Training a Random Forest II

MGMT 638: Data-Driven Investments: Equity

Kerry Back, Rice University

Outline

• We follow 06b-random_forest_train to train the model but include mktvol interactions and set maxdepth=4.
• We save the trained model to forest_ver2.joblib.

Read data

import pandas as pd

url = "https://www.dropbox.com/scl/fi/hjpebns5qv0nzh1ucl4tr/data-2023-11-13.csv?rlkey=ljnn4pf04sewolnxa96t4lpdv&dl=1"
df = pd.read_csv(url)
df.head()

	ticker	date	marketcap	pb	ret	mom	volume	volatility	roe	accruals	agr	sector	mktvol
0	AACC	2011-01-14	188.3	1.4	-0.014634	-0.184615	2.078000e+04	0.071498	-0.118239	-0.182275	0.004383	Financial Services	0.055937
1	AAI	2011-01-14	1012.1	2.0	0.002677	0.438224	2.775580e+06	0.128450	0.108073	-0.220045	0.002237	Industrials	0.055937
2	AAIC	2011-01-14	189.3	1.0	-0.010119	0.684547	3.466000e+04	0.048505	0.136967	0.108055	0.135697	Real Estate	0.055937
3	AAON	2011-01-14	479.4	4.2	0.007778	0.528685	2.817291e+05	0.044912	0.191801	-0.088557	0.011656	Basic Materials	0.055937
4	AATC	2011-01-14	63.3	1.4	-0.013960	0.008216	6.800000e+03	0.049756	0.072269	-0.008792	0.089436	Technology	0.055937

Define model and target variable

from sklearn.ensemble import RandomForestRegressor
forest = RandomForestRegressor(max_depth=4)

df["target"] = df.groupby("date", group_keys=False).ret.apply(
    lambda x: 100*x.rank(pct=True)
)

Define predictors (features)

features = [
    "marketcap", 
    "pb", 
    "mom", 
    "volume", 
    "volatility", 
    "roe", 
    "accruals",
    "agr"
]
features.sort()

for x in features:
    df[x+"_vol"] = df[x]*df.mktvol

features += [x+"_vol" for x in features]

Filter to most recent 3 years

dates = df.date.unique()
dates.sort()
df = df[df.date.isin(dates[-156:])]

Train the model

forest.fit(X=df[features], y=df.target)

RandomForestRegressor(max_depth=4)

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

Save the model

# local machine version

from joblib import dump
dump(forest, "forest_ver2.joblib")

['forest_ver2.joblib']

# colab version 

from joblib import dump
from google.colab import drive
drive.mount('/content/drive')
dump(forest, "/content/drive/My Drive/forest_ver2.joblib")