Pset1, MFIN7034
Dr. Ye Luo
May 29, 2019
Excercise in high-dimensional methods.
You should hand in a copy of code and the results. You should hand in a hard copy
even if you send an e-copy before the deadline for grading purpose.
1. Merge all the factor data from Fame-French, PS and HXZ.Construct lagged factors
with 1,2,3,4,5,6 months of lag of the original factors.
2. Perform a LASSO on the Fama-French regressions for each asset that have data
from 19810131-20121231. You should drop the first a few observations to ensure that
all your factors exist. You may choose rule of thumb (2σ
√
2 log(pn)/n) for some pre-
liminary estimation of σ, e.g., pooled regression, or use cross-validation). Use Training
sample period from the beginning to 20080131, and use testing sample from 20080201
to 20121231.
3. Collect in sample and out of sample prediction errors, averaged by all stock ids.
Report them and briefly discuss.
4. Collect factors selection frequency, i.e., the percentage of a factor being selected
by LASSO over all stocks. For example, if you have 1000 stocks, and 500 of them select
the market factor, then the percentage of selection of the market factor is 50%. Report
the frequencies in a chart.
1