Principal Investigator: Pedro L. Valls Pereira
Institution: São Paulo School of Economics and CEQEF, FGV
This project aims to introduce new methodologies for econometric modeling and forecasting in high-dimensional and/or high-frequency models, as well as in mixed-frequency models.
It also seeks to develop methodologies for econometric analysis in a complex world, including structural changes using sparse or dense (factor) models.
The project is divided into four areas:
- Asset Allocation in High-Dimensional Models:
This subproject aims to assess the robustness of different portfolio construction rules based exclusively on risk measures in a high-dimensional context. It employs statistical learning models for portfolio allocation in high dimensions. - Forecasting Inflation, Exchange Rates, and Financial Series Using High-Dimensional Mixed-Frequency Models:
This involves using mixed-frequency models with high dimensionality and sparse-group Least Absolute Shrinkage and Selection Operator (LASSO) regularization. It also compares predictive power with new methodologies, such as an adapted macroeconomic random forest algorithm for mixed-frequency series, which has not yet been applied in inflation forecasting literature.
The project explores forecasting models that can adapt to abrupt structural changes, as seen during the COVID-19 pandemic. Forecasting financial time series and their volatility is one of the most studied areas in financial research, both in academia and in the financial sector, due to its broad applications and substantial impact. Central banks use exchange rate forecasts as a fundamental input in their decision-making process. Exchange rate forecasts are also crucial for macroeconomic analysis and forecasting. - Estimation and Forecasting in High-Dimensional and/or High-Frequency Models, Including Textual Data and Anomaly-Based Portfolios:
This subproject aims to develop econometric methods for estimation and inference in these types of sparse or dense systems, considering a dependence structure that combines different sparsity assumptions. Another approach is to use textual data from major Brazilian newspapers to predict stock volatility in B3 (the Brazilian stock exchange). Specifically, the project will construct systematic volatility indicators based on news, including company-specific factors.
The innovation in this subproject is the use of textual data for forecasting. Additionally, the project aims to evaluate the risk-adjusted performance of portfolios managed using semivariance for a broad set of anomaly-based portfolios, which have not yet been used in the Brazilian stock market. - Modeling Crypto-Finance:
This area focuses on the complex and rapidly growing ecosystem of digital assets. In this new environment, assets have various characteristics and uses—such as Central Bank Digital Currencies (CBDCs), stablecoins, NFTs, DeFi tokens, among others.
Therefore, understanding the interrelationship between these instruments and traditional and alternative assets is essential. Additionally, it is crucial to analyze how the co-movements between these assets evolve over time, their systemic risk and dependency structure, and the impacts of cryptocurrency adoption for different stakeholders.