| 1 |
// --+ options: json=compute, stochastic +-- |
| 2 |
|
| 3 |
var y x z v; |
| 4 |
|
| 5 |
varexo ex ey ez ; |
| 6 |
|
| 7 |
parameters a_y_1 a_y_2 b_y_1 b_y_2 b_x_1 b_x_2 d_y; // VAR parameters |
| 8 |
|
| 9 |
parameters beta e_c_m c_z_1 c_z_2; // PAC equation parameters |
| 10 |
|
| 11 |
a_y_1 = .2; |
| 12 |
a_y_2 = .3; |
| 13 |
b_y_1 = .1; |
| 14 |
b_y_2 = .4; |
| 15 |
b_x_1 = -.1; |
| 16 |
b_x_2 = -.2; |
| 17 |
d_y = .5; |
| 18 |
|
| 19 |
|
| 20 |
beta = .9; |
| 21 |
e_c_m = .1; |
| 22 |
c_z_1 = .7; |
| 23 |
c_z_2 = -.3; |
| 24 |
|
| 25 |
var_model(model_name=toto, structural, eqtags=['eq:x', 'eq:y', 'eq:v']); |
| 26 |
|
| 27 |
pac_model(auxiliary_model_name=toto, discount=beta, kind=dd, growth=diff(v(-1)), model_name=pacman); |
| 28 |
|
| 29 |
model(use_dll); |
| 30 |
|
| 31 |
[name='eq:y'] |
| 32 |
diff(y) = .01 + a_y_1*diff(y(-1)) + a_y_2*diff(x(-1)) + b_y_1*diff(y(-2)) + b_y_2*diff(x(-2)) + ey ; |
| 33 |
|
| 34 |
|
| 35 |
[name='eq:x'] |
| 36 |
diff(x) = .05 + b_x_1*diff(y(-2)) + b_x_2*diff(x(-1)) + ex ; |
| 37 |
|
| 38 |
[name='eq:v'] |
| 39 |
diff(v) = diff(x) + d_y*diff(y) ; |
| 40 |
|
| 41 |
[name='zpac'] |
| 42 |
diff(z) = e_c_m*(v(-1)-z(-1)) + c_z_1*diff(z(-1)) + c_z_2*diff(z(-2)) + pac_expectation(pacman) + ez; |
| 43 |
|
| 44 |
end; |
| 45 |
|
| 46 |
shocks; |
| 47 |
var ex = 1.0; |
| 48 |
var ey = 1.0; |
| 49 |
var ez = 1.0; |
| 50 |
end; |
| 51 |
|
| 52 |
// Initialize the PAC model (build the Companion VAR representation for the auxiliary model). |
| 53 |
pac.initialize('pacman'); |
| 54 |
|
| 55 |
// Update the parameters of the PAC expectation model (h0 and h1 vectors). |
| 56 |
pac.update.expectation('pacman'); |
| 57 |
|
| 58 |
// Set initial conditions to zero. Please use more sensible values if any... |
| 59 |
initialconditions = dseries(zeros(10, M_.endo_nbr+M_.exo_nbr), 2000Q1, vertcat(M_.endo_names,M_.exo_names)); |
| 60 |
|
| 61 |
// Simulate the model for 500 periods |
| 62 |
options_.solve_algo = 9; |
| 63 |
TrueData = simul_backward_model(initialconditions, 500); |
| 64 |
|
| 65 |
TrueData.save('example2.data') |
| 66 |
|
| 67 |
// Print expanded PAC_EXPECTATION term. |
| 68 |
pac.print('pacman', 'eq:pac'); |
| 69 |
|
| 70 |
clear eparams |
| 71 |
eparams.e_c_m = .9; |
| 72 |
eparams.c_z_1 = .5; |
| 73 |
eparams.c_z_2 = .2; |
| 74 |
|
| 75 |
// Define the dataset used for estimation |
| 76 |
edata = TrueData; |
| 77 |
edata.ez = dseries(NaN(TrueData.nobs, 1), 2000Q1, 'ez'); |
| 78 |
pac.estimate.nls('zpac', eparams, edata, 2005Q1:2005Q1+400, 'fmincon'); |
| 79 |
|
| 80 |
e_c_m_nls = M_.params(strmatch('e_c_m', M_.param_names, 'exact')); |
| 81 |
c_z_1_nls = M_.params(strmatch('c_z_1', M_.param_names, 'exact')); |
| 82 |
c_z_2_nls = M_.params(strmatch('c_z_2', M_.param_names, 'exact')); |
| 83 |
|
| 84 |
% Check consistency with disaggregated target |
| 85 |
ts1 = dseries('example1.data.mat'); |
| 86 |
ts2 = dseries('example2.data.mat'); |
| 87 |
|
| 88 |
if max(abs(ts1.z.data-ts2.z.data))>1e-12 |
| 89 |
error('Simulations in example1 and example2 are not consistent.') |
| 90 |
end |
| 91 |
|
| 92 |
e1 = load('example1.estimation.mat'); |
| 93 |
|
| 94 |
if abs(e_c_m_nls-e1.e_c_m_nls)>1e-6 || abs(c_z_1_nls-e1.c_z_1_nls)>1e-6 || abs(c_z_2_nls-e1.c_z_2_nls)>1e-6 |
| 95 |
error('Estimations in example1 and example2 are not consistent.') |
| 96 |
end |
| 97 |
|
| 98 |
delete example1.data.mat |
| 99 |
delete example2.data.mat |
| 100 |
|
| 101 |
delete example1.estimation.mat |