aboutsummaryrefslogtreecommitdiffstats
path: root/python/task_server/insert_tranche_quotes.py
diff options
context:
space:
mode:
Diffstat (limited to 'python/task_server/insert_tranche_quotes.py')
-rw-r--r--python/task_server/insert_tranche_quotes.py413
1 files changed, 207 insertions, 206 deletions
diff --git a/python/task_server/insert_tranche_quotes.py b/python/task_server/insert_tranche_quotes.py
index f16d76c6..5ece93f1 100644
--- a/python/task_server/insert_tranche_quotes.py
+++ b/python/task_server/insert_tranche_quotes.py
@@ -1,206 +1,207 @@
-from sqlalchemy import Table, create_engine, MetaData
-from sqlalchemy.exc import IntegrityError, DataError
-import os
-import csv
-import datetime
-import logging
-
-logger = logging.getLogger(__name__)
-
-engine = create_engine('postgresql://serenitas_user@debian/serenitasdb')
-metadata = MetaData(bind = engine)
-quotes = Table('tranche_quotes', metadata, autoload = True)
-ins = quotes.insert()
-
-def convert(x):
- if x:
- try:
- return float(x)
- except ValueError:
- logger.info("couldn't convert {0} to float".format(x))
- else:
- return None
-
-def convert_int(x):
- if x:
- try:
- return int(float(x))
- except ValueError:
- logger.info("couldn't convert {0} to int".format(x))
- else:
- return None
-
-tenordict = {'3': '3yr', '5': '5yr', '7':'7yr', '10':'10yr'}
-runningdict1 = {0: 500, 3:100, 7:100, 15: 25}
-runningdict2 = {0: 500, 3:500, 7:500, 10:100, 15:100, 30:100}
-
-def insert_quotes(year=2016, quote_dir=None):
- if not quote_dir:
- quote_dir = os.path.join(os.environ['BASE_DIR'], 'Tranche_data', 'Quotes')
- quotefiles = [f for f in os.listdir(quote_dir) if f.endswith('csv')]
- for quotefile in quotefiles:
- with open(os.path.join(quote_dir, quotefile)) as fh:
- reader = csv.DictReader(fh)
- data = []
- for csvdict in reader:
- for p in ["%d %b %Y %H:%M", "%m/%d/%Y %H:%M"]:
- try:
- timestamp = datetime.datetime.strptime(csvdict['Date'], p)
- break
- except ValueError:
- logger.error("%s: Date in the wrong format", csvdict['Date'])
- continue
- else:
- logger.error("%s: Date in the wrong format", csvdict['Date'])
-
- attach = int(csvdict['Attach'])
- detach = int(csvdict['Detach'])
- series = int(csvdict['Series'])
- version = int(csvdict['Version'])
- if csvdict['Ref']=='': #no delta quote maybe
- logger.info('ref missing')
-
- #common values to all indices
- try:
- d = {'quotedate' : timestamp,
- 'tranchedelta': convert(csvdict['Delta']),
- 'quotesource' : csvdict['Source'],
- 'series': series,
- 'version': version,
- 'attach': attach,
- 'detach': detach,
- 'tenor': tenordict[csvdict['Tenor']]
- }
- except KeyError:
- continue
-
- if csvdict['Ticker'] == 'CDX-NAHY':
- trancheupfrontbid = convert(csvdict['Price Bid'])
- trancheupfrontask = convert(csvdict['Price Ask'])
-
- try:
- trancheupfrontmid = (trancheupfrontbid+trancheupfrontask)/2
- except TypeError:
- trancheupfrontmid = None
- running = 0 if series in [9, 10] and attach==10 else 500
- d.update({'indexrefprice': convert(csvdict['Ref']),
- 'indexrefspread': 375 if series==9 else 500,
- 'trancheupfrontbid' : trancheupfrontbid,
- 'trancheupfrontmid' : trancheupfrontmid,
- 'trancheupfrontask' : trancheupfrontask,
- 'trancherunningbid' : running,
- 'trancherunningmid' : running,
- 'trancherunningask' : running,
- 'index': 'HY'
- })
-
- elif csvdict['Ticker'] == 'CDX-NAIG':
-
- if series>=28:
- continue
- trancheupfrontbid = convert(csvdict['Upfront Bid'])
- trancheupfrontask = convert(csvdict['Upfront Ask'])
- try:
- trancheupfrontmid = (trancheupfrontbid + trancheupfrontask)/2
- except TypeError:
- trancheupfrontmid = None
- if series>=25:
- running = 100
- else:
- running = runningdict2[attach] if series<19 else runningdict1[attach]
- try:
- d.update({'indexrefspread': convert_int(csvdict['Ref']),
- 'trancheupfrontbid' : trancheupfrontbid,
- 'trancheupfrontmid' : trancheupfrontmid,
- 'trancheupfrontask' : trancheupfrontask,
- 'trancherunningbid' : running,
- 'trancherunningmid' : running,
- 'trancherunningask' : running,
- 'index' : 'IG'})
- except KeyError:
- continue
- elif csvdict['Ticker'] == 'ITRAXX-Europe':
- if series<=15:
- if attach<=6:
- trancherunningbid = trancherunningmid = trancherunningask = 500 if attach<=3 else 300
- trancheupfrontbid = convert(csvdict['Upfront Bid'])
- trancheupfrontask = convert(csvdict['Upfront Ask'])
- try:
- trancheupfrontmid = (trancheupfrontbid + trancheupfrontask)/2
- except TypeError:
- trancheupfrontmid = None
- else:
- trancherunningbid = convert(csvdict['Bid'])
- trancherunningask = convert(csvdict['Ask'])
- try:
- trancherunningmid = (trancherunningbid+trancherunningask)/2
- except TypeError:
- trancherunning = None
- trancheupfrontbid = trancheupfrontmid = trancheupfrontask = 0
- if series in [19, 21, 22, 24, 26]:
- if attach<=3:
- trancherunningbid = trancherunningask = trancherunningmid = 500 if series==19 else 100
- trancheupfrontbid = convert(csvdict['Upfront Bid'])
- trancheupfrontask = convert(csvdict['Upfront Ask'])
- try:
- trancheupfrontmid = (trancheupfrontbid + trancheupfrontask)/2
- except TypeError:
- trancheupfrontmid = None
- else:
- trancherunningbid = convert(csvdict['Bid'])
- trancherunningask = convert(csvdict['Ask'])
- try:
- trancherunningmid = (trancherunningbid+trancherunningask)/2
- except TypeError:
- trancherunning = None
- trancheupfrontbid = trancheupfrontmid = trancheupfrontask = 0
-
- d.update({'indexrefspread': convert_int(csvdict['Ref']),
- 'trancheupfrontbid' : trancheupfrontbid,
- 'trancheupfrontmid' : trancheupfrontmid,
- 'trancheupfrontask' : trancheupfrontask,
- 'trancherunningbid' : trancherunningbid,
- 'trancherunningmid' : trancherunningmid,
- 'trancherunningask' : trancherunningask,
- 'index' : 'EU'})
- elif csvdict['Ticker'] == 'ITRAXX-Xover':
- if attach<35:
- ## the quote is sometimes in Price Bid, sometimes in Bid/Ask
- ## so we try both
- trancheupfrontbid = convert(csvdict['Upfront Bid'])
- trancheupfrontask = convert(csvdict['Upfront Ask'])
- try:
- trancheupfrontmid = (trancheupfrontbid + trancheupfrontask)/2
- except TypeError:
- trancheupfrontmid = None
- trancherunningbid = trancherunningask = trancherunningmid = 500
- else:
- trancherunningbid= convert(csvdict['Bid'])
- trancherunningask= convert(csvdict['Ask'])
- try:
- trancherunningmid = (trancherunningbid + trancherunningask)/2
- except TypeError:
- trancherunningmid = None
- trancheupfrontbid = trancheupfrontmid = trancheupfrontask = 0
- d.update({'indexrefspread': convert_int(csvdict['Ref']),
- 'trancheupfrontbid' : trancheupfrontbid,
- 'trancheupfrontmid' : trancheupfrontmid,
- 'trancheupfrontask' : trancheupfrontask,
- 'trancherunningbid' : trancherunningbid,
- 'trancherunningmid' : trancherunningmid,
- 'trancherunningask' : trancherunningask,
- 'index' : 'XO'})
- data.append(d)
- with engine.connect() as conn:
- for i, l in enumerate(data):
- with conn.begin() as t:
- try:
- conn.execute(ins, l)
- except (IntegrityError, DataError) as e:
- logging.debug(e.orig)
- t.rollback()
-
- os.unlink(os.path.join(quote_dir, quotefile))
-
-if __name__=="__main__":
- insert_quotes()
+from sqlalchemy import Table, create_engine, MetaData
+from sqlalchemy.exc import IntegrityError, DataError
+import os
+import csv
+import datetime
+import logging
+
+logger = logging.getLogger(__name__)
+
+engine = create_engine('postgresql://serenitas_user@debian/serenitasdb')
+metadata = MetaData(bind = engine)
+quotes = Table('tranche_quotes', metadata, autoload = True)
+ins = quotes.insert()
+
+def convert(x):
+ if x:
+ try:
+ return float(x)
+ except ValueError:
+ logger.info("couldn't convert {0} to float".format(x))
+ else:
+ return None
+
+def convert_int(x):
+ if x:
+ try:
+ return int(float(x))
+ except ValueError:
+ logger.info("couldn't convert {0} to int".format(x))
+ else:
+ return None
+
+tenordict = {'3': '3yr', '5': '5yr', '7':'7yr', '10':'10yr'}
+runningdict1 = {0: 500, 3:100, 7:100, 15: 25}
+runningdict2 = {0: 500, 3:500, 7:500, 10:100, 15:100, 30:100}
+
+def insert_quotes(year=2016, quote_dir=None):
+ if not quote_dir:
+ quote_dir = os.path.join(os.environ['BASE_DIR'], 'Tranche_data', 'Quotes')
+ quotefiles = [f for f in os.listdir(quote_dir) if f.endswith('csv')]
+ for quotefile in quotefiles:
+ with open(os.path.join(quote_dir, quotefile)) as fh:
+ reader = csv.DictReader(fh)
+ data = []
+ for csvdict in reader:
+ for p in ["%d %b %Y %H:%M", "%m/%d/%Y %H:%M"]:
+ try:
+ timestamp = datetime.datetime.strptime(csvdict['Date'], p)
+ break
+ except ValueError:
+ logger.error("%s: Date in the wrong format", csvdict['Date'])
+ continue
+ else:
+ logger.error("%s: Date in the wrong format", csvdict['Date'])
+
+ attach = int(csvdict['Attach'])
+ detach = int(csvdict['Detach'])
+ series = int(csvdict['Series'])
+ version = int(csvdict['Version'])
+ if csvdict['Ref']=='': #no delta quote maybe
+ logger.info('ref missing')
+
+ #common values to all indices
+ try:
+ d = {'quotedate' : timestamp,
+ 'tranchedelta': convert(csvdict['Delta']),
+ 'quotesource' : csvdict['Source'],
+ 'series': series,
+ 'version': version,
+ 'attach': attach,
+ 'detach': detach,
+ 'tenor': tenordict[csvdict['Tenor']]
+ }
+ except KeyError:
+ continue
+
+ if csvdict['Ticker'] == 'CDX-NAHY':
+ trancheupfrontbid = convert(csvdict['Price Bid'])
+ trancheupfrontask = convert(csvdict['Price Ask'])
+
+ try:
+ trancheupfrontmid = (trancheupfrontbid+trancheupfrontask)/2
+ except TypeError:
+ trancheupfrontmid = None
+ running = 0 if series in [9, 10] and attach==10 else 500
+ d.update({'indexrefprice': convert(csvdict['Ref']),
+ 'indexrefspread': 375 if series==9 else 500,
+ 'trancheupfrontbid' : trancheupfrontbid,
+ 'trancheupfrontmid' : trancheupfrontmid,
+ 'trancheupfrontask' : trancheupfrontask,
+ 'trancherunningbid' : running,
+ 'trancherunningmid' : running,
+ 'trancherunningask' : running,
+ 'index': 'HY'
+ })
+
+ elif csvdict['Ticker'] == 'CDX-NAIG':
+
+ if series >= 30:
+ logger.info("series doesn't exist")
+ continue
+ trancheupfrontbid = convert(csvdict['Upfront Bid'])
+ trancheupfrontask = convert(csvdict['Upfront Ask'])
+ try:
+ trancheupfrontmid = (trancheupfrontbid + trancheupfrontask)/2
+ except TypeError:
+ trancheupfrontmid = None
+ if series >= 25:
+ running = 100
+ else:
+ running = runningdict2[attach] if series<19 else runningdict1[attach]
+ try:
+ d.update({'indexrefspread': convert_int(csvdict['Ref']),
+ 'trancheupfrontbid' : trancheupfrontbid,
+ 'trancheupfrontmid' : trancheupfrontmid,
+ 'trancheupfrontask' : trancheupfrontask,
+ 'trancherunningbid' : running,
+ 'trancherunningmid' : running,
+ 'trancherunningask' : running,
+ 'index' : 'IG'})
+ except KeyError:
+ continue
+ elif csvdict['Ticker'] == 'ITRAXX-Europe':
+ if series <= 15:
+ if attach <= 6:
+ trancherunningbid = trancherunningmid = trancherunningask = 500 if attach<=3 else 300
+ trancheupfrontbid = convert(csvdict['Upfront Bid'])
+ trancheupfrontask = convert(csvdict['Upfront Ask'])
+ try:
+ trancheupfrontmid = (trancheupfrontbid + trancheupfrontask)/2
+ except TypeError:
+ trancheupfrontmid = None
+ else:
+ trancherunningbid = convert(csvdict['Bid'])
+ trancherunningask = convert(csvdict['Ask'])
+ try:
+ trancherunningmid = (trancherunningbid+trancherunningask)/2
+ except TypeError:
+ trancherunning = None
+ trancheupfrontbid = trancheupfrontmid = trancheupfrontask = 0
+ if series in [19, 21, 22, 24, 26]:
+ if attach <= 3:
+ trancherunningbid = trancherunningask = trancherunningmid = 500 if series==19 else 100
+ trancheupfrontbid = convert(csvdict['Upfront Bid'])
+ trancheupfrontask = convert(csvdict['Upfront Ask'])
+ try:
+ trancheupfrontmid = (trancheupfrontbid + trancheupfrontask)/2
+ except TypeError:
+ trancheupfrontmid = None
+ else:
+ trancherunningbid = convert(csvdict['Bid'])
+ trancherunningask = convert(csvdict['Ask'])
+ try:
+ trancherunningmid = (trancherunningbid+trancherunningask)/2
+ except TypeError:
+ trancherunning = None
+ trancheupfrontbid = trancheupfrontmid = trancheupfrontask = 0
+
+ d.update({'indexrefspread': convert_int(csvdict['Ref']),
+ 'trancheupfrontbid' : trancheupfrontbid,
+ 'trancheupfrontmid' : trancheupfrontmid,
+ 'trancheupfrontask' : trancheupfrontask,
+ 'trancherunningbid' : trancherunningbid,
+ 'trancherunningmid' : trancherunningmid,
+ 'trancherunningask' : trancherunningask,
+ 'index' : 'EU'})
+ elif csvdict['Ticker'] == 'ITRAXX-Xover':
+ if attach < 35:
+ ## the quote is sometimes in Price Bid, sometimes in Bid/Ask
+ ## so we try both
+ trancheupfrontbid = convert(csvdict['Upfront Bid'])
+ trancheupfrontask = convert(csvdict['Upfront Ask'])
+ try:
+ trancheupfrontmid = (trancheupfrontbid + trancheupfrontask)/2
+ except TypeError:
+ trancheupfrontmid = None
+ trancherunningbid = trancherunningask = trancherunningmid = 500
+ else:
+ trancherunningbid= convert(csvdict['Bid'])
+ trancherunningask= convert(csvdict['Ask'])
+ try:
+ trancherunningmid = (trancherunningbid + trancherunningask)/2
+ except TypeError:
+ trancherunningmid = None
+ trancheupfrontbid = trancheupfrontmid = trancheupfrontask = 0
+ d.update({'indexrefspread': convert_int(csvdict['Ref']),
+ 'trancheupfrontbid' : trancheupfrontbid,
+ 'trancheupfrontmid' : trancheupfrontmid,
+ 'trancheupfrontask' : trancheupfrontask,
+ 'trancherunningbid' : trancherunningbid,
+ 'trancherunningmid' : trancherunningmid,
+ 'trancherunningask' : trancherunningask,
+ 'index' : 'XO'})
+ data.append(d)
+ with engine.connect() as conn:
+ for i, l in enumerate(data):
+ with conn.begin() as t:
+ try:
+ conn.execute(ins, l)
+ except (IntegrityError, DataError) as e:
+ logging.debug(e.orig)
+ t.rollback()
+
+ os.unlink(os.path.join(quote_dir, quotefile))
+
+if __name__=="__main__":
+ insert_quotes()