Chapter 23 Further Reading: Time Series Analysis for Betting

The following annotated bibliography provides resources for deeper exploration of the time series methods introduced in Chapter 23. Entries are organized by category and selected for their relevance to applying time series analysis in sports betting contexts.

Books: Foundational Time Series

1. Hamilton, James D. Time Series Analysis. Princeton University Press, 1994. The definitive graduate-level textbook on time series analysis. Hamilton covers ARIMA, state-space models, unit root tests, and spectral analysis with mathematical rigor. Chapters 3-5 on ARMA models and Chapter 17 on unit root testing are directly relevant to this chapter. Dense but authoritative; best used as a reference rather than a first read.

2. Hyndman, Rob J. and Athanasopoulos, George. Forecasting: Principles and Practice. 3rd edition, OTexts, 2021. A freely available online textbook (otexts.com/fpp3/) that covers modern forecasting methods with clear explanations and R code. The chapters on ARIMA, exponential smoothing, and forecast evaluation are particularly accessible. While written for R, the concepts translate directly to Python's statsmodels. An excellent starting point for readers new to time series forecasting.

3. Shumway, Robert H. and Stoffer, David S. Time Series Analysis and Its Applications: With R Examples. 4th edition, Springer, 2017. A well-balanced textbook that bridges theory and application. Chapters on ARIMA, spectral analysis, and state-space models are relevant. The companion R package and datasets make it easy to work through examples. The chapter on spectral methods offers an alternative perspective on seasonality not covered in Chapter 23.

4. Brockwell, Peter J. and Davis, Richard A. Introduction to Time Series and Forecasting. 3rd edition, Springer, 2016. A rigorous but accessible introduction to time series theory. Particularly strong on the mathematical foundations of stationarity, autocorrelation, and ARIMA modeling. The treatment of prediction theory provides the formal underpinning for the forecast intervals used in betting applications.

Books: Applied and Sports-Specific

5. Kovalchik, Stephanie. "Searching for the GOAT of Tennis Win Prediction." Journal of Quantitative Analysis in Sports, 12(3), 2016. While technically a journal article, this comprehensive comparison of prediction methods for tennis includes time series approaches. Kovalchik evaluates Elo, Glicko, logistic regression, and point-by-point models, providing context for how time series methods compare to alternatives in sports prediction.

6. Albert, Jim, Glickman, Mark E., Swartz, Tim B., and Koning, Ruud H., eds. Handbook of Statistical Methods and Analyses in Sports. CRC Press, 2017. A comprehensive reference covering statistical methods across multiple sports. Several chapters address time series aspects of sports data, including streakiness, hot hand effects, and performance dynamics. The chapter on sequential testing is relevant to the changepoint detection material in this chapter.

Academic Papers: Time Series in Sports

7. Glickman, Mark E. and Stern, Hal S. "A State-Space Model for National Football League Scores." Journal of the American Statistical Association, 93(441), 1998, pp. 25-35. A foundational paper applying state-space (dynamic linear) models to NFL game scores. Glickman and Stern model team strengths as time-varying latent states updated via the Kalman filter. This approach is more flexible than ARIMA for tracking team quality and is a natural extension of Chapter 23's methods.

8. Baker, Rose D. and McHale, Ian G. "Forecasting Exact Scores in National Football League Games." International Journal of Forecasting, 29(1), 2013, pp. 122-130. Demonstrates how time series models for team offensive and defensive strengths can be used to forecast exact scores. The paper uses a bivariate Poisson model with time-varying parameters, showing how temporal dynamics improve prediction over static models.

9. Manner, Hans. "Modeling and Forecasting the Outcomes of NBA Basketball Games." Journal of Quantitative Analysis in Sports, 12(1), 2016, pp. 31-41. Applies ARMA-type models and dynamic regression to NBA game outcomes. Manner shows that accounting for temporal autocorrelation in team strength leads to measurably better predictions than static models. Directly relevant to the ARIMA modeling section of Chapter 23.

10. Boshnakov, Georgi, Kharrat, Tarak, and McHale, Ian G. "A Bivariate Weibull Count Model for Forecasting Association Football Scores." International Journal of Forecasting, 33(2), 2017, pp. 458-466. Proposes a dynamic model for soccer scores that evolves over time, providing a sophisticated example of time series thinking applied to count data in sports.

Academic Papers: Mean Reversion and Changepoints

11. Boulier, Bryan L. and Stekler, Herman O. "Predicting the Outcomes of National Football League Games." International Journal of Forecasting, 19(2), 2003, pp. 257-270. Examines the predictability of NFL outcomes and implicitly addresses mean reversion by comparing model predictions to market lines over time. The paper's finding that simple models are difficult to beat highlights the importance of properly calibrating mean reversion expectations.

12. Killick, Rebecca, Fearnhead, Paul, and Eckley, Idris A. "Optimal Detection of Changepoints with a Linear Computational Cost." Journal of the American Statistical Association, 107(500), 2012, pp. 1590-1598. The original PELT paper. Introduces the pruned exact linear time algorithm for optimal changepoint detection. Killick et al. show that pruning dramatically reduces computation while maintaining exact optimality. The ruptures Python library implements this algorithm, making it directly applicable to sports data.

13. Adams, Ryan P. and MacKay, David J.C. "Bayesian Online Changepoint Detection." arXiv preprint, arXiv:0710.3742, 2007. The foundational paper for BOCPD. Adams and MacKay present an elegant recursive algorithm for updating the posterior distribution of the current run length (time since the last changepoint). This is the method of choice for real-time changepoint detection in a betting context, where you need to react to regime changes as they happen rather than in retrospect.

14. Page, E.S. "Continuous Inspection Schemes." Biometrika, 41(1-2), 1954, pp. 100-115. The original CUSUM paper. Page introduced the cumulative sum method for detecting shifts in sequential data, motivated by quality control in manufacturing. The simplicity and elegance of CUSUM make it a useful first-pass tool for changepoint detection in sports data, even though more sophisticated methods are available.

Software and Tutorials

15. statsmodels Documentation: Time Series Analysis (statsmodels.org) The official documentation for Python's statsmodels library, which provides ARIMA, SARIMA, exponential smoothing, and stationarity test implementations. The "User Guide > Time Series Analysis" section includes worked examples and API documentation. Essential for implementing the methods in Chapter 23.

16. ruptures: Change Point Detection in Python (centre-borelli.github.io/ruptures-docs/) Documentation for the ruptures library, which implements PELT, binary segmentation, window-based, and kernel-based changepoint detection. The tutorials section provides clear examples with synthetic and real data. This is the primary Python tool for the changepoint detection methods discussed in Chapter 23.

17. Facebook Prophet (facebook.github.io/prophet/) Prophet is a forecasting library that handles time series with strong seasonal effects and multiple changepoints automatically. While designed for business forecasting rather than sports, its decomposition approach (trend + seasonality + holidays + regression) maps well to sports data. The automatic changepoint detection feature is particularly relevant.

Data Sources

18. Basketball Reference / Pro Football Reference / Baseball Reference (sports-reference.com family) The Sports Reference family of websites provides comprehensive game-by-game logs for major North American sports, going back decades. These logs include dates, scores, opponents, and box score statistics. This is the primary data source for constructing the time series datasets used in Chapter 23's exercises.

19. FiveThirtyEight's Elo Ratings and Forecasts (projects.fivethirtyeight.com) While primarily a rating system (covered in Chapter 26), FiveThirtyEight's published Elo ratings for the NFL, NBA, and MLB provide a useful benchmark for time series models. The daily-updated ratings implicitly incorporate temporal dynamics and can be compared against custom ARIMA-based approaches.

20. Covers.com Historical Odds and ATS Data Covers maintains historical against-the-spread records and line history for major sports. This data is essential for testing calendar effects and time series strategies against actual betting lines. The ATS records can be used directly as a binary time series for testing system profitability over time.

How to Use This Reading List

For readers working through this textbook sequentially, the following prioritization is suggested:

  • Start with: Hyndman and Athanasopoulos (entry 2) for an accessible grounding in modern forecasting methods.
  • Go deeper on ARIMA: Hamilton (entry 1) chapters 3-5, or Brockwell and Davis (entry 4).
  • Go deeper on changepoints: Killick et al. (entry 12) and Adams and MacKay (entry 13) for the PELT and BOCPD methods.
  • Go deeper on sports applications: Glickman and Stern (entry 7) for state-space models in the NFL, and Manner (entry 9) for ARMA in the NBA.
  • For implementation: statsmodels (entry 15) and ruptures (entry 16) documentation.
  • For data: Sports Reference (entry 18) for game logs and Covers (entry 20) for ATS data.

Many of these resources will be referenced again in later chapters, particularly Chapter 26 (Ratings) and Chapter 27 (Advanced Regression), where time series concepts are integrated into broader modeling frameworks.