Date of Degree


Document Type


Degree Name





Merih Uctum

Committee Members

Nadia Doytch

Chu-Ping C. Vijverberg

Subject Categories

International Economics


NAIRU, NAIRO, New Keynesian Phillips Curve, LASSO, Machine Learning, Multiple Breaks


Chapter 1. Non-accelerating inflation rate of unemployment and Non-accelerating inflation rate of output We followed Ball and Mankiw (2002) to estimate the natural rates of output and unemployment. The primary purposes of this paper are to provide more accurate estimates of a varying non-accelerating inflation rate of unemployment (NAIRU) than currently exist and to nd a new measure for the nonaccelerating inflation rate of output so we can estimate the output gap more accurately. Our contributions are adding time-varying coefficients estimated with a break test and finding more accurate measurements for the natural rate of unemployment. We also estimated the time-varying natural rate of output and the output gap because of Okun's law, which shows proportionality and a negative relationship between the output gap and unemployment. We showed that our measure of the output gap predicts all the loss criteria and all periods better than the CBO output gap and the Mankiw output gap. We also showed that our measure of the unemployment gap predicted all the loss criteria and all periods better than the CBO unemployment gap and the Mankiw unemployment gap.

Chapter 2. Inflation Dynamics Unemployment, output, labor share (and other measures of slack), and inflation have been disconnected in recent years, especially after the global crisis. Some argued that the Phillips curve had disappeared. Estimating the output gap and the unemployment gap might not give accurate results because they are unobservable natural rates. We developed a structural model of inflation that allows for a fraction of firms that use a backward-looking rule to set prices, as Gali and Gertler (2000) did. The most crucial difference between their paper and ours is that they used measures of marginal cost as the relevant determinant of inflation, and we used the output gap, which we calculated with the new methodology we explained in Chapter 1. Gali and Gertler considered real marginal costs to be a significant and quantitatively important determinant of inflation, while they considered the output gap to be negative and insignificant. We have shown that an accurate measure of the output gap is significant and positive. We also used the CBO output gap as a check for robustness, but the coefficient is negative and insignificant, as the literature shows. Thus, we have concluded that the New Keynesian Phillips curve provides a good first approximation of the dynamics of inflation. Our results suggest that the hybrid new Keynesian Phillips curve may explain inflation dynamics.

Chapter 3. Economics fundamentals and exchange rates, can machine learning help? Joint with Andi Cupallari For the final essay, we studied the relationship between economic fundamentals and nominal bilateral exchange rates. We built on previous studies and evaluated machine learning models to forecast exchange rates using economic fundamentals. The last chapter is part of a growing body of literature that, in recent years, has evaluated and called into question the ability of economic fundamentals to forecast exchange rates. We followed precisely the methodology used by Li et al. (2014) to extend the data. Moreover, we introduce a new method to estimate the output gap based on the Nonaccelerating inflation Rate of Output method we introduced in Chapter 1 as an alternative to the HP-filter approach used by Li et al. (2014). We confirm their finding that machine learning outperforms all the other models out of sample. However, unlike Li et al. (2014), our results show that there is no absolute winner among the elastic let, Lasso, and ridge regressions. We find that Lasso performs better than elastic net and ridge, resulting in positive R2OOS for more currencies compared to the other two methods. We find that calculating the output gap using a different approach than Li et al. (2014) improves the out of sample performance of the model.