THESIS
2017
vii, 38 pages : color illustrations ; 30 cm
Abstract
Do anomalies still exist? Which characteristics provide independent information? To
address these questions, I use Fama-Macbeth and machine learning techniques (Lasso,
elastic net, tree, boosting) to select characteristics.
I construct 40 characteristics in finance and accounting from 1980 to 2016. First perform
Fama-Macbeth regressions to identify those with t3. Next run Lasso regressions, and
identify a few significant characteristics: B/M, Invest, Size, turnover, spread and illiquid.
When loosen the penalty, profit, SUE, ROA, cash, mom1m, and IPO are selected.
Other supervised learning methods are also implemented. Linear model, penalized
models (lasso, elastic net) and boosting perform best out of sample.
Finally, I take a closer look at the Ivol anomaly, by examining chara...[
Read more ]
Do anomalies still exist? Which characteristics provide independent information? To
address these questions, I use Fama-Macbeth and machine learning techniques (Lasso,
elastic net, tree, boosting) to select characteristics.
I construct 40 characteristics in finance and accounting from 1980 to 2016. First perform
Fama-Macbeth regressions to identify those with t>3. Next run Lasso regressions, and
identify a few significant characteristics: B/M, Invest, Size, turnover, spread and illiquid.
When loosen the penalty, profit, SUE, ROA, cash, mom1m, and IPO are selected.
Other supervised learning methods are also implemented. Linear model, penalized
models (lasso, elastic net) and boosting perform best out of sample.
Finally, I take a closer look at the Ivol anomaly, by examining characteristics which are
supposed to explain the puzzle.
Post a Comment