> ML_LITERATURE // BOX-JENKINS-1970-TIME-SERIES-ANALYSIS-FORECASTING-AND-CONTROL_v1.0
Time Series Analysis: Forecasting and Control
George E. P. Box, Gwilym M. Jenkins · Holden-Day, San Francisco (1970)
foundational1970foundationalthirdPartyReproduced
Principal Contribution
Established the systematic three-stage Box-Jenkins methodology (Identification, Estimation, Diagnostic Checking) for ARIMA time-series models.
Operational Relevance
The universal statistical benchmark required for evaluating all deep learning and machine learning time-series models.
Assumptions
- Stationary stochastic processes can be modeled as linear filters driven by white noise; differencing resolves non-stationarity
Limitations
- Linear autocorrelation assumptions break down on non-linear dynamics, structural regime shifts, and multi-seasonal high-frequency data