Industries

Definition of Industries

• Industry groupings are usually built on SIC codes
• We’ll use a couple of examples: Fama-French 12 industries and Fama-French 48 industries
• The mappings SIC code \(\mapsto\) industry can be found on French’s website
• Here are the 12 industries:

Dummy variables

• We can use categorical variables in numerical models by creating dummy variables.
• We create a dummy variable for each industry defined as: =1 if the firm is in the industry and =0 otherwise.
• By including dummy variables in a linear regression, we allow each industry to have a different intercept.

Example

• Get data as in the simple backtest, except
  • include siccd in the select statement
  • include ” where date=‘2020-01’ ” to make a smaller example
• Define a function get_industry(siccd) that selects the right industry for each SIC code using the Fama-French classification.
• Define the industry as “Other” if an SIC code doesn’t fit any of the ranges.

Head of the dataframe

Creating dummies

• pandas has a get_dummies function.
• scikit-learn has OneHotEncoder.
• We’ll use scikit-learn’s LinearRegression for both.

from sklearn.linear_model import LinearRegression

get_dummies

• Create dummies, add to dataframe and include in features.

d = pd.get_dummies(df.industry)
ind_names = d.columns.to_list()
features = ["bm", "roeq"] + ind_names
df2 = df.join(d)

model = LinearRegression(fit_intercept=False)
model.fit(df2[features], df2["ret"])
pd.Series(model.coef_, index=features)

bm                     -0.005865
roeq                   -0.007340
Business Equipment      0.004721
Chemicals              -0.064114
Consumer Durables      -0.042278
Consumer Nondurables   -0.036132
Energy                 -0.163661
Finance                -0.032657
Healthcare              0.023304
Manufacturing          -0.037615
Other                   0.014954
Shops                  -0.043528
Telecommunications      0.020724
Utilities               0.027766
dtype: float64

One Hot Encoder

• Add to a pipeline with LinearRegression, then fit the pipeline.
• Dummies are created as a preprocessing step within the pipeline but not added to the dataframe.

from sklearn.compose import make_column_transformer
from sklearn.preprocessing import OneHotEncoder
from sklearn.pipeline import make_pipeline