My area of research is econometrics, and I am particularly interested in the intersection between econometric/statistical theory and cutting-edge machine learning tools. I focus on theoretical developments and empirical applications in finance, macroeconomics, and the evaluation of public policies, among other areas.

 

The Distant Past (2000-2015)

I have a bachelor's degree in electrical (Systems) engineering and MSc and Ph.D. degrees in electrical engineering, particularly in optimization, control theory, and statistics. Since my early academic life, I have been sure about coming a researcher. I have always aimed to use mathematical tools to solve practical and diverse real-life problems. My graduate studies gave me a unique combination of the required mathematical foundations to pursue my goal. My life as a faculty member at the Department of Economics at PUC-Rio provided interesting and relevant questions in broad areas of economics and finance. I am deeply grateful to my former and current colleagues and students for their vivid and illuminating discussions.

I finished my Ph.D. studies in 2000 after spending one year at the Stockholm School of Economics. My dissertation consisted of four research papers. Three of them help to bridge the gap between traditional nonlinear time-series models and neural networks, which were popular in computer science but primarily despised in Economics. The fourth paper proposed a combinatorial optimization algorithm to estimate multiple-regime nonlinear time-series models. The papers have been published in the IEEE Transactions of Neural Networks (2), the Journal of Time Series Analysis, and the Journal of Computational and Graphical Statistics. I pursued the same line of research during my first years at the Department of Economics at PUC-Rio by exploring the connections between econometrics and machine learning. At that time, this was not a popular line of research.

In 2010, Essie Massoumi (Emory University) and I edited a special issue of Econometric Reviews on "The Link Between Statistical Learning Theory and Econometrics: Applications in Economics, Finance, and Marketing.” This was probably one of the first attempts to bring these two areas together before the machine-learning boom in Economics years later.

During these first years, I wish to highlight the following publications:

1.   Medeiros, Marcelo C. and Álvaro Veiga (2005). A Flexible Coefficient Smooth Transition Time Series Model. IEEE Transactions on Neural Networks, 16, 97 – 113.

2.   Medeiros, Marcelo C., Timo Teräsvirta and Gianluigi Rech (2006). Building Neural Network Models for Time Series: A Statistical Approach. Journal of Forecasting, 25, 49-75.

3.   McAleer, Michael and Marcelo C. Medeiros (2008). A Multiple Regime Smooth Transition Heterogeneous Autoregressive Model for Long Memory and Asymmetries. Journal of Econometrics, 147, 104-119.

 

The Recent Past (2015-2021)

Around 2012, I started to get interested in high-dimensional datasets to the boom of big data applications and my research moved in this direction. Initially, my interest was still in predictive/forecasting models for time series data in high dimensions. However, a few years later, I got interested in methods to estimate counterfactuals to assess the impact of interventions in aggregate data. For example, what is the causal effect of a change in monetary policy on a country's inflation and economic growth? This turned out to be a very fruitful area of research. I have published the following papers on this subject:

1.  Carvalho, Carlos V., Ricardo P. Masini and Marcelo C. Medeiros (2018). ArCo: An Artificial Counterfactual Approach for High-Dimensional Panel Time-Series Data. Journal of Econometrics, 207, 353-380.

2.   Masini, Ricardo P. and Marcelo C. Medeiros (2022). Counterfactual Analysis and Inference with Nonstationary Data. Journal of Business and Economic Statistics, 40, 227–239.

3.  Masini, Ricardo P. and Marcelo C. Medeiros (2021+). Counterfactual Analysis with Artificial Controls: Inference, High Dimensions, and Nonstationarity. Journal of the American Statistical Association.

4. Fan, Jianqing, Ricardo P. Masini and Marcelo C. Medeiros (2021). Do We Exploit all Information for Counterfactual Analysis? Benefits of Factor Models and Idiosyncratic Correction. Journal of the American Statistical Association, 116, 1773–1788.

During these years, I have also had some success by doing theoretical work on estimating large dimensional time-series models. The most relevant published papers in this respect are:

1.  Medeiros, Marcelo C. and Eduardo F. Mendes (2016). L1-Regularization of High-dimensional Time-Series Models with Non-Gaussian and Heteroskedastic Innovations. Journal of Econometrics, 191, 255-271.

2. Caner, Mehmet, Marcelo C. Medeiros and Gabriel Vasconcelos (2022+). Sharpe Ratio Analysis in High Dimensions: Residual-Based Nodewise Regression in Factor Models. Journal of Econometrics, forthcoming.

I also wish to highlight the empirical work I did on forecasting in data-rich environments. Certainly, the most relevant empirical paper during this period is:

1.  Medeiros, Marcelo C., Gabriel F. Vasconcelos, Alvaro Veiga and Eduardo Zilberman (2021). Forecasting Inflation in a Data-Rich Environment: The Benefits of Machine Learning Methods. Journal of Business and Economic Statistics, 39, 98-119.

Two other research activities are worth mentioning. The first one is the NASDA/D-Lab@PUC-Rio. In 2014 I created a research laboratory funded by Lojas Americanas S.A., one of the major retail chains in Brazil. The project’s goal was to create a research environment to bring together students, faculty members, and the industry. I have been the head of the D-Lab since its creation until 2021, when I decided to follow a different path.

In early 2020, with the outburst of the Covid-19 pandemic, some colleagues and I created Covid19Analytics.com.br, a web portal with daily analysis and forecasts of new cases and deaths in Brazil. The project was very successful, with large media coverage. The methodology behind our forecast was recently published in the International Journal of Forecasting. Together with other members of the group. I have written a couple of other papers related to Covid19, which are currently being under revision for potential publication.

 

Present and Future (2022- )

I am continuously challenging myself to find interesting research topics with, whenever possible, a clear contribution from the empirical side. When thinking about the future, my goal is to keep working hard to publish at the very best outlets.

Currently, I have the following ongoing research projects:

1.    Bridging factor and sparse models

(joint with Jianqing Fan (Princeton) and Ricardo P. Masini (Princeton)

Factor and sparse models are two widely used methods to impose a low-dimensional structure in high-dimension. They are seemingly mutually exclusive. In this project, we propose a lifting method that combines the merits of these two models in a supervised learning methodology that allows to efficiently explore all the information in high-dimensional datasets. The method is based on a flexible model for high-dimensional panel data, called factor-augmented regression (FarmPredict) model with both observable or latent common factors and idiosyncratic components. This model not only includes both principal component (factor) regression and sparse regression as specific models but also significantly weakens the cross-sectional dependence and facilitates model selection and interpretability.

 

2.   Global inflation forecasting: Benefits from machine learning methods

(Joint with Erik Christian Montes-Schütte (Aarhus University, Denmark) and Tobias Skipper Nielsen (Aarhus University, Denmark))

Forecasting inflation is an essential and challenging task. Most papers usually focus on a single or a small set of countries. This paper considers the problem of simultaneously forecasting inflation from a large panel of countries. Our strategy is to explore potential (nonlinear) links among countries, and we do not rely on any additional variables apart from inflation and deterministic components, such as seasonal dummies. We also discuss the potential economic mechanisms behind the forecasting gains obtained.

 

3.   Returns forecasting and network effects in ultra-high frequency

(Joint with Erik Christian Montes-Schütte (Aarhus University, Denmark), Christian Brownlees (Pompeu Fabra University, Spain) and Daniel Borup (Aarhus University, Denmark))

This ambitious research agenda will probably result in more than one manuscript. We have collected transaction intraday price data of more than 12,000 firms over more than ten years. Our goals are as follows: (1) construct anomaly factors in ultra-high-frequency and evaluate how these factors explain the dynamics of returns; (2) evaluate potential network links among firms after controlling for common factors; (3) explore the potential for high-frequency forecasting of returns and portfolio construction; and (4) construct alternative models for realized volatility.    

 

4.   Theory for Autoeconders

(Joint with Jianqing Fan (Princeton University) and Ricardo P. Masini (Princeton University))

Autoenconder is a popular machine learning tool for nonlinear dimension reduction. It can be interpreted as a nonlinear version of the well-known principal component analysis. However, although its great popularity, little is known about its theoretical properties. The goal of this project is to establish theoretical guarantees for the use of autoenconders in dimension reduction and to provide an asymptotic theory for a new class of nonlinear factor models.