Date of Degree

9-2022

Document Type

Dissertation

Degree Name

Ph.D.

Program

Economics

Advisor

Christos Giannikos

Committee Members

Sebastiano Manzan

Thom B. Thurston

Subject Categories

Econometrics | Finance | Macroeconomics

Keywords

big data, machine learning models, Federal funds rates, futures markets, Interbank Equilibrium Interest Rate

Abstract

Chapter 1 - Big Data And Machine Learning To Predict Overnight Interest Rates. This paper is a brief introduction to the two main pieces I have elaborated as part of the dissertation. Here, I explain the reasons why I have done my research about predicting the overnight interest rates for Mexico and the United States using big data and machine learning models. I explain the connection between the two research papers, I define some basic concepts such as future contracts and the overnight funding rate for Mexico. There is a summary about the data I use, and the machine learning models I apply. Finally, I conclude with some further research that can be done based on the results found.

Chapter 2 - The Fed Funds Futures Rate As A Forecast Of The Fed Funds Rate: A Machine Learning Approach. In this paper I study and predict the Federal funds rate using the Fed Funds futures and other macroeconomic and financial variables applying machine learning models. From 2000 through 2019, I apply shrinkage methods, dimension reduction techniques, and support vector regressions to a monthly dataset. I find that machine learning models outperform the Fed funds futures when considering extra indicators for the one-, two, and three-month horizons. A post-Lasso analysis shows that leading indicators, and financial variables are significant, weakening the rationality assumption. Strong results given by the Support Vector Regressions (SVR) suggest a non-linear relation between explanatory variables and the Fed funds rate. These findings provide additional information that can improve hedging against changes in monetary policy rates as well as making higher profits when lending money. Finally, when splitting the data in two monetary policy regimes, before and after the 2008-09 financial crisis, mixed results are found.

Chapter 3 - Forecasting Mexico’s Overnight Funding Rate. This paper uses financial and macroeconomic data to predict Mexico’s overnight funding rate. I find that machine learning models yield robust results when forecasting Mexico’s interest rate using big data. The Lasso and ElasticNet provide the lowest out-of-sample forecast errors. Forecast errors are smaller the shorter the forecast horizon. The Feature Importance algorithm and post-Lasso analysis show the Federal funds futures are relevant to forecast the overnight funding rate. Not only domestic indicators are relevant for predicting the Mexican rate, but other type of financial and macroeconomic variables such as leading indicators, monetary aggregates, or business confidence for the U.S show to be important as well. These findings suggest there is a degree of integration between Mexico and the U.S. in terms of monetary policy decisions and that the Bank of Mexico might be applying a forward-looking strategy when making monetary policy decisions.

Share

COinS