Bayesian Change Point Analaysis
Date of Creation
5-13-2019
Document Type
Departmental Honors Thesis - Restricted Access
Department
Mathematics
First Advisor
Eric Ruggieri
Abstract
In regression analysis, change points help divide long datasets into smaller and more manageable subsets. There are a variety of approaches used in identifying change points in datasets. Yet, few of the approaches give uncertainty bounds on both the number and locations of change points. Using Bayes’ Rule, we created a probabilistic model capable of giving us the number and locations of change points at the same time. Knowing that our algorithm is sensitive to prior assumptions, we created various kinds of test datasets and applied our model with different parameter settings to check how well our model is performing. Using training data, we implemented a model selection and model averaging technique to maximize the objectivity in terms of parameter selection. We then applied our model to real datasets including the S&P 500 index, oil and gold prices. The change points we identified in these financial time series match many of the market breakdowns we observed in the last few decades.
Recommended Citation
Qiang, Rui, "Bayesian Change Point Analaysis" (2019). Math and Computer Science Honors Theses. 59.
https://crossworks.holycross.edu/math_honor/59