tools

1
OLS: Ordinary Least Squares
Félix Casares Conforme
The estimator¿ it is linear:

Since the parameters are a linear function of the

random variable, as well as the parameters are
they consider it as a weighted average of the observations
individuals in the dependent variable, according to this
preamble we have:

Starting from the linear model in its basic form,

͓Ɣ ƍͥ ͒ͦ ƍ ͥ ͙ $

In its matrix form we have,

µ Ɣ ´ì ƍ »

Where we obtain the estimator through matrix manipulation,

ìȩ Ɣ ´´´
ʚ ʛͯ ̊
ʚ ´´µʛ

Since µ Ɣ ´ì ƍ » is replaced in that term,

ìȩ Ɣ ´´´
ʚ ʛͯ ̊
´´´ìʚ ƍ » ʛ
ȩ ʚ
ì Ɣ ´´´ ʛͯ ̊
´´´ì ƍ ´´´ ʚ ʛͯ ̊
´´»

Where se can to observe the property of the matrix inverse

by original is equal to the identity

̻ ͯͥ . ̻ Ɣ ̓
ʚ´´´ ʛͯ ̊ ´´´ Ɣ ¥

Therefore,

ìȩ Ɣ ¥ì ƍ ´´´ʚ ʛͯ ̊ ´´»
ìȩ Ɣ ì ƍ ´´´ʚ ʛͯ ̊ ´´»

Where ʚ ʛͯ ̊
´´» Ɣ ³,

ìȩ Ɣ ì ƍ ³»

It can be clearly seen that parameter B is a

combination linear of the perturbations random y what
if they follow the normal distribution then e ~ NID

Félix Casares Conforme

fundamental to the moment of to carry out the tests
statistics.

The parameter¿ it is unbiased

This means that its average or expected value$ Ʒ Ɣ $ ɸ

is equal to the true value, but it should be considered
also what this no guarantees how close or far
is found from real values because according to the
theory the sample of variables X is fixed in the
sampling repetitive for what thus es in Economy for so much
more emphasis should be placed on the variance matrix
covariance, meanwhile, the proof:

ìȩ Ɣ ´´´
ʚ ʛͯ ̊
ʚ ´´µʛ

Since µ Ɣ ´ì ƍ » is replaced in that term,

ìȩ Ɣ ´´´
ʚ ʛͯ ̊
´´´ìʚ ƍ » ʛ
ìȩ Ɣ ´´´
ʚ ʛͯ ̊
´´´ì ƍ ´´´ ʚ ʛͯ ̊
´´»

Where you can to observe the property of the matrix inverse

by original is equal to the identity

̻ ͯͥ . ̻ Ɣ ̓
ʚ´´´ ʛͯ ̊ ´´´ Ɣ ¥

Therefore,

ìȩ Ɣ ¥ì ƍ ´´´ʚ ʛͯ ̊
´´»

Applying the Expectation operator E and taking in

consideration that the 'x' are fixed in the sampling
repetitive and that B, being a parameter, is a constant
in such a way that their hope is equal to themselves.

Ȫ Ɣ ì ƍ ´´´
¡ʚ ìʛ ʚ ʛͯ ̊
´´¡ʚ »ʛ

3Félix Casares According

Where E(e) = 0 because according to the properties of OLS,
the errors suman zero, always what haya one constant in
the model yes no exists constant, this principle already of
to be true; Then, the matrix and the result are nullified
it goes as follows:

Ȫ Ɣì
¡ʚ ìʛ

Another consideration that must be taken into account is that the

criterion del younger Error Quadratic Medium him adjust better
to the demands at the time of modeling because this
take into account the variance of the parameter.

3) The parameter¿ it is the best (optimal)

It is understood as better since it has minimum variance and

ensures that it is less than any other estimator with
different variance, emphasis is placed on the model that has
younger variance by so much deviation standard, position what
the to be one measure of variability en statistics this he
also adjust for the parameters as it is optimal
to have a small variance-covariance matrix.

In its matrix form we have,

µ Ɣ ´ì ƍ »

Where through matrix manipulation we obtain the estimator,

ìȩ Ɣ ´´´
ʚ ʛͯ ̊
ʚ ´´µʛ

Since µ Ɣ ´ì ƍ » is substituted in that term,

ìȩ Ɣ ´´´
ʚ ʛͯ ̊
´´´ìʚ ƍ » ʛ
ȩ ʚ
ì Ɣ ´´´ ʛͯ ̊
´´´ì ƍ ´´´ ʚ ʛͯ ̊
´´»

Where if can observe the property of the matrix inverse

by original is equal to the identity

4 Félix Casares Conform

 ̻ ͯͥ . ̻ Ɣ ̓
ʚ´´´ ʛͯ ̊ ´´´ Ɣ ¥

Therefore,

ìȩ Ɣ ¥ì ƍ ´´´ʚ ʛͯ ̊ ´´»
ìȩ Ɣ ì ƍ ´´´ʚ ʛͯ ̊ ´´»

ìȩ Ǝ ì Ɣ ´´´ʚ ʛͯ ̊
´´»

By definition we have that

͐ ̻͌ Ǝ ͉̽ ͐ ȩ ì Ǝ ìƷ ȩ´ʡ
Ɣ ̿ ʠƳ ì Ǝ ìƷƳ
ʚʚ ʛͯ ̊ ʛʚʚ ʛͯ ̊
Ɣ ̿ ´´´ ´´» ´´´ ´´»´ʛʛ

Transposing,

ʚʚ ʛͯ ̊ ʚ ʛͯ ̊ ʛ
Ɣ ̿ ´´´ ´´»»´´ ´´´

Ordering the terms, y applying the operator E we have

what
̊ ̊
Ɣ ʚʚ´´´ ʛͯ ʚ
´´´¡ʚ »»´ʛ ´´´ ʛͯ ʛ

Where,

̻ ͯͥ . ̻ Ɣ ̓
ʚ´´´ ʛͯ ̊ ´´´ Ɣ ¥

So that matrix is nullified and where = ü ¥ ̋

̋ ʚ ʛͯ ̊
² ®Ǝ «² Ɣ ü ´´´

Where the following should be considered:

• It is symmetric and positive because """. it is ʚ ʛ

• Must to be small position what yes
ʚthen,ʛ the determinant
ʚ of ʛ
this matrix will be larger so it will increase the
VAR-COV and the accuracy of estimation or forecasting
decrease

5Felix Casares As It Is
But we have to compare if it is true that it is the smallest.
variance for which we need to calculate the variance of
another estimator by adding a non-null matrix P.

ìȩ Ɣ ´´´
ʚʚ ʛͯ ̊
´´ƍ ¬ µʛ

Since µ Ɣ ´ì ƍ » is replaced in that term,

ìȩ Ɣ ´´´
ʚʚ ʛͯ ̊ ´´ƍ ¬ ´ìʛʚƍ » ʛ
ìȩ Ɣ ´´´
ʚ ʛͯ ̊ ´´´ì ƍ ´´´ʚ ʛͯ ̊ ´´» ƍ ¬´ì ƍ ¬»
¡ƳìƷȩ Ɣ ¡ ´´´ʚʚ ʛͯ ̊ ´´´ì ƍ ´´´ʚ ʛͯ ̊ ´´¡ʚ »ʛ ƍ ¬´ì ƍ ¬¡ʚ » ʛ ʛ
ìȩ Ɣ ì ƍ ´´´ʚ ʛͯ ̊ ´´» ƍ ¬´ì

Where PX=O (bias condition), E(e) =0

ìȩ Ɣ ì

The VAR-COV matrix of the new one is calculated.

estimator according to the condition of unbiasedness

ìȩ Ɣ ì ƍ ´´´ʚ ʛͯ ̊ ´´» ƍ ¬»
ìȩ Ǝ ì Ɣ ´´´ʚ ʛͯ ̊ ´´» ƍ ¬»
² ® Ǝ «² Ɣ Ƴì ȩƎ ìƷƳ ì ȩƎ ìƷ ´
Ɣ ʞʚʚ´´´ ʛͯ ̊ ´´» ƍ ¬» ´´´
ʛʚʚ ʛͯ ̊ ´´» ƍ ¬» ´ʛ ʟ
Ɣ ʚʚ´´´ ʛͯ ̊ ´´» ƍ ¬» ´»´´´´
ʛʚ ʚ ʛͯ ̊ ƍ »´¬´ ʛ
Developing the operations we have,
̊ ̊ ̊ ̊
Ɣ ʚ´´´ ʛͯ ´´»´»´´´´ʚ ʛͯ ƍ ʚ´´´ ʛͯ ´´»»´¬´ƍ ¬»´»´´´´ ʚ ʛͯ ƍ ¬»»´¬´

Since PX=0,

Ɣ ʚ´´´ ʛͯ ̊ ´´»´»´´´´ʚ ʛͯ ̊
ƍ ¬»»´¬´
Ɣ ʚ´´´ ʛͯ ̊ ´´´»»´´´´ʚ ʛͯ ̊
ƍ ¬»»´¬´
ʚ
Ɣ »»´´´´ ʛͯ ̊ ƍ ¬»»´¬´

whereee' ̋

̋ ʚ ̊
² ®Ǝ «² Ɣ ü ´´´ ʛͯ ƍ ü ̋ ¬¬´

6Félix Casares Conform

 ̋ ʞʚ ʛͯ ̊
² ®Ǝ «² Ɣ ü ´´´ ƍ ¬¬´ ʟ

It can clearly be observed that the VAR-COV matrix of this

estimator es mayor what the first by what is verified
through the Gauss-Markov theorem that the OLS are
MELI estimators
̋ ʞʚ ʛͯ ̊ ʟ ü ´´´ ̋ ʚ ʛͯ ̊
² ®Ǝ «² Ɣ ü ´´´ ƍ ¬¬´Ƙ

7Félix Casares Conforme

Bibliography:

• Econometric Analysis by Greene, 3rd edition

• Econometrics by Damodar Gujarati 5th Edition

cap.3

• Models and Forecasts by Pindyck and Rubinfeld 4th

editing chapters 3 and 4

• 100 Econometrics Exercises

• www.uam.es Faculty of Economics University

Autonomous Community of Madrid

• Probability and Statistics was editorial edition

Shawm

• Statistics and Econometrics by Salvatore Domenick

2nd edition

• Introduction to Econometrics, an approach

modern by Wooldridge

Félix Casares According

9Félix Casares According