diff options
Diffstat (limited to 'python/exploration/VaR.py')
| -rw-r--r-- | python/exploration/VaR.py | 113 |
1 files changed, 84 insertions, 29 deletions
diff --git a/python/exploration/VaR.py b/python/exploration/VaR.py index cf47c71c..1e6a02a3 100644 --- a/python/exploration/VaR.py +++ b/python/exploration/VaR.py @@ -1,34 +1,89 @@ -import datetime -from analytics.curve_trades import curve_pos, on_the_run -from analytics.index_data import index_returns +from analytics.curve_trades import on_the_run +from analytics.index_data import index_returns, get_index_quotes, index_returns +from analytics import Portfolio +from db import dbengine, dbconn +from pandas.tseries.offsets import BDay, BMonthEnd +from copy import deepcopy + import numpy as np import pandas as pd +import math +import datetime + +dawndb = dbengine('dawndb') +serenitasdb = dbengine('serenitasdb') + +def hist_var(portf, index_type='IG', quantile=.05, years=5): + df = index_returns(index=index_type, years=years, + tenor=['3yr', '5yr', '7yr', '10yr']) + df = (df.reset_index(['index'], drop=True). + reorder_levels(['date', 'series', 'tenor'])) + returns = df.spread_return.dropna().reset_index('series') + returns['dist_on_the_run'] = (returns. + groupby('date')['series']. + transform(lambda x: x.max() - x)) + del returns['series'] + returns = returns.set_index('dist_on_the_run', append=True).unstack('tenor') + returns.columns = returns.columns.droplevel(0) + portf.reset_pv() + + otr = on_the_run(index_type) + spreads = pd.DataFrame({'spread': portf.spread, + 'tenor': [ind.tenor for ind in portf.indices], + 'dist_on_the_run': [otr - ind.series for ind in portf.indices]}) + spreads = spreads.set_index(['dist_on_the_run', 'tenor']) + r = [] + for k, g in returns.groupby(level='date', as_index=False): + shocks = g.reset_index('date', drop=True).stack('tenor') + shocks.name = 'shocks' + portf.spread = spreads.spread * (1 + spreads.join(shocks).shocks) + r.append((k, portf.pnl)) + pnl = pd.DataFrame.from_records(r, columns=['date', 'pnl'], index=['date']) + return pnl.quantile(quantile) * math.sqrt(12) + +def rel_spread_diff(report_date = datetime.date.today(), index='HY', rolling=10): + otr = on_the_run(index) + ## look at spreads + df = get_index_quotes(index, list(range(otr - rolling, otr + 1)), + tenor=['3yr', '5yr', '7yr', '10yr'], years=5) + + duration = df.duration.xs((report_date,'5yr', otr), level = ['date','tenor', 'series'])[-1] + + df = df.xs('5yr', level='tenor')['closespread'].groupby(['date']).last() + df = df.loc['2013-01-15':report_date] + curr_spread = df.iloc[-1] + df = df.pct_change(freq='22B').dropna() + + return df.groupby('date').last(), curr_spread, duration + +def get_pos(report_date): + strats = ["HEDGE_CSO", "HEDGE_MBS", "HEDGE_CLO", "HEDGE_MAC", + "SER_IGINX", "SER_HYINX", "SER_IGCURVE", "SER_ITRXCURVE", + "IGOPTDEL", "HYOPTDEL"] + r = {} + for st in strats: + r[st] = pd.read_sql_query("SELECT * from list_cds_marks(%s, %s)", + dawndb, params=(report_date, st)) + return pd.concat(r, names=['strategy']).reset_index() + +def cleared_cds_margins(report_date=datetime.date.today()): -index_type = "IG" -portf = curve_pos(datetime.date(2018, 5, 3), index_type) + df = get_pos(report_date) -df = index_returns(index=index_type, years=5, - tenor=['3yr', '5yr', '7yr', '10yr']) -df = (df.reset_index(['index'], drop=True). - reorder_levels(['date', 'series', 'tenor'])) -returns = df.spread_return.dropna().reset_index('series') -returns['dist_on_the_run'] = (returns. - groupby('date')['series']. - transform(lambda x: x.max() - x)) -del returns['series'] -returns = returns.set_index('dist_on_the_run', append=True).unstack('tenor') -returns.columns = returns.columns.droplevel(0) -portf.reset_pv() + #Cap Allocation for Deltas + percentile = .95 #monthly 90%tile case... + shocks, widen, tighten, onTR_dur, onTR_spread = {}, {}, {}, {}, {} + for ind in ['IG', 'HY', 'EU']: + shocks[ind], onTR_spread[ind], onTR_dur[ind] = rel_spread_diff(report_date, index=ind) + widen[ind] = shocks[ind].quantile(percentile) + tighten[ind] = shocks[ind].quantile(1-percentile) -otr = on_the_run(index_type) -spreads = pd.DataFrame({'spread': portf.spread, - 'tenor': [ind.tenor for ind in portf.indices], - 'dist_on_the_run': [otr - ind.series for ind in portf.indices]}) -spreads = spreads.set_index(['dist_on_the_run', 'tenor']) -r = [] -for k, g in returns.groupby(level='date', as_index=False): - shocks = g.reset_index('date', drop=True).stack('tenor') - shocks.name = 'shocks' - portf.spread = spreads.spread * (1 + spreads.join(shocks).shocks) - r.append((k, portf.pnl)) -pnl = pd.DataFrame.from_records(r, columns=['date', 'pnl'], index=['date']) + df['onTR_notional'] = df.apply(lambda df: + df.notional * df.factor * df.duration / onTR_dur[df.p_index], axis=1) + df['widen'] = df.apply(lambda df: + df.onTR_notional * onTR_spread[df.p_index] * onTR_dur[df.p_index] * widen[df.p_index]/10000, axis=1) + df['tighten'] = df.apply(lambda df: + df.onTR_notional * onTR_spread[df.p_index] * onTR_dur[df.p_index] * tighten[df.p_index]/10000, axis=1) + delta_alloc = df.groupby('strategy').sum() + delta_alloc['total'] = delta_alloc.apply(lambda df: max(abs(df.widen), abs(df.tighten)), axis=1) + return delta_alloc |