Deterministic seasonal not possible anymore in statsmodels unobserved components

[rklees]

In the current version of statsmodels it is not possible to choose a deterministic seasonal using 'stochastic_seasonal': False.

Example:

frequencies = [{'period': 12, 'harmonics': 1}]

model_settings = {
    'irregular': True, 
    'level': True, 
    'stochastic_level': False, 
    'trend': True, 
    'stochastic_trend': True,
    'cycle': True, 
    'damped_cycle': False, 
    'stochastic_cycle': True,
    'cycle_period_bounds': [5.1, 5.2],
    'freq_seasonal': frequencies,
    'stochastic_seasonal': False,
    'use_exact_diffuse': True}

# ssm = state space model
ssm = sm.tsa.UnobservedComponents(data['meas'], **model_settings)

Fitting the model to the data provides, among others a ML fit of the seasonal disturbance variance,

sigma2.freq_seasonal_12(1) = 0.0809

though this parameter must not be estimated using ML. Strange enough, when using

'stochastic_freq_seasonal': False,

it works. So, what is wrong with 'stochastic_seasonal' ? Moreover, if I have two seasonal terms, one deterministic, the other one stochastic, e.g.,

frequencies = [{'period': 12, 'harmonics': 1}, {'period': 5.347, 'harmonics': 1}]

stochastic_seasonal = [False, True],

gives an error message TypeError: unsupported operand type(s) for +: 'int' and 'list' whereas

stochastic_freq_seasonal = [False, True],

works fine.