Vous êtes sur la page 1sur 11

Accelerating

Portfolio-Level Analysis
with Third Party Analytics
Accelerating Portfolio-Level Analysis
with Third Party Analytics
Introduction

This white paper will discuss portfolio-level analysis in general, and how to efficiently produce
a portfolio analysis solution that is easy to maintain, extend, and scale. Given rising in-house devel-
opment costs, many companies are faced with the challenge of how to cost-effectively perform risk
analysis and portfolio-level analysis. This white paper will outline methods for achieving significant
efficiency gains in performing these calculations.

Portfolio-Level Analysis
Portfolio-level analysis is defined, for the purposes of this paper, Both total value and aggregated cash flow are relatively simple
as any calculation that needs to be done at the portfolio level, as analyses and do not require sophisticated calculations at the
opposed to, for example, calculations done at the trade level. This portfolio level, as long as they are carried out at the trade level.
distinction is important when the aggregation is not trivial. But in other types of analysis, the aggregations are not so simple,
and most require information about the risk factors that are
shared by multiple positions within the portfolio.
Types of Portfolio Analysis

Risk Sensitivity
Total Value
This analysis is the rate of change of the value of the portfolio
Total Value is one of the simplest types of portfolio aggregation.
with respect to the market data. Being able to calculate this first
Using this analysis, calculating the value of the portfolio simply
order sensitivity is a vital component of any risk system. For
requires iterating through the positions within the portfolio,
example it can be used to calculate DV01 which is the sensitivity
valuing each one, and
of the portfolio value to a small shift in interest rates a parallel
summing them. In other Portfolio P shift of the whole yield curve. It can also be used to calculate
words the value of the
hedge factors which are the positions in liquid vanilla instruments
portfolio, f(P), equals
that would be required to add to the portfolio to provide an
the sum of the individual Trade T1 Trade T2
instantaneous hedge of all market and credit risk.
trades, f(Ti).
Trade T3 Trade T4 Trade T5

However, if f(Ti) rep- Trade T6


Stress Testing
resents the result of a This analysis involves revaluing the portfolio under a number of
particular analysis applied different scenarios for the market data to quantify the exposure
to an individual trade, then to more extreme market movements than are captured using
it is not usually true that the result for the portfolio, f(P), equals first order risk sensitivity.
the sum of f(Ti).
Credit Value Adjustment
and Potential Future Exposure
Aggregated Cash Flows These are calculations performed on a set of trades with the
The future cash flows for a portfolio sorted in chronological order
same counterparty CVA is an adjustment to the value of a
are sometimes required. This aggregation is not difficult if the
portfolio to reflect the credit risk of the counterparty; PFE is a
cash flows for each trade are available; it involves marshaling the
calculation of the maximum possible loss that would be realized
data and applying a sort algorithm.
if the counterparty were to default, for a given confidence level,
such as 95%.

fincad.com ACCELERATING PORTFOLIO-LEVEL ANALYSIS WITH THIRD PARTY ANALYTICS Page 2


Value at Risk (VaR) Enable Better Business Decisions
This is the maximum possible loss over a given time horizon that
is suffered as a result of market fluctuations, again for a given Better business decisions are only possible when up to date and
confidence level. It is a measure of market risk whereas PFE is a timely information is available. The challenge is to gather the
measure of counterparty credit risk. knowledge and insight needed to make more informed decisions.

There are many other examples of portfolio-level analysis. The The ability to evaluate all possible outcomes, even if the user only
next issue to solve is: how to choose the right portfolio needs or wishes to analyze a small subset, is a key capability for
analysis solution. any portfolio analysis solution. Decision makers need complete
visibility into the inputs and analysis performed. All projected
impacts on a portfolio should be fully quantified and available in
Choosing the Right Solution multiple timeframes and, where appropriate, currencies.

There are a wide number of factors when choosing the right solu- Any aggregated report should allow the user to drill down into
tion, but at a high level a portfolio analysis system should meet the details of how the rolled-up values were derived. Transparen-
the following criteria: cy in how numbers are generated is critical in building confidence
in the accuracy and reliability of those outputs.
Support better business decisions based on timely and accu- Finally, in order to make better business decisions, users need
rate forecasts, even when historical data is scarce timely reporting, often in real or near real time.
Provide the flexibility to rapidly assess the impact on the
users business resulting from changing market conditions
Be cost effective for both the developer and the users or Adaptability to Changes
customers in Market Conditions
Provide risk analysis risk parameters are often required,
even for solutions whose primary purpose is not
With modern portfolio solutions, it is now possible to create new
risk management
derivative products that can be integrated into existing trade
processing and portfolio analysis systems as soon as they have
These four aspects are discussed further below.
been created, without a single line of software being written. This
level of flexibility is unprecedented and was, only a few years ago,
thought unattainable.
Improve
Business The advent of Service Oriented Architectures promised to make
Decisions business services broadly available and re-usable. The aim was to
eliminate boundaries between business functions by making inte-
gration of the disparate and often legacy systems found in many
customer sites possible. But a real-world SOA , not to forget the
latest SOA-in-the-Cloud, can be challenging. Customers systems
Adapt often span departments and geographies.
Analyze & Portfolio Analysis
to Market
Manage Risk Solutions
Conditions
An effective portfolio analysis solution must integrate data sourc-
es and applications, and provide visibility and business insight
across many different platforms, both internally and externally.
The system should not only be fast, secure and always available
but also extensible and capable of coping easily with the inevita-
Control bility of a continually changing landscape.
Solution Costs

Criteria for an effective


portfolio-level analysis solution

fincad.com ACCELERATING PORTFOLIO-LEVEL ANALYSIS WITH THIRD PARTY ANALYTICS Page 3


Cost Effectiveness
The ability for users to cache, reuse and transform models of
their financial world provides them with the means to perform
It is often tempting to build analytics solutions from scratch,
rapid simulations of large portfolios without the need to recon-
thinking that if a solution can be tailored to meet specific needs, it
struct the model for each trade. An effective solution should be
must be quicker that way. This thinking is usually proven
able to meet specific performance needs, whether for a small
to be wrong.
portfolio of simple vanilla products or a large mixed portfolio
containing complex derivatives in multiple asset classes. Finally,
Industry experience shows that building a best of breed solution
the solution should be configurable, grid enabled and computa-
is more likely to be achieved by using best of breed components
tionally efficient.
and concentrating development effort on areas of unique
expertise and value.

In todays more cost sensitive market, as little time as possible Components of a Good Solution
should be spent in development. Therefore, the careful choice of
off-the-shelf components is critical to ensure that any solution An effective portfolio analysis solution requires the use of
built is delivered in a timely and cost effective manner. technology including data repositories, architectural components,
The cost of a solution over its lifetime is often overlooked. It is UI components, and analytics. Data repositories are needed for
almost always less expensive to buy components, particularly datasets such as the Trades in the various portfolios, and for
specialist components, instead of building them. This is especially Market Data. A solution may have live market data as well as
true as companies move further and further away from their historical data, depending on the purpose of the solution.
core competencies.
Architectural components required include computational serv-
Risk Analysis ers, grids, web services, and data services.

Even for solutions whose primary purpose is not risk manage- UI components are needed to organize the workspace. For
ment, it is often the case that risk sensitivities are required, or example, a Portfolio Manager can be used to organize the trade
that the portfolio must be revalued under various scenarios. In and portfolio data, while Reporting Tools can be used to present
order to make what-if and scenario analysis straightforward, and organize the results of the calculations.
the ability to re-value a portfolio by simply selecting a different
model is a critical capability. Here, the term model refers to Finally, an Analytics piece is necessary. The Analytics is a small
an object that represents the set of market data, curves, and component from an architectural perspective, but a very high
mathematical model parameters, that describe a given scenario. cost component when building a one-off analytics library
It is important to be able to re-value the portfolio without having from scratch.
to change any other inputs other than the model object.

Portfolio Manager Reporting Tools

Analytics Library

Market Data Feeds

Market Data Trade/Portfolio Data

fincad.com ACCELERATING PORTFOLIO-LEVEL ANALYSIS WITH THIRD PARTY ANALYTICS Page 4


Cost of In-House Characteristics of
Analytics Development an Effective Analytics Library
A recent reporti from the analyst firm Celent states: As previously discussed, the analytics library is one of the critical
components of an effective portfolio solution. What
Our findings suggest that firms pursuing in-house efforts for characteristics would enable a portfolio valuation or risk solution
derivatives analytics require an upfront investment of at least $9 to be built efficiently and extended easily?
million. Moreover, depending on the aggressiveness of an
institution, recurring annual costs can range between 25% and
50% of initial investment to keep pricing and risk analytics rele- 1. An analytics library for portfolio solutions must:
vant. This translates to $11 million to $22 million over a five-year
production lifecycle to enhance and keep libraries current with 2. Offer powerful portfolio specification
ongoing market requirements across multiple asset classes. .
Aggregated over the total software lifecycle, firms adopting in- 3. Provide portfolio-level analysis
house strategies for OTC pricing will require investments between
$25 million and $36 million alone to build, maintain, and enhance 4. Decouple trades and models
a complete derivatives library.
5. Re-use objects
As evidenced in the report above, it makes more economic sense
for this type of development effort to benefit many systems 6. Meet IT criteria
developers, not just one.
These five characteristics are discussed below in turn.

USD$, TOTAL COST OF DERIVATIVES ANALYTICS


million

50
Unknown costs
represents
45 the cost of
complexity and
Ranges between a drag on the
40 $11 - $22m
Integration costs
35 with internal and
Enhancements Inefficiencies
30 Upgrades & drag due
to fragmented 22.5
25 data, analytics,
platforms Known costs
over 5 years
20 If there are silos,
data disparities & 11.3
15 $4.5m quality issues will Between ~$25
continue to exist to $36 million
Bug Fixes
4.5
10 $9m Support In an ideal world,
this would be
Initial zero.
5 specification, 9.0
build and test
0
Evolve/ 3rd party Total
Development Support/ Indirect
Enhance apps/

One-off,
Recurring costs totaled over a 5-year production lifecycle
initial costs

fincad.com ACCELERATING PORTFOLIO-LEVEL ANALYSIS WITH THIRD PARTY ANALYTICS Page 5


Powerful Portfolio Specification Portfolio-Level Analysis Provision
There are several issues in the area of portfolio specification, Once the portfolio has been represented, we need to perform
which is all about defining the portfolio in terms of its some analyses on it, such as calculate its value, its cash flows, its
constituent trades and positions. There are no modeling issues Potential Future Exposure, its Value at Risk (VaR), etc.
here, or any analytics; its just a question of how to represent
the trade information on the deal sheets in a form that can be It is important that these portfolio-level analyses are available
processed by the analytics engine. within the analytics library, and that the library is not constrained
to perform trade-level calculations only. If the analytics library
Firstly, its important to define a portfolio as a single coherent were only capable of performing trade-level calculations, then
object; this makes it easy to implement calculations at the port- these aggregations need to be implemented externally to the
folio level, and operate on the portfolio as easily as if it were a library. This is inefficient, and sometimes impossible if informa-
single trade. Every calculation that can be performed at the trade tion about the relationships between the trades is not available.
level should be available at the portfolio level. It should be easy to
form a portfolio from a collection of trades, or as a union of other The following example will examine the sensitivity of
portfolios, to an arbitrary level of recursion. a swap portfolio.

It should be easy to support new types of trade within a portfolio In the following diagram, the swap value is almost entirely
system, and the analytics library should enable that. If the dependent on the 5-year swap rate, which is quoted in the
analytics library has a trade class, then by creating an instance of swap markets.
a trade with the new type, all analyses should automatically be
available both at the trade level and also at the portfolio level.
The ability to leverage portfolio-level calculations that are avail- USD 3m LIBOR, quarterly
able within the library, for portfolios that contain the new trade
type, is a very powerful feature. As new trade types are added to
Swap 1
USD 8mio notional Party A Party B
a portfolio system, the cost of adding support for an extra trade
Matures to +5y
type should decrease for each new trade type that is added.
2.4% semi-annually
These efficiency gains are possible by taking advantage of the
trade infrastructure that would be initially built to support the
Swap 1, dependent on 5-year swap rate
generic trade class.

Thirdly, it should be possible to represent any trade. For vanilla


trades, there should be templates available to conveniently create
them. On the other hand, there should be no restriction on the
complexity of trades that can be supported this can be achieved
by using a Domain Specific language for trade representation for
exotic trades.

Finally, it must be possible for a portfolio to contain trades from


more than one asset class. For example, it must be possible for a
portfolio to contain any mixture of FX trades, equity trades, fixed
income, credit derivatives, commodity derivatives, and so on.
Such mixed portfolios are common and require analysis.

fincad.com ACCELERATING PORTFOLIO-LEVEL ANALYSIS WITH THIRD PARTY ANALYTICS Page 6


The value of that swap as a function of the 5-year swap rate is The brown line in the graph below is the sensitivity of the value
shown in the following graph. The blue dot is the current value of of Swap 2 to the 5-year swap rate. Both swaps would also have
the swap, for the current value of the 5-year swap rate. The slope some sensitivity to other parts of the yield curve, although to
of that line at the blue dot is the sensitivity of the value to the a much lesser extent. The value of the portfolio is given by the
5-year swap rate. green dashed line. It is not quite horizontal because there is still
some dependency on the 5-year swap rate the swaps are not
exactly offsetting.

1,500,000 1,500,000

1,000,000 1,000,000 1
ap
1 ap
Sw Sw
500,000 500,000
Value

0 0
0% 1% 2% 3% 4% 5% 0% 1% 2% 3% 4% 5%
-500,000 -500,000

-1,000,000 -1,000,000

-1,500,000
Swap 1 as a function of 5-year swap Swap 1, Swap 2, and Portfolio value as a function of 5-year

Now consider adding a second swap to the portfolio, Swap 2. This But how is that sensitivity to be calculated? The value of each
swap is almost a perfect offset to Swap 1. However, the notional, swap, and the value of the portfolio, can be calculated for the
maturity date, and fixed rate do not quite match. In Swap 2, Party current value of the market quote for the 5-year swap rate. That
A is paying fixed, in contrast to Swap 1 in which Party A is receiv- gives the position of the 3 dots. The question is how to calculate
ing fixed. So Swap 2 is almost offsetting Swap 1 but not quite. the slope of the green dashed line, or total portfolio value, as it
This portfolio is idealized to get the point across. passes through the green dot, or current market value.

A common way of calculating the slope is by so-called bump-


USD 3m LIBOR, quarterly
ing. This is done by bumping the 5-year swap rate up by a small
Swap 1 amount, and recalculating the portfolio value, giving the green
USD 8mio notional
Matures to +5y Party A Party B dot to the right. Then, the 5-year swap rate is bumped down by
a small amount, and the portfolio value recalculated giving the
2.4% semi-annually green dot to the left. The slope is then obtained by taking the
difference in portfolio value and dividing by the small change in
USD 3m LIBOR, quarterly swap rate. Using the right-hand and left hand dots is more accu-
rate than using a one-sided difference.
Swap 2
USD 9mio notional
Matures to +5y1m Party A Party B

2.6% semi-annually

Swap 2 is added to offset swap 1

fincad.com ACCELERATING PORTFOLIO-LEVEL ANALYSIS WITH THIRD PARTY ANALYTICS Page 7


1,500,000 1,500,000

1,000,000 1,000,000
ap
Sw ap
Sw
500,000

Portfolio
Value

0 0
1% 2% 3% 5% 1% 2% 3% 5%
Sw 5 Year Swap
-500,000 ap

-1,000,000 -1,000,000

-1,500,000 -1,500,000
Using bumping method to determine
portfolio sensitivity to 5-year swap rate Calculating sensitivity using calculus

This is a common approach, but not very efficient. It requires the


value of the portfolio to be calculated three times altogether: The previous first order risk example was used to illustrate the
once in the middle to get the current value, and twice more to get availability of portfolio-level analyses within the analytics library.
the slope. This can be very computationally expensive, especially
if model recalibration had been involved. These ideas are particularly powerful when applied to more
complex portfolios, whose pricing requires model calibration.
A more efficient analytics library would operate at the The ability to propagate first order risk sensitivity to market data
portfolio-level it would know the relationships between the through the model calibration process generates huge efficiency
trades within that portfolio and automatically be aware of which savings, compared with bumping the market data and recalibrat-
market data they were exposed to. For example, Swap 1 and ing the model.
Swap 2 both depend on the 5-year swap rate, and not much on
anything else. Also, if the portfolio-level analysis is performed within the
analytics library, then the need for additional external
The library would be able to calculate that sensitivity analytically infrastructure to build a portfolio system is reduced; infrastruc-
since it would know all the mathematical operations involved in ture such as external mapping of sensitivities to pre-defined risk
pricing. In other words, it would be able to calculate the slopes on factors or curve points would potentially no longer be necessary.
this chart using calculus, rather than by finite difference or bump-
ing, and without the user having to specify which risk factors to These ideas are also very important when applied to other types
bump, as shown below. of analysis, such as CVA or VaR. These analyses are highly non-lin-
ear, and it is just not possible to derive their values for a portfolio
simply by knowing their values for each constituent trade. For
example, its not possible to calculate the CVA for a portfolio just
by knowing the CVA for each individual trade in the portfolio
since there may be offsetting positions.

fincad.com ACCELERATING PORTFOLIO-LEVEL ANALYSIS WITH THIRD PARTY ANALYTICS Page 8


Decoupled Trades and Models For example, for a vanilla European option in the Black-Scholes
model, many solution methods would be available, such as closed
form, Monte Carlo, or backward evolution on a finite difference
The third characteristic of an effective analytics library is the
grid or on a tree. However, for exotic trades, or for complex mod-
idea of decoupled trades and models. Separating the concepts
els, only Monte Carlo would be available.
of trade, model and solution method provides many advantages.
For example, it is possible to control model risk by substituting
Once these three objects have been defined, and an output
different models and solution methodologies. And if the analytics
request has been formed, it is then just a question of making a
library has a trade or product class, then consistent calcula-
call to a single universal function that can process those objects
tion types will be available on all instruments, leading to more
with a single interface.
efficient solution implementations. Here is a high-level view of a
natural approach to separating the concepts.

Model
+ Product / Trade
+ Method Model
+ Product / Trade
+ Method

Market Models Options Default Closed Form Market Models Options Default Closed Form
Black Scholes Swaps Monte Carlo Black Scholes Swaps Monte Carlo
Libor Market Model Portfolios Backward Evolution Libor Market Model Portfolios Backward Evolution

Achieve more efficient solutions


by decoupling trade and model calculations
Value
Output Risk
Cash Flows
Starting with the Product (or Trade) class in the middle, this object
represents the financial contract, or a set of trades in the case Value a portfolio or trade with a single function call
of a portfolio. This is information that is found on a term sheet,
or collection of term sheets. This object knows nothing about
quantitative finance.
This approach has many advantages. For example, it is possible to
assess model risk by substituting different models and solution
The Model on the left is a set of mathematical equations describ-
methodologies. For the same product/trade object, or portfolio
ing the dynamics of certain random variables. It also includes
object in the middle, different instances of model and solution
market data and the values of the parameters embedded in the
method could easily be substituted, and the effect on the value
definition of the mathematical models. The Model includes the
could be quantified.
market data as well as calibrated or bootstrapped parameter
values, such as discount factor curves, implied rate curves, CDS
spreads, default probability curves, and the covariance matrix for
forward rates in the Libor Market Model.

The solution Method on the right represents the numerical


technique used to value the trade with the given model. Examples
of this are closed form expressions (such as the Black-Scholes for-
mula), Monte Carlo simulation, or backwards evolution (trees or
finite difference methods). This object would also contain the data
needed to run these algorithms, such as step-size for a grid, or
the number of Monte Carlo paths to be simulated. Not all solution
methods would necessarily be applicable for all combinations of
trade and model.

fincad.com ACCELERATING PORTFOLIO-LEVEL ANALYSIS WITH THIRD PARTY ANALYTICS Page 9


Object Re-Use

Continuing with the analytics library characteristics, an area related to the concept of decoupled trades and models is that of object re-use. For
example, since the Model object contains the calibrated parameters and market data, then it can be cached and reused for various different
portfolios or types of output request. This can represent a huge savings in computational speed, since model calibration is a very computational-
ly intensive step. If new market data became available, it would only be necessary to recalibrate the affected components of the model.

Model Product / Trade Method Output

Market Models Value


Options Default Closed Form
Black Scholes Risk
Swaps Monte Carlo
Libor Market Model Cash Flows
Portfolios Backward Evolution
Heston Par Rate / Greeks

Improve computational speed by re-using objects

Other classes that would benefit greatly from caching and re-use Conclusions
would be curves, market conventions, and portfolio definitions.
For example, stress testing a portfolio would be accomplished
This white paper has examined portfolio-level analysis in general
by setting up the portfolio object once, and revaluing it under
and the criteria for an effective portfolio analysis solution. In-
several different scenarios, each of which would be represented
creasing costs of developing in-house solutions, which can reach
by a different instance of a model.
up to $36 million over the production lifecycle, continue to drive
the need for industry-standard third-party analytics solutions.
IT Considerations An effective solution should enable better business decisions,
have the ability to adapt to changing market conditions, be cost
The final set of considerations for the analytics library is in the effective, and be able to efficiently perform risk and sensitivity
area of IT. No matter how powerful the analytics platform is, in analysis.
order to be useful, it must be capable of being deployed into a
production environment. Supporting multiple operating systems Decoupling of trades and pricing models, treating portfolios as co-
and deployment options is critical, allowing flexible applications herent objects, and performing first order risk sensitivity instead
to be built that can adapt to different users IT infrastructure, of finite difference all provide significant savings in computational
regardless of what that might be. speed and accuracy. The use of object-oriented architecture also
greatly simplifies the valuation process. Using these methods, it is
Solutions should have the ability to be integrated quickly and possible to perform complex, portfolio-level analysis in a fraction
easily, while providing a long term and supportable solution that of the time required for traditional methods.
minimizes the effort and cost required to add new trade and
model constructs to the software.

fincad.com ACCELERATING PORTFOLIO-LEVEL ANALYSIS WITH THIRD PARTY ANALYTICS Page 10


Disclaimer FINCAD makes no warranty either express or implied, including, but not limited to, anyimplied warranties
of merchantability or fitness for a particular purpose regarding thesematerials, and makes such materials
available solely on an as-is basis. In no event shallFINCAD be liable to anyone for special, collateral,
incidental, or consequential damages inconnection with or arising out of purchase or use of these materials.
This information issubject to change without notice. FINCAD assumes no responsibility for any errors in
thisdocument or their consequences, and reserves the right to make improvements andchanges to this
document without notice.

Copyright FinancialCAD Corporation. All rights reserved

F3, FinancialCAD and FINCAD are registeredtrademarks of FinancialCAD Corporation. Other trademarks
Trademarks are the property of theirrespective holders.

Revisions
Every effort has been made to ensure the accuracy of this document. FINCAD regrets anyerrors and
omissions that may occur and would appreciate being informed of any errorsfound. FINCAD will correct
any such errors and omissions in a subsequent version, asfeasible. Please contact us at:

Vancouver USA/Canada
Vancouver
1 800 304 0702
1 604 957 1200

New York New York


Europe
1 646 435 5920
00 800 304 07020
London 44 207 464 4190
London Dublin 3 531 400 3100
fincad.com
Dublin Fax 1 604 957 1201
Email info@fincad.com

Copyright FinancialCAD Corporation. All rights reserved. FinancialCAD, UAD and FINCAD are trademarks of FinancialCAD Corporation. Other trademarks are the
fincad.com ACCELERATING PORTFOLIO-LEVEL ANALYSIS WITH THIRD PARTY ANALYTICS Page 11
property of their respective holders. This is for information purposes only. FINCAD MAKES NO WARRANTIES, EXPRESSED OR IMPLIED, IN THIS SUMMARY.

Vous aimerez peut-être aussi