Developing A Data-Driven Financial Model for Decision Support in Evaluating Investment Portfolio Performance
| dc.contributor.author | Nghilundwa, Pendapala | |
| dc.date.accessioned | 2026-02-09T12:04:45Z | |
| dc.date.available | 2026-02-09T12:04:45Z | |
| dc.date.issued | 2025-09-12 | |
| dc.description.abstract | Contemporary financial markets demonstrate heightened complexity and volatility, necessitating sophisticated instruments for the precise assessment of investment portfolios. This research explores the application of machine learning (ML) models to predict Month-to-Date (MTD) returns, aiming to enhance financial decision-making. Conventional models frequently exhibit insufficient flexibility to fluctuating market conditions, highlighting the necessity for data-driven approaches that prioritise portfolio-specific metrics, such as Market Value Dirty and Year-to-Date (YTD) returns. Meanwhile, macroeconomic variables such as gross domestic product (GDP) growth, inflation, and interest rates played a secondary role. The study employed a quantitative method, using secondary data from January 2017 to August 2024, which comprised financial measures and macroeconomic variables. Four machine learning models were developed, namely Random Forest, Gradient Boosting, XGBoost and Long Short-Term Memory (LSTM). Data preprocessing and feature engineering played a critical role in model development. Feature engineering involved creating cumulative MTD returns, moving averages (5-day and 10-day), and volatility metrics to capture market trends and risk dynamics. These features were derived from daily returns, grouped by year and month, and calculated using rolling operations. Data normalisation was applied to standardise input variables, and missing values resulting from rolling operations were filled to ensure dataset completeness. The dataset was then divided into training and testing datasets using a 1:1 ratio. Model performance was evaluated using mean squared error (MSE) and R-squared (R²) metrics, with cross-validation assuring robustness. Among the models, Gradient Boosting attained the lowest mean squared error (MSE: 2.39 × 10⁻⁶) and the highest R² (0.922), outperforming Random Forest (MSE: 2.72×10⁻⁶, R²: 0.911), XGBoost (MSE: 3.13×10⁻⁶, R²: 0.898), and LSTM (MSE: 3.98×10⁻⁶, R²: 0.870). Feature importance analysis highlighted Market Value Dirty, YTD Return, and Benchmark (BM) Size as the most influential predictors. At the same time, macroeconomic variables such as interest rates and inflation contributed minimally to short-term forecasting. This demonstrates the dominance of portfolio-specific metrics in predicting MTD returns. While LSTM excelled in capturing temporal relationships, its predictive accuracy lagged due to volatility during high-risk periods. The results affirm the effectiveness of machine learning models in enhancing financial decision-making. Gradient Boosting and Random Forest models offer accurate predictions and valuable insights into key portfolio-specific factors, underscoring their utility for risk management and strategic planning. The dissertation recommends further exploration of hybrid models, the inclusion of additional macroeconomic variables, and the integration of real-time data to enhance predictive accuracy and robustness. These improvements establish data-driven approaches as essential instruments for financial firms operating in unpredictable markets. | |
| dc.identifier.citation | Nghilundwa, P. (2025). Developing A Data-Driven Financial Model for Decision Support in Evaluating Investment Portfolio Performance [Master’s thesis, Namibia University of Science and Technology]. | |
| dc.identifier.uri | http://hdl.handle.net/10628/1119 | |
| dc.language.iso | en | |
| dc.publisher | Namibia University of Science and Technology | |
| dc.title | Developing A Data-Driven Financial Model for Decision Support in Evaluating Investment Portfolio Performance | |
| dc.type | Thesis |
Files
Original bundle
1 - 1 of 1
No Thumbnail Available
- Name:
- Pendapala_Nghilundwa_Revised_DISSERTATION - Final.pdf
- Size:
- 2.13 MB
- Format:
- Adobe Portable Document Format
License bundle
1 - 1 of 1
No Thumbnail Available
- Name:
- license.txt
- Size:
- 1.71 KB
- Format:
- Item-specific license agreed upon to submission
- Description: