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
228
229
230
231
232
|
import pandas as pd
from db import dbengine
import numpy as np
import statsmodels.api as sm
from statsmodels.formula.api import ols
from pandas.tseries.offsets import MonthEnd
from scipy.interpolate import interp1d
import datetime
from pandas.tseries.offsets import *
import bbg_helpers
#from PyQt4 import QtGui #has conflict with matplotlib
#import matplotlib
#matplotlib.use("Q4Agg") #to prevent conflict with PyQT4
def monthlySpreadDiff(index="IG", tenor="5yr"):
date = (pd.Timestamp.now() - pd.DateOffset(years=4)).date()
sql_str = "SELECT date, series, version, closespread AS spread FROM index_quotes " \
"WHERE index=%s and date>=%s and tenor = %s"
df = pd.read_sql_query(sql_str, dbengine('serenitasdb'), parse_dates=['date'], index_col=['date', 'series'],
params=(index.upper(), date, "5yr"))
df1 = df.groupby(level = ['date','series']).last()
df1 = df1.unstack(level = 1).resample('1m').last().diff()
df1 = df1.stack(level = 1).groupby(level = 'date').last()['spread']
return df1
def nearestDate(base, dates):
nearness = { abs(base.timestamp() - date.timestamp()) : date for date in dates }
return nearness[min(nearness.keys())]
def interpolate_rates(s, hist_data):
date, rate = s.BDay, s.moddur
v = hist_data.loc[date].values
return interp1d(v[:,1], v[:,0], fill_value='extrapolate')(rate)
def aux(df):
#assume shock to be 40
df['adjpv'] = (df.pv+df.delta_ir*df.yielddiff/40)
return df
#Df = Read all external Marks
df = (pd.read_sql_table('external_marks_mapped',dbengine('dawndb'),
parse_dates=['date']).
sort_values('date'))
BBG_IP = ['192.168.9.65']
#BBG_IP = ['192.168.9.61', '192.168.9.65', '192.168.0.10', '192.168.0.12']
bbgstartdate = pd.datetime(2013, 1, 1)
hist_securities = ["US0006M Index", "USSWAP2 Curncy", "USSWAP3 Curncy", "USSWAP4 Curncy", "USSWAP5 Curncy", "USSWAP7 Curncy", "USSWAP10 Curncy"]
hist_fields = ['PX_LAST']
with bbg_helpers.init_bbg_session(BBG_IP) as session:
hist_data = bbg_helpers.retrieve_data(session, hist_securities, hist_fields, start_date=bbgstartdate)
hist_data1 = pd.concat(hist_data,ignore_index=False, names = ['Rate Index', 'numIndex']).reset_index()
mapping = {"US0006M Index": .5, "USSWAP10 Curncy": 10, "USSWAP2 Curncy": 2, "USSWAP3 Curncy": 3, "USSWAP4 Curncy": 4, "USSWAP5 Curncy": 5, "USSWAP7 Curncy": 7}
mapping = pd.Series(mapping, name = 'tenor')
mapping.index.name = 'Rate Index'
mapping = mapping.reset_index()
hist_data1 = hist_data1.merge(mapping, on='Rate Index')
hist_data1 = hist_data1[['date','PX_LAST','tenor']].set_index('date')
#rate = interpolateRates(hist_data1, '2016-12-20', 3.4)
#Ddf1 = reformat dates, get rid of duplicates of external marks
df1 = df
df1['date'] = pd.to_datetime(df['date'], errors= 'coerce')
df1 = df1.sort_values('date')
df1 = df1.set_index(['date','identifier','source'])
df1 = df1.groupby(level=['date','identifier','source'])['mark'].mean()
df1 = df1.reset_index()
df1 = df1.set_index(['date'])
#df2 = mapping of monthend to the closest available model date
mapping = pd.read_sql_query("select distinct date(timestamp) as model_date from priced",dbengine('rmbs_model'), parse_dates = ['model_date']).sort_values('model_date')
df2 = df.set_index(['date'])
df2 = df2.groupby(level=['date'])['mark'].mean()
for dt in df2.index:
df2[dt]= nearestDate(dt, mapping['model_date'])
df2 = pd.DataFrame(df2)
df2.columns = ['model_date']
#merge the mapped model dates back to the external marks table
df1 = df1.join(df2)
#Df3 = all the model runs on the available model dates
df3=pd.DataFrame()
sql_str = "select timestamp as date, pv, cusip, moddur, delta_ir from priced " \
"where date(timestamp) = %s and model_version = 1"
for date in df2['model_date']:
#Note: Need to add .to_pydatetime() to be converted into a datetime passable to MYSQL. Not required for Postgres
df3= df3.append(pd.read_sql_query(sql_str,dbengine('rmbs_model'), parse_dates = ['date'], params=(date.to_pydatetime(), )))
df3= df3.rename(columns = {"cusip": "identifier"})
df3 = df3.sort_values('date')
df3['date'] = df3['date'].dt.date
df3 = df3.set_index(['date', 'identifier'])
#Now merge external monthend marks with the available model dates
df1 = df1.reset_index()
df1= df1.rename(columns = {"model_date": "date", "date": "monthend_date"})
df1 = df1.set_index(['date','identifier'])
df1 = df1.merge(df3, left_index = True, right_index= True)
df1 = df1.reset_index().sort_values('monthend_date')
availratesdate = pd.DataFrame(sorted(pd.unique(hist_data1.index)), columns=['date'])
df1['BDay'] = pd.merge_asof(df1, availratesdate, left_on = 'monthend_date', right_on = 'date', allow_exact_matches=True)['date_y']
df1temp = df1.groupby(['identifier','BDay','moddur']).mean().reset_index()
df1temp['yield'] = df1temp.apply(interpolate_rates, 1, args=(hist_data1,))
df1temp['yielddiff'] = df1temp.groupby(['identifier'])['yield'].diff()*100
df1 = pd.merge(df1, df1temp[['identifier', 'BDay', 'moddur','yielddiff']], on= ['identifier', 'BDay', 'moddur'])
#doing in roundabout way just as notes. no need to groupby unless we are doing a diff along the way
df1 = df1.groupby('identifier').apply(aux)
df1.adjpv= df1.adjpv.where(df1.adjpv.notnull(), df1.pv)
df1['spread'] = (df1['adjpv']*100-df1['mark'])/df1['moddur']*100
df1 = df1.set_index(['monthend_date','identifier','source'])
#difference = difference in spread
#Filter
SpreadFilter = 1000
SpreadDiffFilter = 250
DurationFilter = 1.5
begindate = '2014-01-01'
enddate = '2017-01-01'
interval = 1
HYDifference = monthlySpreadDiff(index = 'HY')
difference = df1[abs(df1['spread'])<SpreadFilter].groupby(level=['identifier','source'])['pv','moddur','spread'].diff(periods = interval)
difference = difference[abs(difference['spread'])<SpreadDiffFilter]
difference = difference[abs(difference['moddur'])<DurationFilter]
difference.spread = difference.spread.astype(np.float64)
#method 1: for all sources - average through Time
avgchange1 = pd.DataFrame(difference.groupby(level=['monthend_date']).mean()['spread'])
avgchange1 = avgchange1.join(HYDifference, rsuffix = 'HY')
avgchange1 = avgchange1[begindate:enddate]
regmodel1 = sm.OLS(avgchange1['spread'],avgchange1['spreadHY']).fit()
ratio1 = (avgchange1['spreadHY']/avgchange1['spread']).mean()
#method 2: look at it source by source
avgchange2 = pd.DataFrame(difference.groupby(level=['monthend_date','source']).mean()['spread'])
avgchange2 = avgchange2.unstack(level =1)
HYDifference = HYDifference.to_frame()
HYDifference.columns = pd.MultiIndex.from_arrays([['Index'],['HY']])
avgchange2 = avgchange2.merge(HYDifference, left_index = True, right_index= True)
regmodel2 ={}
ratio2 = {}
for source in avgchange2['spread']:
regmodel2[source] = sm.OLS(avgchange2['spread'][source],avgchange2['Index']['HY'], missing= 'drop').fit()
ratio2[source] = (avgchange2['spread'][source]/avgchange2['Index']['HY']).mean()
Rsquared2 = {}
params2 = {}
for source in regmodel2:
Rsquared2[source] = regmodel2[source].rsquared
params2[source] = regmodel2[source].params
#method 3: just regress mark spread change (x) to HY change (y) regardless of Time
#has filter to do JUST priceserve and pricingdirect
difference1 = difference.reset_index().set_index('monthend_date')
#difference1 = difference1.loc[difference1['source'].isin(['PRICINGDIRECT','PRICESERVE'])]
#difference1 = difference1[(difference1['source'] == 'PRICESERVE')]
HYDifference.columns = pd.MultiIndex.from_arrays([('HY',)])
difference1 = difference1.merge(HYDifference, left_index = True, right_index= True)
regmodel3 = {}
Rsquared3 = {}
params3 = {}
for source, g in difference1.groupby('source'):
regmodel3[source] = sm.OLS(g['spread'],g['HY'], missing = 'drop').fit()
Rsquared3[source] = regmodel3[source].rsquared
params3[source] = regmodel3[source].params
#regmodel3.summary()
#method 4: Let's just look at how spread changes within each source (i.e. see if spread widening ==> lower marks for most/all bonds?)
#difference.xs('PRICESERVE', level = 'source', drop_level=False)
#difference2[difference2['source']=='BVAL'].to_clipboard()
difference4 = difference1.rename(columns = {('HY'): "HY Spread"})
difference4['HY vs mark'] = difference4['spread'] - difference4['HY Spread']
difference4.index = difference4.index.rename('monthend_date')
difference4 = difference4.set_index('source', append = True)
meanmarkchange= difference4.groupby(level = ['monthend_date', 'source'])['spread'].mean().unstack(level = -1)
StdDevmarkchange = difference4.groupby(level = ['monthend_date', 'source'])['spread'].std().unstack(level = -1)
#failed code
#df1 = df
#df2 = df
#df1 = df1.set_index(['date','identifier','source'])
#df2['prevdate'] = df2['date']- MonthEnd(n=1)
#df2 = df2.rename(columns={'date': 'nextdate', 'prevdate':'date'})
#df2 = df2.set_index(['date','identifier','source'])
#import pdb; pdb.set_trace()
#df3 = df1.mark - df2.mark
#now let's get HY spreads
# index = 'IG'
# def readtenorSpreadold(index):
# serenitasdb = dbengine('serenitasdb')
# maxindexnumber = pd.read_sql_query("select max(series) from index_version where index= %s",serenitasdb, params=(index,))['max'][0]
# d={}
# seriesstart = {}
# df = pd.DataFrame()
# for i in range(maxindexnumber-8, maxindexnumber+1):
# #import pdb; pdb.set_trace()
# d[i]= pd.read_sql_query("select date, closeprice, closespread from index_quotes where index= %s and series = %s and tenor = '5yr' order by date desc",serenitasdb, parse_dates = ['date'], params=(index,i))
# seriesstart[i] = d[i].min(axis=0).date
# seriesstart[maxindexnumber+1]= pd.Timestamp(2030,1,1)
# for i in range(maxindexnumber-8, maxindexnumber+1):
# df = df.append(d[i][(d[i].date>seriesstart[i]) & (d[i].date<seriesstart[i+1])])
# df = df.sort_values('date')
# df = df.set_index('date')
# df = df.diff()
# return df.dropna()
# def interpolateRates(df, date, duration):
# set1 = False
# set2 = False
# #import pdb; pdb.set_trace()
# for tenor in sorted(df[date]['tenor']):
# if duration < tenor and set1 == False:
# lowertenor = tenor
# lowerrate = df[date][df[date]['tenor'].isin([tenor])]['PX_LAST']
# set1 = True
# elif duration < tenor and set2 == False and set1 == True:
# uppertenor = tenor
# upperrate = df[date][df[date]['tenor'].isin([tenor])]['PX_LAST']
# set2 = True
# rate = (upperrate[0]-lowerrate[0])/(uppertenor-lowertenor)*(duration - lowertenor)+lowerrate[0]
# else:
# rate = df[date][df[date]['tenor'].isin([tenor])]['PX_LAST'][0]
# return rate
|
