LAGGED BAYESIAN ONLINE CHANGEPOINT DETECTION FOR MULTIVARIATE STOCK PRICE DATA

• Main Author: Susanto, Julius
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70810
• Institution: Institut Teknologi Bandung

In early 2020, the Covid-19 virus began to spread throughout the world. As a result, the government was forced to make social restrictions mandatory on national and international scales, and the economy started to collapse. Many companies are experiencing losses from this incident which impacts their share price, so the price shares experience a behaviour change. This case example is one case of changes in stock price behaviour due to regulatory changes. When a change occurs in a company that has a significant impact, the difference in stock price behaviour can be seen in the stock price plot. But there are times when change behaviour changes little by little, so it is challenging to know the changepoints. So, there is a need for a method to detect the changepoints in stock price behaviour so that investors can think about the following action to be taken. One of the methods to detect the point of changepoints in time series online data, like stocks, is Bayesian online changepoint detection (BOCPD) or according to (Byrd, et al.,2018) call it EXO. This method can predict the location of the changepoints of time series data that increases over time (online). This method can detect the time of change behaviour quickly, but the result of this method could be more stable. Therefore, this thesis will use the lagged exact online Bayesian changepoint form for detection or LEXO. This method incorporates an advanced recursive algorithm with past information, and a recursive algorithm backtracks from the information in the future. Knowing some fore before the decision is hoped that this method’s accuracy will be better than the EXO method. This study will apply the EXO and LEXO methods to multivariate data. It is expected that the EXO and LEXO methods can detect the change points caused by changes in covariance. By knowing the changes in the covariance of stock data, investors can see how good their stock portfolio is. So that drawing conclusions, investors become more thorough. Therefore the normal multivariate model will be used. According to the simple stock price movement model, the natural logarithm of the geometric stock price returns is normally distributed. In addition, the LEXO method this model will also use to detect skewed data and stock price data with natural logarithms of geometric returns not normally distributed multivariate.