STA 364 Spring 2025
Outline
Outline
The basic steps and concepts of forecasting. Examples: Closer forecasts will be better, and predictable patterns will be easy to forecast.
Time series graphics
- Time plots
- Patterns, trend, season, cycle
- Interpreting
gg_season
,gg_subseries
graphs. - Autocorrelation, lag plots, correlogram
- White noise
- Connecting/cross-referencing information from these plots.
- Decomposition
- Transformations and adjustments - multiplicative to additive, box_cox use.
- TS components, Trend, season, remainder, seasonally adjusted series
- Moving average
- STL decomposition (that it uses loess at multiple stages)
- Toolbox
- EDA -> Define model -> Train model -> Check performance -> produce forecasts
- Train/ Test. Why. Using test sets to compare models but using metrics on either to evaluate an individual model.
- Simple forecasting methods: MEAN, NAIVE, SNAIVE, DRIFT
- When to use which, and how to model with them.
- Residuals and residual diagnostic (check for autocorrelation)
- Prediction intervals
- Forecasting with decomposition
- TSLM
- Linear model in TS
- Regression model assessment. AIC, AICc, BIC, R2, and R2Adj
- Picking useful predictors: dummy, harmonic, step, spike, lagged values
- Nonlinear: piece-wise, log-log