如何解决从 Statsmodels 运行 ARIMA 时出现的 ValueError？

Question

I use the following code when building ARIMA

arima = ARIMA(ts.astype(float), freq = 'M',order=(4, d, 3)).fit()

I get the following error when I run the above:

ValueError: The given frequency argument could not be matched to the given index.

My data frame looks like this:

Date        A          B           C        D       E           F
2020-04-01  118     10932.54    203.617     1   0.296610    0.382666
2020-05-01  144     11645.20    167.575     1   0.361111    0.401953
2020-06-01  89      8545.86     196.084     1   0.449438    0.524768
2020-07-01  117     10512.30    194.442     1   0.384615    0.472443
2020-08-01  75      6613.11     189.289     1   0.280000    0.332995

The Date column is my index. It does not have a frequency label.

I tried to give it a frequency label with the following

df.asfreq('M')

However, this turned the values in every column, except the Date column, to NaN .

I tried to run the ARIMA model without a frequency argument; however, I received the following warning message:

ValueWarning: No frequency information was provided, so inferred frequency MS will be used.

As well, I received this warning:

ValueError: The computed initial MA coefficients are not invertible
You should induce invertibility, choose a different model order, or you can pass your own start_params.

I'm not sure what to do here. Any help would be greatly appreciated.

Answer 1

First, you can use

df = df.resample("M").last()

to get a clean month-end df without introducing NaN values.

Second, you should be using either SARIMAX or statsmodels.tsa.arima.model.ARIMA rather than statsmodels.tsa.arima_model.ARIMA (Note the _ in the latter is a . in the former). statsmodels.tsa.arima.model.ARIMA is essentially a special case of SARIMAX that generally performs better. You will need statsmodels 0.12.0rc0 or a build of the master branch to use the new ARIMA .

In this new model, you will not have the issues with MA invertibility.

Answer 2

'MS' means 'Month-Start' whereas 'M' is for 'Month-End'.

My data is for the beginning of each month, so I am supposed to pass 'MS' into the frequency argument.

I solve the not invertible error with a try/except.