diff options
| -rw-r--r-- | python/analytics/index.py | 3 | ||||
| -rw-r--r-- | python/analytics/option.py | 2 | ||||
| -rw-r--r-- | python/exploration/option_trades.py | 91 | ||||
| -rw-r--r-- | python/mark_backtest_underpar.py | 1 | ||||
| -rw-r--r-- | python/option_trades_et.py | 95 |
5 files changed, 126 insertions, 66 deletions
diff --git a/python/analytics/index.py b/python/analytics/index.py index 8137f63f..3628721f 100644 --- a/python/analytics/index.py +++ b/python/analytics/index.py @@ -95,6 +95,7 @@ class Index(object): self._step_in_date, self._value_date, [self.end_date], np.array([self._spread]), np.zeros(1), np.array([self.recovery])) + self._risky_annuity = self._fee_leg.pv(self.trade_date, self._step_in_date, self._value_date, self._yc, self._sc, False) @@ -279,6 +280,8 @@ class Index(object): accrued_str = "Accrued ({} Days)".format(self.days_accrued) else: accrued_str = "Accrued ({} Day)".format(self.days_accrued) + if not self.spread: + raise ValueError("Market spread is missing!") s = ["{:<20}\t{:>15}".format("CDS Index", colored(self.name, attrs = ['bold'])), "", "{:<20}\t{:>15}".format("Trade Date", ('{:%m/%d/%y}'. diff --git a/python/analytics/option.py b/python/analytics/option.py index e2177d6d..4a61e90f 100644 --- a/python/analytics/option.py +++ b/python/analytics/option.py @@ -148,7 +148,7 @@ class Swaption(ForwardIndex): args = (self.forward_pv, self.exercise_date, self.exercise_date_settle, self.index, self._forward_yc, tilt, self._w) eta = 1.05 - a = self.index.spread + a = self.index.spread * 0.99 b = a * eta while True: if calib(*((b,) + args)) > 0: diff --git a/python/exploration/option_trades.py b/python/exploration/option_trades.py index 611fdc83..f9685fbb 100644 --- a/python/exploration/option_trades.py +++ b/python/exploration/option_trades.py @@ -8,11 +8,12 @@ from pandas.tseries.offsets import BDay from arch import arch_model from db import dbengine, dbconn from scipy.interpolate import interp1d -from analytics import Index +from analytics import Index, ForwardIndex from index_data import index_returns + serenitasdb = dbengine('serenitasdb') -def realized_vol(index, series, tenor='5yr', date=None, years=None): +def realized_vol(index, series, tenor='5yr', date=None, years=None, return_type='spread'): """computes the realized spread volatility""" if date is None: if years is None: @@ -20,7 +21,7 @@ def realized_vol(index, series, tenor='5yr', date=None, years=None): date = (pd.Timestamp.now() - pd.DateOffset(years=years)).date() returns = index_returns(index=index, series=series, tenor=tenor, years=None) # GARCH(1,1) volatility process with constant mean - am = arch_model(returns) + am = arch_model(returns[return_type+'_return']) res = am.fit(update_freq=0, disp='off') return (res.conditional_volatility * math.sqrt(252), res) @@ -37,45 +38,67 @@ def lr_var(res): var = res.params[names[0]]/(1 - res.params[names[1:]]) return math.sqrt(var * 252) -def atm_vol_fun(v, ref_is_price=False, moneyness=0.2): +def atm_vol_fun(v, moneyness=0.2, index=None): f = interp1d(v.strike.values, v.vol.values, fill_value='extrapolate') - atm_val = v['fwdspread'].iat[0] - otm_val = atm_val * (1 + moneyness) ## doesn't make sense for HY + if index is None: + atm_val = v['fwdspread'].iat[0] + otm_val = atm_val * (1 + moneyness) + else: + if 'HY' in index.name: + index.price = v['ref'].iat[0] + atm_val = ForwardIndex(index, v.index.get_level_values('expiry')[0], ref_is_price=True).forward_spread + otm_val = atm_val * (1 + moneyness) + index.spread = atm_val + atm_val = index.price + index.spread = otm_val + otm_val = index.price + else: + index.spread = v['ref'].iat[0] + atm_val = ForwardIndex(index, v.index.get_level_values('expiry')[0], ref_is_price=False).forward_spread + otm_val = atm_val * (1 + moneyness) return pd.Series(f(np.array([atm_val, otm_val])), index = ['atm_vol', 'otm_vol']) -def atm_vol(index, series, moneyness=0.2): - df = pd.read_sql_query('SELECT quotedate, expiry, strike, vol from swaption_quotes ' \ - 'WHERE index = %s and series = %s', - serenitasdb, index_col=['quotedate', 'expiry'], - params = (index.upper(), series)) - index_data = pd.read_sql_query( - 'SELECT quotedate, expiry, fwdspread from swaption_ref_quotes ' \ - 'WHERE index= %s and series = %s', - serenitasdb, index_col = ['quotedate', 'expiry'], - params = (index.upper(), series)) - - df = df.join(index_data) - df = df.groupby(level=['quotedate', 'expiry']).filter(lambda x: len(x)>=2) - df = df.groupby(level=['quotedate', 'expiry']).apply(atm_vol_fun, index=="HY", moneyness) - df = df.reset_index(level=-1) #move expiry back to the column - return df - -def atm_vol_date(index, date): - df = pd.read_sql_query('SELECT quotedate, series, expiry, strike, vol ' \ +def atm_vol(index, date, series = None, moneyness=0.2): + if series is None: + df = pd.read_sql_query('SELECT quotedate, series, expiry, strike, vol ' \ 'FROM swaption_quotes ' \ 'WHERE index = %s and quotedate >= %s', serenitasdb, index_col=['quotedate', 'expiry', 'series'], params=(index.upper(), date), parse_dates=['quotedate']) - index_data = pd.read_sql_query( - 'SELECT quotedate, expiry, series, fwdspread FROM swaption_ref_quotes ' \ - 'WHERE index= %s and quotedate >= %s', - serenitasdb, index_col=['quotedate', 'expiry', 'series'], - params = (index.upper(), date), parse_dates=['quotedate']) + index_data = pd.read_sql_query( + 'SELECT quotedate, expiry, series, ref, fwdspread FROM swaption_ref_quotes ' \ + 'WHERE index= %s and quotedate >= %s', + serenitasdb, index_col=['quotedate', 'expiry', 'series'], + params=(index.upper(), date), parse_dates=['quotedate']) + else: + df = pd.read_sql_query('SELECT quotedate, series, expiry, strike, vol from swaption_quotes ' \ + 'WHERE index = %s and series = %s and quotedate >= %s', + serenitasdb, index_col=['quotedate', 'expiry', 'series'], + params = (index.upper(), series, date)) + index_data = pd.read_sql_query( + 'SELECT quotedate, series, expiry, ref, fwdspread from swaption_ref_quotes ' \ + 'WHERE index= %s and series = %s and quotedate >= %s', + serenitasdb, index_col = ['quotedate', 'expiry', 'series'], + params = (index.upper(), series, date)) + df = df.join(index_data) - df = df.groupby(df.index).filter(lambda x: len(x)>=2) - df = df.groupby(level=['quotedate', 'expiry', 'series']).apply(atm_vol_fun) - df = df.reset_index(level=['expiry', 'series']) #move expiry and series back to the columns + df = df.groupby(df.index).filter(lambda x: len(x) >= 2) + + df1 = atm_vol_calc(df, index) + return df1 + +def atm_vol_calc(df, index): + g_temp = {} + for s, g1 in df.groupby(level='series'): + index_obj = Index.from_name(index, s, '5yr') + for date, g2 in g1.groupby(g1.index.get_level_values(0)): + index_obj.trade_date = date.date() + for expiry, g3 in g2.groupby(g2.index.get_level_values(1)): + g_temp[(date, expiry, s)] = atm_vol_fun(g3, index=index_obj) + df = pd.concat(g_temp, names=['quotedate', 'expiry', 'series']) + df = df.unstack(-1) + df = df.reset_index(level=['expiry', 'series']) return df def rolling_vol(df, col='atm_vol', term=[3]): @@ -93,7 +116,7 @@ def rolling_vol(df, col='atm_vol', term=[3]): return df.dropna() def vol_var(percentile=0.99, index='IG'): - df = atm_vol_date("IG", datetime.date(2014, 6, 11)) + df = atm_vol(index, datetime.date(2014, 6, 11)) df = rolling_vol(df, term=[1,2,3]) df = df.sort_index() df = df.groupby(df.index.date).nth(-1) diff --git a/python/mark_backtest_underpar.py b/python/mark_backtest_underpar.py index c6128b13..b17cc26f 100644 --- a/python/mark_backtest_underpar.py +++ b/python/mark_backtest_underpar.py @@ -7,6 +7,7 @@ import statsmodels.api as sm from statsmodels.formula.api import gls import seaborn as sb + df = pd.read_sql_table('external_marks_mapped', dbengine('dawndb'), parse_dates=['date']) df = df[df.source.notnull()] diff --git a/python/option_trades_et.py b/python/option_trades_et.py index 9f27e4b3..8b64e7b7 100644 --- a/python/option_trades_et.py +++ b/python/option_trades_et.py @@ -59,27 +59,21 @@ def beta_calc(): ltvar = lr_var(resHY) graphit = compute_allocation(all_tenors) -def build_swaption(index = 'IG', series = 27, expiry = datetime.date(2017, 4, 19), strike = 65, ref = 62, trade_date = datetime.date(2017, 2, 23), t_range = None): +def build_swaption(index, series, expiry, strike, ref, trade_date, t_range= None, spread_range = None): index_obj = Index.from_name(index, series, '5yr', trade_date) swap_obj = opt.Swaption(index_obj, expiry, strike, option_type="payer") swap_obj.notional = 100000000 if t_range is None: t_range = pd.bdate_range(trade_date, expiry- BDay(), freq = '5B') + if spread_range is None: + spread_range = pd.Series(np.arange(ref - 10, ref +19, 5)) + vol_range = pd.Series(np.arange(25, 60, 5)) #not inclusive of end point - spread_range = pd.Series(np.arange(ref - 10, ref +19, 5)) df = pd.DataFrame(index = pd.MultiIndex.from_product([t_range, spread_range, vol_range], names = ['date', 'spread', 'vol']), columns = ['pv']) df = df.reset_index() - def manual_index_update(index, date): - index._yc = index._yc.expected_forward_curve(date) - index._trade_date = date - index._step_in_date = index.trade_date + datetime.timedelta(days=1) - index._accrued = index._fee_leg.accrued(index._step_in_date) - index._value_date = (pd.Timestamp(index._trade_date) + 3* BDay()).date() - index._update() - def aux(row, index, swap): index.spread = row.spread manual_index_update(index, row.date.date()) @@ -91,7 +85,7 @@ def build_swaption(index = 'IG', series = 27, expiry = datetime.date(2017, 4, 19 #calculate mapped vol df['moneyness'] = (strike- df.spread)/df.spread - df['days_to_expiry'] = (expiry - df.date) / np.timedelta64(1,'D') + df['days_to_expiry'] = (expiry - df.date).dt.days vol_surface = build_vol_surface_functions(trade_date, index, series) df['mapped_vol'] = df.apply(vol_from_surface, axis = 1, args=(vol_surface[0], vol_surface[1])) df['mapping_shift'] = pd.to_numeric(df.vol/100 - df.mapped_vol, errors = 'ignore') @@ -99,12 +93,37 @@ def build_swaption(index = 'IG', series = 27, expiry = datetime.date(2017, 4, 19 return df +def calc_delta_pnl(index, series, ref, trade_date, notional, t_range, spread_range): + + index_obj = Index.from_name(index, series, '5yr', trade_date) + index_obj.spread = ref + index_obj.notional = notional + startingpv = -index_obj.clean_pv + + index_pv = {} + for date in t_range: + for spread in spread_range: + index_obj.spread = spread + manual_index_update(index_obj, date) + #import pdb; pdb.set_trace() + index_pv[(date, spread)] = -index_obj.clean_pv + notional* (date.date()-trade_date).days/360* index_obj.fixed_rate/10000 + + df = pd.DataFrame.from_dict(index_pv, orient = 'index').reset_index() + df['date'] = df['index'].apply(lambda x: x[0]) + df['spread'] = df['index'].apply(lambda x: x[1]) + del df['index'] + df = df.set_index(['date','spread']).sort_index() + df = (df - startingpv).unstack(-1) + df.columns = df.columns.droplevel() + + return df + def find_mapped_pv(bought, sold, date): sold = sold.xs(date).reset_index() bought = bought.xs(date).reset_index() - #Bivariate B-Spline, instead of interp2d. Interp2d doesn't behave well.... + #Bivariate B-Spline, instead of interp2d. Interp2d doesn't behave well and complains a lot. annoying x = bought.spread.unique() y = sorted(bought.mapping_shift.unique()) grid = np.meshgrid(x,y) @@ -112,7 +131,6 @@ def find_mapped_pv(bought, sold, date): f_sold = SmoothBivariateSpline(sold.spread, sold.mapping_shift, sold.pv, kx = 4, ky = 4) intp_buy = f_buy.ev(grid[0],grid[1]) intp_sold = f_sold.ev(grid[0],grid[1]) - #import pdb; pdb.set_trace() df = pd.DataFrame(intp_buy, index = grid[1][0:,0], columns = grid[0][0]) df1 = pd.DataFrame(intp_sold, index = grid[1][0:,0], columns = grid[0][0]) @@ -151,8 +169,6 @@ def plot_color_map(df, val_date): #rows are spread, columns are volatility surface shift fig, ax = plt.subplots() - #import pdb; pdb.set_trace() - #Different ways to do a colormap: imshow and pcolormesh. using imshow here midpoint = 1 - df.max().max()/(df.max().max() + abs(df.min().min())) shifted_cmap = shiftedColorMap(cm.RdYlGn, midpoint=midpoint, name='shifted') @@ -166,8 +182,6 @@ def plot_color_map(df, val_date): fig.colorbar(chart, shrink = .8) - #import pdb; pdb.set_trace() - fig.savefig("/home/serenitas/edwin/PythonGraphs/payer_swap_" + str(val_date.date()) + ".png") def build_vol_surface_functions(date = datetime.date(2017, 2, 23), index = 'IG', series = '27'): @@ -187,7 +201,7 @@ def build_vol_surface_functions(date = datetime.date(2017, 2, 23), index = 'IG', df1 = df1.groupby(df1.index).filter(lambda x: len(x) >= 2) df1 = df1.reset_index() #once the dates are in the columns you need the use .dt to access dates functions - df1['days_to_expiry'] = (df1.expiry - df1.quotedate.dt.normalize().dt.tz_localize(None)) / np.timedelta64(1,'D') + df1['days_to_expiry'] = (df1.expiry - df1.quotedate.dt.normalize().dt.tz_localize(None)).dt.days df1['moneyness'] = (df1.strike - df1.ref)/df1.ref df1 = df1.groupby(['days_to_expiry','moneyness']).nth(-1).vol df1 = df1.reset_index() @@ -201,6 +215,21 @@ def vol_from_surface(row, f, g): vol = g(row.days_to_expiry, row.moneyness) return vol +def calc_and_plot(bought, sold, traded_price, week, lowerbound, upperbound, deltaPNL=None): + + if week > len(bought.index.get_level_values(0).unique()): + week = len(bought.index.get_level_values(0).unique())-1 + + date = bought.index.get_level_values(0).unique()[week] + + PNL = find_mapped_pv(bought, sold, date) - traded_price + deltaPNL.loc[date.date()] + + PNL = PNL[lowerbound:upperbound].sort_index(ascending = False) + + plot_color_map(PNL, date) + + return PNL + def full_analysis(): index = 'IG' series = 27, @@ -210,37 +239,41 @@ def full_analysis(): sell_strike = 72 ref = 62 trade_date = datetime.date(2017, 2, 23) + delta_notional = 13000000 t_range = pd.bdate_range(trade_date, buy_expiry- BDay(), freq = '5B') + spread_range = pd.Series(np.arange(ref - 10, ref +19, 5)) - bought = build_swaption(index, series, buy_expiry, buy_strike, ref, trade_date, t_range) - sold = build_swaption(index, series, sell_expiry, sell_strike, ref, trade_date, t_range) + bought = build_swaption(index, series, buy_expiry, buy_strike, ref, trade_date, t_range, spread_range) + sold = build_swaption(index, series, sell_expiry, sell_strike, ref, trade_date, t_range, spread_range) + + delta_PNL = calc_delta_pnl(index, series, ref, trade_date, delta_notional, t_range, spread_range) #Calc PNL and Plot: traded_price = 5000 lowerbound = -.05 #parallel shift down 5% vol upperbound = .1 #parallel shift up 10% vol - week = -3 #negative to count backwards + week = 1 #negative to count backwards - PNL = calc_and_plot(bought, sold, traded_price, week, lowerbound, upperbound) + PNL = calc_and_plot(bought, sold, traded_price, week, lowerbound, upperbound, delta_PNL) + + return (bought, sold, PNL, delta_PNL) - return (bought, sold, PNL) -def calc_and_plot(bought, sold, traded_price, week, lowerbound, upperbound): - if week > len(bought.index.get_level_values(0).unique()): - week = len(bought.index.get_level_values(0).unique())-1 - date = bought.index.get_level_values(0).unique()[week] - PNL = find_mapped_pv(bought, sold, date) - traded_price - PNL = PNL[lowerbound:upperbound].sort_index(ascending = False) - plot_color_map(PNL, date) - return PNL +def manual_index_update(index, date): #index as Index Object + index._yc = index._yc.expected_forward_curve(date) + index._trade_date = date + index._step_in_date = index.trade_date + datetime.timedelta(days=1) + index._accrued = index._fee_leg.accrued(index._step_in_date) + index._value_date = (pd.Timestamp(index._trade_date) + 3* BDay()).date() + index._update() import numpy as np |