Methods and guidelines for effective model calibration


Proof 1: Parameters estimated by linear regression are unbiased



Download 0,49 Mb.
Pdf ko'rish
bet54/55
Sana28.05.2022
Hajmi0,49 Mb.
#613965
1   ...   47   48   49   50   51   52   53   54   55
Bog'liq
EffectiveCalibration WRIR98-4005

Proof 1Parameters estimated by linear regression are unbiased.
Take the expected value of the optimized parameters, as calculated using equation (C7):
E(b

) = (X
T
ω
X)
-1
X
T
ω
E(y) = (X
T
ω
X)
-1
X
T
ω
X
β
(C8)
If
X
= X, 
(X
T
ω
X)
-1
X
T
ω
X
= I (C9)
where I is an identity matrix. Substituting C9 into C8 yields: 
E(b) = 
β
(C10)


87
Thus, if
X
= X, the expected values of the estimates equal the true values, which means that the 
estimates are unbiased. In nonlinear models, the equality is unlikely to be true, so that unbiasedness 
is not guaranteed for nonlinear models, even if the mdoel if correct.
Proof 2: The weight matrix needs to be defined in a particular way for the parameter esti-
mates to have the smallest variance.
It is desirable to estimate parameters with the smallest variance and, therefore, the greatest 
precision. The variance of the parameter estimates occur as the diagonal terms in the variance-co-
variance matrix of the parameters, which is calculated using the equation defined above as:
V(b) = E[(b-E(b)) (b-E(b))
T
].
(C11)
replacing b with equation (C7) and E(b) with equation (C10) yields:
V(b) = E[ ((X
T
ω 
X)
-1
X
T
ω
y - 
β
) ((X
T
ω
X)
-1
X
T
ω
y - 
β
)
T
].
(C12)
Expanding the product on the right-hand side produces an equation with four terms:
V(b) = E[ ((X
T
ω
X)
-1
X
T
ω
y) ((X
T
ω
X)
-1
X
T
ω
y)
T
-
(C13)
((X
T
ω
X)
-1
X
T
ω
y) 
β
T
-
β 
((X
T
ω
X)
-1
X
T
ω
y) 
T
+
β β

]
Use the matrix property (AB)
T
=B
T
A
T
to rearrange the first term as:
((X
T
ω
X)
-1
X
T
ω
y) (( X
T
ω
y)
-1
X
T
ω
y)
T

(C14)
( X
T
ω
y)
-1
X
T
ω
y y
T
X(X
T
ω
X)
-1
Take the expected value of each term and note that only y is stochastic to obtain:
V(b) = (X
T
ω
X)
-1
X
T
ω
E[y y
T
]
ω
X( X
T
ω
X)
-1

(C15)
( X
T
ω
X)
-1
X
T
ω
E[y]) 
β
T
-
β
(( X
T
ω
X)
-1
X
T
ω
E[y]) 
T
+
β β
T
In the first term, apply y = 
X
β

ε
, so that:
E[y y
T
] = E[(
X
β

ε
) (
X
β

ε
)
T
] (C16)
= E[(
X
β
)(
X
β
)
T
+ (
X
β

ε
T

ε
X
β
T

ε ε
T
]
Taking the expected value of each term, and noting that only 
ε
is stochastic and that the second 
and third terms of equation C16 equal zero because E[
ε
]=0 produces:


88
E[y y
T
] = (
X
β
)(
X
β
T
) + E[
ε ε
T
] = 
X
β
β
T
X
T
+ E[
ε ε
T
]
(C17)
Note that E[
ε ε
T
] = V(
ε
), the variance-covariance matrix of the true errors. This can be derived by 
applying the standard equation for calculating the variance-covariance matrix of a vector, so that 
V(
ε
) = E[(
ε
-E(
ε
)) (
ε
-E(
ε
))
T
], and noting that E(
ε
)=0. 
Substituting these results into equation (C15) yields: 
V(b) = (X
T
ω
X)
-1
X
T
ω
X
β β
T
X
T
ω
X (X
T
ω
X)
-1
(C18)
+ (X
T
ω
X)
-1 
X
T
ω
E[
ε ε
T

ω
X (X
T
ω 
X)
-1
- ((X
T
ω
X)
-1
X
T
ω
X
β

β
T
-
β
((X
T
ω
X)
-1 
X
T
ω
X
β

T
+
β β
T
If 
X
= X, then (X
T
ω
X)
-1 
X
T
ω
X
= I, which gives the following:
V(b) = 
β
β
T
+ (X
T
ω
X)
-1
X
T
ω
E[
ε
ε
T

ω
X(X
T
ω
X)
-1
(C19)

β
β
T

β
β
T

β
β
T
The
β
β
T
terms cancel, leaving:
V(b) = (X
T
ω
X)
-1
X
T
ω
E[
ε
ε
T

ω
X (X
T
ω
X)
-1
(C20)
If the weight matrix is defined such that
E[
ε
ε
T
] = V(
ε
) = 
σ
2
ω
-1
,
(C21)
where 
σ
2
is the true common error variance, equation C20 reduces to:
V(b) = 
σ
2
(X
T
ω
X)
-1
= s
2
(X
T
ω
X)
-1
(C22)
where the last equals sign is approximate and s2, the calculated error variance, approximates the 
unknown true common error variance. Equation C22 is the expression commonly used to calculate 
the variance-covariance matrix for the parameter values, but really only applies if 
X
= X, and C21 
applies.
If the equation for V(b) cannot be simplified to equation C22, equations of the form C18 or 
C20 should be used to calculate the variance-covariance matrix of the of the parameter estimates
although it is unclear how to evaluate C18 because 
β
is unknown. For linear problems, equarion 


89
C19 always produces a larger variance for the parameters and simulated predictions than is pro-
duced by otehr possible equations (Bard, 1974; Beck and Arnold, p. 232-234). Thus, the smallest 
variance parameter estimates are those for which equation C21 applies and, therefore, for which 
X
= X and the weighting is defined such that 
ω 
= V(
ε
)
-1
(the weighting is closely related to the vari-
ance-covariance matrix of the true, unknown errors). Although not always valid, linear theory pro-
vides the only available guidance for defining the weight matrix for nonlinear problems.

Download 0,49 Mb.

Do'stlaringiz bilan baham:
1   ...   47   48   49   50   51   52   53   54   55




Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©hozir.org 2024
ma'muriyatiga murojaat qiling

kiriting | ro'yxatdan o'tish
    Bosh sahifa
юртда тантана
Боғда битган
Бугун юртда
Эшитганлар жилманглар
Эшитмадим деманглар
битган бодомлар
Yangiariq tumani
qitish marakazi
Raqamli texnologiyalar
ilishida muhokamadan
tasdiqqa tavsiya
tavsiya etilgan
iqtisodiyot kafedrasi
steiermarkischen landesregierung
asarlaringizni yuboring
o'zingizning asarlaringizni
Iltimos faqat
faqat o'zingizning
steierm rkischen
landesregierung fachabteilung
rkischen landesregierung
hamshira loyihasi
loyihasi mavsum
faolyatining oqibatlari
asosiy adabiyotlar
fakulteti ahborot
ahborot havfsizligi
havfsizligi kafedrasi
fanidan bo’yicha
fakulteti iqtisodiyot
boshqaruv fakulteti
chiqarishda boshqaruv
ishlab chiqarishda
iqtisodiyot fakultet
multiservis tarmoqlari
fanidan asosiy
Uzbek fanidan
mavzulari potok
asosidagi multiservis
'aliyyil a'ziym
billahil 'aliyyil
illaa billahil
quvvata illaa
falah' deganida
Kompyuter savodxonligi
bo’yicha mustaqil
'alal falah'
Hayya 'alal
'alas soloh
Hayya 'alas
mavsum boyicha


yuklab olish