We need to select the appropriate return series (we will again use a 10-day holding period) and remove outliers. We will also convert returns to log returns as follows:

target = 'Returns10D'
outliers = .01
model_data = pd.concat([y[[target]], X], axis=1).dropna().reset_index('asset', drop=True)
model_data = model_data[model_data[target].between(*model_data[target].quantile([outliers, 1-outliers]).values)]

model_data[target] = np.log1p(model_data[target])
features = model_data.drop(target, axis=1).columns
dates = model_data.index.unique()

DatetimeIndex: 45114 entries, 2014-01-02 to 2015-12-16
Columns: 183 entries, Returns10D to stock_YELP INC
dtypes: float64(183)