Three Essays on Data-Driven Methods in Asset Pricing and Forecasting
| Author: | Barbora Gregor (12.10.2022) |
|---|---|
| Year: | 2022 - winter |
| Leaders: | doc. PhDr. Jozef Baruník Ph.D. |
| Consultants: | |
| Work type: | Dissertations |
| Language: | English |
| Pages: | 145 |
| Awards and prizes: | |
| Link: | |
| Abstract: | This dissertation thesis consists of three papers focusing on applications of data-driven methods in asset pricing and forecasting. In the first paper, we decompose the term structure of crude oil futures prices using dynamic Nelson-Siegel model and propose to forecast them with the generalized regression framework based on neural networks. We find the neural networks to produce significantly more accurate forecasts as compared to several benchmark models. The second paper demonstrates how time-varying coefficients model can help to explore dynamics in risk-return trade-off on sovereign bond market across entire term structure. Our extensive 12-year dataset of high-frequency data of U.S. and German sovereign bond prices of 2-year, 5-year, 10-year and 30-year tenors allows us to construct realized measures of risk as well as exploring risk-return relationship under various market conditions. In addition to realized volatility, we find realized kurtosis to be priced in bond returns. Importantly, we detect the risk factor captured by realized kurtosis to have positive effect on returns in crisis turning to negative values in calm periods. In the third paper, we use time-varying coefficients methodology and higher realized moments in bond volatility forecasting challenging the HAR model. We detect realized kurtosis to be valuable volatility predictor across the term structure especially for the shorter tenors. Time-varying coefficient models are found to bring significant out-of-sample forecasting accuracy gain at the short end of the term structure. |
