📊 MSc Thesis - Volatility Forecasting & Time Series Analysis in R

This repository is part of my Master's thesis in Finance at the University of Minho, titled:
📖 "Forecasting FTSE-100 Volatility Using HAR-Type Models".

The research explores 16 variations of HAR-based models to enhance volatility forecasting, incorporating jumps, signed jumps, realized semivariance, and leverage effects. The models were tested using multiple loss functions and statistical validation techniques to evaluate predictive performance.

📄 Full Thesis Available Here:
🔗 Forecasting FTSE-100 Volatility Using HAR-Type Models

📂 Available Files

1️⃣ Thesis-Related Scripts

• har_rv_model.R → Implementation of HAR-RV and its variations.
• har_rv_j_model.R → HAR-RV-J model (incorporating jumps).
• lhar_rv_model.R → LHAR-RV model (leverage effect included).

2️⃣ Additional Studies (Time Series & Volatility Models)

• volatility_models.R → Overview of various volatility models (ARCH, GARCH, etc.). (Not part of the thesis but included as a complementary study.)
• time_series_analysis.R → Time series modeling techniques beyond HAR. (Not part of the thesis but included to demonstrate broader knowledge.)

3️⃣ Data Processing Scripts

• data_cleaning.R → Preprocessing and cleaning of financial data.

📊 Research Overview

This study aimed to evaluate the efficiency of HAR-based models in predicting market volatility. The methodology involved:

• Developing and testing 16 variations of HAR-Type models, incorporating different decompositions of realized volatility.
• Analyzing the impact of jumps, signed jumps, and leverage effects in volatility forecasting.
• Testing the predictive power of HAR variations using multiple loss functions and Model Confidence Set (MCS) methodology.

HAR-Type Models Used in the Thesis

The thesis explored 16 variations of the HAR model, including:

• HAR-RV → Standard HAR model using realized volatility.
• HAR-RV-J → Incorporates jumps as an explanatory variable.
• HAR-CJ → Separates realized volatility into continuous and jump components.
• HAR-RSV → Uses realized semivariance instead of realized volatility.
• HAR-RSV-J → Adds jumps to the realized semivariance model.
• HAR-RV-SJ → Uses signed jumps to differentiate between positive and negative impacts.
• HAR-RV-SSJ (I) & (II) → Further decomposition of signed jumps into lagged weekly and monthly effects.
• LHAR Variants → Each of the above models was also tested with the leverage effect (negative returns).

The results showed that weekly volatility had the strongest predictive power, and that negative realized semivariance and signed jumps provided more information than their positive counterparts. However, the Model Confidence Set (MCS) analysis indicated that no single model significantly outperformed all others in out-of-sample forecasting.

⚠ Data Disclaimer

The dataset used in this research contains proprietary financial data and cannot be shared. However, the R scripts in this repository can be used with any similar time series dataset. Users are encouraged to apply the models with their own data.

📚 References

G. M. Freitas (2020). Forecasting FTSE-100 Volatility Using HAR-Type Models.
University of Minho Repository. Available at: https://repositorium.sdum.uminho.pt/handle/1822/69474

🔧 How to Use

1. Clone the repository:

   git clone https://github.com/gustavo-m-freitas/MSc-Thesis-R.git

2. Navigate to the R_Scripts/ directory and run the .R scripts with your dataset.
3. Install required R packages before running the scripts: install.packages(c("dplyr", "ggplot2", "tidyverse", "lmtest", "slider"))