statsmodels: [ARMA] Predict model gives "Wrong number of items passed X, placement implies Y"

Hi,

I fitted my ARMA model using the Dataframe below, and try to predict the next 60 days:

>>> print(endog_products)
2014-06-17    1.0
2014-06-19    0.0
2014-06-20    0.0
2014-06-21    0.0
2014-06-22    0.0
2014-06-23    0.0
2014-06-24    0.0
2014-06-25    2.0
2014-06-26    2.0
2014-06-27    3.0
2014-06-28    3.0
2014-06-29    0.0
2014-06-30    0.0
2014-07-01    3.0
2014-07-02    2.0
2014-07-03    2.0
2014-07-04    1.0
2014-07-05    1.0
2014-07-06    0.0
2014-07-07    1.0
2014-07-08    1.0
2014-07-09    1.0
2014-07-10    1.0
2014-07-11    4.0
2014-07-12    1.0
2014-07-13    1.0
2014-07-14    0.0
2014-07-15    2.0
2014-07-16    1.0
2014-07-17    0.0
2014-07-18    1.0
2014-07-19    1.0
2014-07-20    0.0
2014-07-21    0.0
2014-07-22    3.0
2014-07-23    3.0
2014-07-24    2.0
2014-07-25    2.0
2014-07-26    0.0
2014-07-27    1.0
2014-07-28    0.0
2014-07-29    1.0
2014-07-30    0.0
2014-07-31    1.0
2014-08-01    0.0

This is my code to fit the model, and it works great:

arma_order = arma_order_select_ic(endog_products, ic=['aic', 'bic', 'hqic'], trend='c', fit_kw=dict(method='css-mle'))
p = arma_order.aic_min_order[0]
q = arma_order.aic_min_order[1]
arma_model = ARMA(endog_products, freq='D', order=(p, 0, q))
arma_res = arma_model.fit() 

When I try to predict the next 60 days using this code (prediction_values = arma_res.predict(start=prelearning_date, end=max_date)), Python returns this error: Error when computing ARMA model for product 3035668: Wrong number of items passed 106, placement implies 107.

I don’t understand what this error means. I checked on Google and on issues but I don’t find some issues that are corresponding to my problem.

Thanks a lot for your answers.