Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
Designing a system to analyze portfolio risks and to determine optimum margin requirements
1. Designing a system to analyze portfolio risks
and to determine optimum margin
requirements
Serkan KABA & Murat ACAR
skaba@takasbank.com.tr macar@takasbank.com.tr
ISE Settlement and Custody Bank Inc.
Istanbul/TURKEY
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 1
2. Designing a system to analyze portfolio
risks and to determine optimum margin
requirements
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 2
3. Overview
• Objective
• Problem statement
• Background information
• Description of the system
• Methods
• Results
• Conclusions
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 3
4. Objective
• Designing a real-time risk management system to
evaluate portfolios
• Developing a responsive system to analyze portfolios
and orders in real-time
• Updating the portfolio risks accordingly
• Determining the best method to estimate margin
levels for every asset class
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 4
5. Problem Statement
• Uncertainty and risk are everywhere in finance
• Determining the value loss that the derivatives
portfolio could hypothetically suffer with some given
probability and assumptions
• Defining margins required by brokers from their
customers for certain kinds of transactions
• Protecting brokers from losses that may result from
adverse price changes affecting the customer's net
balance
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 5
6. ISE & ISE Data - 1
• Istanbul Stock Exchange (ISE) is a dynamic and
growing emerging market with an increasing number of
publicly traded companies, state-of-the-art technology
and strong foreign participation
• The derivatives market began operation in 2001. In
2005, Turkish Derivatives Exchange (TURKDEX)
started its operation as a successor
• ISE National 100 Index data is used to estimate PSR
values for ISE100 derivatives.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 6
8. Components of the System
• CME SPAN (http://www.cmegroup.com/clearing/risk-
management/span-overview.html)
• SPAN parameters and parameter estimation
• Realtime risk management system using SPAN
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 8
9. CME SPAN
• Developed and implemented in 1988 by Chicago
Mercantile Exchange (CME)
• Portfolio based risk and margin calculation.
• Scenario based: 16 scenarios taken into consideration
and worst loss is taken into account.
• Currently used by more than 50 exchanges and
clearinghouses.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 9
10. Realtime Risk Management System using SPAN - 1
• Trades are fed to clearing system as soon as they’re
matched.
• Trades are processed in clearing system and new
positions are calculated.
• New positions are sent to SPAN daemon for margin
calculations. For efficiency multiple SPAN calculation
engines are used and multiple portfolios are evaluated
concurrently.
• Margin requirements are sent back to clearing system
and account risk is updated accordingly.
• Notification is sent to exchange if the account becomes
risky.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 10
11. Realtime Risk Management System using SPAN - 2
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 11
12. Realtime Risk Management System using SPAN - 3
• Similar mechanism is used for processing orders of risky
accounts.
• Single order for risky accounts are fed to clearing system
as soon as they’re issued.
• Orders are processed in clearing system and hypothetical
positions are calculated.
• Hypothetical positions are sent to a separate SPAN
daemon for margin calculations.
• Margin requirement (if the order is matched) are sent
back to clearing system and the new account risk is
determined.
• Notification is sent to exchange if the account becomes
non-risky.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 12
13. Realtime Risk Management System using SPAN - 4
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 13
14. SPAN Parameters
• Price scan range (PSR)
• Volatility scan range (VSR)
• Inter-commodity spread credit
• Short option minimum (SOM)
• Intra-commodity spread charge
• Spot month charge
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 14
15. Variables used to judge methods
• All variables used to estimate VaR and judge methods we use. We
start by calculating the daily return series and estimating VaR's.
Then we judge methods based on characteristics of these VaR
estimates.
xt ISE National 100 index series
xt ISE National 100 index return series
r t= −1
xt − 1
Exceedances
T
∑ et Total Exceedances
t=1
T
∑ et Exceedance rate
t= 1
T
Average Margin
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 15
17. Methods discussed
• Extreme Value Theory
• Historical Simulation
• GARCH
• EGARCH
• Asymmetric CaViaR
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 17
18. Extreme Value Theory
• Models possibility of big losses.
• Focuses on tails instead of the entire distribution.
• Our approach uses Peaks Over Threshold method
which uses values above a high threshold and
Generalized Pareto Distribution (GPD) as limiting
distribution of these values.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 18
19. Historical Simulation
• Non-parametric method
• Directly uses historical data to estimate the current
market conditions
• Determines nth worst loss in historical data depending
on the window size and confidence level.
• 750 day window is used as opposed to other methods
which we used a 500 day window.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 19
20. GARCH
• Established by Tim Bollerslev in 1986
• Models current volatility by recent volatility values
and returns.
• We calculate the VaR using the volatility estimate by
assuming that the returns are normally distributed.
• We use GARCH(1,1) since higher lags don't improve
the results.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 20
21. EGARCH
• GARCH model considers positive and negative
returns equally.
• In practice, positive and negative returns have
asymmetric effect on volatility. For equities, volatility
increases more with a negative return shock where as
foreign exchange volatilities exhibit an opposite
relationship.
• EGARCH model established by Nelson is a form of
GARCH which considers these asymmetries.
• We calculate the VaR using the volatility estimate
using Student's t distribution.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 21
22. Asymmetric CAViaR
• Conditional autoregressive value at risk (CAViaR)
methods introduced by Engle and Manganelli.
• Focus on behavior of a quantile.
• They model quantiles autoregressively relationship.
• We use the asymmetric variation which takes into
account asymmetric effect of returns on volatility.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 22
25. Comparing PSR Estimation Methods - 1
• When the methods tested against near term data, none of the
methods is able to meet the 5‰ exceedance level.
• Extreme value and EGARCH, come close to this performing
below 1% exceedance rate.
• Next best performers are CAViaR asymmetric and historical
simulation performing below 1.5% exceedance rate.
• Results obtained from long term data show that exceedance
rates for extreme value, EGARCH and historical simulation
decreased slightly, whereas that of CAViaR asymmetric
decreased drastically for long term.
• We can conclude that CAViaR asymmetric could have
performed better for earlier years of ISE and other 4 slightly
underperformed in the near term.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 25
26. Comparing PSR Estimation Methods - 2
• When average margins are taken into account we see that
extreme value comes with a price, higher margin rates.
• Long term margin averages of the best perfoming models,
extreme value and EGARCH, are 11% and 8.9% respectively,
way above the 7.5% maintenance margin rate currently used.
• Near term averages are more acceptable 8.8% and 5.6%
respectively.
• Analyzing average margins, since both methods cause
exceedances below 1%, approximately 2 exceedances per year,
we can say that EGARCH can be more usable since it produces
much lower margin rates.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 26
27. Graph of PSR Estimates vs. Current Margin Rate
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 27
28. Conclusions
• We presented a methodology for estimating PSR
parameter for SPAN because it's important for a risk
management system to calculate optimum margin rates.
• PSR estimation methods can be further compared by
analyzing other characteristics such as margin variance.
• Margins changes can be decreased by smoothing, in
other words discarding tolerable changes in margin
levels.
• The other parameters of SPAN that we mentioned can be
estimated using similar methods.
“RACR-2011, 22-25 May, 2011, Laredo, TX USA” “S. KABA & M. ACAR” 28