Department of Mathematical and Computational Sciences
Permanent URI for this community
Browse
Browsing Department of Mathematical and Computational Sciences by Author "Chagwiza, Wilbert"
Now showing 1 - 4 of 4
Results Per Page
Sort Options
Item Open Access Fundamental Analysis for Stocks using Extreme Gradient Boosting(2022-11-10) Gumani, Thanyani Rodney; Chagwiza, Wilbert; Kubjana, TlouWhen it comes to stock price prediction, machine learning has grown in popularity. Accurate stock prediction is a very difficult activity as financial stock markets are unpredictable and non-linear in nature. With the advent of machine learning and improved computational capabilities, programmed prediction methods have proven to be more effective in stock price prediction. Extreme gradient boosting(XGBoost) is the variant of the gradient boosting machine. XGBoost, an ensemble method of classification trees, is investigated for the prediction of stock prices based on the fundamental analysis. XGBoost outperformed the competition and had higher accuracy. The developed XGBoost model proved to be an effective model that accurately predicts the stock market trend, which is considered to be much better than conventional non-ensemble learning techniques.Item Open Access Modelling volatility, equity risk and extremal dependence of the BRICS Stock Markets(2022-07-15) Mukhodobwane, Rosinah Mphedziseni; Sigauke, Caston; Chagwiza, Wilbert; Garira, WinstonWith the use of empirical data of the BRICS (Brazil, Russia, India, China, and South Africa) stock markets, this thesis focuses on solving three main nancial and investment issues involving returns volatility, risk and extremal dependence via robust statistical modelling. The rst issue involves modelling nancial returns volatility (when the true distribution is unknown) using the univariate GARCH model under the assumptions of seven error distributions. The ndings, using two of the error distributions, show that the Chinese market has the highest volatility persistence, followed by the South African, Russian, Indian and Brazilian markets in that order. For risk modelling and analysis, the ndings show that the Russian market has the highest risk level, followed by the South African, Chinese, Brazilian and Indian markets, respectively. For the extremal dependence modelling, using the bivariate point process and conditional multivariate extreme value (CMEV) models, the ndings show varied levels of low extremal dependence structure whose outcomes are highly bene cial to investors, portfolio managers and other market participants who are interested in maximising their investment returns and nancial gains. However, it is observed that the point process was able to model many more extreme observations or exceedances that contribute to the likelihood estimation and it gives more information than the threshold excess method of the CMEV model.Item Open Access Predicting an Economic Recession Using Machine Learning Techniques(2022-11-10) Molepo, Mashaka Ruth; Chagwiza, Wilbert; Kubjana, Tloufew economic downturns were predicted months in advance. This research has the ability to give the best performing models to assist businesses in navigating prior recession periods. The study address the subject of identifying the most important variables to improve the overall performance of the algorithm that would effectively predict recessions. The primary aim of this study was to improve economic recession prediction using machine learning (ML) techniques by developing an inch-perfect and efficient prediction model in order to avoid greater government deficits, growing inequality, significantly decreased income, and higher unemployment. The study objective was to establish the relevant method for addressing imbalance data with suitable features selection strategy to enhance the performance of the machine learning algorithm developed. Furthermore, artificial neural network(ANN) and Random Forest (RF) were used in predicting economic recession using ML techniques. This study would not have been possible without the publicly available data from the online open source Kaggle, which provided ordinal categorical data for the specific data utilized. The major findings of this study were that the ML algorithm RF performed better at recession prediction than its rival ANN. Due to the fact that two ML algorithms in this research were employed , further ML tools can be used to improve the statistical components of the study.Item Open Access A stochastic programming framework for financial intermediaries liquidity in South Africa(2015-05) Chagwiza, Wilbert; Garira, W.; Moyo, S.We provide a thorough overview of the current liquidity problem, dis cussing what are considered to be state-of-the-art approaches in both in dustry and in academia, and clearly establish our motivation to depart from current liquidity management standards. We noted that during 2007 to 2009 financial crisis different countries throughout the whole world faced serious challenges that ranged from chronic liquidity problems, deep-rooted risk management deficiencies and poor corporate governance practices. The liq uidity crisis, especially in the USA, was caused by low real interest rates stimulating an asset price bubble fuelled by new financial products that were not stress-tested and that failed in the downturn. Large depository institu tions face the question of how to optimally utilise funds and manage liquid ity problems. The existing liquidity models are not standardised, and do not take into consideration the complexity and the nature of the bank. As evi denced by the late 2000 financial crisis, it is evident that nowadays there is not a metric that seems to be completely adequate to prevent liquidity crises. Our contribution is fourfold. First we investigate appropriate scenario gener ation methods and perform a rigorous investigation on how to generate mul tistage scenario trees. Further, we investigate the inclusion of all qualifying liquid asset instruments into the portfolio optimisation. We propose a novel multistage stochastic programming methodology for liquid asset control. Thus we define how to construct and solve stochastic programming models for liquidity needs-driven sub-portfolios. Our approach is based on scenario trees and makes no assumption on the distributions of random variables. Fi nally, we investigate the inclusion of liquidity needs-driven strategies which are core liquidity, cash cushion, operational cash and discretionary liquidity into the overall liquidity portfolio. Stochastic programming is a multi-faceted problem, and even the most fo cused treatment necessarily incorporates techniques from a wide range of disciplines. Through the development of fully coherent models and a suffi ciently robust solution methodology, we provide a thorough overview of the problem at hand, discussing approaches in both industry and in academia. We introduce practical and theoretical advances that are to our best knowl edge unexplored in the current literature, and document the usefulness of these avenues through a systematic series of increasingly complex applica tions and experiments. The data used is from the South African Reserve Bank (SARB) and International Monetary Fund (IMF) from January 1988 to May 2014. Most corporations such as commercial banks generally hold a unitary liquidity portfolio and not necessarily segmented to focus on bank specific cash and liquidity needs. Banks do have different cash needs at different times. Financial intermediaries to manage liquidity efficiently and effectively, segmenting the optimal liquid asset portfolio is the best manage ment strategy. We construct different stochastic programming models based on decision making under risk and provision of powerful paradigm for de cision making under uncertainty. The stochastic programming models with recourse are clear, easy to implement, very efficient and provide the optimal solution according to future possible set of scenarios. In defining and con structing stochastic programming models with recourse, sensitivity analysis should be carried out to increase the decision maker's understanding of the problem and to show the effect of different assumptions. Constructing an SP model with more stages provides better approximations compared to a single-period model. We found that the use of the proposed models through segmenting the unitary liquidity portfolio improves the management of liq uidity compared to current techniques which are based on simulation, ex perience and trial-and-error. Thus high quality liquidity buffers insure the bank against any adverse liquidity risk. We found that liquidity buffer is best modelled as a stochastic process than deterministic. High net cash flow enhance effective and efficient liquidity management. Finally, excess re serves improve country monetary policy and broaden the scope of central bank lending programmes to address conditions in credit markets but on the other hand, may lead banks to loosen standards by weakening lending crite ria in an attempt to increase returns. In designing the sub-portfolios, liquidity determinants such as liquid asset returns, loan returns, payment flow, interest rates and gross income are un certain, and to avoid liquidity problems caused by these variables, we need to include randomness on them. To effectively and efficiently manage liq uidity, a bank may be required to follow certain key liquidity principles. The first is that the bank needs to understand and categorise its cash needs into at least four liquidity sub-portfolios as explained in the research. In this study, we strongly recommend that the banks' management should design the following sub-portfolios; liquidity buffer, cash cushion, operational cash and discretionary liquidity. The bank should clearly maintain the holdings of high quality liquid assets that can provide reliable reserves under all con ditions. In addition, strict and relatively comprehensive disclosure practices in relation to liquidity risk management objectives should be submitted to central banks. Finally, there should be improvement on funding markets and public confidence by broadening the scope of bank guarantees to ensure future financial stability. We need to emphasise the point that bank liquid ity is restricted by capital adequacy, required reserves, liability and deposit insurance defined by the central bank. Research can therefore be done on models that can be used to efficiently and accurately forecast future cash outflow taking into consideration the behavioural cash flows and estimating the discretionary and excess reserves. Further research can also be done on comparing the benefits of the calculated strategies in the stochastic programming context to more traditional methods.