def do_one_robustness(df):
[betas_soleshared, mgr_keys,
permno_keys] = matlab_sparse(df.mgrno,
df.permno,
df.beta_soleshared,
compress=False)
[betas_sole, mgr_keys, permno_keys] = matlab_sparse(df.mgrno,
df.permno,
df.beta_sole,
compress=False)
[betas, mgr_keys, permno_keys] = matlab_sparse(df.mgrno,
df.permno,
df.beta,
compress=False)
kappa_sole = raw_kappa(betas, betas_sole)
kappa_soleshared = raw_kappa(betas, betas_soleshared)
kappa_all = raw_kappa(betas, betas)
#kappa_drop=raw_kappa(betas_drop,betas_drop)
idx = kappa_all.nonzero()
out_df = pd.DataFrame({
'from': permno_keys[idx[0]],
'to': permno_keys[idx[1]],
'kappa_all': kappa_all[idx].flatten(),
'kappa_sole': kappa_sole[idx].flatten(),
'kappa_soleshared': kappa_soleshared[idx].flatten()
})
out_df['quarter'] = df.quarter.iloc[0]
return out_df
def do_one_merger_breakup(df2):
# breakup in three blocks
blockA = df2.loc[~df2['InvestorName'].isnull(),
['mgrno', 'permno', 'beta']]
blockB = df2.loc[df2['InvestorName'].isnull(), ['mgrno', 'permno', 'beta']]
blockA.beta = 0.5 * blockA.beta
blockC = blockA.copy()
blockC.mgrno = -blockC.mgrno
df3 = pd.concat([blockA, blockB, blockC], axis=0, ignore_index=True)
# first do the regular case
[betas, mgr_keys, permno_keys] = matlab_sparse(df2.mgrno, df2.permno,
df2.beta)
k1 = calc_kappa(betas)
# now do the breakup case using the augmented data
[betas_b, mgr_keys_b,
permno_keys_b] = matlab_sparse(df3.mgrno, df3.permno, df3.beta)
k2 = calc_kappa(betas_b)
df4 = df2.groupby(['mgrno_merger', 'permno']).sum().reset_index()
# finally do the merger using the merger mgrno's instead of the real ones
[betas_m, mgr_keys_m,
permno_keys_m] = matlab_sparse(df4.mgrno_merger, df4.permno, df4.beta)
k3 = calc_kappa(betas_m)
# Ignore BlackRock+Vanguard
df4 = df2[~(df2['InvestorName'].isin(['BlackRock', 'Vanguard']))]
[betas_drop, mgr_keys_drop,
permno_keys_drop] = matlab_sparse(df4.mgrno,
df4.permno,
df4.beta,
compress=False)
k4 = calc_kappa(betas_drop)
# put it all together and return
idx = k1.nonzero()
out_df = pd.DataFrame({
'from': permno_keys[idx[0]],
'to': permno_keys[idx[1]],
'kappa': k1[idx].flatten(),
'kappa_breakup': k2[idx].flatten(),
'kappa_merger': k3[idx].flatten(),
'kappa_drop': k4[idx].flatten()
})
out_df['quarter'] = df2.quarter.iloc[0]
return out_df
def do_one_l1(df):
[betas, mgr_keys, permno_keys] = matlab_sparse(df.mgrno, df.permno,
df.beta)
l1_measure = calc_l1_measure(betas)
idx = l1_measure.nonzero()
out_df = pd.DataFrame({
'from': permno_keys[idx[0]],
'to': permno_keys[idx[1]],
'l1_measure': l1_measure[idx].flatten()
})
out_df['quarter'] = df.quarter.iloc[0]
return out_df
def do_one_investor_similarity(df):
[betas, mgr_keys, permno_keys] = matlab_sparse(df.mgrno, df.permno,
df.beta)
# Market portfolio weights
(aum, l2, l1) = investor_helper(betas)
out_df = pd.DataFrame({
'mgrno': mgr_keys.astype(int),
'aum_weight': aum,
'l2_similarity': l2,
'l1_similarity': l1,
'cov_aum_l1': np.cov(l1, aum)[1][0]
})
out_df['quarter'] = df.quarter.iloc[0]
return out_df
def do_one_firm_similarity(df):
[betas, mgr_keys, permno_keys] = matlab_sparse(df.mgrno, df.permno,
df.beta)
(aum, l2, l1) = investor_helper(betas)
norm_l2 = y = (l2 @ (betas / betas.sum(0)))
norm_l1 = y = (l1 @ (betas / betas.sum(0)))
nonnorm_l2 = y = (l2 @ betas)
nonnorm_l1 = y = (l1 @ betas)
out_df = pd.DataFrame({
'permno': permno_keys.astype(int),
'normalized_l1': norm_l1,
'nonnormalized_l1': nonnorm_l1,
'normalized_l2': norm_l2,
'nonnormalized_l2': nonnorm_l2
})
out_df['quarter'] = df.quarter.iloc[0]
return out_df
def do_one_period(df):
[betas, mgr_keys, permno_keys] = matlab_sparse(df.mgrno, df.permno,
df.beta)
kappa = calc_kappa(betas)
kappa2 = calc_kappa(betas, 2)
kappa3 = calc_kappa(betas, 3)
kappa4 = calc_kappa(betas, 0.5)
kappa5 = calc_kappa(betas, 'CLWY')
cosine = cosine_similarity(betas.transpose())
idx = kappa.nonzero()
out_df = pd.DataFrame({
'from': permno_keys[idx[0]],
'to': permno_keys[idx[1]],
'kappa': kappa[idx].flatten(),
'kappa_pow2': kappa2[idx].flatten(),
'kappa_pow3': kappa3[idx].flatten(),
'kappa_sqrt': kappa4[idx].flatten(),
'kappa_CLWY': kappa5[idx].flatten(),
'cosine': cosine[idx].flatten()
})
out_df['quarter'] = df.quarter.iloc[0]
return out_df
def do_one_period(df):
[betas, mgr_keys, permno_keys] = matlab_sparse(df.mgrno, df.permno,
df.beta)
kappa = calc_kappa(betas)
kappa2 = calc_kappa(betas, 2)
kappa3 = calc_kappa(betas, 3)
kappa4 = calc_kappa(betas, 0.5)
kappa5 = calc_kappa(betas, 'CLWY')
cosine = cosine_similarity(betas.transpose())
# this is a bit slow
l1_measure = calc_l1_measure(betas)
idx = kappa.nonzero()
return pd.DataFrame({
'from': permno_keys[idx[0]],
'to': permno_keys[idx[1]],
'kappa': kappa[idx].flatten(),
'kappa_pow2': kappa2[idx].flatten(),
'kappa_pow3': kappa3[idx].flatten(),
'kappa_sqrt': kappa4[idx].flatten(),
'kappa_CLWY': kappa5[idx].flatten(),
'cosine': cosine[idx].flatten(),
'l1_measure': l1_measure[idx].flatten()
})