Exactly, price series are emphatically nonstationary. That's why one needs to transform or take linear combinations or trade volatility or some other novel tradable derivative combination that is stationary.
Where does this idea come from that the emphatically nonstationary can be made stationary via some magical combinatrion of transformations? Is there evidence to back this up?
Exactly, price series are emphatically nonstationary. That's why one needs to transform or take linear combinations or trade volatility or some other novel tradable derivative combination that is stationary.
Where does this idea come from that the emphatically nonstationary can be made stationary via some magical combinatrion of transformations? Is there evidence to back this up?
Haven't been working with ARIMA specifically but I'm close to a system that uses time series models + microstructure considerations. I think the combination is likely a good way for retailers to compete with the big boys.
Haven't been working with ARIMA specifically but I'm close to a system that uses time series models + microstructure considerations. I think the combination is likely a good way for retailers to compete with the big boys.
Just out of curiosity, how many data points does your model require to work? For example, all the experts seem to agree that a minimum of 50 data points is required to create an acceptable ARIMA model.
Just out of curiosity, how many data points does your model require to work? All the experts seem to agree that a minimum of 50 data points is required to create an acceptable ARIMA model.
Like I said I haven't used ARIMA models so I can't help you there. In general, you can get confidence levels on your parameter estimates for your model and that should tell you whether you have a good estimate or bad, not a universal cutoff on the number of training points.
