Cao and Nie (2017): Amplification and Asymmetry without Collateral Constraint

Cao and Nie (2017) studies a two-agent model with a borrowing constraint tied to the price of an asset with fixed supply. It is well known that the existence of a collateral constraint amplifies aggregate productivity shocks when the shock pushes the constraint to bind. They show that the incomplete-markets structure of the model by itself can deliver amplified and asymmetric responses of the model to symmetric productivity shocks, even without a collateral constraint.

The model is an example to show that getting the initial starting point for policy iterations and the solution to the boundary of the state space right is crucial in solving the model. These features can be implemented very conveniently with the toolbox.

Please see the paper for the details of the model.

The gmod File

CaoNie2017.gmod

% Parameters
parameters H beta2 beta1 sigma1 sigma2 sigmah eta m phi nu ABad AGood;
H = 1;
beta1 = 0.98;
beta2 = 0.99;
sigma1 = 1;
sigma2 = 1;
sigmah = 1;
eta = 1;
m = 0.89;
phi = 0.057143;
nu = 0.041224;

PrintFreq = 100;
SaveFreq = 100;

% States
var_state X;
XPts = 701;
XMin=0.0;
XMax=1;
X=linspace(XMin,XMax,XPts);

% Shocks
shock_num=3;

var_shock A;
ABad = 0.97;
AGood = 1.03;
A = [ABad 1 AGood];

% From Equation (17)
shock_trans = [
  0.7204 1-0.7204 0.0
  (1-0.872)/2 0.872 (1-0.872)/2
  0.0 1-0.7204 0.7204
  ];

% Tensor variables
var_tensor;

% Variable for the last period
var_policy_init h L;
inbound_init h 1e-20 H;
inbound_init L 1e-20 2;

var_aux_init c cp q;
model_init;
  Y = A*h^nu * L^(1-nu);
  q = nu*A*h^(nu-1)*L^(1-nu);
  w = (1-nu)*A*h^nu*L^(-nu);
  
  c = q*H*X;
  cp = Y-q*H*X;
  hp = H-h;
  
  foc_hp = -1 + phi*hp^(-sigmah)/(q*cp^(-sigma2));
  foc_L = -1 + w*cp^(-sigma2)/(L^(eta-1));
  
  equations;
    foc_hp;
    foc_L;
  end;
end;

% Endogenous variables, bounds, and initial values
var_policy p q c h a cp L mu Xp[3];
inbound p 0 2 adaptive(1.5);
inbound q 0 2 adaptive(1.5);
inbound c 0.0 2 adaptive(1.5);
inbound h 1e-20 1;
inbound a 1e-20 2 adaptive(1.5);
inbound L 1e-20 2 adaptive(1.5);
inbound mu 0.0 1;
inbound cp 0.0 2 adaptive(1.5);
inbound Xp 0.0 1.2;

var_aux Y YTilde b bp hp;

var_interp cFuture cpFuture qFuture;
cFuture = c;
cpFuture = cp;
qFuture = q;
% Initialize using model_init
initial cFuture c;
initial cpFuture cp;
initial qFuture q;

%
var_output Xp Y YTilde p q;

model(X>0);
  Y = A*h^nu * L^(1-nu);
  % eq9
  w = (1-nu)*A * h^nu * L^(-nu);
  profit = (Y-w*L)/h;
  
  [cNext',cpNext',qp'] = GDSGE_INTERP_VEC'(Xp');
  
  Eqp = GDSGE_EXPECT{qp'};
  
  % Transform of variable
  b = a - m*Eqp*h;
  
  % Implied by market clear
  bp = -b;
  hp = H-h;
  
  % Implied by budget constraint
  budget1 = c-(q*X*H + Y-w*L - q*h - p*b);
  budget2 = cp-(q*(1-X)*H + w*L - q*hp - p*bp);
  
  YTilde = Y + phi*hp^(-sigmah)/(cp^(-sigma2)) * hp;

  % Equations in text
  eq10 = -1 + mu*m*Eqp*p/(q-profit) + beta1*GDSGE_EXPECT{qp'*cNext'^(-sigma1)} / ((q-profit)*c^(-sigma1));
  eq11 = mu*a;
  eq12 = -1 + mu + beta1 * GDSGE_EXPECT{ cNext'^(-sigma1) } / (p*c^(-sigma1));
  eq13 = -1 + phi*hp^(-sigmah) / (q*cp^(-sigma2)) + beta2*GDSGE_EXPECT{ qp'*cpNext'^(-sigma2) } / (q*cp^(-sigma2));
  eq14 = -1 + beta2 * GDSGE_EXPECT{ cpNext'^(-sigma2) } / (p*cp^(-sigma2));
  eq15 = -1 + L^(eta-1) / (w*cp^(-sigma2));
  
  % Consistency
  consis1 = (qp(1)*h+b) / (qp(1)*H) / Xp(1) - 1;
  consis2 = (qp(2)*h+b) / (qp(2)*H) / Xp(2) - 1;
  consis3 = (qp(3)*h+b) / (qp(3)*H) / Xp(3) - 1;
  
  equations;
    % budget
    budget1;
    budget2;
  
    % Optimality condition
    eq10;
    eq11;
    eq12;
    eq13;
    eq14;
    eq15;
    
    % Consistency
    if A==AGood
      Xp(1);
    else
      consis1;
    end
    
    consis2;
    
    if A==ABad
      Xp(3);
    else
      consis3;
    end
  end;
end;

model(X==0);
  Y = A*h^nu * L^(1-nu);
  w = (1-nu)*A * h^nu * L^(-nu);
  profit = (Y-w*L)/h;
  
  [cNext',cpNext',qp'] = GDSGE_INTERP_VEC'(Xp');
  
  Eqp = GDSGE_EXPECT{qp'};
  
  % Transform of variable
  b = a - m*Eqp*h;
  
  % Implied by market clear
  bp = -b;
  hp = H-h;
  
  % Implied by budget constraint
  budget2 = cp-(q*(1-X)*H + w*L - q*hp - p*bp);
  
  YTilde = Y + phi*hp^(-sigmah)/(cp^(-sigma2)) * hp;

  % Equations in Guangyu's new note
  eq7 = -1 + phi*hp^(-sigmah) / (q*cp^(-sigma2)) + beta2 * GDSGE_EXPECT{ qp'*cpNext'^(-sigma2) } / (q*cp^(-sigma2));
  eq8 = -1 + beta2 * GDSGE_EXPECT{ cpNext'^(-sigma2) } / (p*cp^(-sigma2));
  eq9 = -1 + L^(eta-1) / (w*cp^(-sigma2));
  eq11 = -1 + (q - profit)/(m*Eqp*p);
  
  % Consistency
  consis1 = (qp(1)*h+b) / (qp(1)*H) / Xp(1) - 1;
  consis2 = (qp(2)*h+b) / (qp(2)*H) / Xp(2) - 1;
  consis3 = (qp(3)*h+b) / (qp(3)*H) / Xp(3) - 1;
  
  equations;
    % budget constraint
    eq11; % This is equivalent to enterpreneur's budget when c and X = 0
    budget2;
  
    % Optimality condition
    c-0;
    mu-0; % I have set mu=0, though mu is not known ex-ante.
    a-0;
    eq7;
    eq8;
    eq9;
    
    % Consistency
    if A==AGood
      Xp(1);
    else
      consis1;
    end
    
    consis2;
    
    if A==ABad
      Xp(3);
    else
      consis3;
    end
  end;
end;

simulate;
  num_periods = 10000;
  num_samples = 100;
  initial X 0.5
  initial shock 1;
  var_simu Y YTilde;
  X' = Xp';
end;