def test_second_order_accuracy(self):
# This solves the optimal growth example at second order
# and computes the second order correction to the steady-state
# We test that both the statefree method and the perturbation to states
# yield the same result.
from dolo.misc.yamlfile import yaml_import
model = yaml_import('examples/global_models/rbc.yaml', compiler=None)
from dolo.numeric.perturbations import solve_decision_rule
from dolo.numeric.perturbations_to_states import approximate_controls
coeffs = approximate_controls(model, order=2, return_dr=False)
state_perturb = coeffs[0]
dr = solve_decision_rule(model)
statefree_perturb = dr['ys'] + dr['g_ss'] / 2.0
ctls = model.symbols_s['controls']
ctls_ind = [model.variables.index(v) for v in ctls]
# the two methods should yield exactly the same result
from numpy.testing import assert_almost_equal
A = statefree_perturb[ctls_ind]
B = state_perturb
assert_almost_equal(A, B) # we compare the risk-adjusted constants
def test_second_order_accuracy(self):
# This solves the optimal growth example at second order
# and computes the second order correction to the steady-state
# We test that both the statefree method and the perturbation to states
# yield the same result.
from dolo.misc.yamlfile import yaml_import
model = yaml_import('examples/global_models/rbc.yaml', compiler=None)
from dolo.numeric.perturbations import solve_decision_rule
from dolo.numeric.perturbations_to_states import approximate_controls
coeffs = approximate_controls(model,order=2, return_dr=False)
state_perturb = coeffs[0]
dr = solve_decision_rule(model)
statefree_perturb = dr['ys'] + dr['g_ss']/2.0
ctls = model.symbols_s['controls']
ctls_ind = [model.variables.index(v) for v in ctls]
# the two methods should yield exactly the same result
from numpy.testing import assert_almost_equal
A = statefree_perturb[ctls_ind]
B = state_perturb
assert_almost_equal(A, B) # we compare the risk-adjusted constants
def test_perturbation(self):
# This solves the optimal growth example at second order
# and computes the second order correction to the steady-state
# We test that both the statefree method and the perturbation to states
# yield the same result.
from dolo.misc.yamlfile import yaml_import
model = yaml_import('../examples/global_models/optimal_growth.yaml')
from dolo.numeric.perturbations_to_states import approximate_controls
[Xbar, X_s, X_ss] = approximate_controls(model, 2)
state_perturb = Xbar
from dolo.numeric.perturbations import solve_decision_rule
dr = solve_decision_rule(model)
statefree_perturb = dr['ys'] + dr['g_ss'] / 2.0
ctls = model['variables_groups']['controls'] + model[
'variables_groups']['expectations']
ctls_ind = [model.variables.index(v) for v in ctls]
# the two methods should yield exactly the same result
from numpy.testing import assert_almost_equal
A = statefree_perturb[ctls_ind]
B = state_perturb
assert_almost_equal(A, B)
def test_perturbation(self):
# This solves the optimal growth example at second order
# and computes the second order correction to the steady-state
# We test that both the statefree method and the perturbation to states
# yield the same result.
from dolo.misc.yamlfile import yaml_import
model = yaml_import('../examples/global_models/optimal_growth.yaml')
from dolo.numeric.perturbations_to_states import approximate_controls
[Xbar, X_s,X_ss] = approximate_controls(model,2)
state_perturb = Xbar
from dolo.numeric.perturbations import solve_decision_rule
dr = solve_decision_rule(model)
statefree_perturb = dr['ys'] + dr['g_ss']/2.0
ctls = model['variables_groups']['controls'] + model['variables_groups']['expectations']
ctls_ind = [model.variables.index(v) for v in ctls]
# the two methods should yield exactly the same result
from numpy.testing import assert_almost_equal
A = statefree_perturb[ctls_ind]
B = state_perturb
assert_almost_equal(A, B)
def perturbation_solution(self,with_bounds=False):
"""
Returns perturbation solution around the steady-state
Result is [X,Y,Z] representing :
s_t = X s_{t-1} + Y e_t
x_t = Z s_t
"""
model = self.model
data = self.read_model()
controls_i = [model.variables.index(v) for v in data['controls'] ]
states_i = [model.variables.index(v) for v in data['states_vars'] ]
from dolo.numeric.perturbations import solve_decision_rule
dr = solve_decision_rule(model,order=1)
A = dr['g_a'][states_i,:]
B = dr['g_a'][controls_i,:]
[Z,err,rank,junk] = np.linalg.lstsq(A.T,B.T)
Z = Z.T
X = A[:,states_i] + np.dot( A[:,controls_i], Z )
Y = dr.ghu[states_i,:]
# steady_state values
s_ss = dr['ys'][states_i,]
x_ss = dr['ys'][controls_i,]
# We also compute bounds for distribution at a given probability interval
if with_bounds:
M = np.linalg.solve( 1-X, Y)
Sigma = np.array(model.covariances).astype(np.float64)
[V,P] = np.linalg.eigh(Sigma)
# we have Sigma == P * diag(V) * P.T
# unconditional distribution is ( P*diag(V) ) * N
# where N is the normal distribution associated
H = np.dot(Y,np.dot(P,np.diag(V)))
n_s = Sigma.shape[0]
I = np.eye(n_s)
points = np.concatenate([H,-H],axis=1)
lam = 2 # this coefficient should be more cleverly defined
p_infs = np.min(points,axis = 1) * lam
p_max = np.max(points,axis = 1) * lam
bounds = np.row_stack([
s_ss + p_infs,
s_ss + p_max
])
else:
bounds = None
return [[s_ss,x_ss],[X,Y,Z],bounds]
def perturbation_solution(self,with_bounds=False):
"""
Returns perturbation solution around the steady-state
Result is [X,Y,Z] representing :
s_t = X s_{t-1} + Y e_t
x_t = Z s_t
"""
model = self.model
data = self.read_model()
controls_i = [model.variables.index(v) for v in data['controls'] ]
states_i = [model.variables.index(v) for v in data['states_vars'] ]
from dolo.numeric.perturbations import solve_decision_rule
dr = solve_decision_rule(model,order=1)
A = dr['g_a'][states_i,:]
B = dr['g_a'][controls_i,:]
[Z,err,rank,junk] = np.linalg.lstsq(A.T,B.T)
Z = Z.T
X = A[:,states_i] + np.dot( A[:,controls_i], Z )
Y = dr.ghu[states_i,:]
# steady_state values
s_ss = dr['ys'][states_i,]
x_ss = dr['ys'][controls_i,]
# We also compute bounds for distribution at a given probability interval
if with_bounds:
M = np.linalg.solve( 1-X, Y)
Sigma = np.array(model.covariances).astype(np.float64)
[V,P] = np.linalg.eigh(Sigma)
# we have Sigma == P * diag(V) * P.T
# unconditional distribution is ( P*diag(V) ) * N
# where N is the normal distribution associated
H = np.dot(Y,np.dot(P,np.diag(V)))
n_s = Sigma.shape[0]
I = np.eye(n_s)
points = np.concatenate([H,-H],axis=1)
lam = 2 # this coefficient should be more cleverly defined
p_infs = np.min(points,axis = 1) * lam
p_max = np.max(points,axis = 1) * lam
bounds = np.row_stack([
s_ss + p_infs,
s_ss + p_max
])
else:
bounds = None
return [[s_ss,x_ss],[X,Y,Z],bounds]
def check(self):
var_string = self.ui.lineEdit_var.text()
shocks_string = self.ui.lineEdit_shocks.text()
parameters_string = self.ui.lineEdit_parameters.text()
params_definition = self.ui.plainTextEdit.toPlainText()
initval_string = self.ui.plainTextEdit_2.toPlainText()
l = [el.get_text() for el in self.widgets]
l = [e for e in l if e]
content = str.join(';\n', l)
if len(l) > 0:
content += ';'
simple_mod = '''
var {vars};
varexo {varexo};
parameters {parms};
{params_definition}
model;
{equations}
end;
initval;
{initval_string}
end;
'''.format(vars=var_string,
varexo=shocks_string,
parms=parameters_string,
equations=content,
params_definition=params_definition,
initval_string=initval_string)
from dolo.misc.modfile import parse_dynare_text
try:
model = parse_dynare_text(simple_mod)
model.check(verbose=True)
info = model.info
info['name'] = model.fname
txt = '''
Model check {name}:
Number of variables : {n_variables}
Number of equations : {n_equations}
Number of shocks : {n_shocks}'''.format(**info)
except Exception as e:
txt = '\nModel is not valid'
self.ui.textEdit.setText(txt)
return
try:
from dolo.numeric.perturbations import solve_decision_rule
dr = solve_decision_rule(model, order=1)
txt += '\nBlanchard-Kahn conditions are met.'
except ImportError as e:
txt += '\nImpossible to solve the model (lapack could not be imported)'
except Exception as e:
txt += '\nImpossible to solve the model (yet).'
print e
self.ui.textEdit.setText(txt)
def check(self):
var_string = self.ui.lineEdit_var.text()
shocks_string = self.ui.lineEdit_shocks.text()
parameters_string = self.ui.lineEdit_parameters.text()
params_definition = self.ui.plainTextEdit.toPlainText()
initval_string = self.ui.plainTextEdit_2.toPlainText()
l = [ el.get_text() for el in self.widgets]
l = [ e for e in l if e]
content = str.join(';\n',l)
if len(l)>0:
content += ';'
simple_mod = '''
var {vars};
varexo {varexo};
parameters {parms};
{params_definition}
model;
{equations}
end;
initval;
{initval_string}
end;
'''.format(vars = var_string,varexo = shocks_string,
parms=parameters_string,equations = content,
params_definition=params_definition, initval_string=initval_string)
from dolo.misc.modfile import parse_dynare_text
try:
model = parse_dynare_text(simple_mod)
model.check(verbose=True)
info = model.info
info['name'] = model.fname
txt ='''
Model check {name}:
Number of variables : {n_variables}
Number of equations : {n_equations}
Number of shocks : {n_shocks}'''.format(**info)
except Exception as e:
txt = '\nModel is not valid'
self.ui.textEdit.setText(txt)
return
try:
from dolo.numeric.perturbations import solve_decision_rule
dr = solve_decision_rule( model, order = 1 )
txt += '\nBlanchard-Kahn conditions are met.'
except ImportError as e:
txt += '\nImpossible to solve the model (lapack could not be imported)'
except Exception as e:
txt += '\nImpossible to solve the model (yet).'
print e
self.ui.textEdit.setText(txt)
