We will introduce the Fama—French data to estimate the exposure of assets to common risk factors using linear regression in Chapter 8, Time Series Models. The five Fama—French factors, namely market risk, size, value, operating profitability, and investment have been shown empirically to explain asset returns and are commonly used to assess the risk/return profile of portfolios. Hence, it is natural to include past factor exposures as financial features in models that aim to predict future returns.

We can access the historical factor returns using the pandas-datareader and estimate historical exposures using the PandasRollingOLS rolling linear regression functionality in the pyfinance library as follows:

factors = ['Mkt-RF', 'SMB', 'HML', 'RMW', 'CMA']
factor_data = web.DataReader('F-F_Research_Data_5_Factors_2x3', 
              'famafrench', start='2000')[0].drop('RF', axis=1)
factor_data.index = factor_data.index.to_timestamp()
factor_data = factor_data.resample('M').last().div(100)
factor_data.index.name = 'date'
factor_data = factor_data.join(data['return_1m']).sort_index()

T = 24
betas = (factor_data
         .groupby(level='ticker', group_keys=False)
         .apply(lambda x: PandasRollingOLS(window=min(T, x.shape[0]-1), y=x.return_1m, x=x.drop('return_1m', axis=1)).beta))

We will explore both the Fama—French factor model and linear regression in Chapter 7, Linear Models in more detail. See the notebook for additional examples.