#!/usr/bin/env python
import numpy as np
import os, sys
path2Pymods = os.path.join(os.path.dirname(__file__), '../../')
if not sys.path.count(path2Pymods):
sys.path.append(path2Pymods)
from pymods.support.wrapLogbook import wrapLogger
verb = 'INFO'
logger = wrapLogger(loggerName='MkovM.log', streamVerb=verb, logFile=None)
class MkovM:
def __init__(self, S, T):
self.S = np.asarray(S)
self.T = np.asarray(T)
self.states = np.size(S, 0)
self.vars = np.size(S, 1)
self.lmbda = self.getlmbda()
self.E = self.getE()
self.Et = self.getEt()
self.V = self.getV()
self.E2 = self.getE2()
self.E2_lag = self.getE2_lag()
self.getCov = self.getV
def __getitem__(self, key):
if key == 'E':
return self.E
if key == 'Et':
def mcInterface(method, uncMean, persistence, varCov, nval, verb = 'DEBUG'):
'''
Inputs:
1) method
2) uncMean as column vector
3) persistence as matrix
4) varCov as matrix
5) nval as column vector
'''
# Get a logger instance
logger = wrapLogger(loggerName = 'mcInterface', streamVerb = verb, logFile = '')
# Make sure that input matrices are not arrays to allow for * notation
# Also enforce column vectors
nVars = len(nval)
uncMean = np.mat(uncMean).reshape(len(uncMean), 1)
persistence = np.mat(persistence)
varCov = np.mat(varCov)
nval = np.mat(nval).reshape(len(nval), 1)
if nval[0, 0] == 1:
ShockMatrix = uncMean.T
TransMatrix = 1
return
# Map unconditional mean into an intercept of a VAR process
I = np.mat(np.eye(len(nval)))
intercept = ( I - persistence ) * uncMean
# Derivation: varEps
# see wave Variance Covariance Vector Operations for rules
# Y_t = alpha + theta Y_{t-1} + eps_{t+1}
# V(Y_t) = V(theta Y_{t-1} + V(eps_{t+1})
# V(Y) = theta V(Y) theta' + V(eps)
# V(eps) = (I-theta I theta') V(Y)
varEps = ( I - persistence * I * persistence.T ) * varCov
logger.debug('uncMean %s' % uncMean)
logger.debug('persistence %s' % persistence)
logger.debug('varCov %s' % varCov)
if method.lower() == 'Tauchen1987Joh'.lower():
Ns = nval[0, 0]
bandwidth = 1
retDict = Tauchen1987Joh(persistence, intercept, varCov, Ns, bandwidth, verb)
ShockMatrix = retDict['y']
TransMatrix = retDict['P']
elif method.lower() == 'KnotekTerry'.lower():
'''
needs 3 shock states to match standard deviation and/or persistence.
'''
A0 = np.eye(nVars)
A1 = intercept
A2 = persistence
sigma = varEps
N = nval
random_draws = 10000
method = 1
[P,states,zbar] = fn_var_to_markov(A0,A1,A2,sigma,N,random_draws,method)
ShockMatrix = states.T
TransMatrix = P
elif method.lower() == 'Tauchen1987fortran'.lower():
'''
doesn't match persistence
'''
pathToFile = os.path.dirname( __file__ )
if not os.path.isfile(os.path.join(os.getcwd(), 'GHQUAD.DAT')):
shutil.copy(os.path.join(pathToFile, 'GHQUAD.DAT'), os.getcwd())
markovInputFile = open('PARMS', 'w')
print >> markovInputFile, len(nval)
print >> markovInputFile, 1
for ele in nval.flat:
print >> markovInputFile, ele
for ele in uncMean.flat:
print >> markovInputFile, ele
for ele in persistence.flat:
print >> markovInputFile, ele
for ele in varCov.flat:
print >> markovInputFile, ele
markovInputFile.close()
states = nval.prod()
vars = len(nval)
TransMatrix, ShockMatrix = markov.markov(states, vars)
os.remove('GHQUAD.DAT')
os.remove('PARMS')
os.remove('dog')
elif method.lower() == 'momentMatching'.lower():
dims = {}
# all variables have the same # of states.
dims['states'] = np.prod(nval)
dims['vars'] = len(nval)
shockTarget = {}
shockTarget['E'] = np.ravel(uncMean)
shockTarget['Std'] = np.asarray(np.sqrt(np.diag(varCov)))
shockTarget['Cor'] = np.asarray([varCov[0, 1] / np.sqrt(varCov[0,0]) * np.sqrt(varCov[1,1])])
S = optShock(dims, shockTarget)
# sort S
ind = np.lexsort((S.T[1], S.T[0]))
ShockMatrix = S[ind].copy()
transTarget = {}
transTarget['Theta'] = persistence
TransMatrix = optTrans(dims, transTarget, S)
else:
print('method not implemented. exiting... ')
os._exit(1)
logger.debug('ShockMatrix %s' % ShockMatrix)
logger.debug('TransMatrix %s' % TransMatrix)
return ShockMatrix, TransMatrix
#!/usr/bin/env python
import numpy as np
import os, sys
path2Pymods = os.path.join(os.path.dirname(__file__), '../../')
if not sys.path.count(path2Pymods):
sys.path.append(path2Pymods)
from pymods.support.wrapLogbook import wrapLogger
verb = 'INFO'
logger = wrapLogger(loggerName = 'MkovM.log', streamVerb = verb, logFile = None)
class MkovM:
def __init__(self, S, T):
self.S = np.asarray(S)
self.T = np.asarray(T)
self.states = np.size(S, 0)
self.vars = np.size(S, 1)
self.lmbda = self.getlmbda()
self.E = self.getE()
self.Et = self.getEt()
self.V = self.getV()
self.E2 = self.getE2()
self.E2_lag = self.getE2_lag()
self.getCov = self.getV
def __getitem__(self, key):
if key == 'E':
return self.E
if key == 'Et':
Et = self.getEt()
def mcInterface(method, uncMean, persistence, varCov, nval, verb='DEBUG'):
'''
Inputs:
1) method
2) uncMean as column vector
3) persistence as matrix
4) varCov as matrix
5) nval as column vector
'''
# Get a logger instance
logger = wrapLogger(loggerName='mcInterface', streamVerb=verb, logFile='')
# Make sure that input matrices are not arrays to allow for * notation
# Also enforce column vectors
nVars = len(nval)
uncMean = np.mat(uncMean).reshape(len(uncMean), 1)
persistence = np.mat(persistence)
varCov = np.mat(varCov)
nval = np.mat(nval).reshape(len(nval), 1)
if nval[0, 0] == 1:
ShockMatrix = uncMean.T
TransMatrix = 1
return
# Map unconditional mean into an intercept of a VAR process
I = np.mat(np.eye(len(nval)))
intercept = (I - persistence) * uncMean
# Derivation: varEps
# see wave Variance Covariance Vector Operations for rules
# Y_t = alpha + theta Y_{t-1} + eps_{t+1}
# V(Y_t) = V(theta Y_{t-1} + V(eps_{t+1})
# V(Y) = theta V(Y) theta' + V(eps)
# V(eps) = (I-theta I theta') V(Y)
varEps = (I - persistence * I * persistence.T) * varCov
logger.debug('uncMean %s' % uncMean)
logger.debug('persistence %s' % persistence)
logger.debug('varCov %s' % varCov)
if method.lower() == 'Tauchen1987Joh'.lower():
Ns = nval[0, 0]
bandwidth = 1
retDict = Tauchen1987Joh(persistence, intercept, varCov, Ns, bandwidth,
verb)
ShockMatrix = retDict['y']
TransMatrix = retDict['P']
elif method.lower() == 'KnotekTerry'.lower():
'''
needs 3 shock states to match standard deviation and/or persistence.
'''
A0 = np.eye(nVars)
A1 = intercept
A2 = persistence
sigma = varEps
N = nval
random_draws = 10000
method = 1
[P, states, zbar] = fn_var_to_markov(A0, A1, A2, sigma, N,
random_draws, method)
ShockMatrix = states.T
TransMatrix = P
elif method.lower() == 'Tauchen1987fortran'.lower():
'''
doesn't match persistence
'''
pathToFile = os.path.dirname(__file__)
if not os.path.isfile(os.path.join(os.getcwd(), 'GHQUAD.DAT')):
shutil.copy(os.path.join(pathToFile, 'GHQUAD.DAT'), os.getcwd())
markovInputFile = open('PARMS', 'w')
print >> markovInputFile, len(nval)
print >> markovInputFile, 1
for ele in nval.flat:
print >> markovInputFile, ele
for ele in uncMean.flat:
print >> markovInputFile, ele
for ele in persistence.flat:
print >> markovInputFile, ele
for ele in varCov.flat:
print >> markovInputFile, ele
markovInputFile.close()
states = nval.prod()
vars = len(nval)
TransMatrix, ShockMatrix = markov.markov(states, vars)
os.remove('GHQUAD.DAT')
os.remove('PARMS')
os.remove('dog')
elif method.lower() == 'momentMatching'.lower():
dims = {}
# all variables have the same # of states.
dims['states'] = np.prod(nval)
dims['vars'] = len(nval)
shockTarget = {}
shockTarget['E'] = np.ravel(uncMean)
shockTarget['Std'] = np.asarray(np.sqrt(np.diag(varCov)))
shockTarget['Cor'] = np.asarray(
[varCov[0, 1] / np.sqrt(varCov[0, 0]) * np.sqrt(varCov[1, 1])])
S = optShock(dims, shockTarget)
# sort S
ind = np.lexsort((S.T[1], S.T[0]))
ShockMatrix = S[ind].copy()
transTarget = {}
transTarget['Theta'] = persistence
TransMatrix = optTrans(dims, transTarget, S)
else:
print('method not implemented. exiting... ')
os._exit(1)
logger.debug('ShockMatrix %s' % ShockMatrix)
logger.debug('TransMatrix %s' % TransMatrix)
return ShockMatrix, TransMatrix