Professional Documents
Culture Documents
October 2020
When the COVID-19 outbreak became a global 2020, banks should aim to optimize their trading
pandemic, financial-markets volatility hit its books and risk positions. This ambition requires
highest level in more than a decade, amid pervasive more accurate and timely valuations. With those
uncertainty over the long-term economic impact. priorities in mind, more advanced models and
Calm has returned to markets in recent months, but sufficient computational power are imperatives.
volatility continues to trend above its long-term Indeed, “speed” is of the essence.
average. Amid persistent uncertainty, financial
institutions are seeking to develop more advanced In response, some leading institutions have started
quantitative capabilities to support faster and more to incorporate advanced techniques into their
accurate decision making. quantitative armories. In pricing, an area that has
experienced a spike in recent activity, several banks
As financial markets gyrated in recent months, are applying machine learning (ML) to enhance
banks faced particular problems calculating traditional models—for example, by calibrating
value at risk (VAR) across asset classes. Many parameters more efficiently. In particular, banks
institutions experienced elevated levels of VAR have used neural networks, a type of ML focused
back-testing exceptions, leading to higher on nonlinear and complex data relationships.
regulatory-capital multipliers. Increases of as much Advanced machine-learning techniques can do
as 30 percent were reported, prompting regulators the following:
to apply exemptions in some cases. There were
also challenges with valuation adjustments, as — speed up calculations, reducing operational
derivatives faced snowballing collateral calls costs and allowing real-time risk management of
and increasing funding costs. Where credit- complex products
value-adjustment (CVA) risks were excluded from
market risk models, CVA hedges sat “naked” on — animate more complex models that may
the balance sheet, leading to significant uplifts in currently be unusable in practice, and unlock
exposures, and therefore in risk-weighted assets more accurate valuations
(RWAs). One large US dealer was hit with a loss of
$950 million stemming from a valuation adjustment — generate high volumes of synthetic but market-
(XVA) in the first quarter of 2020. Elsewhere, rising consistent data, helping, for example, to offset
gap risk in illiquid securities catalyzed painful fair- the disruptive impact of COVID-19-related
value losses—as high as $200 million in the case of market moves
a major Europe-based bank.
One way to implement neural networks is to
In an unpredictable environment, financial modelers apply them to pricing, where they can “learn” how
were required to come up with solutions, but were to price vanilla calibration instruments under a
often stymied by inadequate models or the need given (possibly complex) model, and then act as
for huge computational power that was not always pricing engines for new model calibration. The
available. Given the speed of response required, approach obviates one of the most significant
models in some cases were rendered unusable. The challenges associated with ML, which is
inevitable result was an increase in risk exposures parameter interpretability. In this case, there is no
and opacity from valuations, sometimes in absolute interpretability issue because the network uses
value and other times relating to the reason for the original model’s parameters. This means that
specific model outputs. there is no ML “black box,” and the key calibrated
parameters can be interpreted in the original
model’s context.
An imperative to act
Since forecasting institutions expect the global
economy to have contracted by about 5 percent in
2 Applying machine learning in capital markets: Pricing, valuation adjustments, and market risk
Neural networks can also support future exposure processing (Exhibit 2). That saves banks from using
modeling for valuation adjustments (Exhibit 1). time-consuming nested Monte Carlo approaches
and less accurate analytical approximations or “least
The network can be trained on established samples, squares”—style regressions.
such as those relating to the evolution of risk factors
and corresponding cash flows for the products being There are equally promising applications in real-time
modeled. The additional efficiency provided by the portfolio valuation, risk assessment, and margining.
network makes for improved accuracy and faster
Web <2020>
<Machine learning capital markets>
Exhibit
Exhibit <1>1of <2>
Neural
Neural networks
networks can
can support
support future-exposure modeling.
Web <2020>
<Machine learning capital markets>
Exhibit <2> of <2>
Exhibit 2
Neural networks
Neural networks can
can enable fast and
enable fast and accurate
accurateexposure
exposurecalculations.
calculations.
Problem: Future-exposure modeling is main bottleneck in Solution: Neural-network approach proposed to achieve
current valuation-adjustment models; portfolio-valuation all 3 goals at same time:
and risk calculations must be fast, accurate, and consistent 1. Perform portfolio-valuation and risk calculations via
with FO1-pricing models, but typical approaches fail to neural networks
meet all 3 goals at same time 2. Train neural networks using simulated risk-factor paths
and pathwise-evaluated cash flows—no extra cost to
Nested Monte Analytical approximations/ generate training sample
Carlo method LSM2-style regressions 3. Use differential regularization to optimize accuracy for
Fast both pricing and risk while achieving fast training
Accurate
Consistent with
FO-pricing models
Front office.
1
Least-squares method.
2
Applying machine learning in capital markets: Pricing, valuation adjustments, and market risk 3
Financial institutions cannot maximize
the machine-learning opportunity
without acquiring the capabilities to build,
maintain, and apply ML-enabled models.
4 Applying machine learning in capital markets: Pricing, valuation adjustments, and market risk
• build capabilities via learning on the job • updating risk-management practices, such
as model governance and risk assessment,
• understand typical challenges and pitfalls and to monitor and control new risks introduced
how to solve them by ML
Juan Aristi Baquero is a partner in McKinsey’s New York office, Akos Gyarmati and Pedro J Silva are consultants in the
London office, Marie-Paule Laurent is a partner in the Brussels office, and Torsten Wegner is an associate partner in the
Berlin office.
Applying machine learning in capital markets: Pricing, valuation adjustments, and market risk 5