cds beta hedge back-testing
DESCRIPTION
CDS Beta Hedge Back-testing. CDS Back-test Portfolio – Base Case. The back-test portfolio consists of 131 CDS names selected on the basis of liquidity and continuous data availability. All CDS contracts are of 5 years. - PowerPoint PPT PresentationTRANSCRIPT
CDS Beta Hedge Back-testing
2
CDS Back-test Portfolio – Base Case
The back-test portfolio consists of 131 CDS names selected on the basis of liquidity and continuous data availability.
All CDS contracts are of 5 years.
Because of the selection criteria, the portfolio covers most headline names, such as AIG, Ambac, MBIA, FSA, etc.
The portfolio is beta-hedged with 5Y on-the-run CDX IG index, rebalanced every 2 days.
The hedged and unhedged portfolio P/L is tracked from Jan 2008 to Mar 2009.
The period covers 2 major stress events – Bear Sterns merged to JP Morgan in March and Lehman bankruptcy in September, and subsequent market dislocation after that.
All CDS quotes are sourced from MarkIt Partners.
3
Back-test Result
-10%
-8%
-6%
-4%
-2%
0%
2%
2008
0104
2008
0204
2008
0304
2008
0403
2008
0501
2008
0530
2008
0627
2008
0728
2008
0825
2008
0923
2008
1023
2008
1121
2008
1222
2009
0122
2009
0220
Cum
ulat
ive
P/L
Unhedged LogSpread Beta Hedged Level Spread Beta Hedged
Prior to September 2008, the hedged portfolio generally kept cumulative P/L below 1% of portfolio notional.
While the performance deteriorated after Lehman’s bankruptcy, the (log spread) beta hedged portfolio can still limit the cumulative loss to 2% up to mid March 2008. During the same period, the unhedged portfolio incurred a 8% loss.
Log spread beta hedge is more effective than level spread beta.
* Cumulative P/L calculation is described in Appendix.
4
The hedge size is reasonably stable, except during the month of Sept 2008.
As a word of caution, daily P/L swing of the beta hedged portfolio could still be large, even though these fluctuations tend to cancel out over time.
How stable is the hedge ratio?
Perfect Hedge LineDeviation from the perfect hedge line represents residual P/L of the hedged portfolio
0%
20%
40%
60%
80%
100%
120%
140%
160%
2008
0102
2008
0131
2008
0229
2008
0401
2008
0429
2008
0528
2008
0625
2008
0724
2008
0821
2008
0919
2008
1021
2008
1119
2008
1218
2009
0120
2009
0218
2009
0318
Hed
ge A
mou
nt (a
s %
of p
ortf
olio
not
iona
l)
-3%
-2%
-1%
0%
1%
2%
3%
-3% -2% -1% 0% 1% 2% 3%
Index 2-day P/L
Unh
edge
d Po
rtfo
lio 2
-day
P/L
5
Different Hedging Intervals
-10%
-8%
-6%
-4%
-2%
0%
2%
Cum
ulat
ive
P/L
Unhedged log Spread Beta Hedged
-10%
-8%
-6%
-4%
-2%
0%
2%
2008
0109
2008
0207
2008
0307
2008
0408
2008
0506
2008
0604
2008
0702
2008
0731
2008
0828
2008
0926
2008
1028
2008
1126
2008
1226
2009
0127
2009
0225
Cum
ulat
ive
P/L
Unhedged log Spread Beta Hedged
Rebalance Daily
Rebalance Weekly
We don’t observe significant difference in hedging performance as changing to 1-day or 5-day hedging intervals, especially prior to September 2008.
This observation is consistent to the stability of the hedging ratio shown in previous slide.
6
How about a smaller portfolio?
-12%
-10%
-8%
-6%
-4%
-2%
0%
2%
4%
2008
0104
2008
0204
2008
0304
2008
0403
2008
0501
2008
0530
2008
0627
2008
0728
2008
0825
2008
0923
2008
1023
2008
1121
2008
1222
2009
0122
2009
0220
Cum
ulat
ive
P/L
Unhedged log Spread Beta Hedged
We randomly selected 40 names from the universe tested in the base case*. (It includes financial names, such as AIG, Ambac.)
We observed slightly higher volatility of the hedged portfolio, but it is still substantially lower than un-hedged.
* The random names are chosen based on alphabetical order of respective CDS tickers
7
Introducing Tenor Mismatch…
-16%
-14%
-12%
-10%
-8%
-6%
-4%
-2%
0%
2008
0104
2008
0204
2008
0304
2008
0403
2008
0501
2008
0530
2008
0627
2008
0728
2008
0825
2008
0923
2008
1023
2008
1121
2008
1222
2009
0122
2009
0220
Cum
ulat
ive
P/L
Unhedged log Spread Beta Hedged
In real CDS portfolios, contracts may have tenors different from the hedging index.
We created a portfolio of the same 131 names in the universe, but with 10-year maturity. We beta-hedged with 5-year index.
At the end of the back-testing period, the cumulative P/L of the hedged portfolio is -2% vs. -10% of the unhedged portfolio.
8
Stress Test…
-20%
-15%
-10%
-5%
0%
5%
10%
15%
2008
0104
2008
0204
2008
0304
2008
0403
2008
0501
2008
0530
2008
0627
2008
0728
2008
0825
2008
0923
2008
1023
2008
1121
2008
1222
2009
0122
2009
0220
Cum
ulat
ive
P/L
Unhedged log Spread Beta Hedged
We created a small portfolio using 23 financial names (e.g. Ambac, AIG, CIT, etc.)
Not surprisingly, the hedge deteriorated after watershed event of Lehman bankruptcy.
It seems to suggest that financials, as a group, are subject to additional common factor other than the index after Sept 2008.
Appendix
10
Beta of CDS contract i is defined as
where Δln(si)=log spread change of CDS contract i,
Δln(sM)=log spread change of market benchmark, and
Log CDS Beta – Statistical Definition
2)]ln([)]ln()ln([)]ln([)]ln([)]ln()ln([
MMM
MiMii sEssE
sEsEssE
iii
iiii
M DVw
sDVws
01
01
iiii
iiiii
p sDVw
sDVw
01
01
E[.] denotes exponentially weighted expectation. (In our implementation, we use decay factor with 6-month half-life.)
Portfolio beta βp is calculated as
By construction, market beta is equal to 1.
11
MTM change of CDS contract i, Δvi , can be expressed as
Index Equivalent (H) represents the notional amount of benchmark index contract (CDX IG 5Y in our case) that can offset the MTM change Δvi
Index Equivalent for the portfolio is simply the sum of individual Hi
Portfolio index equivalent is interpreted as the notional of index contract to hedge portfolio systematic risk.
Index Equivalent
MiM
iii
iiii
sssDVN
sDVNv
01
01
MM
iiiii
MMiMiM
iii
sDVsDVNH
sDVHsssDVN
0101
0101
12
P/L over a fixed interval (e.g. daily) of contract i is calculated by
Thus, cumulative P/L from time 0 to time T (non-overlapping) is
where
P/L Calculation
iiiiittttsDVsNPnL
01
kt
PnL
k
Tt
t
it
n
k
1
13
Index Intrinsic Value
0
50
100
150
200
250
300
350
-60
-40
-20
0
20
40
60
80
100
120
140
160Composite Spread
Intrinsic Spread
Dif ference
Index Roll Index Roll
Source: MarkIt
The correlation between single name and index decreases after September 08.
In fact, the index trading level started to deviate from its intrinsic value (calculated by MarkIt) from almost 0 to 50 bp since then.
This discrepancy introduces additional volatility to the hedged portfolio.