QUESTION: dealing with missing observations in 1 minute frequency stock market time series

[konradsemsch]

Hello, I'm dealing with a time series problem where stock market data is collected on a 1-minute frequency. The data is available from 9:00 till 17:00. Outside of this time nothing is recorded as there are no trades taking place.

In that situation when I plot the value of the stock against time, I have large time gaps. It's impossible to use any classical forecasting method when these data gaps occur unless I impute them, however, that also doesn't seem like the right thing to do in this case.

What's the right approach when dealing with this data when time continuity is required?

On the other hand I could simply use regular ML methods that don't depend on time so strickly.

[mitchelloharawild]

Support for structurally missing datasets is described in tidyverts/tsibble#18, which will be usable with these models when it is added to tsibble.

Classical forecasting methods are capable of handling atypical time structures and temporal missingness, but this is currently not possible in fable due to the lack of time classes to represent these temporally discontinuous (but in some sense, regular) time series.

For now, you can use a numeric index and manually specify the seasonal periods if you use a seasonal model.

[konradsemsch]

Thank you for a prompt answer! I think that clears it up for the time being 😉