Errors When Running Portfolio-Choice Models

1

Dear Robert,

I have started using your VFI toolkit again, and it’s great! I am finishing my PhD and decided to use AWS g4dn to run your codes, which require a GPU. My research investigates the interplay between homeownership and stock demand, so I am particularly interested in portfolio-choice models (models 31-35 and the example CGM2005), especially the one with housing (model 35).

I noticed that the codes for models 31-35 and CGM2005 are ending with errors. Could you please take a look and see what might be going wrong? I have pasted the error message below.

Thank you very much!

1 Like
2

Model 31

Now solve the value function iteration problem, just to check that things are working before we go to General Equilbrium
disp(‘Test ValueFnIter’)
tic;
[V, Policy]=ValueFnIter_Case1_FHorz(n_d,n_a,n_z,N_j,d_grid, a_grid, z_grid, pi_z, ReturnFn, Params, DiscountFactorParamNames, , vfoptions);
toc

% V is now (a,z,j). This was already true, just that previously z was trivial (a single point)
% Compare
size(V)
% with
[n_a,n_z,N_j]
% there are the same.
% Policy is
size(Policy)
% which is the same as
[length(n_d),n_a,n_z,N_j]
% The n_a,n_z,N_j represent the state on which the decisions/policys
% depend, and there is one decision for each decision variable ‘d’
Test ValueFnIter
Unrecognized function or variable ‘N_a2’.

Error in CreateaprimeFnMatrix_RiskyAsset (line 144)
if N_dN_uN_a2<1000000

Error in ValueFnIter_Case1_FHorz_RiskyAsset_noa1_raw (line 213)
[aprimeIndex,aprimeProbs]=CreateaprimeFnMatrix_RiskyAsset(aprimeFn, n_d, n_a, n_u, d_grid, a_grid, u_grid, aprimeFnParamsVec,1); % Note, is actually aprime_grid (but a_grid is anyway same for all ages)

Error in ValueFnIter_Case1_FHorz_RiskyAsset (line 83)
[VKron, PolicyKron]=ValueFnIter_Case1_FHorz_RiskyAsset_noa1_raw(n_d,n_a2,n_z,n_u, N_j, d_grid, a2_grid, z_gridvals_J, u_grid, pi_z_J, pi_u, ReturnFn, aprimeFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, aprimeFnParamNames, vfoptions);

Error in ValueFnIter_Case1_FHorz (line 571)
[V,Policy]=ValueFnIter_Case1_FHorz_RiskyAsset(n_d,n_a1,n_a2,n_z,vfoptions.n_u, N_j, d_grid, a1_grid, a2_grid, z_gridvals_J, vfoptions.u_grid, pi_z_J, vfoptions.pi_u, ReturnFn, vfoptions.aprimeFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, vfoptions);

3

Model 32

Again at the same place:

Unrecognized function or variable ‘N_a2’.

4

Model 33

%% Now solve the value function iteration problem, just to check that things are working before we go to General Equilbrium

disp(‘Test ValueFnIter’)

tic;

[V, Policy]=ValueFnIter_Case1_FHorz(n_d,n_a,n_z,N_j,d_grid, a_grid, z_grid, pi_z, ReturnFn, Params, DiscountFactorParamNames, , vfoptions);

toc

% V is now (a,z,j). This was already true, just that previously z was trivial (a single point)

% Compare

size(V)

% with

[n_a,n_z,N_j]

% there are the same.

% Policy is

size(Policy)

% which is the same as

[length(n_d),n_a,n_z,N_j]

% The n_a,n_z,N_j represent the state on which the decisions/policys

% depend, and there is one decision for each decision variable ‘d’

Test ValueFnIter

Unrecognized function or variable ‘l_a’.

Error in ValueFnIter_Case1_FHorz (line 406)

vfoptions.WarmGlowBequestsFnParamsNames={temp{l_a+1:end}}; % the first inputs will always be aprime

5

Model 34

Now solve the value function iteration problem, just to check that things are working before we go to General Equilbrium
disp(‘Test ValueFnIter’)
tic;
[V, Policy]=ValueFnIter_Case1_FHorz(n_d,n_a,n_z,N_j,d_grid, a_grid, z_grid, pi_z, ReturnFn, Params, DiscountFactorParamNames, , vfoptions);
toc

% V is now (a,z,j). This was already true, just that previously z was trivial (a single point)
% Compare
size(V)
% with
[n_a,n_z,N_j]
% there are the same.
% Policy is
size(Policy)
% which is the same as
[length(n_d),n_a,n_z,N_j]
% The n_a,n_z,N_j represent the state on which the decisions/policys
% depend, and there is one decision for each decision variable ‘d’
Test ValueFnIter
Error using .*
Out of memory on device. To view more detail about available memory on the GPU, use ‘gpuDevice()’. If the problem persists,
reset the GPU by calling ‘gpuDevice(1)’.

Error in ValueFnIter_FHorz_RiskyAsset_EpsteinZin_noa1_raw (line 95)
becareful=logical(isfinite(ReturnMatrix).*(ReturnMatrix~=0)); % finite and not zero

Error in ValueFnIter_Case1_FHorz_EpsteinZin_RiskyAsset (line 165)
[VKron, PolicyKron]=ValueFnIter_FHorz_RiskyAsset_EpsteinZin_noa1_raw(n_d,n_a2,n_z,n_u, N_j, d_grid, a2_grid, z_gridvals_J, u_grid, pi_z_J, pi_u, ReturnFn, aprimeFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, aprimeFnParamNames, vfoptions, sj, warmglow, ezc1,ezc2,ezc3,ezc4,ezc5,ezc6,ezc7,ezc8);

Error in ValueFnIter_Case1_FHorz_RiskyAsset (line 54)
[V, Policy]=ValueFnIter_Case1_FHorz_EpsteinZin_RiskyAsset(n_d,n_a1,n_a2,n_z,n_u,N_j,d_grid,a1_grid, a2_grid, z_gridvals_J, u_grid, pi_z_J, pi_u, aprimeFn, ReturnFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, aprimeFnParamNames, vfoptions);

Error in ValueFnIter_Case1_FHorz (line 571)
[V,Policy]=ValueFnIter_Case1_FHorz_RiskyAsset(n_d,n_a1,n_a2,n_z,vfoptions.n_u, N_j, d_grid, a1_grid, a2_grid, z_gridvals_J, vfoptions.u_grid, pi_z_J, vfoptions.pi_u, ReturnFn, vfoptions.aprimeFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, vfoptions);

6

Model 35 (similar to 31 and 32)

Now solve the value function iteration problem, just to check that things are working before we go to General Equilbrium
disp(‘Test ValueFnIter’)
vfoptions.verbose=1;
tic;
[V, Policy]=ValueFnIter_Case1_FHorz(n_d,n_a,n_z,N_j,d_grid, a_grid, z_grid, pi_z, ReturnFn, Params, DiscountFactorParamNames, , vfoptions);
toc

% V is now (a,z,j). This was already true, just that previously z was trivial (a single point)
% Compare
size(V)
% with
[n_a,n_z,N_j]
% there are the same.
% Policy is
size(Policy)
% which is the same as
[length(n_d)+1,n_a,n_z,N_j]
% The n_a,n_z,N_j represent the state on which the decisions/policys
% depend, and there is one decision for each decision variable ‘d’ plus one
% more for the standard asset
Test ValueFnIter

vfoptions =

struct with fields:

    survivalprobability: 'sj'
             riskyasset: 1
      exoticpreferences: 'EpsteinZin'
      EZpositiveutility: 0
         EZriskaversion: 'phi'
               aprimeFn: @(savings,riskyshare,u,r)LifeCycleModel31_aprimeFn(savings,riskyshare,u,r)
                    n_u: 5
                 u_grid: [5×1 gpuArray]
                   pi_u: [5×1 gpuArray]
                verbose: 1
               parallel: 2
           returnmatrix: 2
              lowmemory: 0
               paroverz: 1
       divideandconquer: 0
        incrementaltype: 0
            polindorval: 1
             outputkron: 0
policy_forceintegertype: 0
                dynasty: 0
        experienceasset: 0
       experienceassetu: 0
          residualasset: 0

vfoptions =

struct with fields:

    survivalprobability: 'sj'
             riskyasset: 1
      exoticpreferences: 'EpsteinZin'
      EZpositiveutility: 0
         EZriskaversion: 'phi'
               aprimeFn: @(savings,riskyshare,u,r)LifeCycleModel31_aprimeFn(savings,riskyshare,u,r)
                    n_u: 5
                 u_grid: [5×1 gpuArray]
                   pi_u: [5×1 gpuArray]
                verbose: 1
               parallel: 2
           returnmatrix: 2
              lowmemory: 0
               paroverz: 1
       divideandconquer: 0
        incrementaltype: 0
            polindorval: 1
             outputkron: 0
policy_forceintegertype: 0
                dynasty: 0
        experienceasset: 0
       experienceassetu: 0
          residualasset: 0

Finite horizon: 80 of 81
Unrecognized function or variable ‘N_a2’.

Error in CreateaprimeFnMatrix_RiskyAsset (line 144)
if N_dN_uN_a2<1000000

Error in ValueFnIter_FHorz_RiskyAsset_EpsteinZin_raw (line 418)
[a2primeIndex,a2primeProbs]=CreateaprimeFnMatrix_RiskyAsset(aprimeFn, n_d, n_a2, n_u, d_grid, a2_grid, u_grid, aprimeFnParamsVec,2); % Note, is actually aprime_grid (but a_grid is anyway same for all ages)

Error in ValueFnIter_Case1_FHorz_EpsteinZin_RiskyAsset (line 171)
[VKron, PolicyKron]=ValueFnIter_FHorz_RiskyAsset_EpsteinZin_raw(n_d,n_a1,n_a2,n_z,n_u, N_j, d_grid, a1_grid,a2_grid, z_gridvals_J, u_grid, pi_z_J, pi_u, ReturnFn, aprimeFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, aprimeFnParamNames, vfoptions, sj, warmglow, ezc1,ezc2,ezc3,ezc4,ezc5,ezc6,ezc7,ezc8);

Error in ValueFnIter_Case1_FHorz_RiskyAsset (line 54)
[V, Policy]=ValueFnIter_Case1_FHorz_EpsteinZin_RiskyAsset(n_d,n_a1,n_a2,n_z,n_u,N_j,d_grid,a1_grid, a2_grid, z_gridvals_J, u_grid, pi_z_J, pi_u, aprimeFn, ReturnFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, aprimeFnParamNames, vfoptions);

Error in ValueFnIter_Case1_FHorz (line 571)
[V,Policy]=ValueFnIter_Case1_FHorz_RiskyAsset(n_d,n_a1,n_a2,n_z,vfoptions.n_u, N_j, d_grid, a1_grid, a2_grid, z_gridvals_J, vfoptions.u_grid, pi_z_J, vfoptions.pi_u, ReturnFn, vfoptions.aprimeFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, vfoptions);

7

CMG2005

%% Solve for the value function and policy fn
% vfoptions.lowmemory=1;
vfoptions.verbose=1;
tic;
[V, Policy]=ValueFnIter_Case1_FHorz_PType(n_d,n_a,n_z,N_j,Names_i, d_grid, a_grid, z_grid, pi_z, ReturnFn, Params, DiscountFactorParamNames,vfoptions);
toc
Permanent type: 1 of 3
Error using ValueFnIter_Case1_FHorz (line 165)
No longer use vfoptions.e_grid_J, instead just put the age-dependent grid in vfoptions.e_grid (functionality of VFI Toolkit has
changed to make it easier to use)

Error in ValueFnIter_Case1_FHorz_PType (line 145)
[V_ii, Policy_ii]=ValueFnIter_Case1_FHorz(n_d_temp,n_a_temp,n_z_temp,N_j_temp,d_grid_temp, a_grid_temp, z_grid_temp, pi_z_temp, ReturnFn_temp, Parameters_temp, DiscountFactorParamNames_temp, , vfoptions_temp);

8

I will take a look later this week and get back to you.

1 Like
9

@bob_rb, I think they are all fixed (update both VFI Toolkit and the Intro to Life-Cycle models to the latest github). Thanks for bringing to my attention.

I had removed the “Case3” codes (which is what “riskyasset” was called before I decided that “riskyasset” was a better approach) and there were still some vestiges of it in use that error-ed as a result.

1 Like
10

That AWS instance is pretty nice. The T4 is a solid gpu. I see AWS actually let you have 8 T4’s in a single instance. Unfortunately VFI Toolkit is not yet at the point it can exploit multiple GPUs but in principle I think Matlab now can and it would be really nice for solving multiple permanent types across multiple GPUs all at once. Something I want to try if I ever get hold of a setup with multiple GPUs.

2 Likes
11

Thank you!

Models 31, 32, 33, and 35 now work smoothly for me, and I will start studying them now. However, the errors for Model 34 and CGM2005 still persist.

Model 34

From what I have read, there seems to be a memory issue on my device.

Test ValueFnIter
Error using .*
Out of memory on device. To view more detail about available memory on the GPU, use ‘gpuDevice()’. If the problem persists, reset the
GPU by calling ‘gpuDevice(1)’.

Error in ValueFnIter_FHorz_RiskyAsset_EpsteinZin_noa1_raw (line 95)
becareful=logical(isfinite(ReturnMatrix).*(ReturnMatrix~=0)); % finite and not zero

Error in ValueFnIter_Case1_FHorz_EpsteinZin_RiskyAsset (line 165)
[VKron, PolicyKron]=ValueFnIter_FHorz_RiskyAsset_EpsteinZin_noa1_raw(n_d,n_a2,n_z,n_u, N_j, d_grid, a2_grid, z_gridvals_J, u_grid, pi_z_J, pi_u, ReturnFn, aprimeFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, aprimeFnParamNames, vfoptions, sj, warmglow, ezc1,ezc2,ezc3,ezc4,ezc5,ezc6,ezc7,ezc8);

Error in ValueFnIter_Case1_FHorz_RiskyAsset (line 54)
[V, Policy]=ValueFnIter_Case1_FHorz_EpsteinZin_RiskyAsset(n_d,n_a1,n_a2,n_z,n_u,N_j,d_grid,a1_grid, a2_grid, z_gridvals_J, u_grid, pi_z_J, pi_u, aprimeFn, ReturnFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, aprimeFnParamNames, vfoptions);

Error in ValueFnIter_Case1_FHorz (line 572)
[V,Policy]=ValueFnIter_Case1_FHorz_RiskyAsset(n_d,n_a1,n_a2,n_z,vfoptions.n_u, N_j, d_grid, a1_grid, a2_grid, z_gridvals_J, vfoptions.u_grid, pi_z_J, vfoptions.pi_u, ReturnFn, vfoptions.aprimeFn, Parameters, DiscountFactorParamNames, ReturnFnParamNames, vfoptions);

Error in LifeCycleModel34 (line 140)
[V, Policy]=ValueFnIter_Case1_FHorz(n_d,n_a,n_z,N_j,d_grid, a_grid, z_grid, pi_z, ReturnFn, Params, DiscountFactorParamNames, , vfoptions);

I am using an AWS g4dn.xlarge instance with the following specifications:

I do not know where is the problem. Should I use stronger instance?

Model CGM2005

It runs for a few minutes, and then I receive the following message:

CoccoGomesMaenhout2005
Warning: Colon operands must be real scalars. This warning will become an error in a future release.
In CoccoGomesMaenhout2005 (line 100)

Warning: Colon operands must be real scalars. This warning will become an error in a future release.

In CoccoGomesMaenhout2005 (line 103)

Warning: Colon operands must be real scalars. This warning will become an error in a future release.

In CoccoGomesMaenhout2005 (line 106)

discretizeLifeCycleAR1_Kirkby: 1 moment in 0.00 cases, 2 moments in 0.13 cases, 3 moments in 0.03 cases, 4 moments in 0.84 cases (target was 4 moments)
discretizeLifeCycleAR1_Kirkby: 1 moment in 0.00 cases, 2 moments in 0.12 cases, 3 moments in 0.03 cases, 4 moments in 0.85 cases (target was 4 moments)
discretizeLifeCycleAR1_Kirkby: 1 moment in 0.00 cases, 2 moments in 0.12 cases, 3 moments in 0.03 cases, 4 moments in 0.85 cases (target was 4 moments)
Permanent type: 1 of 3
Error using ValueFnIter_Case1_FHorz (line 165)
No longer use vfoptions.e_grid_J, instead just put the age-dependent grid in vfoptions.e_grid (functionality of VFI Toolkit has
changed to make it easier to use)

Error in ValueFnIter_Case1_FHorz_PType (line 145)
[V_ii, Policy_ii]=ValueFnIter_Case1_FHorz(n_d_temp,n_a_temp,n_z_temp,N_j_temp,d_grid_temp, a_grid_temp, z_grid_temp, pi_z_temp, ReturnFn_temp, Parameters_temp, DiscountFactorParamNames_temp, , vfoptions_temp);

Error in CoccoGomesMaenhout2005 (line 292)
[V, Policy]=ValueFnIter_Case1_FHorz_PType(n_d,n_a,n_z,N_j,Names_i, d_grid, a_grid, z_grid, pi_z, ReturnFn, Params, DiscountFactorParamNames,vfoptions);

12

Yes, so far, it seems to work very well. I am using an AWS g4dn.xlarge instance with the following specifications:

gpuDevice()

ans =

CUDADevice with properties:

                  Name: 'Tesla T4'
                 Index: 1
     ComputeCapability: '7.5'
        SupportsDouble: 1
 GraphicsDriverVersion: '560.94'
           DriverModel: 'TCC'
        ToolkitVersion: 12.2000
    MaxThreadsPerBlock: 1024
      MaxShmemPerBlock: 49152 (49.15 KB)
    MaxThreadBlockSize: [1024 1024 64]
           MaxGridSize: [2.1475e+09 65535 65535]
             SIMDWidth: 32
           TotalMemory: 15928852480 (15.93 GB)
       AvailableMemory: 15793710766 (15.79 GB)
           CachePolicy: 'balanced'
   MultiprocessorCount: 40
          ClockRateKHz: 1590000
           ComputeMode: 'Default'
  GPUOverlapsTransfers: 1
KernelExecutionTimeout: 0
      CanMapHostMemory: 1
       DeviceSupported: 1
       DeviceAvailable: 1
        DeviceSelected: 1

This is the weakest GPU solution available on AWS. I recommend it to those who want to run your codes with the VFI toolkit but lack a GPU. Although this solution is not free, it is reasonably priced if you need it.

1 Like
13

Is the CGM2005 code you ran from

(just making sure I check the right code. if not, where did you get the code?)

14

Yeah, that is a GPU out of memory error in life-cycle model 34 (you can make the grids smaller just editing n_d and n_a to be smaller numbers; should run then but obviously won’t be so accurate).

1 Like
15

Yes, it is the code in your link. I run it now again and get the same error.

clear
CoccoGomesMaenhout2005
Warning: Colon operands must be real scalars. This warning will become an error in a future release.
In CoccoGomesMaenhout2005 (line 100)

Warning: Colon operands must be real scalars. This warning will become an error in a future release.

In CoccoGomesMaenhout2005 (line 103)

Warning: Colon operands must be real scalars. This warning will become an error in a future release.

In CoccoGomesMaenhout2005 (line 106)

discretizeLifeCycleAR1_Kirkby: 1 moment in 0.00 cases, 2 moments in 0.13 cases, 3 moments in 0.03 cases, 4 moments in 0.84 cases (target was 4 moments)
discretizeLifeCycleAR1_Kirkby: 1 moment in 0.00 cases, 2 moments in 0.12 cases, 3 moments in 0.03 cases, 4 moments in 0.85 cases (target was 4 moments)
discretizeLifeCycleAR1_Kirkby: 1 moment in 0.00 cases, 2 moments in 0.12 cases, 3 moments in 0.03 cases, 4 moments in 0.85 cases (target was 4 moments)
Permanent type: 1 of 3
Error using ValueFnIter_Case1_FHorz (line 165)
No longer use vfoptions.e_grid_J, instead just put the age-dependent grid in vfoptions.e_grid (functionality of VFI Toolkit has
changed to make it easier to use)

Error in ValueFnIter_Case1_FHorz_PType (line 145)
[V_ii, Policy_ii]=ValueFnIter_Case1_FHorz(n_d_temp,n_a_temp,n_z_temp,N_j_temp,d_grid_temp, a_grid_temp, z_grid_temp, pi_z_temp, ReturnFn_temp, Parameters_temp, DiscountFactorParamNames_temp, , vfoptions_temp);

Error in CoccoGomesMaenhout2005 (line 292)
[V, Policy]=ValueFnIter_Case1_FHorz_PType(n_d,n_a,n_z,N_j,Names_i, d_grid, a_grid, z_grid, pi_z, ReturnFn, Params, DiscountFactorParamNames,vfoptions);

16

I will fix CGM2005 sometime tomorrow (likely) and let you know once done.

1 Like
17

@bob_rb CGM2005 should work now.

1 Like
18

Thank you, but unfortunately, I receive the following error after running the updated CGM2005 code:

CoccoGomesMaenhout2005
Warning: Colon operands must be real scalars. This warning will become an error in a future release.
In CoccoGomesMaenhout2005 (line 100)

Warning: Colon operands must be real scalars. This warning will become an error in a future release.

In CoccoGomesMaenhout2005 (line 103)

Warning: Colon operands must be real scalars. This warning will become an error in a future release.

In CoccoGomesMaenhout2005 (line 106)

Unrecognized function or variable ‘discretizeLifeCycleAR1_KFTT’.

Error in CoccoGomesMaenhout2005 (line 163)
[z_grid_J.College, pi_z_J.College,jequaloneDistP,otheroutputs] = discretizeLifeCycleAR1_KFTT(zeros(1,N_j.College),Params.rho_zones(1,N_j.College),Params.sigma_epsilonz.Collegeones(1,N_j.College),n_z,N_j.College,kirkbyoptions);

19

Ah okay, apparently that “Colon operands” warning is something to do with the latest Matlab version. I will update and see sometime in next few weeks.

“Unrecognized function or variable ‘discretizeLifeCycleAR1_KFTT’.”
This error is probably that you need to update VFI Toolkit to latest github version.

2 Likes
20

OK, I will be able to try the code on the updated VFI Toolkit tomorrow morning.