-
Notifications
You must be signed in to change notification settings - Fork 0
/
NestedCV.py
227 lines (193 loc) · 6.78 KB
/
NestedCV.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
# Import required libraries
import numpy as np
import pandas as pd
from pandas.plotting import register_matplotlib_converters
register_matplotlib_converters()
import datetime
import math
import seaborn as sns
import matplotlib as mpl
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
plt.rcParams["figure.figsize"] = (12, 8)
import sklearn.metrics as metrics
from sklearn import linear_model
from sklearn.metrics import confusion_matrix
from sklearn.preprocessing import StandardScaler
# Read csv fle from github
url = 'https://raw.githubusercontent.com/Brent-Morrison/Misc_scripts/master/econ_fin_data.csv'
df_raw = pd.read_csv(url)
# Convert date to datetime and create month end date
df_raw['date'] = pd.to_datetime(df_raw['date'], format='%Y-%m-%d')
df_raw['me_date'] = pd.Index(df_raw['date']).to_period('M').to_timestamp('M')
# Set date as index in new df
df = df_raw.set_index('me_date')
# Creation of attributes
df = df.assign(
CRED_SPRD = df.BAA - df.AAA
,YLD_SPRD = df.GS10 - df.FEDFUNDS
,LOAN_GROWTH = np.log(df.LOANS / df.LOANS.shift(6))
)
# Inspect csv data - first and last records
df.iloc[np.r_[0:4, len(df) - 4:len(df)],]
# List variables
# Dependent variable
dep_var = ['y1']
# Predictors/ independent variables
ind_vars = ['CRED_SPRD', 'YLD_SPRD', 'LOAN_GROWTH', 'rtn_6m']
# Other variables to be used in the plot
oth_vars = ['close']
vars = dep_var + ind_vars + oth_vars
df = df[vars]
# Drop na' when variables are not null / Nan
df = df.dropna(subset = vars)
# Inspect csv data - first and last records
df.iloc[np.r_[0:4, len(df) - 4:len(df)],]
# Set training and testing ranges
train_length = 300
test_length = 4
loops = math.floor((len(df) - train_length) / test_length)
start = len(df) - (loops * test_length + train_length)
stop = math.floor((len(df) - train_length) / test_length) * test_length
# Empty objects
y_pred_prob = None
model_coef = None
# Training loop
for i in range(start, stop, test_length):
# Model data
y_train_raw = np.array(df.iloc[i:i + train_length, 0])
x_train_raw = np.array(df.iloc[i:i + train_length, 1:len(vars) - 1])
y_test_raw = np.array(df.iloc[i + train_length:i + train_length + test_length, 0])
x_test_raw = np.array(df.iloc[i + train_length:i + train_length + test_length, 1:len(vars) - 1])
# Scale for model ingestion
sc = StandardScaler()
x_train = sc.fit_transform(x_train_raw)
# Apply mean and standard deviation from transform applied to training data to test data
x_test = sc.transform(x_test_raw)
# Specify model
sgd = linear_model.SGDClassifier(
loss = 'log'
,penalty = 'elasticnet'
,max_iter = 2500
,n_iter_no_change = 500
,tol = 1e-3)
# Train model
sgd.fit(x_train, y_train_raw)
# Predict on test data and write to array
y_pred = sgd.predict_proba(x_test)
if y_pred_prob is None:
y_pred_prob = y_pred
else:
y_pred_prob = np.concatenate((y_pred_prob, y_pred))
# Capture co-efficients and write to array
coef = np.repeat(
np.append(
sgd.intercept_[0],
sgd.coef_).reshape((1, -1)),
test_length,
axis = 0
)
if model_coef is None:
model_coef = coef
else:
model_coef = np.concatenate((model_coef, coef))
# Create predictions dataframe with date index
df_preds = pd.DataFrame(
data = y_pred_prob
,index = pd.date_range(
start = df.index[train_length + start]
,periods = stop
,freq = 'M')
,columns = [0, 1]
)
# Theshold for hard prediction, to populate confusion matrix
df_preds = df_preds.assign(pred = np.where(df_preds[1] > 0.25, 1, 0))
# Join predictions to df & rename prediction to pred_prob
df_preds = df_preds.join(df, how = 'inner')
df_preds = df_preds.rename(columns = {1:'pred_prob'}).drop(columns = 0)
df_preds.y1 = df_preds.y1.astype(int)
df_preds.close = np.log(df_preds['close'])
# Create co-efficients dataframe with date index
ind_vars.insert(0, 'Int')
ind_vars = [x + '_coef' for x in ind_vars]
df_model_coef = pd.DataFrame(
data = model_coef
,index = pd.date_range(
start = df.index[train_length + start]
,periods = stop
,freq = 'M')
,columns = ind_vars
)
# Join predictions & co-efficients df's
df_preds_coefs = df_preds.join(df_model_coef, how = 'inner')
# Inspect dataframe of prediction probability
df_preds_coefs.iloc[np.r_[0:4, len(df_preds_coefs) - 4:len(df_preds_coefs)],]
# Set plot style
sns.set_style('white', {"xtick.major.size": 2, "ytick.major.size": 2})
flatui = ["#c5b4cc", "#3498db", "#95a5a6", "#e74c3c", "#34495e", "#2ecc71","#f4cae4"]
sns.set_palette(sns.color_palette(flatui,7))
# Plot timeseries of SP500, prediction %, and y label shading
fig1, (ax1, ax2) = plt.subplots(nrows = 2)
fig1.suptitle('Prediction probability and S&P 500', size = 16).set_y(1.05)
fig1.subplots_adjust(top = 0.85)
ax1.plot(df_preds_coefs.index, df_preds_coefs['pred_prob'], 'k-',
color = sns.xkcd_rgb['grey'])
ax1.fill_between(
df_preds_coefs.index,
df_preds_coefs['pred_prob'],
y2 = 0,
where = df_preds_coefs['y1']
)
ax1.set_ylabel('Probability')
ax2.plot(df_preds_coefs.index, df_preds_coefs['close'], 'k-',
color = sns.xkcd_rgb['grey'])
ax2.fill_between(
df_preds_coefs.index,
df_preds_coefs['close'],
y2 = 0,
where = df_preds_coefs['y1']
)
ax2.set_ylim(bottom = 4.5)
ax2.set_ylabel('S&P500 (log scale)')
fig1.tight_layout()
# Plot parameters
fig2, (ax1, ax2, ax3, ax4) = plt.subplots(nrows = 4)
fig2.suptitle('Rolling regression parameters', size = 16).set_y(1.05)
fig2.subplots_adjust(top = 0.85)
ax1.plot(df_preds_coefs.index, df_preds_coefs['Int_coef'], 'k-',
color = sns.xkcd_rgb['grey'])
ax1.fill_between(
df_preds_coefs.index,
df_preds_coefs['Int_coef'],
y2 = df_preds_coefs['Int_coef'].min(),
where = df_preds_coefs['y1']
)
ax1.set_ylabel('Intercept')
ax2.plot(df_preds_coefs.index, df_preds_coefs['CRED_SPRD_coef'], 'k-',
color = sns.xkcd_rgb['grey'])
ax2.fill_between(
df_preds_coefs.index,
df_preds_coefs['CRED_SPRD_coef'],
y2 = df_preds_coefs['CRED_SPRD_coef'].min(),
where = df_preds_coefs['y1']
)
ax2.set_ylabel('Credit spread')
ax3.plot(df_preds_coefs.index, df_preds_coefs['YLD_SPRD_coef'], 'k-',
color = sns.xkcd_rgb['grey'])
ax3.fill_between(
df_preds_coefs.index,
df_preds_coefs['YLD_SPRD_coef'],
y2 = df_preds_coefs['YLD_SPRD_coef'].min(),
where = df_preds_coefs['y1']
)
ax3.set_ylabel('Yield spread')
ax4.plot(df_preds_coefs.index, df_preds_coefs['LOAN_GROWTH_coef'], 'k-',
color = sns.xkcd_rgb['grey'])
ax4.fill_between(
df_preds_coefs.index,
df_preds_coefs['LOAN_GROWTH_coef'],
y2 = df_preds_coefs['LOAN_GROWTH_coef'].min(),
where = df_preds_coefs['y1']
)
ax4.set_ylabel('Loan growth')
fig2.tight_layout()