#### Thesis Title

#### Date of Graduation

5-2017

#### Level of Access

Open Access

#### Degree Name

Bachelor of Arts

#### Department or Program

Economics

#### Number of Pages

65

#### First Advisor

Tefft, Nathan

#### Abstract

I develop three new types of vector autoregressions that use supervised

machine learning models to estimate coefficients in place of ordinary least

squares. I use these models to estimate the effects of monetary policy on the

real economy. Overall, I find that the machine learning vector autoregressions

produce impulse responses that are well behaved and similar to their ordinary

least squares counterparts. In practice, the machine learning vector autoregressions

produce more conservative estimates than the traditional ordinary

least squares vector autoregressions. Additionally, I establish a simulation

scheme to compare the relative efficiency of impulse responses generated from

machine learning and ordinary least squares vector autoregressions. To calculate

condence intervals, I use a bias corrected bootstrapping method from

Politis and Romano (1994) called the stationary bootstrap. In future work, I

intend to compare these impulse responses using simulated data from Killian

and Kim (2011).

#### Components of Thesis

1 pdf file

#### Recommended Citation

Varner, Michael Allen, "Vector Autoregressions with Machine Learning" (2017). *Honors Theses*. 216.

http://scarab.bates.edu/honorstheses/216