skip to main navigation skip to secondary navigation skip to content
Board of Governors of the Federal Reserve System
skip to content

Occasional Staff StudiesThe Anderson-Moore Algorithm (AMA)


Package 'AMA'

February 15, 2012

 

Type: Package

Title: Anderson-Moore Algorith

Version: 1.0.8

Depends: rJava

Date: 2010-10-13

Author: Gary Anderson, Aneesh Raghunandan

Maintainer: Gary Anderson >> gary.anderson@frb.gov

Description: Implements Anderson-Moore algorithm for solving linear rational expectations models. For information about the algorithm and its uses, please see http://www.federalreserve.gov/pubs/oss/oss4/aimindex.html. This version works on both Unix and Windows. For questions about the algorithm and its implementation/applications, please contact gary.anderson@frb.gov. If you are having technical issues with the package, please contact aneesh.raghunandan@yale.edu.

License: GPL 2.0

Copyright: The "C" code, sparseAim.c, implementing the basic algorithm is in the public domain and may be used freely. However, the authors would appreciate acknowledgement of the source by citation of any of the following papers:

Anderson, G. and Moore, G. ``A Linear Algebraic Procedure For Solving Linear Perfect Foresight Models.'' Economics Letters, 17, 1985.

Anderson, G. ``Solving Linear Rational Expectations Models: A Horse Race.'' Computational Economics, 2008, vol. 31, issue 2, pp. 95-113

Anderson, G. ``A Reliable and Computationally Efficient Algorithm for Imposing the Saddle Point Property in Dynamic Models.'' Journal of Economic Dynamics and Control, 2010, vol 34, issue 3, pp. 472-489.

SystemRequirements: Java

LazyLoad: yes

R topics documented:

callAMA - Anderson-Moore Algorithm

Description: Calls the Anderson-Moore Algorithm for the given inputs. callSparseAim is a wrapper to the C and FORTRAN functions which carry out the computation.

Usage:

callAMA(cofh, neq, leads, lags, qmax)

Arguments:

cofh >> Full H matrix, passed to R columnwise. The H matrix is the matrix [H_-T ... H_0 ... H_theta]. Must have neq rows and neq*(leads+lags+1) columns.

neq >> Number of equations in the model. Must equal number of rows in H matrix.

leads >> Number of leads in the model. Must be a positive integer.

lags >> Number of lags in the model. Must be a positive integer.

qmax >> Number of elements to allocate for sparse matrix storage. Optional input. Default is neq * leads * neq * (leads + lags).

Details:

Note: for details about what the H, B, Q, S, script A, script B etc. matrices are, please see [Anderson 2010]. You will need to separately load in the data file in the form of the "H" matrix that is, the matrix [H_-T ... H_0 ... H_theta]. For details on how to obtain this matrix given a model and data, see the papers referenced and use the genHmat function. This utility may be added to this R package in the future. Output will be a list of "list" objects i.e. the B, S, etc. matrices will be returned as one-dimensional list objects, embedded within a list. It is pretty straightforward to get these into matrix form (use outObject[k][[1]] to get the kth output where outObject is the result of callSparseAim). This applies to the B (structural coefficients) and Q (asymptotic linear constraints) matrices. Alternatively, you can also use the genBmat and genQmat functions to get these matrices directly, as matrix objects. You can also obtain the S (observable structure) matrix as well as the stochastic transition matrices, labelled script A and script B in [Anderson 2010].

Value:

The following list object: [cofh, hrows, hcols, neq, leads, lags, nstate, qmax, ptr ,cofb, qmat] Where: The first six elements (cofh, ..., lags) are the same as the inputs nstate = leads + lags + 1 qmax is same as on input ptr is a pointer to the integer corresponding to the error code; 0 denotes successful output. See getReturnCode() to "translate". cofb = reduced form coefficients matrix qmat = asymptotic linear constraints matrix

Note:

It is preferable to ensure that the matrix H is as you want it before passing it as an argument to this function (and to genBmat, genQmat, etc.) However, if you wish to simply load the matrix H from a text file with neq*(nlag+nlead+1) lines, make sure to store the entries column-wise as this is R's default method of storage.

If you are only interested in, for example, the B matrix or the S matrix, then it is not necessary to call this function. Use genBmat or genScof instead--both call this function as well. To extract a matrix from the jth element of the list returned by this function, use result[j][[1]] (where result is the name of the list).

Authors:

Gary Anderson and Aneesh Raghunandan

References:

Gary S. Anderson. "A reliable and computationally efficient algorithm for imposing the saddle point property in dynamic models." Journal of Economic Dynamics & Control, 2010.

Gary Anderson and George Moore. "An Efficient Procedure for Solving Linear Perfect Foresight Models." Unpublished manuscript, Board of Governors of the Federal Reserve System. 1983.

See Also:

genBmat, genScof, genQmat, genHmat, getStochTrans, getFactorMatrices

Examples:

hmat = read.table(system.file("extdata/Lcofbob2.cof",package="AMA"))

hrows = 4

hcols = 12

neq = 4

leads = 1

lags = 1

output <- callAMA(hmat, neq, leads, lags)


example7 - example7.mod - Text input for example

Description:

Contains model definition

Usage:

example7.mod

Format:

Uses modelez format

hmat<-genHmat(system.file("extdata/example7.mod",package="AMA"),system.file("extdata/example7params.prm",package="AMA"))

bmat<-genBmat(hmat,4,1,1)


example7params - example7params.txt - Text input for example

Description:

Contains parameter assignments for the example7 matrix assignments.

Format:

Rows of assignment statements for model parameters.

Examples:

hmat<-genHmat(system.file("extdata/example7.mod",package="AMA"),system.file("extdata/example7params.prm",package="AMA"))

bmat<-genBmat(hmat,4,1,1)


genBmat - Reduced-Form Coefficients Matrix

Description:

Computes the reduced-form coefficients matrix ("B") output by the Anderson-Moore Algorithm. Arguments are identical to those passed to callSparseAim.

Usage:

genBmat(cofh, neq, leads, lags, qmax)

Arguments:

cofh >> Full H matrix, passed to R columnwise. The H matrix is the matrix [H_-T ... H_0 ... H_theta]. Must have neq rows and neq*(leads+lags+1) columns.

neq >> Number of equations in the model. Must equal number of rows in H matrix.

leads >> Number of leads in the model. Must be a positive integer.

lags >> Number of lags in the model. Must be a positive integer.

qmax >> Number of elements to allocate for sparse matrix storage. Optional input. Default is neq * leads * neq * (leads + lags).

Value:

cofb ("B" matrix)

Note:

Alternatively, can be called from the object that callAMA returns; i.e. if res is the output of callAMA, then use "bmat = res[10][[1]]".

References:

Gary S. Anderson. "A reliable and computationally efficient algorithm for imposing the saddle point property in dynamic models." Journal of Economic Dynamics & Control, 2010.

Gary Anderson and George Moore. "An Efficient Procedure for Solving Linear Perfect Foresight Models." Unpublished manuscript, Board of Governors of the Federal Reserve System. 1983.

See Also:

callAMA

Examples:

hmat <- read.table(system.file("extdata/Lcofbob2.cof",package="AMA")) bmat <- genBmat(hmat, 4, 1, 1) hmat <-genHmat(system.file("extdata/example7.mod",package="AMA"),system.file("extdata/example7params.prm",package="AMA")) bmat <-genBmat(hmat,4,1,1)


genHmat - Structural Coefficients Matrix Scripts

Description:

Creates the structural coefficients matrix H, given the model file and parameter values as inputs.

Usage:

genHmat(modelFileNameFull, params, wantParamVec = FALSE)

Arguments:

modelFileNameFull

String object containing the name of the model file. Model file must be in MODELEZ syntax.

params

Either: -A string object containing the name of the parameters file (One parameter per line, with each line in the format "param = ") OR: -A vector containing the parameters. Vector must have names associated with each entry (i.e. names(params) should be correct and complete)

wantParamVec

If TRUE, then the output of the function is a list object containing as its first element the H matrix and as its second element the vector of parameters. Default is FALSE. You should probably set this to TRUE if you're reading in from a parameter file, and leave it as FALSE if you already have a parameter vector.

Details:

Model input file must be in MODELEZ syntax.

Value:

Structural coefficients matrix H.

References:

For the details of MODELEZ syntax, see for example \url{http://www.federalreserve.gov/pubs/oss/oss4/papers/habitMatlab/habitMatlab.html}

For details about the algorithm:

Gary S. Anderson. "A reliable and computationally efficient algorithm for imposing the saddle point property in dynamic models." Journal of Economic Dynamics \& Control, 2010.

Gary Anderson and George Moore. "An Efficient Procedure for Solving Linear Perfect Foresight Models." Unpublished manuscript, Board of Governors of the Federal Reserve System. 1983.

See Also:

callAMA

Examples:

hmat<-genHmat(system.file("extdata/example7.mod",package="AMA"),system.file("extdata/example7params.prm",package="AMA")) bmat<-genBmat(hmat,4,1,1)


genHmatrixScripts - Structural Coefficients Matrix Scripts

Description:

Creates an R file in the present working directory which, when run, creates the matrix H.

Usage:

genHmatrixScripts(modelFileNameFull)

Arguments:

modelFileNameFull >> String object containing the name of the model file. Model file must be in MODELEZ syntax. Details:

Model input file must be in MODELEZ syntax.

Value:

An R script, in the current directory, called "\_sparseAimMatrices.r", which when run generates the matrix H.

References:

For the details of MODELEZ syntax, see for example \url{http://www.federalreserve.gov/pubs/oss/oss4/papers/habitMatlab/habitMatlab.html}

Gary S. Anderson. "A reliable and computationally efficient algorithm for imposing the saddle point property in dynamic models." Journal of Economic Dynamics & Control, 2010.

Gary Anderson and George Moore. "An Efficient Procedure for Solving Linear Perfect Foresight Models." Unpublished manuscript, Board of Governors of the Federal Reserve System. 1983.

See Also:

callAMA Examples:

genHmatrixScripts(system.file("extdata/example7.mod",package="AMA"))

genQmat - Asymptotic Linear Constraints

Description:

Computes the asymptotic linear constraints matrix ("Q") used by the Anderson-Moore Algorithm. Arguments are identical to those passed to callSparseAim.

Usage:

genQmat(cofh, neq, leads, lags, qmax)

Arguments:

 

cofh >> Full H matrix, passed to R columnwise. The H matrix is the matrix [H_-T ... H_0 ... H_theta]. Must have neq rows and neq*(leads+lags+1) columns.

neq >> Number of equations in the model. Must equal number of rows in H matrix.

leads >> Number of leads in the model. Must be a positive integer.

lags >> Number of lags in the model. Must be a positive integer.

qmax >> Number of elements to allocate for sparse matrix storage. Optional input. Default is neq * leads * neq * (leads + lags).

Value:

qmat (asymptotic linear constraints matrix)

Note:

Alternatively, can be called from the object that callAMA returns; i.e. if res is the output of callAMA, then use "qmat = res[11][[1]]".

References:

Gary S. Anderson. "A reliable and computationally efficient algorithm for imposing the saddle point property in dynamic models." Journal of Economic Dynamics \& Control, 2010.

Gary Anderson and George Moore. "An Efficient Procedure for Solving Linear Perfect Foresight Models." Unpublished manuscript, Board of Governors of the Federal Reserve System. 1983.

See Also:

callAMA

Examples:

hmat = read.table(system.file("extdata/Lcofbob2.cof",package="AMA")

) hrows = 4

hcols = 12

neq = 4

leads = 1

lags = 1

Qmatrix <- genQmat(hmat, neq, leads, lags)


genScof - Observable Structure Matrix

Description:

Computes the observable structure matrix ("S") using outputs from the Anderson-Moore Algorithm.

Usage:

genScof(hmat, bmat, neq, leads, lags)

Arguments:

hmat >> Full H matrix, passed to R columnwise. The H matrix is the matrix [H_-T ... H_0 ... H_theta].

bmat >> The B (reduced-form coefficients) matrix.

neq >> Number of equations in the model.

leads >> Number of leads in the model. Must be a positive integer.

lags >> Number of lags in the model. Must be a positive integer.

Value:

cofS (observable structure matrix)

Note:

Alternatively, can be called from the object that callAMA returns; i.e. if res is the output of callAMA, then use "bmat = res[10][[1]]".

References:

Gary S. Anderson. "A reliable and computationally efficient algorithm for imposing the saddle point property in dynamic models." Journal of Economic Dynamics & Control, 2010.

Gary Anderson and George Moore. "An Efficient Procedure for Solving Linear Perfect Foresight Models." Unpublished manuscript, Board of Governors of the Federal Reserve System. 1983.

See Also:

callAMA, genBmat

Examples:

hmat = read.table(system.file("extdata/Lcofbob2.cof",package="AMA"))

hrows = 4

hcols = 12

neq = 4

leads = 1

lags = 1

qmax = 150000

bmat <- genBmat(hmat, neq, leads, lags)

Scoef <- genScof(hmat, bmat, neq, leads, lags)


getFactorMatrices - Shock Factor Matrices for Inhomogeneous Solutions

Description:

Computes the exogenous shock scaling (phi) and transfer (F) matrices used in the inhomogeneous case (i.e.

Usage:

getFactorMatrices(hmat, bmat, neq, leads, lags)

Arguments:

hmat >> Full H matrix, passed to R columnwise. The H matrix is the matrix [H_-T ... H_0 ... H_theta].

Must have neq rows and neq*(leads+lags+1) columns.

bmat >> Full B matrix, as generated either by callSparseAim or genBmat.

neq >> Number of equations in the model. Must equal number of rows in H matrix.

leads >> Number of leads in the model. Must be a positive integer.

lags >> Number of lags in the model. Must be a positive integer.

Value:

list object containing phi as first element and F as second element.

References:

Gary S. Anderson. "A reliable and computationally efficient algorithm for imposing the saddle point property in dynamic models." Journal of Economic Dynamics \& Control, 2010.

Gary Anderson and George Moore. "An Efficient Procedure for Solving Linear Perfect Foresight Models." Unpublished manuscript, Board of Governors of the Federal Reserve System. 1983.

See Also:

callAMA, genBmat

Examples:

hmat = read.table(system.file("extdata/Lcofbob2.cof",package="AMA"))

hrows = 4

hcols = 12

neq = 4

leads = 1

lags = 1

bmat <- genBmat(hmat, neq, leads, lags)

hmat = matrix(as.matrix(hmat), neq)

factorMats <- getFactorMatrices(hmat, bmat, neq, leads, lags)

phi = factorMats[1][[1]]

F = factorMats[2][[1]]


getReturnCode - Return Code

Description:

Computes the asymptotic linear constraints matrix ("Q") used by the Anderson-Moore Algorithm. Arguments are identical to those passed to callSparseAim.

Usage:

getReturnCode(sparseAimObject)

Arguments:

sparseAimObject >> Output object from callSparseAim.

Details:

[to be filled in]

Value:

Return code (corresponds to 0 if call to SparseAIM is successful; prints error message otherwise)

Note:

Integer corresponding to error code can be extracted directly from the object that callSparseAim returns; i.e. if res is the output of callSparseAim, then use "returnCode = res[9][[1]]".

References:

Gary S. Anderson. "A reliable and computationally efficient algorithm for imposing the saddle point property in dynamic models." Journal of Economic Dynamics & Control, 2010.

Gary Anderson and George Moore. "An Efficient Procedure for Solving Linear Perfect Foresight Models." Unpublished manuscript, Board of Governors of the Federal Reserve System. 1983.

See Also:

callAMA

Examples:

hmat = read.table(system.file("extdata/Lcofbob2.cof",package="AMA"))

hrows = 4

hcols = 12

neq = 4

leads = 1

lags = 1

output <- callAMA(hmat, neq, leads, lags)

errorMsg = getReturnCode(output)


getStochTrans - Stochastic Transition Matrices

Description:

Computes the stochastic transition matrices ("script A" and "script B") used by the Anderson-Moore Algorithm. Matrices H and S must already be matrix objects of proper dimension.

Usage:

getStochTrans(hmat, scof)

Arguments:

hmat >> Full H matrix object. The H matrix is the matrix [H_-T ... H_0 ... H_theta]. (i.e. T lags and theta leads)

scof >> Matrix S ("observable structure") .

Value:

A and B are the solution to [x_t-T+1 ... x_t]' = A[x_t-T ...xt-1]' + B[e_t + psi*(E[z_t|I_t] - E[z_t|I_t-1)]

References:

Gary S. Anderson. "A reliable and computationally efficient algorithm for imposing the saddle point property in dynamic models." Journal of Economic Dynamics & Control, 2010.

Gary Anderson and George Moore. "An Efficient Procedure for Solving Linear Perfect Foresight Models." Unpublished manuscript, Board of Governors of the Federal Reserve System. 1983.

See Also:

callAMA, genScof, genBmat

Examples:

hmat = read.table(system.file("extdata/Lcofbob2.cof",package="AMA"))

hrows = 4

hcols = 12

neq = 4

hmatrix = matrix((as.matrix(hmat)), neq)

leads = 1

lags = 1

bmat <- genBmat(hmat, neq, leads, lags)

Scoef <- genScof(hmat, bmat, neq, leads, lags)

stochMatrices <- getStochTrans(hmatrix, Scoef)

scriptA = stochMatrices[1][[1]]

scriptB = stochMatrices[2][[1]]


Lcofbob2.cof - Lcofbob2.cof Example Input data

Description:

Input matrix for example

Usage:

Lcofbob2.cof

Format:

Matrix elements columnwise.

Examples:

hmat = read.table(system.file("extdata/Lcofbob2.cof",package="AMA"))

hrows = 4

hcols = 12

neq = 4

leads = 1

lags = 1

output <- callAMA(hmat, neq, leads, lags)

Last update: August 2, 2013