Python中的Statsmodels包 - 检索ARIMA模型的样本外预测的问题

Question

I'm trying to retrieve an out-of-sample prediction for ARIMA model. However, I constantly receive errors and I'm not sure how should I proceed now:( The code is the following:

    from statsmodels.tsa.arima_model import ARIMA
    fit = ARIMA(endog, (1,1,1)).fit()
    params = fit.params
    forecast = fit.predict(params.all(), start='2015-9-21', end='2016-9-21', typ='levels')

It worked well (ie giving me a result, but not an out-of-sample one...) when I only used

    forecast = fit.predict(params.all(), typ='levels')

but when I added "start" and "end" dates (or only "start") it doesn't want to work, I constantly get errors. In case of the first cited chunk of code its: "TypeError: predict() got multiple values for keyword argument 'start'". I also tried with datetime type and it also didn't work. Can anyone help me with it?

Answer 1

I was getting a similar error to the one reported above:

"AttributeError: 'NoneType' object has no attribute 'get_loc' "

But I realised this was because I was passing an array (or list) without a datetime index, eg if you use pandas dataframes and input it as df.values , then you drop the time index, and ARMA doesn't have the dates information (so dates is None) which triggers this error. I suggest you feed in a pd.DataFrame or pd.Series object with a datetime index. See also this thread http://pystatsmodels.narkive.com/rhX3T509/arma-predict-throws-attributeerror-with-start-and-end-dates

Answer 2

You can use

fit.forecast(steps, exog=None, alpha=0.05)

where steps=365 according to your start and end parameter if its on monthly basis. Refer to the this answer