More than four state variables with time iteration

[TimMunday]

Hello -

I'm trying to solve a non-linear model which has some lagged variables in it. I was getting an error message putting them in directly into the 'arbitrage' section, so I created new state variables, and added them to the 'transition' section.

Whilst this seems to be fine for perturbation methods, once more than 4 state variables are in the transition section I get a complicated error message. Is this because time iteration doesn't support more than 4 states?

Since github won't let me attach a .yaml, below is a very simple minimum workable example. Its a 3 eq linear new keynesian model where the taylor rule uses some lagged variables (the actual model I'm solving is non-linear, hence the desire to use time iteration). I also include the error message below as well.

Any help much appreciated. Great library

Tim

# Minimum workable example for lagged vars

name: MWE

symbols:
  states: [pi_s, r_s, x_1, i_1, i_2]
  controls: [pi, x, i]
  exogenous: [eps_pi, eps_r]
  parameters: [bbeta, kkappa, ssigma, rho_pi]

equations:

  arbitrage:
    - pi = bbeta*pi(+1) + kkappa*x + pi_s 
    - x = -(1/ssigma)*(i - pi(+1)) + x(+1) + r_s
    - i = 2.5*pi + x_1 + 0.25*i_1 + 0.1*i_2

  transition:
    - pi_s = rho_pi*pi_s(-1) + eps_pi
    - r_s = 0.7*r_s(-1) + eps_r
    - x_1 = x(-1)
    - i_1 = i(-1)
    - i_2 = i_1(-1)

calibration:
  bbeta : 0.9
  kkappa : 0.4
  ssigma : 2
  rho_pi : 0.7

  # ss vals
  x_1 : 0
  i_1 : 0
  i_2 : 0
  pi : 0
  x : 0
  i : 0
  pi_s : 0
  r_s : 0
  volrs : 0

exogenous: !Normal
  Sigma: [[0.1, 0], [0, 0.1]]

domain:
  pi_s : [-0.1, 0.1]
  r_s : [-0.1, 0.1]
  x_1 : [-0.3, 0.3]
  i_1 : [-0.3, 0.3]
  i_2 : [-0.3, 0.3]

options:
  grid: !CartesianGrid
    orders: [5, 5, 5, 5, 5]

The error message takes the form:

TypingError: Failed in nopython mode pipeline (step: nopython frontend)
Invalid use of Function(<function _filter_cubic at 0x0000024A0C8E6168>) with argument(s) of type(s): (UniTuple(Tuple(float64, float64, int32) x 5), array(float64, 6d, C))
 * parameterized
In definition 0:
    TypeError: __filter_cubic() missing 1 required positional argument: 'C'
    raised from C:\Users\tmund\Anaconda3\envs\Taylor\lib\site-packages\numba\core\typing\templates.py:579
In definition 1:
    TypeError: __filter_cubic() missing 1 required positional argument: 'C'
    raised from C:\Users\tmund\Anaconda3\envs\Taylor\lib\site-packages\numba\core\typing\templates.py:579
In definition 2:
    UnboundLocalError: local variable '___filter_cubic' referenced before assignment
    raised from C:\Users\tmund\Anaconda3\envs\Taylor\lib\site-packages\interpolation\splines\prefilter_cubic.py:293
In definition 3:
    UnboundLocalError: local variable '___filter_cubic' referenced before assignment
    raised from C:\Users\tmund\Anaconda3\envs\Taylor\lib\site-packages\interpolation\splines\prefilter_cubic.py:293
This error is usually caused by passing an argument of a type that is unsupported by the named function.
[1] During: resolving callee type: Function(<function _filter_cubic at 0x0000024A0C8E6168>)
[2] During: typing of call at C:\Users\tmund\Anaconda3\envs\Taylor\lib\site-packages\interpolation\splines\prefilter_cubic.py (298)


File "..\..\..\..\Anaconda3\envs\Taylor\lib\site-packages\interpolation\splines\prefilter_cubic.py", line 298:
def prefilter_cubic(*args):
    return _filter_cubic(*args)
    ^

[albop]

yes, interpolation.py doesn't support prefiltering for more than four dimensions . It is issue EconForge/interpolation.py#40 .
What you can probably do is change the interpolation method, try passing interp_method='linear' to time_iteration.

[TimMunday]

Thanks for the quick reply - much appreciated.