Forecasting methods for finance

• Grant number: 25/03453-3
• Support Opportunities: Scholarships in Brazil - Doctorate (Direct)
• Start date: May 01, 2025
• End date: April 30, 2029
• Field of knowledge: Applied Social Sciences - Economics - Quantitative Methods Applied to Economics
• Principal Investigator: Marcelo Fernandes
• Grantee: Matheus Lopes Carrijo
• Host Institution: Escola de Economia de São Paulo (EESP). Fundação Getúlio Vargas (FGV). São Paulo , SP, Brazil
• Associated research grant: 23/01728-0 - Econometric modeling and forecasting in high dimensional models, AP.TEM

Abstract

Forecasting models are used not only in academic settings but also in commercial environments. Their applications include planning and budgeting, risk management, strategic decision-making, resource allocation, among others. These models can be studied and employed to learn from empirical evidence, improve forecasting accuracy, and explore how this learning could be used to advance both the theory and practice of forecasting. See, for example, Makridakis et al. (2020).The use of forecasting models is a continuously growing field of study. Several classic works address this topic; see, for example, Clemen (1989); Engle and Yoo (1987); Geweke et al. (1983); Andersen and Bollerslev (1998). One measure of the growth and dissemination of forecasting research is the increasing impact factor of one of the leading journals focused on forecasting, the International Journal of Forecasting. As an illustration, its JCR (Journal Citation Reports) impact factor was 1.4 in 2013 and increased to 7.02 in 2023, with a rising trend each year. Furthermore, the use and development of machine learning methods, as well as their comparisons with econometric models, have led to a field with a growing number of studies. See examples such as Gu et al. (2020), Goulet Coulombe et al. (2022), and Masini et al. (2023).Covid-19The Covid-19 pandemic brought forecasting into the spotlight, placing epidemiological models against extrapolative time series devices. Abrupt structural changes were a significant characteristic of pandemic data due to measurement errors, changes in definitions and testing, policy interventions, technological advancements, and rapidly shifting trends. In this context, the need for the development of adaptable predictive models emerged. Forecasting models for Covid-19 have been studied in: Fanelli and Piazza (2020); Grasselli et al. (2020); Castle et al. (2021); Doornik et al. (2021); Foroni et al. (2022).High-Dimensional ModelsHigh-dimensional forecasting involves predicting future values of complex datasets with a large number of variables or features. This requires specialized techniques and methods to capture the interdependence and interactions among variables. Common approaches include dimension reduction techniques such as principal component analysis (PCA) (see: He et al. (2021); Taylor and McSharry (2007); Davò et al. (2016)) or factor analysis (see: Agarwal et al. (2010); Castle et al. (2013)), LASSO (least absolute shrinkage and selection operator) and its variants, Random Forests, deep learning neural networks (see: Hastie et al. (2009); Garcia et al. (2017); Medeiros et al. (2021); Goulet Coulombe et al. (2022)), and the Autometrics algorithm (see: Castle et al. (2013); Hendry and Pretis (2023)) to reduce the number of variables while retaining the most relevant information.Financial Time SeriesFinancial time series forecasting is undoubtedly one of the most studied areas by financial researchers, both in academia and the financial sector, due to its wide range of implementations and substantial impact.This project will consider stylized facts in finance, such as the presence of heteroskedasticity and structural breaks. Robust methods such as those studied by Hwang and Valls Pereira (2006); Hwang et al. (2007); Trucíos et al. (2019); and Trucíos et al. (2021) will be considered. The forecasting of financial series volatility follows the models of Wink Junior and Pereira (2011). (AU)