Events Calendar

PhD Thesis Proposal

Date:
Friday, January 18, 2019
Time:
10:00 am
Location:
Western Science Centre (WSRC)
Room: WSC-248
Cost:
Free

Title: HMM-modulated early warning mechanisms for the detection of financial instability

Regulators’ early intervention is crucial when the financial system is experiencing difficulties. Financial stability must be preserved to avert banks’ bail outs, which hugely drains government’s financial resources. Detecting in advance periods of financial crisis entails developing accurate and robust quantitative techniques. We employ statistical models in our financial-crisis analysis.

In the literature of financial crisis analyses, many studies were based simply on one-state stochastic processes. However, such processes might not be able to precisely capture the behaviours of the underlying data series, especially during the periods of market uncertainty and abrupt fluctuations. We approach this problem by employing a regime-switching paradigm to improve model performance not only in attaining very good model fitting of the data but also in making reliable short-term predictions. The latter consideration plays an important role in pre-crisis warning signal detection.

The main objectives of this research comprise the following: (i) Construct multivariate hidden Markov models (HMM) and higher-order hidden Markov models (HOHMM) in the hybridised Ornstein-Uhlenbeck (OU) and Geometric Brownian Motion (GBM) modelling framework to capture dynamics of multidimensional data series. (ii) Develop self-calibrating algorithms and dynamic estimation procedures for our proposed models. (iii) Derive recursive filters for quantities under the HOHMM setting its special case, the HMM, and then obtain their optimal parameter estimates. (iv) Construct on-line supervised learning algorithm based on filters’ outcome to generate early-warning signals in advance of the financial-crisis episodes.

Certain statistical methods from sensitivity, principal-component, and sequential analyses with the addition of some machine learning techniques will be considered. We concentrate on estimating quantities akin to practitioners’ implementation of early-warning signals and appropriate measures. The ultimate goal is to translate the models’ results and implications into concrete regulatory policies that build resilience in the financial system.

This research consists of four related projects, the first two of which were already completed and will be presented.

Supervisors:

  • Dr. Rogemar Mamon
  • Dr. Hao Yu

Contact:
Event Type:


Powered by Blackbaud
nonprofit software