Page 122 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Business and Management
Department of Information Systems, Business Statistics & Operations Management

Supervisor: KWOK James Sai Ho / ISOM
Student: POON Benjamin Cheuk Bun / GBUS Course: UROP1100, Fall

Throughout the Fall semester, our team has explored several aspects of deep learning and investigated into its
application in stock market prediction. We examined multiple areas in the field where there were potential research
opportunities. We also took into consideration our team’s strengths and weaknesses in the decision of where to focus
our research efforts on. After a few failed attempts, our research focus is now settled on volatility prediction. We are
hoping to continue with the project and this report will hopefully serve as a reminder of the past mistakes that we
should avoid in the future.

Risk Management
Supervisor: SO Mike Ka Pui / ISOM
Student: CUI Ningshan / MATH-SFM Course: UROP1100, Fall

With the arrival of the era of Big Data, the explosion of discrete data online stimulates the researchers to create and
improve a variety of text processing model with sophisticated statistical structures. As one of the most important
milestones, Latent Dirichlet Allocation Model was born with solid theoretical foundation and great practicability to serve
as an important tool for topic-based text document classification. In this paper, I summarized the features of a few
precursors model, and for sure LDA itself, a three-level hierarchical generative probabilistic model. Also, when it comes
to applications, it needs to use variational inference and EM algorithm to approximate several unknown but essential
parameters to enable the model to process un-seen documents. The variational inference procedure is also presented in
this report, followed by a discussion of the application on Wikileaks Afghanistan War Logs. At last, another model called
Correlated Topic Model is presented, serving as the derived model from LDA with improved features.

Supervisor: SO Mike Ka Pui / ISOM
Student: HE Jianyao / QFIN Course: UROP2100, Fall

This project is follow-up research on implementation of latent space model in dynamic network continuing from
UROP1100. Having obtained fundamental knowledge on latent space model, Bayesian analysis and Markov Chain Monte
Carlo sampling, the project moves forward to application of the above theories and analyse the dynamic network. Over
the past few weeks, I first reviewed the MCMC sampling scheme, especially Metropolis-Hastings to have a thorough
understanding. Then, I revisited Latent Space Models for Dynamic Networks and its supplementary material to fully
understand the whole process of how to implement MCMC sampling and Bayesian analysis to study the changes of
dynamic network relation with respect to time.

Supervisor: SO Mike Ka Pui / ISOM
Student: WANG Ziyi / RMBI Course: UROP3100, Spring

Professor Mike So has proposed a cross-sectional vine copula model to incorporate correlations among high dimensional
high frequency stock return series and market factors, giving a new dimension of risk management. During the past two
sessions, we have built the research frame, organized the statistics rationale and successfully testified the model
estimation feasibility. As the third session of the research project, I mainly worked on preparing for the research paper
and further explore the real life impacts and applications of the model. The report will summary my research process in
mainly two parts: Understand financial return dependencies and explore the application of the model.

112