Registration now open for OODAcon 2024.

Home > Analysis > Is Deep Learning the Future of Financial Stability (or Volatility and Crisis)?

Financial network complexity, biased algorithms, regulatory considerations, and predictive analytics (based on faulty foundational models) are only a few of the risk variables that will determine the future of AI in the finance sector.  Securities and Exchange Commission Chair Gary Gensler, in his previous life as an MIT Sloan School of Management Professor, wrote a seminal paper on deep learning and financial stability.  Find the still prescient insights from the paper here.  

Background

What is Deep Learning?

According to Deloitte Insights, “deep learning algorithms are a powerful general-purpose form of machine learning that has moved into the mainstream. It uses a variant of neural networks to perform high-level abstractions such as voice or image pattern recognition. Pioneered in 2006 by Geoffrey Hinton, a professor at the University of Toronto and a former researcher at Google, deep learning has proved its metal across a diverse spectrum of challenges—from new drug development to translating human conversations from English to Chinese.

Google is using deep learning with its Android phone to recognize voice commands and on its social network to identify and tag images. Facebook uses deep learning as a means to target ads and identify faces and objects. DeepMind [a London-based AI company which Google acquired in 2014]…[developed] a Neural Turing Machine. This computing architecture utilizes a form of recurrent neural networks that can do end-to-end processing from a sensory perception data stream to interpretation and action. This system has been used to learn games from scratch, and in some cases has demonstrated better-than-human-level performance.” 

In 2016, AlphaGo (developed by DeepMind) defeated the world champion Go player, marking a significant milestone in AI and deep learning’s application to complex games.  Deep learning is increasingly integrated into various industries, including autonomous vehicles, healthcare, finance, and more. Transfer learning, where pre-trained models are fine-tuned for specific tasks, has become popular, enabling faster development of AI applications.  Deep learning continues to evolve, with ongoing research in areas like self-supervised learning, generative adversarial networks (GANs), and reinforcement learning. Ethical and societal considerations surrounding AI and deep learning have also become a significant focus.

AI, Finance and Public Policy

SEC’s Gary Gensler on how artificial intelligence is changing finance

“There’s a risk that the crisis of 2027 or the crisis of 2034 is going to be embedded somewhere in predictive data analytics.”

AI has the ability to transform finance, but caution is needed, says the Securities and Exchange Commission chair. Here are four takeaways from Gary Gensler’s comments at the 2022 MIT AI Policy Forum:  

AI in finance is especially complex:  Having solid predictive models is crucial in AI, whether it’s in social media or in driverless cars. The difference with finance is that “the robustness of the network itself” matters just as much as the model…if predictive data analytics is a “new tool” for capital markets, the question becomes how to bring that under the realm of public policy.  

Unbiased algorithms are important:  Gensler highlighted the importance of having neutral algorithms that don’t put a platform or a business’ revenue or profit ahead of fiduciary duty, to make sure people don’t get steered toward higher margin products or trading options:  “My call to action, maybe to the academics and computer scientists, is to help people think through this — how you could have a neutral algorithm that’s not putting a platform or a business’ revenue or profits ahead of the investing public,” [Gensler said]. 

The U.S. doesn’t need AI-specific regulations:  Should there be specific rules that are tailored toward artificial intelligence in finance? Gensler said no. When new tools have come along previously, “We generally don’t write new laws or regulations,” he said. In finance, “We’ve come to some consensus through our legislative bodies, and we’ve adopted laws to protect the public” across investor protection and financial stability. These are “tried and true public policies” and less “about a new law or a new rule about artificial intelligence.”

Keep an eye on predictive analytics:  Gensler believes that predictive analytics are revolutionizing the financial industry but are an “emerging risk” and must be watched closely…He cited a possible problem with foundation models, as an example. In AI, foundational models are trained on large amounts of data that can be adapted and used for a wide range of cases; but they can easily become a “concentrated risk” if people rely on them too much.  Predictive data analytics have “remarkable abilities to predict things,” he said, but “I do think that there’s a risk that the crisis of 2027 or the crisis of 2034 is going to be embedded somewhere in predictive data analytics.”

What Next? 

This 2020 paper details SEC chair’s concerns about AI and finance

Highlights from Gary Gensler and Lily Bailey’s “Deep Learning and Financial Stability” outline five ways AI could lead to future financial crises.

From the MIT Sloan School of Management Ideas that Matter Blog:    “Artificial intelligence has the potential to transform finance, but it could spell trouble without proper oversight, according to Gary Gensler, chair of the U.S. Securities and Exchange Commission…Gensler became one of the first regulators to propose rules for AI to address conflict-of-interest concerns raised by financial firms’ use of predictive data analytics. In recent weeks, he has continued to talk about the need to minimize AI’s risk to investors.  

While a professor at MIT Sloan, Gensler explored these issues in-depth in a 2020 paper, “Deep Learning and Financial Stability,” written with then-research assistant Lily Bailey, who is now special assistant to the chief of staff at the SEC. The paper outlines five pathways whereby broad adoption of deep learning, a subset of AI, could increase fragility in the financial system.   Here are the authors’ areas of concern:  

  • Data:  Across different sectors of the economy, Gensler and Bailey noted coalescence around important datasets. For example, Google Maps, Google Earth, and affiliate Waze dominate traffic datasets and the route optimization business. Having this kind of uniformity presents a huge risk.

  • Model design:  Models can lead to systemic risks like those that caused the 2008 financial crisis…the authors caution that the unique attributes and construction of deep learning models increase sensitivity and could lead to a higher prominence of black swan events. When models coordinate and communicate with each other to optimize results, it’s possible that they will execute the same strategies, increasing volatility.

  • Regulation:  Existing financial sector regulations will probably fall short in addressing the risks posed by deep learning, the authors write.  The adoption of deep learning in finance is likely to be uneven among different parties, with less-regulated fintech startups likely to move faster than both large, regulated financial institutions and smaller community institutions that don’t have the resources to independently adopt deep learning.

  • Algorithmic coordination:  The authors raise the possibility that algorithmic coordination could lead to increased network interconnectedness if models at different financial firms communicate with each other, again increasing the likelihood of herding behavior.  The authors also write that the regulatory tools used to identify algorithmic coordination might not be able to discern deep learning coordination until after the fact.

  • User interfaces:  Deep learning is widely used in the user interface space and in customer interaction. This includes platforms that provide automated advice and recommendations for investing, lending, and insurance.  The authors caution that the standardization of virtual assistant software, such as chatbots that provide investment advice, could cause herding across client decision-making — potentially across an entire asset class or sector…while there is plenty of opportunity for AI to help companies better serve their clients, AI can easily mask who’s at fault and what exactly happened in the event of a crisis, the authors write in their conclusion.

Additional Resources

(Generating using OpenAI’s ChatGPT 3.5) 

“What is Deep Learning?”

Deep learning is a subfield of artificial intelligence (AI) and machine learning (ML) that focuses on training artificial neural networks to perform tasks that typically require human-like intelligence. It is called “deep” learning because it involves the use of deep neural networks, which are composed of multiple layers of interconnected nodes, known as neurons. These networks can automatically learn to extract relevant features and representations from raw data, making them highly effective for a wide range of complex tasks.  Key characteristics and concepts of deep learning include:

  1. Neural Networks: Deep learning models are built using artificial neural networks, which are inspired by the structure and function of the human brain. These networks consist of layers of interconnected nodes, with each connection having an associated weight.
  2. Deep Architectures: Deep learning models are characterized by their depth, meaning they have many hidden layers between the input and output layers. These hidden layers enable the network to learn hierarchical representations of data, with each layer capturing increasingly abstract features.
  3. Representation Learning: Deep learning excels at learning meaningful representations of data directly from raw input. It can automatically discover features and patterns that are useful for a given task without the need for manual feature engineering.
  4. End-to-End Learning: Deep learning models are capable of learning end-to-end solutions for complex tasks. Instead of breaking down a problem into multiple handcrafted stages, deep learning systems can directly map inputs to desired outputs.
  5. Training with Backpropagation: Deep neural networks are trained using a process called backpropagation. During training, the network adjusts its internal weights to minimize the difference between its predictions and the ground truth labels in a training dataset. This process is typically guided by an optimization algorithm, such as gradient descent.
  6. Large Datasets: Deep learning models often require large labeled datasets for effective training. Having access to extensive data allows these models to generalize well to new, unseen examples.
  7. Hardware Acceleration: The training and inference of deep learning models can be computationally intensive. Graphics Processing Units (GPUs) and specialized hardware like Tensor Processing Units (TPUs) are commonly used to accelerate deep learning computations.

Deep learning has achieved remarkable success in various domains, including image and speech recognition, natural language processing, autonomous driving, recommendation systems, healthcare, and more. Notable deep learning architectures include Convolutional Neural Networks (CNNs) for image processing, Recurrent Neural Networks (RNNs) for sequential data, and Transformers for natural language understanding and generation.  The continuous advancement of deep learning techniques has led to breakthroughs in AI applications and has played a pivotal role in the development of intelligent systems that can perceive, understand, and make decisions from complex data.

“What are use cases, case studies and deep Learning innovations in the Financial Sector?”

Deep learning has found numerous applications in the financial sector, offering innovative solutions for various tasks. Here are some use cases, case studies, and innovations in the financial industry involving deep learning:

  1. Algorithmic Trading and Predictive Analytics:

   – Use Case: Deep learning models are used to analyze large volumes of financial data, identify patterns, and make real-time trading decisions.

   – Case Study: Renaissance Technologies, a prominent hedge fund, has successfully employed deep learning and AI techniques to drive their trading strategies.

  1. Fraud Detection and Prevention:

   – Use Case: Deep learning is applied to detect fraudulent transactions by analyzing transaction data for anomalies and suspicious patterns.

   – Case Study: PayPal uses deep learning models to identify fraudulent activities and protect their customers from unauthorized transactions.

  1. Credit Scoring and Risk Assessment:

   – Use Case: Deep learning models help financial institutions assess creditworthiness by analyzing customer data, including credit history, income, and behavior.

   – Case Study: ZestFinance uses deep learning to provide more accurate credit scoring and improve lending decisions.

  1. Customer Service and Chatbots:

   – Use Case: Deep learning-powered chatbots and virtual assistants are used to provide customer support, answer queries, and assist with account management.

   – Case Study: Capital One’s Eno is an AI-powered chatbot that helps customers with their finances.

  1. Portfolio Management and Asset Allocation:

   – Use Case: Deep learning is employed to optimize investment portfolios and asset allocation strategies based on market data and risk preferences.

   – Case Study: Wealthfront uses deep learning algorithms to automate portfolio management for individual investors.

  1. Market Sentiment Analysis:

   – Use Case: Deep learning models analyze news articles, social media, and other textual data to gauge market sentiment and make trading decisions.

   – Case Study: Hedge funds like Two Sigma use sentiment analysis models based on deep learning techniques to inform their trading strategies.

  1. Options Pricing and Derivative Trading:

   – Use Case: Deep learning is used to develop complex models for options pricing and risk management in derivative trading.

   – Case Study: JP Morgan has explored the use of deep learning for options pricing and risk management.

  1. Anti-Money Laundering (AML) and Know Your Customer (KYC) Compliance:

   – Use Case: Deep learning helps in automating AML and KYC checks by analyzing customer information and transaction data.

   – Case Study: Chainalysis uses deep learning to identify suspicious cryptocurrency transactions for AML compliance.

  1. Market Prediction and Forecasting:

   – Use Case: Deep learning models are applied to predict market trends, stock prices, and currency exchange rates.

   – Case Study: AlphaSense uses deep learning to analyze financial documents and predict market-moving events.

  1. Customer Personalization and Recommendations:

    – Use Case: Deep learning-driven recommendation systems offer personalized financial product suggestions, such as investment options, insurance, and loans.

    – Case Study: Netflix’s recommendation system is an example of deep learning applied to content personalization, which can be analogous to financial product recommendations.

These use cases and case studies demonstrate how deep learning innovations are transforming various aspects of the financial industry, from trading and risk management to customer service and compliance. By leveraging the power of deep learning, financial institutions can improve decision-making, reduce risks, enhance customer experiences, and stay competitive in a rapidly evolving sector.

“What are the implications and case studies of deep learning applications?”

Deep learning has had a profound impact on various industries and has led to numerous applications across different domains. Here are some implications of deep learning and case studies that showcase its applications:

  1. Image and Video Analysis

   – Case Study: Image Classification – Deep learning has revolutionized image classification tasks. For example, convolutional neural networks (CNNs) are used in autonomous vehicles to identify objects on the road.

   – Case Study: Facial Recognition – Deep learning has enabled facial recognition systems used in security, smartphone authentication, and even social media tagging.

  1. Natural Language Processing (NLP)

   – Case Study: Chatbots and Virtual Assistants – NLP models like GPT (Generative Pre-trained Transformer) have been employed in chatbots, virtual assistants, and customer support applications, such as OpenAI’s GPT-3.

   – Case Study: Language Translation – Deep learning models like Transformer have greatly improved machine translation services like Google Translate.

  1. Healthcare

   – Case Study: Medical Imaging – Deep learning is used for detecting diseases in medical images, such as X-rays and MRIs, aiding in early diagnosis.

   – Case Study: Drug Discovery – Deep learning accelerates drug discovery by predicting potential drug candidates and their interactions with biological systems.

  1. Autonomous Systems

   – Case Study: Autonomous Vehicles – Deep learning plays a crucial role in self-driving cars, helping them perceive their surroundings and make real-time decisions.

   – Case Study: Drones and Robotics – Drones and robots use deep learning for navigation, object detection, and task execution.

  1. Finance

   – Case Study: Algorithmic Trading – Deep learning models are used in financial institutions to analyze market trends, predict stock prices, and optimize trading strategies.

   – Case Study: Fraud Detection – Deep learning helps identify fraudulent transactions by analyzing large volumes of financial data.

  1. Entertainment

   – Case Study: Content Recommendation – Streaming services like Netflix and music platforms like Spotify use deep learning to personalize content recommendations for users.

   – Case Study: Video Games – Deep reinforcement learning is applied to create intelligent agents in video games, enhancing their realism and challenge.

  1. Manufacturing and Industry

   – Case Study: Predictive Maintenance – Deep learning models predict equipment failures in manufacturing plants, reducing downtime and maintenance costs.

   – Case Study: Quality Control – Deep learning is used to inspect and ensure the quality of products on assembly lines.

  1. Climate and Environmental Science

   – Case Study: Climate Modeling – Deep learning assists in climate modeling and predicting extreme weather events.

   – Case Study: Conservation – Deep learning helps in tracking and monitoring endangered species using image and sound recognition.

  1. Security and Cybersecurity

   – Case Study: Intrusion Detection – Deep learning algorithms are used to detect anomalies and potential security breaches in computer networks.

   – Case Study: Malware Detection – Deep learning helps in identifying and classifying malware threats.

  1. Art and Creativity

    – Case Study: DeepArt – Deep learning has been used to create artistic pieces, such as neural style transfer, which merges the style of one image with the content of another.

    – Case Study: Music Composition – Deep learning models can compose music and generate original compositions.

These case studies illustrate the versatility and transformative potential of deep learning across a wide range of fields. However, it’s essential to consider ethical, privacy, and bias-related implications in the deployment of deep learning systems to ensure responsible and fair use of AI technologies.

“What is the History and Timeline of the development of Deep Learning?”

The history and timeline of the development of deep learning can be traced back several decades, with significant milestones and breakthroughs along the way. Here’s a concise overview of key developments:

  • 1940s – 1950s: The foundations of neural networks, the basis for deep learning, were laid in the 1940s with the work of Warren McCulloch and Walter Pitts, who created a mathematical model of a neuron. In the 1950s, Frank Rosenblatt developed the Perceptron, an early form of a single-layer neural network.
  • 1960s – 1970s: Neural networks gained some popularity in the 1960s and 1970s, but their limitations were soon discovered, leading to the “AI Winter.” Researchers found that single-layer perceptrons had limitations in solving complex problems.
  • 1980s – 1990s: The resurgence of neural networks began in the 1980s with the development of multi-layer neural networks, often referred to as “deep” networks. The backpropagation algorithm, which allowed for efficient training of these networks, was developed. However, due to limited computational power and data, progress was slow.
  • 2000s – Early 2010s: Deep learning research continued but was relatively niche until the mid-2000s when computational power, larger datasets, and new techniques began to advance the field. In 2006, Geoffrey Hinton and his team published a paper on deep belief networks, reigniting interest in deep learning.
  • 2012: The breakthrough moment for deep learning came in 2012 when a deep convolutional neural network (CNN) called AlexNet, developed by Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton, won the ImageNet Large Scale Visual Recognition Challenge, significantly outperforming other methods.
  • 2013 – 2015: Deep learning continued to make rapid progress, with neural networks achieving state-of-the-art results in various tasks, including speech recognition and natural language processing. Google’s DeepMind developed deep Q-networks for reinforcement learning, further demonstrating the power of deep learning.
  • 2016: AlphaGo, developed by DeepMind, defeated the world champion Go player, marking a significant milestone in AI and deep learning’s application to complex games.
  • Late 2010s: Deep learning became increasingly integrated into various industries, including autonomous vehicles, healthcare, finance, and more. Transfer learning, where pre-trained models are fine-tuned for specific tasks, became popular, enabling faster development of AI applications.
  • 2020s and Beyond: Deep learning continues to evolve, with ongoing research in areas like self-supervised learning, generative adversarial networks (GANs), and reinforcement learning. Ethical and societal considerations surrounding AI and deep learning have also become a significant focus.

This timeline provides a high-level overview of the development of deep learning, but it’s important to note that many researchers and organizations have contributed to its growth over the years. Deep learning has transformed various industries and remains a vibrant and evolving field of study.

Tagged: AI Finance Sector
Daniel Pereira

About the Author

Daniel Pereira

Daniel Pereira is research director at OODA. He is a foresight strategist, creative technologist, and an information communication technology (ICT) and digital media researcher with 20+ years of experience directing public/private partnerships and strategic innovation initiatives.