Reducing Monte-Carlo Noise in Complex Path-Dependent Trades: Francesco Chiminello Barclays Capital

Reducing Monte-Carlo Noise in
Complex Path-Dependent Trades

Francesco Chiminello
Barclays Capital
This presentation has been prepared by Barclays Capital - the investment banking division of Barclays Bank PLC and its affiliates
worldwide (Barclays Capital). This publication is provided to you for information purposes, any pricing in this report is indicative and is
not
intended as an offer or solicitation for the purchase or sale of any financial instrument. The information contained herein has been
obtained from sources believed to be reliable but Barclays Capital does not represent or warrant that it is accurate and complete. The
views reflected herein are those of Barclays Capital and are subject to change without notice. Barclays Capital and its respective officers,
directors, partners and employees, including persons involved in the preparation or issuance of this document, may from time to time
act
as manager, co-manager or underwriter of a public offering or otherwise deal in, hold or act as market-makers or advisors, brokers or
commercial and/or investment bankers in relation to the securities or related derivatives which are the subject of this report. Neither
Barclays Capital, nor any officer or employee thereof accepts any liability whatsoever for any direct or consequential loss arising from
any
use of this publication or its contents. Any securities recommendations made herein may not be suitable for all investors. Past
performance is no guarantee of future returns. Any modelling or back-testing data contained in this document is not intended to be a
statement as to future performance. Investors should seek their own advice as to the suitability of any investments described herein for
their own financial or tax circumstances. This communication is being made available in the UK and Europe to persons who are
investment professionals as that term is defined in Article 19 of the Financial Services and Markets Act 2000 (Financial Promotion Order)
2001. It is directed at persons who have professional experience in matters relating to investments. The investments to which is relates
are
available only to such persons and will be entered into only with such persons. Barclays Capital - the investment banking division of
Barclays Bank PLC, authorized and regulated by the Financial Services Authority (FSA) and member of the London Stock Exchange.
Copyright in this report is owned by Barclays Capital ( Barclays Bank PLC, 2005) - no part of this report may be reproduced in any
manner without the prior written permission of Barclays Capital. Barclays Bank PLC is registered in England No. 1026167. Registered
office 1 Churchill Place, London E14 5HP.

Risks by MC: re-simulation
Re-simulation approach
take a mkt parameter q
bump it by an amount dq
re-run the simulation
risks by finite differences: P
X
(q+dq) P
X
(q-dq)

Pros:
simple to implement
unbiased by construction
consistent with PL by construction

Cons:
computational requirements
MC noise
Risks by MC: re-simulation
MC noise in risks
amplified by taking small differences
larger bumps than in analytic/PDE prices

reduced by the high correlation between
simulations
reuse of same random numbers
theta can problematic
affected by local vol surface

compute risks by finite differences

bootstrap/term structure noise
Risks by MC: other approaches
Pathwise derivatives
derivation on option-by-option basis
no discontinuous payouts
difficult to compute transition probabilities

Likelihood ratio
works well with single-step MC
works very badly for daily steps MC

Benchmark instruments
similar to weighted MC
need a complete base of benchmarks
a lot of benchmarks to be chosen and computed
incomplete basis will give nonsensical results
Risks by MC
In the following, it is assumed that computation by
re-simulation is the only practically available
methodology for risks.

How can the adverse effects of MC noise be
mitigated, under the constraint of finite
computational power?

Note that by the sqrt(N) law, even modest
improvements can be significant.
Example TARN
An Energy TARN will be used for the numerical
examples.

Rationale of the choice:

complex path-dependent trade justifies use of MC
realistic commodities trade
payout is defined in terms of Asian averages
underlying is a complex portfolio of options
digital component to the payout
not Longstaff-Schwartz

Example trade: TARN
Monthly payments, 1 year duration
first payment is evaldate+2months

Underlying portfolio of options is a 3-way:
short 2 puts at 65 $/bbl
long 1 call at 90 $/bbl
short 1 call at 105 $/bbl
observing the prompt contract at all times

Early termination
if cumulative sum of call spreads reaches 40$, deal
terminates
maximum total positive payout is 40$
adjustment to last coupon in case of early
termination
Example trade: market, model, simulation
simplified forward curve: all forwards 85 $/bbl

realistic implied volatilities from the market
ATM ranging from 42% down to 32%
Skews ranging from -2% to 7%

2-factors model with term structure of volatility

all simulations are run with 10000 antithetic paths
interested in MC noise

3 simulations per test (with different seeds)

bumps: delta, vega 0.2$; vega 1%
Basic results
Seed 1 Seed 2 Seed 3 Average St Dev
MC PV -28.7252 -28.5957 -31.1591 -29.4933 1.4440
MC Error 1.0041 1.0026 1.0383 1.0150
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
May-10 Aug-10 Nov-10 Feb-11 Jun-11 Sep-11
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Basic results
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Basic results
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Gamma
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Control variates
Add and subtract to exotic payout a portfolio of
vanillas whose price is known analytically

The weights are computed so to minimise the MC
variance (linear algebra)

The vanillas are defined for the example TARN as
each of the monthly 3-ways
Basic v. control variates results
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Gamma
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Gamma
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Synthetic asset
Build an asset representing the asian averages
forward level equal to the average observed in the
TARN
volatility computed by analytic technique for Asians
(e.g., moment-matching)
Note: different vol term structure

Build a TARN going with the synthetic asset
European observations instead of Asian ones

Performance improvement by using longer MC
steps
Basic v. synthetic asset results
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Delta
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Gamma
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Gamma
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Vega
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
Weighted MC
Choose a set of analytic control trades
One size fits all or trades-specific?

Run MC normally, price exotic trade and controls

Alter each paths probability to match all controls
undetermined problem: add a penalty to deviation
form uniform (entropy, quadratic)

Price exotic trade with weighted paths

In this example: controls are fwds, ATM straddles
Basic v. weighted MC results
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Gamma
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Gamma
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Scattered TARN
How to handle the noise from TARNs cumulative
cap?

Obvious approach: soft cap
similar to digital options as call spreads
choose a range around cap level
fractionally exercise within the range
need to guess correlation of consecutive hits:
multiplication (not realistic)
minimum (not smooth)
Does not seem to work well
Scattered TARN
How to handle the noise from TARNs cumulative
cap?

Other approach: scattered TARN
choose a range around cap level
this example: 8.92 $
choose a number of evenly-spread levels
price a portfolio of TARNs, one for each level
average out the result

No need to guess correlation
Works quite well

Scattered TARN
Refinements:

Use the above approach to decide if any of the
scattered TARNs knocks out, but compute the KO
payout according to the original cap level (smaller
bias)

No need to actually price all TARNs literally, just
need to keep a counter of the highest breached
cap level

Use actual trade definition during lifecycling

Can use a range below the original cap level to
achieve path-wise conservative pricing
Basic v. scattered results
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Gamma
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Gamma
-0.07
-0.06
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Combining approaches: control variates and
scattered TARN
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
scattered TARN
Gamma
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Gamma
-0.07
-0.06
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
scattered TARN
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Bias comparison and risks recap
Seed 1 Seed 2 Seed 3 Average Stdev
Basic -28.73 -28.60 -31.16 -29.49 1.02
Control variates -29.47 -29.56 -30.57 -29.87 0.43
Synthetic asset -28.08 -28.02 -30.27 -28.79 0.91
Weighted MC -29.77 -29.41 -29.73 -29.64 0.14
Scattered -28.68 -28.49 -31.06 -29.41 1.02
Scattered+CV -29.42 -29.45 -30.48 -29.78 0.42
Delta Avg Delta stdev Gamma Avg Gamma stdev Vega Avg Vega stdev
Basic 0.2445 0.0089 -0.0402 0.0687 -0.1821 0.0358
Control variates 0.2452 0.0084 -0.0416 0.0692 -0.1832 0.0197
Synthetic asset 0.2446 0.0054 -0.0259 0.0571 -0.1815 0.0271
Weighted MC 0.2456 0.0067 -0.0340 0.0845 -0.1822 0.0387
Scattered 0.2446 0.0025 -0.0100 0.0130 -0.1828 0.0353
Scattered+CV 0.2452 0.0036 -0.0106 0.0125 -0.1839 0.0188
Conservative pricing
Why:
Bid-ask spread
Risk margins
Overhedge
as a way to embed risk margins
simpler to manage than exact hedge
reduce leverage of pin risk

How:
Inequalities on individual paths

Example: scattered TARN with cap levels moved
below the original trades
highest scattered TARN = 40$
same spread as unbiased example
Control variates and scattered TARN:
centered v. below
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Delta
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
centered v. below
Gamma
-0.07
-0.06
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
Gamma
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
centered v. below
Vega
-0.45
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
Vega
-0.4
-0.35
-0.3
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
centered v. below
Basic -29.47 -29.56 -30.57 -29.87 0.61
Centered -29.42 -29.45 -30.48 -29.78 0.42
Below -29.01 -29.01 -30.08 -29.37 0.44
Delta Avg Delta stdev Gamma Avg Gamma stdev Vega Avg Vega stdev
Basic 0.2452 0.0084 -0.0416 0.0692 -0.1832 0.0197
Centered 0.2452 0.0036 -0.0106 0.0125 -0.1839 0.0188
Below 0.2471 0.0034 -0.0020 0.0208 -0.1808 0.0154
Conservative pricing: approaching autocall
Modest differences at inception

What happens when
nearing the TARN cap
market puts TARN in the money?

All past fixings now 100$
Forward curve now 105$

centered v. below: approaching autocall
Delta
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
Delta
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
Gamma
-0.05
0
0.05
0.1
0.15
0.2
0.25
Gamma
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1
Vega
-0.06
-0.05
-0.04
-0.03
-0.02
-0.01
0
Vega
-0.018
-0.016
-0.014
-0.012
-0.01
-0.008
-0.006
-0.004
-0.002
0
0.002
Centered 13.10 13.20 13.10 13.14 0.06
Below 14.60 14.65 14.59 14.61 0.03
Delta Avg Delta stdev Gamma Avg Gamma stdev Vega Avg Vega stdev Theta Avg Theta stdev
Centered 0.0184 0.0013 0.0154 0.0108 -0.0132 0.0015 0.0989 0.0322
Below 0.0048 0.0005 0.0023 0.0066 -0.0029 0.0005 0.0291 0.0132
Acknowledgements
Thanks to Barclays Capital for providing the data and the
opportunity.
Thanks to Mircea Marinescu and Yuri Zhestkov for useful
discussions.

Reducing Monte-Carlo Noise in Complex Path-Dependent Trades: Francesco Chiminello Barclays Capital

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Reducing Monte-Carlo Noise in Complex Path-Dependent Trades: Francesco Chiminello Barclays Capital

Hochgeladen von

Copyright:

Verfügbare Formate

Reducing Monte-Carlo Noise in

Complex Path-Dependent Trades

Das könnte Ihnen auch gefallen