SU+ Digital Repository

SU+ is an online repository for the preservation and promotion of assorted digital content at Strathmore University

Off-Campus Access to restriced resources (including the ExamsBank) now requires registration using an @strathmore.edu email address

Authentication is NOT required for On-Campus Access to content

Photo by @inspiredimages

Communities in DSpace

Select a community to browse its collections.

Now showing 1 - 5 of 7

Conferences / Workshops / Seminars + 474
Documents and Proceedings of Conferences, Seminars, Workshops (and more) held at Strathmore University
Digital Archives 178
Assorted collections of resources covering various subject themes contributed by Faculty and Library Staff
Reports / Policies + 8
Public reports and policy documents
Research / Researchers / Publications 4342
Researcher Profiles / Conference presentations / Published research articles / Faculty and Corporate research outputs
Strathmore Heritage Collection 41
A digital chronicle of the History of the University presented through a mix of pictures, videos and digitized publications

Recent Submissions

Analysis on the convergence of the Least Squares Monte Carlo method and the Finite Differences Method on the valuation of American options
(Strathmore University, 2025) Waweru, T. K.
American options, which permit early exercise, pose considerable challenges in financial valuation due to the intricacies involved in identifying optimal exercise strategies. Conventional analytical models are often inadequate for these options, particularly in high-dimensional contexts, thereby necessitating the use of robust numerical techniques such as the Least Squares Monte Carlo (LSM) and Finite Difference (FD) methods. The LSM method is distinguished by its adaptability in addressing high-dimensional issues, employing a combination of Monte Carlo simulations and regression analysis to estimate continuation values, which renders it particularly effective for complex, path-dependent options. Conversely, the FD method utilizes a grid-based framework to solve the partial differential equations (PDEs) that dictate option pricing, providing stable and dependable solutions, especially in one-dimensional scenarios. This study compares two numerical methods—Least Squares Monte Carlo and Finite Difference Methods —for pricing American options under a SABR-based volatility calibration. LSMC offers flexibility and moderate runtime in higher-dimensional contexts, while FDM provides a deterministic, systematically refined solution. Both approaches are benchmarked against binomial-lumps and Bjerksund Stensland references for a short-dated American call, with emphasis on computational efficiency, accuracy, and the practical implications of adopting a lognormal SABR model to ensure market-consistent volatility. We analyze convergence rates, computational runtimes, and pricing accuracy across varying grid resolutions, simulation sizes, and basis selections. Our results delineate the trade-offs between flexibility, dimensional scalability, numerical stability and model run time, offering investors clear guidelines on choosing the optimal technique, LSMC for high-dimensional, path-dependent payoffs and FDM for one-dimensional problems demanding tight error control under SABR-driven market dynamics. Keywords: American options, Implied Volatility (IV), Least Squares Monte Carlo (LSMC), Finite Differences Methods (FDM), Stochastic Alpha, Beta, Rho (SABR) Model, Control Variate (CV), Antithetic Variate (AV)
Modeling non-life insurance claims using Exponential Log-Logistics distribution
(Strathmore University, 2025) Eileen, C.
Determining the amount of insurance claim in the insurance industry is a challenging task. This study aims at modelling and estimating loss claims from an insurance company, with a particular focus on extreme event risks. The main objective was to develop a distribution using the density hazard approach, characterized by skewness and heavy tail, to effectively capture both high-frequency small claims and low-frequency large claims. Various parametric distributions have been applied to the insurance loss claims to determine the model that best fits the data. The Exponential Log-Logistic distribution, characterized by parameters gives the best fit for the data, as indicated by the lowest negative log-likelihood (NLL) value. Comparative analysis highlights the Exponential Log-Logistic distribution’s superior performance in modeling heavy-tailed and skewed data, essential for accurate risk assessment in insurance. In contrast, the Log-Normal distribution, while achieving the lowest Bayesian Information Criterion (BIC) value does not fit the data as well in terms of NLL. The Value at Risk (VaR) and Expected Shortfall (ES) metrics further support the effectiveness of the proposed model in predicting risk management. The Exponential Log-Logistic distribution’s higher VaR values, and reasonably high ES values underscore its robustness in managing insurance risks. The findings recommend the Exponential Log-Logistic distribution as the most suitable model for insurance claims data, due to its balance of complexity and fit, offering significant improvements over simpler models like the Exponential and Lomax distributions. This model’s ability to accurately represent the distribution of claims supports better risk management and pricing strategies in the insurance industry. Keywords Skewness, heavy-tailed, density hazard distribution, Insurance Claims, Parametric distributions.
Stochastic methods for virtual asset pricing and risk management in Kenya
(Strathmore University, 2025) Rotich, S. C.
The increasing adoption of Bitcoin (BTC) and Ethereum (ETH) in global financial markets has raised critical questions regarding their valuation, volatility, liquidity, and regulatory oversight. This study investigates the effectiveness of stochastic models including Geometric Brownian Motion (GBM), Heston, Ornstein-Uhlenbeck (O-U), and Jump-Diffusion in capturing the unique price dynamics and volatility patterns of BTC and ETH. Using historical market data, the research applies these models alongside Auto-regressive Conditional Heteroskedasticity (ARCH/GARCH) models to analyze volatility persistence and risk characteristics. The findings indicate that while GBM provides a basic framework for price evolution, it fails to account for volatility clustering and market shocks. The Heston model captures stochastic volatility, whereas Jump-Diffusion models effectively incorporate sudden price jumps. GARCH (1,1) models confirm significant volatility clustering in both BTC and ETH. To assess risk exposure, the study computes Value at Risk (VaR) and Conditional VaR (CVaR) at 95% and 99% confidence levels. Results show that ETH exhibits higher tail risk than BTC, implying greater vulnerability to extreme losses. Furthermore, liquidity analysis, measured through market depth reveals that BTC has stronger liquidity and lower relative volatility risk compared to ETH. The research also applies Monte Carlo simulations to price BTC and ETH derivatives, demonstrating that stochastic models significantly influence option valuation by incorporating market uncertainties. Stress-testing scenarios highlight vulnerabilities in price stability, underscoring the need for margin requirements, volatility controls, and liquidity monitoring to mitigate systemic risks. The study’s findings contribute to the growing discussions on risk management and policy formulation of the VA ecosystem, offering recommendations to enhance market stability while fostering innovation. Keywords: Bitcoin, Ethereum, Volatility, ARCH/GARCH, VaR, CVaR, Liquidity, Derivatives, Risk Management, Regulation, Kenya
Dealing with missing data under joint modelling: application to HIV data
(Strathmore University, 2025) Akoko, T. A.
Background: Addressing missing data poses a significant challenge in clinical research, particularly in studies like those focused on HIV. Traditional methods employed by researchers to tackle this issue often yield biased estimates and unreliable recommendations. In this study, we assess the effectiveness of an innovative strategy that integrates both longitudinal and survival processes to handle missing data, applied specifically to HIV data from Kenya obtained from the Kenya Health Information System (KHIS). Methods: We conducted simulation studies by generating five datasets with varying percentages of missingness: 0% (representing a complete simulated dataset), 10%, 20%, 30%, and 40%. Subsequently, multiple imputation was performed on the four datasets containing missing values. This was followed by fitting a joint model to the imputed datasets. After the simulation studies, we applied the analysis to the real HIV data and conducted some diagnostic tests to the fitted joint models. Results: The results indicate that there is no discernible difference in the models post imputation across different percentages of missingness. Additionally, the joint model exhibits a good fit for our data compared to individual sub-models for longitudinal and survival analysis. Conclusion: Joint modeling integrating survival and longitudinal models, emerges as a powerful statistical approach in clinical research, particularly in HIV studies, to address complex data structures and missing data challenges. This study concludes with insights into its significant contributions and future directions in clinical research. Keywords: Missingness; Joint model; Time-to-event; Imputation; Simulations.
Optimizing credit risk assessment with ensemble sampling and hybrid machine learning models
(Strathmore University, 2025) Mucheru, N.
Accurate credit risk modeling is essential for minimizing financial losses, but class imbalance, where defaulters make up a small fraction of the data, remains a challenge. This study tackles the issue using ensemble sampling and hybrid machine learning models. A Kaggle dataset with 32,582 entries was used in this study. SMOTE + Random Under sampling, ADASYN + Random Under sampling, Borderline-SMOTE + Random Under sampling, SVM-SMOTE + Random Under sampling, and SMOTE-TOMEK, were applied before training. Our findings reveal that Random Forest with Borderline-SMOTE + Random Under sampling achieved the highest recall, while SMOTE + Random Under sampling with Random Forest achieved highest AUC. While hybrid machine learning models improved precision, they sacrificed recall. This study reinforces the power of ensemble sampling and hybrid approaches in credit risk modeling, with future research focusing on dynamic thresholding and advanced ensemble strategies to refine predictions. Keywords: Credit risk modeling, Class Imbalance, Ensemble sampling, Hybrid machine learning, Random Forest