Application of Histogram-valued time series data for econometric models

Abstract

In financial markets, asset prices (stocks, bonds, exchange rates, etc.) are high-frequency data, such as tick-by-tick information recorded every second. Throughout the day, asset prices fluctuate continuously before settling at the closing price. Consequently, the closing price each day does not capture the full extent of price volatility throughout the day. Relying solely on such data may overlook intraday volatility and the distribution patterns of prices, potentially leading to inaccuracies or errors when estimating values from low-frequency data. However, using high-frequency data for analysis also presents challenges, including extreme values and varying distribution patterns. Traditional models may therefore be inefficient for such data. Thus, it is essential to preprocess the data before estimation. One effective method is organizing the data into histograms, which allows for the selection of data that better reflects broader market conditions. Currently, uncertainty plays a crucial role in financial markets, significantly influencing investment decisions. Therefore, the selection of data for analysis is of paramount importance. The considerations have motivated the development of this research. The primary focus of this thesis is on financial econometrics, particularly concerning the application of histogram data to financial econometric models. First, in the context of forecasting asset returns, we face the challenge of estimating asset returns accurately despite inherent uncertainties. Economists and investors often associate this uncertainty with risk, defined as the range of outcomes governed by probability. Hence, the objective of this study is to introduce a quantile forecasting approach to the Capital Asset Pricing Model (CAPM) using histogram data to predict the returns of two stocks, Apple (AAPL) and Microsoft (MSFT). The S&P 500 index and U.S. government bonds are utilized to represent market returns and risk-free rates, respectively. The primary innovation of this study lies in its capability to promptly analyze specific datasets within predefined quantiles using histogram data. Secondly, this study focuses on analyzing the dynamic correlations among the stock markets of ASEAN-5 (Thailand, Malaysia, the Philippines, Singapore, and Indonesia). The study aims to investigate these evolving relationships in the aftermath of the COVID-19 pandemic and the Russia-Ukraine conflict by utilizing histogram data within the framework of the Dynamic Conditional Correlation Generalized Autoregressive Conditional Heteroskedasticity (HVTS-DCC-GARCH) model. Additionally, the study aims to apply this enhanced model to analyze the financial markets of Asian countries. Asia, a region characterized by diverse economies and high interconnectivity, plays a critical role in the global economy. By focusing on Asian markets, the research intends to uncover how these markets react to regional and global events, how their interdependencies evolve over time, and what implications these shifts hold for investors and policymakers. Finally, this study also applies histogram data in machine learning models. The primary objective of this research is to utilize time series data in a hybrid modeling approach to forecast movements in foreign exchange markets. Hybrid modeling, which involves combining forecasts generated from multiple models, is known to enhance forecasting accuracy. This practice has become increasingly widespread due to its improved predictive performance. The application of hybrid models has significant potential to aid policymakers in strategic planning and decision-making, as well as to assist investors in formulating effective financial strategies. By integrating these hybrid models and evaluating them against various metrics, we aim to advance the forecasting techniques for exchange rates within the context of histogram-valued data. Acknowledging the inherent complexity of this field, this study introduces a novel approach by exploring hybrid models that combine multiple models to achieve better forecasting performance. This method of hybrid modeling is relatively new to the literature and leads to improved exchange rate forecasts.