Authors

  • Zharmagambetov Yernar
    Investment portfolio manager, MBA, Almaty, Kazakhstan

DOI:

https://doi.org/10.37547/tajmei/Volume06Issue12-06

Keywords:

Synthetic data generative models volatility

Abstract

The article considers the possibility of using synthetic data to model a portfolio in conditions of instability in financial markets. Methods based on the analysis of historical data need to be revised to account for rare events that are missing from historical records. The use of generative models makes it possible to generate synthetic data, creating conditions for modeling situations that go beyond the known scenarios. The purpose of the article is to consider the impact of synthetic data on asset management methods in conditions of high market volatility. The methods of creating synthetic data using generative adversarial networks and their role in modeling situations with limited access to necessary information are described. The use of synthetic data in scientific papers confirms their effectiveness in adapting asset management strategies, which contributes to improving results. This underscores the need for their application and guarantees the stability of financial systems in the external conditions that determine the present. In conclusion, it is noted that generative models that create synthetic data increase the accuracy and flexibility of financial portfolio management strategies. The considered approaches open up opportunities for forecasting and decision-making. Due to this, the information contained in the work will be useful to investors and bank employees.


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF MANAGEMENT AND ECONOMICS INNOVATIONS (ISSN- 2693-0811)

VOLUME 06 ISSUE12

55

https://www.theamericanjournals.com/index.php/tajmei

PUBLISHED DATE: - 18-12-2024
DOI: -

https://doi.org/10.37547/tajmei/Volume06Issue12-06

PAGE NO.: - 55-62

USING SYNTHETIC DATA TO MODEL A
PORTFOLIO IN CONDITIONS OF HIGH
VOLATILITY. HOW SYNTHETIC DATA
ALLOWS YOU TO TEST STRATEGIES FOR
RARE MARKET EVENTS. EXAMPLES OF
GENERATIVE MODELS APPLICATION


Zharmagambetov Yernar

Investment portfolio manager, MBA, Almaty, Kazakhstan

INTRODUCTION

Modern financial markets are characterized by
high volatility, limiting the applicability of
traditional methods for forecasting and testing
investment strategies under extreme fluctuations.
In such circumstances, it is essential to develop
methods that model asset behavior in rare and
unconventional market situations. These methods
facilitate accurate assessments of portfolio
stability and profitability. One such approach

involves the use of synthetic data generated to
analyze various market scenarios, including
financial crises, sharp asset price fluctuations, and
changes in exchange rates.

According to estimates by the McKinsey Global
Institute, generative AI could deliver between $2.6
trillion and $4.4 trillion annually across various
industries worldwide in the 63 scenarios analyzed.
Among industries, the banking sector is expected

RESEARCH ARTICLE

Open Access

Abstract


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF MANAGEMENT AND ECONOMICS INNOVATIONS (ISSN- 2693-0811)

VOLUME 06 ISSUE12

56

https://www.theamericanjournals.com/index.php/tajmei

to realize some of the greatest benefits, with an
annual potential ranging from $200 billion to $340
billion (equivalent to 9

15 percent of operating

profit), primarily driven by productivity gains. The
economic impact is likely to be positive across all
banking segments and functions, with the highest
absolute benefits projected in the corporate and
retail sectors ($56 billion and $54 billion,
respectively) [11].

These

models

replicate

dependencies

characteristic of real markets, including
extraordinary situations. The application of such
methods enables the testing of strategies under
conditions that are rarely represented in historical
data, improving risk forecasting and enhancing
portfolio resilience.

The use of synthetic data requires the
development of new methods for risk assessment
and portfolio optimization under conditions of
economic instability. Amid environmental changes
and market uncertainty, employing synthetic data
for testing and adapting strategies represents a
critical task, as it expands modeling capabilities
and improves the quality of investment decision-
making.

The purpose of this article is to explore the
potential of using synthetic data for portfolio
modeling under conditions of high volatility, as
well as to examine practical examples.

METHODS

Studies dedicated to forecasting rare events
highlight the necessity of employing generative
adversarial networks (GANs) to create synthetic
data. Articles by Labiad B., Berrado A., Benabbou L.
[1], and Dannels, S. [4] demonstrate the application
of these models for generating market data under
conditions of increased volatility. This approach
enables the modeling of phenomena that are
challenging to capture using traditional methods.
Issues related to the scarcity of data on market

surges are addressed through conditional
generative adversarial networks, as described in
the work of Chen Y. et al. [9].

Research by Tepelyan R., Gopal A. [5], Gibson L.,
Hoerger M., and Kroese D. [6] confirms the
effectiveness of these networks in forecasting
returns. These methods aid in the development of
strategies that account for changing market
conditions.

The article by Naritomi Y. and Adachi T. [2]
discusses the use of generative networks to
improve trading forecasts, while Dogariu M. et al.
[3, 10] explore the potential of generative
networks for testing strategies.

The method proposed by Li J. et al. [7] focuses on
the application of Wasserstein in generating
synthetic flows of market orders, facilitating the
modeling of market liquidity. Coletta A. et al. [8]
use conditional generative adversarial networks to
simulate market scenarios, expanding the
possibilities for modeling various strategies.

The practical examples described in this study
draw on sources [13-14], available on the websites
*masterofcode.com*

and

*www.orientsoftware.com*, which detail the
experiences of companies using generative
networks for portfolio modeling. Source [11],
located on the website *www.mckinsey.com*,
provides statistical data related to the use of
generative intelligence.

The article employs a method of analyzing existing
scientific literature and reviewing real-world
examples in the field.

RESULTS AND DISCUSSIONS

Financial markets in recent decades have been
characterized by high volatility, complicating asset
management. In such conditions, methods that
adequately reflect ongoing changes are essential.
Forecasts based on historical data analysis lose
their

effectiveness

in

predicting

future


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF MANAGEMENT AND ECONOMICS INNOVATIONS (ISSN- 2693-0811)

VOLUME 06 ISSUE12

57

https://www.theamericanjournals.com/index.php/tajmei

developments. Synthetic data generated using
mathematical models offer opportunities for
developing risk management techniques.

Synthetic data are artificially created datasets with
statistical properties similar to real indicators.
They are generated using generative adversarial
networks (GANs) and machine learning methods.
These technologies enable the modeling of rare
market situations that are infrequent in historical
data but significantly impact investment decision-
making [1, 4, 9].

The use of synthetic data facilitates the study of
complex market situations. Traditional methods
do not capture the full spectrum of risks,
necessitating alternative approaches. Artificial
data helps mitigate information deficits by
modeling events not accounted for in historical
analysis.

Various algorithmic approaches are used to create
synthetic data. Generative adversarial networks

consist of two neural models: one generates data,
while the other evaluates their conformity to
predefined parameters. GANs serve as a tool for
generating data, including time series of market
activity. The model architecture includes a
generator that forms information based on
random input signals and a discriminator that
assesses the degree of similarity [6]. The objective
is to achieve a state where the differences between
artificially generated and real data become
imperceptible.

Below is the formula used for modeling data under
conditions of high volatility in financial markets. In
the context of portfolio modeling during high
volatility, GANs are employed to generate realistic
time series that simulate market data, such as
stock prices or indices subject to significant
fluctuations. The generator attempts to create time
series resembling real data, while the
discriminator evaluates their similarity.

𝑚𝑖𝑛

𝐺

𝑚𝑎𝑥

𝐷

𝑉(𝐷, 𝐺) = 𝐸

𝑥~𝑃𝑑𝑎𝑡𝑎(𝑥)

[𝑙𝑜𝑔 𝑙𝑜𝑔

𝐷(𝑥)

] + 𝐸

𝑧~𝑃𝑧 (𝑧)

[𝑙𝑜𝑔 𝑙𝑜𝑔

(1 −

𝐷(𝐺(𝑧))

]

(1)

Where:

-

G

is the generator, which transforms a random

vector

z

from the distribution

P

z

(z)

into data (e.g.,

an image).

-

D

is the discriminator, which evaluates whether

an input example is real (from the data distribution

P

data

(x)

) or generated (from the generator

G(z)

).

-

x

represents data drawn from the distribution

P

data

(x)

.

-

z

represents random data (typically latent

variables or noise) fed into the generator.

-

G(z)

represents the generated data output by the

generator when

z

is provided as input.

The ideal case (equilibrium) is achieved when the
generator produces data that the discriminator
cannot distinguish from real data, meaning D(x) =
0.5 for all x, and the discriminator cannot
differentiate between real and generated
examples. This process models a zero-sum game
between two agents (the generator and the
discriminator), forming the basis for training.

These models are used to create surfaces that
reproduce historical data while incorporating
elements of market behavior. This approach
facilitates stress testing by simulating various


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF MANAGEMENT AND ECONOMICS INNOVATIONS (ISSN- 2693-0811)

VOLUME 06 ISSUE12

58

https://www.theamericanjournals.com/index.php/tajmei

crisis scenarios [3,10]. Table 1 below outlines the
advantages and disadvantages of using generative

networks.

Table 1. Advantages and Disadvantages of Using Generative Networks

(compiled by the author)

Advantage

Disadvantage

Generative networks effectively model complex
nonlinear relationships between assets, useful for
identifying hidden risk factors and market
opportunities.

Forecasts can complicate integration and
justification, posing challenges for asset
managers and investors.

Generative models create synthetic data,
including rare or extreme events (e.g., crises),
critical for risk assessment in highly volatile
conditions.

Reliable training of generative models requires
large amounts of high-quality data, which may
not be available in volatile conditions.

Generative networks can help simulate portfolios
resilient to unexpected market fluctuations by
generating numerous alternative scenarios.

Generative models demand significant
computational power, increasing the cost of their
use.

Generative networks enable the development of
new asset allocation strategies that adapt to
changing market conditions and high volatility.

Improper tuning or insufficient data can lead to
overfitting, resulting in poor generalization and
unstable outcomes.

Generative networks are used to predict future
market states based on trends and historical data,
crucial during periods of instability.

Results are highly dependent on model
hyperparameter selection (e.g., network
architecture, learning rate), requiring careful
tuning and optimization.

Generative models reduce the risk of asset
overvaluation or forecasting errors common in
traditional methods through more complex and
flexible data structures.

Generative models can be sensitive to noise and
unreliable data, potentially leading to incorrect
conclusions and decisions.

Synthetic data expands the range of analyzed
scenarios, including rare market phenomena. This
enhances risk assessment accuracy and identifies
vulnerabilities in management practices.

Generative models create scenarios that allow
testing the resilience of investment strategies in
unstable market environments.

Mastercard actively employs algorithms to combat
fraud and ensure the protection of customer data.
As cyber threats continue to grow, detecting
fraudulent activities becomes increasingly
challenging. Criminals use methods such as
spyware, malware, and skimming to gain access to
card data. Many limit their activities to selling


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF MANAGEMENT AND ECONOMICS INNOVATIONS (ISSN- 2693-0811)

VOLUME 06 ISSUE12

59

https://www.theamericanjournals.com/index.php/tajmei

fragments of information on dark markets, making
detection more difficult.

Algorithms enable financial organizations to
efficiently process real-time data, identifying
patterns indicative of fraud. Predictive models
forecast potential criminal schemes, allowing
timely responses to threats. Mastercard has
reported [12] that these technologies have
doubled the detection rate of compromised cards
while reducing false positives in the analysis of
suspicious

transactions.

Additionally,

the

verification of high-risk merchants has been
significantly accelerated.

JPMorgan Chase developed the LLM Suite AI
assistant system to support its 60,000 employees
[12]. This tool automates routine tasks, enabling
employees to focus on more critical aspects of their

work. The system’s functionality includes

protecting data from external threats and training

staff in effective technology utilization methods.

However,

the

bank

acknowledges

that

implementing such technologies will result in
changes to workforce structures. Automating
certain processes may reduce the number of jobs
in specific areas. It is crucial to emphasize that the
use of AI must be carefully managed, particularly
when handling sensitive data, to prevent the
dissemination of incorrect conclusions or
misinformation [12].

Forrester reports [13] that nearly 70% of decision-
makers in the banking sector believe
personalization is essential for effective customer
service. These efforts are also supported by
company executives, who recognize that a
personalized approach is critical for business
success. However, only 14% of surveyed
consumers currently rate banks as providing
excellent personalized service (Figure 1).

Figure 1. Forrester Survey on Generative System Applications in Banking [13]

Generative AI systems help address issues in the
banking sector by tailoring services to individual
customer characteristics. They analyze goals,
preferences,

and

risk

levels,

providing

recommendations that align with unique

conditions. This approach enables the creation of
precise offerings, improving service quality and
customer satisfaction. The technology responds to
current market trends and considers economic
forecasts, offering critical data for decision-


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF MANAGEMENT AND ECONOMICS INNOVATIONS (ISSN- 2693-0811)

VOLUME 06 ISSUE12

60

https://www.theamericanjournals.com/index.php/tajmei

making.

In investment banking, generative models
significantly enhance the processing of large data
volumes. They uncover hidden patterns that might
otherwise go unnoticed, improving the accuracy of
risk assessments and simplifying the decision-
making process. Models based on trend

information generate forecasts that give analysts a
comprehensive understanding of the potential
outcomes of various actions. Combining advanced
analytics with powerful algorithms creates
opportunities to enhance results for all
stakeholders, from investors to financial
institutions [13].

Table 2. Advantages and Disadvantages of Using Synthetic Data for Portfolio

Modeling (compiled by the author)

Advantages

Disadvantages

Synthetic data enable the creation of rare market
events, such as crises or significant corrections,
which are difficult to observe in real data but are
essential for portfolio stress testing.

Generating synthetic data that accurately reflects
real market conditions is challenging, particularly
when attempting to replicate complex economic
and political factors.

The use of synthetic data allows testing
strategies under rare but possible extreme market
events, such as liquidity crashes or sharp
volatility shifts.

Synthetic data do not always fully model all risks
associated with real markets, potentially leading
to an underestimation of risks during instability.

Generative models facilitate varying market
conditions, creating both highly volatile and
stable market scenarios necessary for analyzing
portfolio behavior in diverse situations.

The quality of synthetic data heavily depends on
the model used; if the model is inadequate,
results may be distorted.

Synthetic data allow the generation of large
datasets in a short time, accelerating the testing
and improvement of strategies.

Synthetic data often lack the long-term historical
context needed for assessing strategy resilience
across different economic cycles.

Generative models enable control over volatility,
correlations, and other key data parameters,
creating idealized conditions for testing various
hypotheses.

If a model generates data too similar to real
conditions, it may lead to overfitting strategies to
synthetic data, reducing their adaptability in real-
world scenarios.

Based on the above, it can be concluded that
generative models enable the simulation of rare
market scenarios that are infrequent in historical
data. This capability supports making informed
investment decisions.

CONCLUSION

The use of synthetic data in forming investment
portfolios under conditions of instability
represents an approach that addresses the
challenges of financial analysis. Generative


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF MANAGEMENT AND ECONOMICS INNOVATIONS (ISSN- 2693-0811)

VOLUME 06 ISSUE12

61

https://www.theamericanjournals.com/index.php/tajmei

algorithms create models of rare market scenarios
that are inaccessible through the study of historical
data. This method incorporates numerous factors
shaping

market

dynamics,

enabling

the

development of tools that account for current
economic realities.

The application of synthetic data facilitates the
creation of accurate models for risk assessment,
allows for adjustments to asset management
strategies under changing market conditions, and
supports the development of solutions for
addressing

unconventional

circumstances.

Practical examples confirm the effectiveness of this
approach in testing investment decisions aimed at
reducing the likelihood of financial losses.

REFERENCES

1.

Labiad B., Berrado A., Benabbu L. (2023),
"Forecasting extreme events in the stock
market

using

generative

adversarial

networks", International Journal of Intellectual
Informatics Achievements, vol. 9, No. 2, pp.
218-230.

2.

Naritomi Yu., Adachi T. (2020), "Increasing the
volume of high-frequency financial data using
a Generating Network", International Joint
Conference on Web Intelligence and Intelligent
Agent Technologies (WI-IAT), pp. 641-648.

3.

D. Fernando, Stefan L., Boteanu B., Lamba S.
And Ionescu. (2021), "Towards creating
realistic financial time series using a
generative link", 29th European Signal
Processing Conference (EUSIPCO), pp. 1341-
1345.

4.

Dannels S. (2023), "Creating Catastrophes:
Predicting recession using synthetic time
series generated by GAN", arXiv, abs / 2302,
10490.

5.

Tepelian R., Gopal A. (2023), "Generative
machine learning for multidimensional equity
returns", Proceedings of the Fourth ACM

International Conference on AI in Finance, pp.
159-166.

6.

Gibson L., Herger M., Kres D. (2023),
"Generative flow-based model for modeling
rare events", arXiv preprint arXiv:2305.07863.

7.

Lee J. et al. (2020), "Generating realistic flows
of stock market applications", Proceedings of
the AAAI Conference on Artificial Intelligence,
vol. 34, No. 01, pp. 727-734.

8.

Colette. et al. (2021), "Towards realistic
market modeling: an approach to generating
interconnecting networks", Proceedings of the
Second ACM International Conference on AI in
Finance, pp.1-9.

9.

Chen Yu. et al. (2021), "Towards the creation of
synthetic multidimensional time series for
forecasting spikes", Artificial Intelligence and
Soft

Computing:

20th

International

Conference, ICAISC 2021, Virtual Event, June
21-23, 2021, Proceedings, Part I, 20.

Springer

International Publishing, pp. 296-307.

10.

D. Fernando. et al. (2022), "Generation of
realistic synthetic financial time series", ACM
Transactions on Multimedia Computing,
Communications, and Applications (TOMM),
vol. 18, No. 4, pp. 1-27.

11.

Using the full capabilities of the AI generator in
the banking sector[Electronic resource]. -
Achievement

mode:

https://www.mckinsey.com/industries/finan
cial-services/our-insights/capturing-the-full-
value-of-generative-ai-in-banking.

12.

Generative AI in banking: transformational
potential and use cases that change the rules of
the game [Electronic resource]. - Achievement
mode:
https://www.orientsoftware.com/blog/gener
ative-AI-in-banking /.

13.

Why is the generating AI in the banking sector


background image

THE USA JOURNALS

THE AMERICAN JOURNAL OF MANAGEMENT AND ECONOMICS INNOVATIONS (ISSN- 2693-0811)

VOLUME 06 ISSUE12

62

https://www.theamericanjournals.com/index.php/tajmei

a secret resource: Your implementation plan
[Electronic resource]. - Achievement mode:

https://masterofcode.com/blog/generative-
ai-in-banking.

References

Labiad B., Berrado A., Benabbu L. (2023), "Forecasting extreme events in the stock market using generative adversarial networks", International Journal of Intellectual Informatics Achievements, vol. 9, No. 2, pp. 218-230.

Naritomi Yu., Adachi T. (2020), "Increasing the volume of high-frequency financial data using a Generating Network", International Joint Conference on Web Intelligence and Intelligent Agent Technologies (WI-IAT), pp. 641-648.

D. Fernando, Stefan L., Boteanu B., Lamba S. And Ionescu. (2021), "Towards creating realistic financial time series using a generative link", 29th European Signal Processing Conference (EUSIPCO), pp. 1341-1345.

Dannels S. (2023), "Creating Catastrophes: Predicting recession using synthetic time series generated by GAN", arXiv, abs / 2302, 10490.

Tepelian R., Gopal A. (2023), "Generative machine learning for multidimensional equity returns", Proceedings of the Fourth ACM International Conference on AI in Finance, pp. 159-166.

Gibson L., Herger M., Kres D. (2023), "Generative flow-based model for modeling rare events", arXiv preprint arXiv:2305.07863.

Lee J. et al. (2020), "Generating realistic flows of stock market applications", Proceedings of the AAAI Conference on Artificial Intelligence, vol. 34, No. 01, pp. 727-734.

Colette. et al. (2021), "Towards realistic market modeling: an approach to generating interconnecting networks", Proceedings of the Second ACM International Conference on AI in Finance, pp.1-9.

Chen Yu. et al. (2021), "Towards the creation of synthetic multidimensional time series for forecasting spikes", Artificial Intelligence and Soft Computing: 20th International Conference, ICAISC 2021, Virtual Event, June 21-23, 2021, Proceedings, Part I, 20. – Springer International Publishing, pp. 296-307.

D. Fernando. et al. (2022), "Generation of realistic synthetic financial time series", ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), vol. 18, No. 4, pp. 1-27.

Using the full capabilities of the AI generator in the banking sector[Electronic resource]. - Achievement mode: https://www.mckinsey.com/industries/financial-services/our-insights/capturing-the-full-value-of-generative-ai-in-banking.

Generative AI in banking: transformational potential and use cases that change the rules of the game [Electronic resource]. - Achievement mode: https://www.orientsoftware.com/blog/generative-AI-in-banking /.

Why is the generating AI in the banking sector a secret resource: Your implementation plan [Electronic resource]. - Achievement mode: https://masterofcode.com/blog/generative-ai-in-banking.