gistfile1.txt

TIME SERIES REVIEW
------------------

- Get Data (value is dependent over time, consistent interval)
- Plot (visualize)
- Break down dataset into components
- Trend (long running average or polynomial 2nd or 3rd)
- Seasonal (regular repeating interval based on normal calendar, christmas, new years, summer)
- Cyclic (based on business cycle, not tied to a season - every 3 weeks)
- Seasonal and Cyclic can get conflated
- Residuals, white noise
- Find all the components of the time series (trend, seasonal, cyclic and residuals)
- Residuals should be stationary (mean of 0 and constant variance) (Homoscedasticity)
- Verify your residuals are stationary using (ADF) test, H0 is not stationary
- How to detrend - linear regression - polynomial regression - rolling average - differencing
- Your trying to model the white noise
- Autocorrelation and Partial Autocorrelation plots
- S - SL1 - SL2 - SL3
- _______, ____, _____
- Help you establish your ARIMA - AR and MA
- ARMIA (auto regressive, integrated, moving average)
- AR - regression with previous time steps
- I - differencing
- MA - moving average - regression of previous residuals
- PACF (relationship between the first 3 lags) - 3 as AR, MA is usually its between (0 and AR - 1)
- Grid search in the general area based on the PACF plot
- Find the model with lowest AIC, BIC (goodness of fit and model simplicity)
- Make predictions (use holdout set - test) compare test set to predictions
- MSE