Some Essays on models in the Bond and Energy Markets

The term structure of interest rates plays a fundamental role as an indicator of economy and market trends, as well as a supporting tool for macroeconomic strategies, investment choices or hedging practices. Therefore, the availability of proper techniques to model and predict its dynamics is of crucial importance for players in the financial markets. Along this path, the dissertation initially examined the reliability of parametric and neural network models to fit and predict the term structure of interest rates in emerging markets, focusing on the Brazilian, Russian, Indian, Chines and South African (BRICS) bond markets. The focus on the BRICS is straightforward: the dynamics of their term structures make tricky the application of consolidated yield curve models. In this respect, BRICS yield curve act as stress testers. The study then examined how to apply the above cited models to energy derivatives, focusing the attention on the Natural Gas and Electricity futures, motivated by the existence of similarity. The research was carried out using ad hoc routines, such as the R package "DeRezende.Ferreira", developed by the candidate and now freely downloadable at the Comprehensive R Archive Network (CRAN) repository*, as well as by means of code written in MatLab 2021a - 2022a and Python (3.10.10) using the open-source Keras (2.4.3) library with TensorFlow (2.4.0) as backend. The dissertation consists of four chapters based on published and/or under submission materials. Chapter 1 is an excerpt of the paper • Castello, O.; Resta, M. Modeling the Yield Curve of BRICS Countries: Parametric vs. Machine Learning Techniques. Risks 2022 The work firstly offers a comprehensive analysis of the BRICS bond market and then investigates and compares the abilities of the parametric Five–Factor De Rezende–Ferreira model and Feed–Forward Neural Networks to fit the yield curves. Chapter 2 is again focused on the BRICS market but investigates a methodology to identify optimal time–varying parameters for parametric yield curve models. The work then investigates the ability of this method both for in–sample fitting and out–of–sample prediction. Various forecasting methods are examined: the Univariate Autoregressive process AR(1), the TBATS and the Autoregressive Integrated Moving Average (ARIMA) combined to Nonlinear Autoregressive Neural Networks (NAR–NN). Chapter 3 studies the term structure dynamics in the Natural Gas futures market. This chapter represents an extension of the paper • Castello, O., Resta, M. (2022). Modeling and Forecasting Natural Gas Futures Prices Dynamics: An Integrated Approach. In: Corazza, M., Perna, C., Pizzi, C., Sibillo, M. (eds) Mathematical and Statistical Methods for Actuarial Sciences and Finance. MAF 2022. After showing that the natural gas and bond markets share similar stylized facts, we exploit these findings to examine whether techniques conventionally employed on the bonds market can be effectively used also for accurate in–sample fitting and out–of–sample forecast. We worked at first in–sample and we compared the performance of three models: the Four–Factor Dynamic Nelson–Siegel–Svensson (4F-DNSS), the Five–Factor Dynamic De Rezende–Ferreira (5F–DRF) and the B–Spline. Then, we turned the attention on forecasting, and explored the effectiveness of a hybrid methodology relying on the joint use of 4F–DNSS, 5F–DRF and B–Splines with Nonlinear Autoregressive Neural Networks (NAR–NNs). Empirical study was carried on using the Dutch Title Transfer Facility (TTF) daily futures prices in the period from January 2011 to June 2022 which included also recent market turmoil to validate the overall effectiveness of the framework. Chapter 4 analyzes the predictability of the electricity futures prices term structure with Artificial Neural Networks. Prices time series and futures curves are characterized by high volatility which is a direct consequence of an inelastic demand and of the non–storable nature of the underlying commodity. We analyzed the forecasting power of several neural network models, including Nonlinear Autoregressive (NAR–NNs), NAR with Exogenous Inputs (NARX–NNs), Long Short–Term Memory (LSTM–NNs) and Encoder–Decoder Long Short–Term Memory Neural Networks (ED–LSTM–NNs). We carried out an extensive study of the models predictive capabilities using both the univariate and multivariate setting. Additionally, we explored whether incorporating various exogenous components such as Carbon Emission Certificates (CO2) spot prices, as well as Natural Gas and Coal futures prices can lead to improvements of the models performances. The data of the European Energy Exchange (EEX) power market were adopted to test the models. Chapter 4 concludes. ____________________________ * https://cran.r-project.org/web/packages/DeRezende.Ferreira/index.html