Parsimony ⇒ The idea of good amount of params so that model don’t overfit
What is the real meaning of it ?
If we have a timeseries of say Y0, Y1…YN with a certian joint distribution. and shifting the data by m units hence Ym+1, Ym+2, … YM+N
If both had same joint probability distribution, means All the aspects remains the same even if went ahead of the time. The its stationaty.
Always pick adjusted close price.
See how data looks like before predicting.
- Trend/Cycle
- Season
- Error
You can adjust the data based on season, delete season from timeline
How this is done, Do time series decomposition?
People use moving averages etc for removing error but morders is
- LOESS Regression Do expenential smoothing models