Thesis Defence – Gökhan Çiftçioğlu

Thesis Defence – Gökhan Çiftçioğlu

Asst. Prof. Dr. Emrah Ahi– Advisor

Date: 31.03.2026

Time: 14:00

Location: Özyeğin University Altunizade Campus - Classroom ALT 101

“ REGIME-BASED DYNAMIC ASSET ALLOCATION: COMPARING FORECASTING MODELS IN A JUMP MODEL FRAMEWORK ”

Asst. Prof. Dr. Emrah Ahi, Özyeğin University

Asst. Prof. Dr. Levent Güntay , Özyeğin University

Assoc. Prof. Dr. Gamze Öztürk Danışman, Bilgi University

Abstract:

Financial markets change over time, and the patterns of risk and return are not always stable. For this reason, investors and researchers have long used regime-switching models to better understand market conditions and improve portfolio performance. However, existing regime-switching models face important limitations, particularly in terms of estimation stability and forecasting fl exibility. This thesis adopts and empirically evaluates a two-step framework that pairs the Jump Model (a stable and transparent unsupervised regime identifi cation method) with a set of supervised forecasting approaches, including Ordinary Least Squares, Logistic Regression, Random Forest, and Extreme Gradient Boosting (XGBoost). Using daily data from 2008 to 2025 across 12 asset classes spanning equities, fi xed income, commodities, and real estate, the framework is evaluated through both asset-level timing strategies and multi-asset portfolio simulations under equally weighted, minimum variance and mean–variance allocation rules. The results show that all regime-aware strategies outperform passive buy-and-hold benchmarks in terms of risk-adjusted returns and maximum drawdown reduction. Among the forecasting approaches, the two tree-based machine learning models Random Forest and XGBoost consistently outperform the linear benchmarks across most asset classes and portfolio construction methods. These fi ndings suggest that forecastable market regimes exist in fi nancial data, and that a real-world investor can achieve meaningfully better risk-adjusted outcomes by incorporating regime awareness into dynamic asset allocation decisions.

Keywords: Regime Switching, Jump Model, Machine Learning, Dynamic Asset Allocation, Portfolio Management

Bio:

Gökhan Çiftçioğlu, having graduated from the Department of Mechanical Engineering at Middle East Technical University (METU).