chapter 11 gmm: general formulas and applicationefinance.org.cn/cn/ap/ap11.pdf · 2018. 5. 24. ·...

56
Chapter 11 GMM: General Formulas and Application

Upload: others

Post on 28-Aug-2020

13 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Chapter 11GMM: General Formulas

and Application

Page 2: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Main Content

General GMM FormulasTesting MomentsStandard Errors of Anything by Delta MethodUsing GMM for RegressionsPrespecified weighting Matrices and Moment ConditionsEstimating on One Group of Moments,Testing on AnotherEstimating the Spectral Density Matrix

Page 3: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

11.1 General GMM Formulas

GMM procedures can be used to implement a host of estimation and testing exercises.To estimate the parameter, you just have to remember (or look up) a few very general formulas, and then map them into your case.Express a model as

Everything is a vector: can represent a vector of L sample moments, can be M data series, can be N parameters.

[ ( , )]tE f x bf

tx b

Page 4: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Definition of the GMM Estimate

We estimate parameters to set some linear combination of sample means of to zero

where

is a matrix that defines which linear combination of will be set to zero.If you estimate by ,the first-order condition are

This is mean

bf

:b set ( ) 0T Ta g b =

1

1( ) ( , )T

T tt

g b f x bT =

= ∑ta

( )Tg b

b min ( ) ( )T Tg b Wg b′( ) 0T

Tg Wg bb′∂

=∂

TT

ga Wb′∂

=∂

Page 5: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Standard Error of Estimate

Hansen (1982), Theorem 3.1 tells us that the asymptotic distribution of the GMM estimate is

where

In practical terms,this means to use

1 1( ) [0,( ) ( ) ]T b b N ad aSa ad− − ′′− →

( )[ ( , )] Tt

f g bd E x bb b∂ ∂

≡ =′ ′∂ ∂

lim Ta p a=

[ ( , ), ( , ) ]t t jj

S E f x b f x b∞

−=−∞

′≡ ∑

1 11var( ) ( ) ( )b ad aSa adT

− − ′′=

Page 6: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Distribution of the moments

Hansen’s Lemma 4.1 gives the sampling distribution of the moments :

The terms account for the fact that in each sample some linear combinationsof are set to zero. Then S is singular.

( )Tg b

1 1( ) [0,( ( ) ) ( ( ) ) ]TT g b N I d ad a S I d ad a− − ′→ − −

1( )I d ad a−−

Tg

Page 7: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Test

A sum of squared standard normals is distributed , so we have

is distributed which has degrees of freedom given by number of nonzero linear combinations of , the number of moments less the number of estimated parameters

2χ1 1 1( ) [( ( ) ) ( ( ) ) ] ( )T TTg b I d ad a S I d ad a g b− − −′ ′− −

Tg

Page 8: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

It does, but with a hitch: The variance-covariance matrix is singular, so you have to pseudo-invert it.For example, you can perform an eigenvalue decomposition and then invert only the non-zero eigenvalues.

'Q Q= Λ∑

Page 9: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Efficient Estimates

Hansen shows that one particular choice is statistically optimal,This choice is the first order condition to

that we studied in the last Chapter.With this weighting matrix, the standard error of b reduces to

1a d S −′=

1

{ }min ( ) ( )T Tb

g b S g b−′

1 1( ) [0,( ) ]T b b N d S d− −′− →

Page 10: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

With the optimal weights the variance of the moments simplifies to

Proof:

1S −

1 11cov( ) ( ( ) )Tg S d d S d dT

− −′ ′= −

1 1

1

1 1

1 1 1 1

1 1 1 1

1 1 1 1

1var( ( )) ( ( ) ) ( ( ) )

( ( ) ) ( ( ) ) )( ( ) ) ( ( ) )

( ( ) )( ( ) )

( ) ( ) ( )

(

Tg b I d ad a S I d ad aT

a d SI d ad a S I d ad a

I d d S d d S S I d d S d d S

S d d S d d I S d d S d d

S d d S d d d d S d d d d S d d

S d d S

− −

− −

− − − −

− − − −

− − − −

′= − −

′=

′− −

′ ′ ′ ′= − −

′′ ′ ′ ′ ′= − −

′′ ′ ′ ′ ′ ′= − − +

′= − 1 1)d d− ′ ′

Page 11: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Using this matrix in a test, there is an equivalent and simpler way to construct this test

1 2( ) ( ) (# # )T TTg b S g b moments parametersχ−

→ −

Page 12: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Alternatively, note that is a pseudo-inverse of the second stageProof: A pseudo inverse times should result in an idempotent matrix of the same rank as

Then, check that the result is idempotent

This derivation not only verifies that has the same distribution as , but that they are numerically the same in every sample.

1S −

cov( )Tg

cov( )Tgcov( )Tg

1 1 1 1 1cov( ) ( ( ) ) ( )TS g S S d dS d d I S d dS d d− − − − −′ ′ ′ ′= − = −

1 1 1 1 1 1 1 1 1( ( ) )( ( ) ) ( ) )I S d d S d d I S d d S d d I S d d S d d− − − − − − − − −′ ′ ′ ′ ′ ′− − = −

TJcov( )T T Tg g g′

Page 13: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Model Comparisons

You often want to compare one model to another. If one model can be expressed as a special or restricted case of the other or unrestricted model we can perform a statistical comparison that looks very much like a likelihood ratio test.

If the restricted model is really true, it should nor rise “much.This is a “ difference” test due to Neweyand West(1987), who call it the “D-test”

2( ) ( ) (# )T TTJ restricted TJ unrestricted restrictionχ− ∼

Page 14: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

11.2 Test Moments

How to test one or a group of pricing error.(1)Use the formula for (2)A difference test We can use the sampling distribution of , to evaluate the significance of individual pricing errors, to construct a t-test(for a single moment) or a test(for groups of moments)

var( )Tg

Tg

Page 15: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Alternatively, you can use the difference approach.Start with a general model that includes all the moments, and form an estimate of the spectral density matrix S.Set to zero the moments you want to test, and denote the vector of moments, including the zeros(s for “smaller)

(#eliminated moments)If moments we want to test truly are zero,the criterion should not be that much lower

( )sTg b

1 1 2( ) ( ) ( ) ( )T T sT s sT sTg b S g b Tg b S g b χ− −′ ′− ∼

Page 16: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

11.3 Standard Errors of Anything by Delta Method

we want to estimate a quantity that is a nonlinear function of sample means

In this case, we have

For example, a correlation coefficient can be written as a function of sample means as

Thus, take

[ ( )] ( )tb E x uφ φ= =

1var( ) [ ] cov( , )[ ]T t t jd db x x

T du duφ φ∞

−−∞

′′= ∑

2 2 2 2

( ) ( ) ( )( , )( ) ( ) ( ) ( )

t t t tt t

t t t t

E x y E x E ycorr x yE x E x E y E y

−=

− −2 2[ ( ), ( ), ( ), ( ), ( )]t t t t t tu E x E x E y E y E x y ′=

Page 17: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

11.4 Using GMM for Regression

Mapping any statistical procedure into GMM makes it easy to develop an asymptotic distribution that corrects for statistical problems such as non-normality, serial correlation and conditional heteroskedasticity.For example, I map OLS regressions into GMM.When errors do not obey the OLS assumptions, OLS is consistent, and often more robust than GLS, but its standard errors need to be corrected.

Page 18: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

OLS picks parameters β to minimize the variance of the residual:

We find from the first order condition, which states that the residual is orthogonal to the right hand variable:

It is exactly identified. We set the sample moments exactly to zero and there is no weighting matrix (a = I). We can solve for the estimate analytically,

This is the familiar OLS formula. But its standard error need to be corrected.

β

( ) [ ( )] 0T t t tg E x y xβ β′= − =

1[ ( )] ( )T t t T t tE x x E x yβ −′=

2

{ }min [( ) ]T t tE y xβ

β ′−

Page 19: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

We can use GMM to obtain the standard errors through , so that

1 1

( )( , ) ( )

1var( ) ( ) [ ( )] ( )

t t

t t t t t t

t t t t t j t j t tj

d E x xf x x y x x

E x x E x x E x xT

β β ε

β ε ε∞

− −− −

=−∞

′=

= − =

′ ′ ′ ′= ∑

1 1( ) [0,( ) ]T b b N d S d− −′− →

Page 20: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Serially Uncorrelated, Homoskedastic Errors

Formally,the OLS assumptions are

The first assumption means that only the j=0 term enter the sum

The second assumption means that

Hence the standard errors reduces to our old form

1 1 2

2 21 1 2

( | , , , ) 0

( | , , , ) tant t t t t

t t t t t E

E x x

E x x cons t

ε ε ε

ε ε ε σ− − −

− − −

=

= =

2( ) ( )t t t j t j t t tj

E x x E x xε ε ε∞

− −=−∞

′′ ′ =∑

2 2( ) ( ) ( )t t t t t tE x x E E x xε ε′ ′=

2 11var( ) ( )X XT εβ σ −′=

Page 21: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Heteroskedastic Errors

If we delete the condition homoskedasticity assumption

The standard errors are

These are known as “heteroskedasticity consistent standard errors” or “white standard errors” after White (1980)

2 21 1 2( | , , , ) tant t t t tE x x cons t εε ε ε σ− − − = =

1 21var( ) ( ) ( ) ( )t t t t t t tE x x E x x E x xT

β ε−′ ′′=

Page 22: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Hansen-Hodrick Errors

When the regression notation is

under the null that one-period returns are unforecastable, we still see correlation in the etdue to the overlapping data. Unforecastablereturns imply

forUnder this condition, the standard errors are

t k k t t ky xβ ε+ +′= +

( ) 0t t jE ε ε − = | |j K≥

1 1

1

1var( ) ( ) [ ( )] ( )k

k t t t t t j t j t tj k

E x x E x x E x xT

β ε ε− −− −

=− +

′ ′ ′ ′= ∑

Page 23: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

11.5 Prespecified Weighting Matrices and Moment Conditions

In the last chapter, our final estimates were based on the “efficinet” weighting matrix. A prespecified weighting matrix lets you specify which moments or linear combination of momentsGMM will value in the minimization.So you can also go one step further and impose which linear combinations of moment conditions will be set to zero in estimation rather than use the choice resulting from a minimization.

1S −

Ta

Page 24: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

For example, if , , but so that the second moment is 10 times more sensitive to the parameter value than the first moment, then GMM with fixed weighting matrix set

If we want GMM to pay equal attention to the two moment, we can fix the matrix directly.Using a prespecified weighting matrix is not the same thing as ignoring correlation of the error in the distribution theory.

1 2[ , ]T T Tg g g= W I= / [1,10]Tg b∂ ∂ =

1 21* 10* 0T Tg g+ =

Ta

tu

Page 25: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

How to Use Prespecified Weighting Matrices

If we use weighting matrix , the first-order conditions to are

So the variance-covariance matrix of the estimated coefficients is

The variance-covariance matrix of the moments

The above equation can be the basis of test for the overidentifying restrictions.

W{ }

min ( ) ( )T Tbg b Wg b′

( ) ( ) ( ) 0TT T

g b Wg b d Wg bb

′∂ ′= =∂

1 11var( ) ( ) ( )b d Wd d WSWd d WdT

− −′ ′ ′=

Tg1 11var( ) ( ( ) ) ( ( ) )Tg I d d Wd d W S I Wd d Wd d

T− −′ ′ ′ ′= − −

Page 26: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

If we interpret to be a generalized inverse, then

If is singular, you can inverting only the nonzero eigenvalues.

1( )−i

1 2var( ) (# # )T T Tg g g moment parametersχ−′ −∼var( )Tg

Page 27: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Motivations for Prespecified Weighting Matrices

Robustness, as with OLS vs. GLSWhen errors are autocorrelated or heteroskedasticand we correctly model the error covariancematrix and the regression is perfectly specified , the GLS procedure can improve efficiency.If the error covariance matrix is incorrectly, the GLS estimates can be much worse than OLS.In these cases, it is often a good idea to use OLS estimates. But we need to correct the standard error of the OLS estimates

Page 28: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

For GMM, first-stage or other fixed weighting matrix estimates may give up something in asymptotic efficiency, standard errors and model fit tests.They are still consistent and more robust to statistical and economic problems.But we use the S matrix in computing standard error.When the parameter estimates have a great different between the first stage and the second stage, we should decide what cause this.It is truly due to efficiency gain or a model misspecification.

Page 29: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Near-Singular S

The spectral density matrix is often nearly singular, since asset returns are highly correlated with each other.As a result, second stage GMM tries to minimize differences and differences of differences of asset returns in order to extract statistically orthogonal components with lowest variance.This feature leads GMM to place a lot of weight on poorly estimated, economically uninteresting, or otherwise non-robust aspects of the data.

Page 30: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

For example, suppose that S is given by

So

We can write

where

Then, the GMM criterion isis equivalent to

11

ρ⎛ ⎞

= ⎜ ⎟⎝ ⎠

12

1111

ρρ− −⎡ ⎤= ⎢ ⎥−− ⎣ ⎦

1C C S −′ =

2 2

11 1

0 1C

ρρ ρ

−⎡ ⎤⎢ ⎥= − −⎢ ⎥⎢ ⎥⎣ ⎦

1min T Tg S g−′

min( )( )T Tg C Cg′ ′

Page 31: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

gives the linear combination of moments that efficient GMM is trying to minimize.As ,for the matrix C, the (2,2) element stay at 1, but the (1,1) and (1,2) elements get very large.If then

This mean that GMM pay a little attention to the second moment, and play three times as much weight on the difference between first and second moment.Through the decomposition of S, we can see what moments GMM is prizing.

TCg

1ρ →

3.20 3.040 1

C−⎡ ⎤

= ⎢ ⎥⎣ ⎦

0.95ρ =

Page 32: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

GMM wants to focus on well-measured moments.In asset pricing applications, the errors are close to uncorrelated over time, so GMM is looking for portfolios with small values of . Those will be assets with small return variance.Then, GMM will pay most attention to correctly pricing the sample minimum-variance portfolio.This cause that sample minimum-variance portfolio may have little to do with the true minimum-variance portfolio.Like any portfolio on the sample frontier, its composition largely reflects luck.

1 1var( )et tm R+ +

Page 33: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation
Page 34: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Economically Interesting Moment

The optimal weighting matrix makes GMM pay close attention to linear combinations of moments with small sampling error in both estimation and evaluation.We want to force the estimation and evaluation to pay attention to economically interesting moments instead.

Page 35: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Level Playing Field

The S matrix changes as the model and as its parameters change.As the S matrix changes, which assets the GMM estimate tries hard to price well changes as well.For example we take a model and create a new model by simply adding noise, unrelated to asset returns (in sample), then the moment condition is unchanged. However, the spectral density matrix rise dramatically. This can reduce the , leading to a false sense of “improvement”.

tm

t t tm m ε′ = +

TE ( )eT t tg m R′=

2(( ) )e et t t tS E m R Rε= +

TJ

Page 36: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Level Playing Field

If the sample contains a nearly riskfree portfolio of the test assets, or a portfolio with apprently small variances of mt+1Re

t+1, then JT test will focus on the pricing of this portfolio and will lead to a false rejection, since there is an eigenvalue of S that is too small.Some stylized facts, such as the RMSE, pricing errors, are as interesting as the statistical tests.

Page 37: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Some Prespecified Weighting MatricesWhen the second-moment matrix of payoffs

in place of S (Hansen and Jagannathan (1997)). The minimum distance (second moment) between a candidate discount factor y and the space of true discount factors is the same as the minimum value of the GMM criterion with as weighting matrix.Why is this true?

1( )W E xx −′=

1( )W E xx −′=

Page 38: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation
Page 39: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Proof :The distance between y and the nearest valid m is the same as the distance between is the same as the distance between and . From the OLS formula,

is the portfolio of that price x

Then, the distance between y and is

( | )proj y X x∗

1( | ) ( ) ( )proj y X E yx E xx x−′ ′=

x∗ x1( )x p E xx x∗ −′ ′=

x∗

1 1

1

1

1

( | )( ) ( ) ( )

( ( ) ) ( )[ ( ) ] ( ) [ ( ) ]

( )T T

y x proj y X xE yx E xx x p E xx xE yx p E xx x

E xy p E xx E xy pg E xx g

∗ ∗

− −

− = −

′ ′ ′ ′= −

′ ′ ′= −′ ′ ′ ′= − −

′ ′=

Page 40: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Identity Matrix

Using the identity matrix weights has a particular advantage with large systems in which S is nearly singular.It avoids most of the problems associated with inverting a near-singular S matrix.

Page 41: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Comparing the Second-Moment and Identity Matrices

The second moment matrix and the optimal weighting matrix S give an objective that is invariant to the initial choice of assets or portfolios.

It is not true of the identity or other fixed matrices. The results depend on the initial choice of portfolios.The second-moment matrix is often even more nearly singular than the spectral density matrix. It is no help on overcoming the near singularity of S.

1

1[ ( ) ] ( ) [ ( ) ]

[ ( ) ] ( ) [ ( ) ]E yAx Ap E Axx A E yAx Ap

E yx p E xx E yx p

′ ′ ′− −′ ′= − −

Page 42: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Estimating on One Group of Moment, Testing on Another

We can use one set of moment for estimate and another for testingWe can also using one set of asset returns and then see how the model does “out of sample” on another set of assetWe can do all this very simply by using an appropriate weighting matrix or a prespecified moment matrix , for example Ta

[ ,0 ]T N N Ma I +=

Page 43: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

11.7 Estimating the Spectral Density Matrix

The optimal weighting matrix S depend on population moments, and depend on the parameters b.

There are a lot of parameters.How do we estimate this matrix in practice?

1( ), ( ( ) )t t j t t tj

S E u u u m b x p∞

− −=−∞

′= ≡ −∑

Page 44: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Use a sensible first-stage W, or transform the data

In the first-stage b estimates, we should use a sensible weighting matrix. Sometimes, some moments will have different unit than other moment. For example the moment formed by and the moment formed by .It is also useful to start with moments that are not horrendously correlated with each other.

1 *tt

dRp+

1 *1tR +

Page 45: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

For example, you might consider andrather than and .

aRbR

a bR R−

aR

1 1 1 0 2 10 1 1 1 1 1

W− −⎡ ⎤ ⎡ ⎤ ⎡ ⎤

= =⎢ ⎥ ⎢ ⎥ ⎢ ⎥− −⎣ ⎦ ⎣ ⎦ ⎣ ⎦

Page 46: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Remove meansUnder the null, whether we estimate the covariance matrix by removing mean.

Hansen and Singleton (1982) advocate removing the means in sample.But this mothed also make that estimate S matrices are often nearly singular. Since

is a singular matrix

( ) 0tE u =

1 1

1 1[( )( ) ],T T

t t tt t

u u u u u uT T= =

′− − =∑ ∑

( ) cov( , ) ( ) ( )E uu u u E u E u′ ′ ′= +

( ) ( )E u E u′

Page 47: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Downweight highter-order correlations

When we estimate S, we want to construct consistent estimates that are automatically positive definite in every sample.For example the Newey and West estimate, it is

The Newey-West estimator is base on the variance of kth sums. So it is positive definite

1

| | 1ˆ ( ) ( )k T

t t jj k t

k jS u uk T −

=− =

− ′= ∑ ∑

1 11

var( ) ( ) ( 1)[ ( ) ( )]

| |[ ( ) ( )] ( )

k

t j t t t t t tj

k

t t k t k t t t jj k

u kE u u k E u u E u u

k jE u u E u u k E u uk

− − −=

− − −=−

′ ′ ′= + − + +

−′ ′ ′+ + =

Page 48: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

What value of k, or how wide a window if of another shape, should you use?The rate at which k should increase with sample size, but not as quickly as the sample size increases.

Page 49: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Consider parametric structures for autocorrelation and heteroskedasticity

GMM is not inherently tied to “nonparametric” covariance matrix estimates.We can impose a parametric structure on the S matrix.For example, if we model a scalar u as an AR(1) with parameter and then

So we only need to estimate two parameter

2uσ ρ

2 | | 2 1( )1

jt t j u u

j jS E u u ρσ ρ σ

ρ

∞ ∞

−=−∞ =−∞

+= = =

−∑ ∑

Page 50: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Use the null limit correlations

In the asset pricing setup, the null hypothesis specifies that as well asIn this situation, you can get

However, the null might not be correct, if the null is not correct, you have a inconsistent estimate.If the null is not correct ,estimating extra lags that should be zero under the null only loses a little bit of power.

1 1 1( ) ( 1) 0t t t t tE u E m R+ + += − = 1( ) 0tE u + =

1

1ˆ .T

t tt

S u uT =

′= ∑

Page 51: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Monte Carlo evidence suggest that adding null hypothesis can help with the power of test statistics.Small-sample performance of the nonparametric estimators with many lags is not very good.We can test the autocorrelated of to decide whether the model is right.If there is a lot of correlation, this is an indication that something is wrong with the estimate.

tu

1

1ˆ .T

t tt

S u uT =

′= ∑

Page 52: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Size problems; consider a factor or other parametric cross-sectional structure

when the number of moments is more than around 1/10 the number of data points, S estimates tend to become unstable and near-singular.It might be better to estimate an S imposing a factor structure on all the primitive assets.One might also use a highly structured estimate of S as weighting matrix, while using a less constrained estimate for the standard errors.

Page 53: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

Alternatives to the two-stage procedure: iteration and one-step.

Iterate: we can use this formula

Where is a first-stage estimate, held fixed in the minimization over , then use to find , find

and so on. We can find this estimate serial converge to one value.

12 1{ }ˆ min ( ) ( ) ( )T Tbb g b S b g b−′=

1b2b 2̂b

2̂( )S b1

3 2{ }ˆ min[ ( ) ( ) ( )T Tbb g b S b g b−′=

Page 54: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

This procedure is also likely to produce estimates that do not depend on the initial weighting matrix.Pick b and S simultaneously.When search for b, the S also change. Then the object become into

The first-order conditions are

1

{ }min[ ( ) ( ) ( )]T Tb

g b S b g b−′

11 ( )2*( ) ( ) ( ) ( ) ( ) 0T

T T Tg S bS b g b g b g bb b

−−∂ ∂′ ′+ =

∂ ∂

Page 55: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

In the iteration method, each step involves a numerical search over , may be much quicker to minimize once over

.On the other hand, the latter is not a locally quadratic form, so the search may run into greater numerical difficulties.

( ) ( )T Tg b Sg b′

( ) ( ) ( )T Tg b S b g b′

Page 56: Chapter 11 GMM: General Formulas and Applicationefinance.org.cn/cn/ap/AP11.pdf · 2018. 5. 24. · 11.1 General GMM Formulas GMM procedures can be used to implement a host of estimation

The End

Thank you!