AI-Powered Finance: Boom or Bust? you decide.

AI-Powered Finance: Boom or Bust? you decide

AI in Finance

Depending on what you do for a living, where you are now, and your age, you might classify it as boom or bust, but to me, after "playing" with neural nets back in the early 90s as Director of Quantitative analysis for a New York based, long-short hedge fund, we are definitely in the boom stage of AI, where great wealth will be generated in the next 1-3 years for those who have the vision to create products that the market needs, or maybe create entirely new markets, and/or services. AI is bringing about profound and sweeping transformations across the finance industry.

With decades of cumulative experience in data science, quantitative analysis, and machine learning, industry veterans have witnessed firsthand how AI unlocks greater speed, precision, creativity and customization for financial institutions. At the same time, these industry veterans are visualizing the next stage of this AI revolution years ahead of institutional implementation, with applications is risk management, insurance, entertainment, publishing and even in the legal world. AI is here to stay.

In finance, AI is driving a revolution impacting analytics, modeling, trading, risk management, and core processes. In a nutshell, AI is already benefiting from 4 mains aspects:

• Imperfect / asymmetric information
• Behavioral effects / biases
• Microstructure
• Regulation

One area experiencing significant disruption but with some mixed results is financial forecasting, with the exception of Renaissance Technologies' Medallion Fund which keeps an exceptional and impressive track record in its returns. By comprehensively detecting subtle and complex patterns across massive, highly granular datasets, AI empowers more accurate forecasts of revenues, asset valuations, credit risks, and other key financial metrics.

Back in 2012, with my startup Ttwick, I was making accurate predictions for some companies that traded based on top line revenues, analyzing millions of data points coming from public sources (including weather) and analyzing the data with the aid of machine learning algos, and selling specific trade recommendations to hedge funds. I even got offers from Point72 to acquire my data collection, however, if you are just a computer science guy or programmer, you might have sold your data to a hedge fund for peanuts because you did not know how to profit from it. That was not my case and rejected Point72's offer.

Today, what I call "low hanging fruits" in alternative datasets are not there any more, but some other opportunities have emerged. Needless to say, those were exciting days, if you took options "bets" against giants like Tiger Global Management, and saw a nice return in your investment because your data was truly capturing alpha, and your models were better than theirs in very particular opportunities. Fun at least while it lasted.

Ttwick was capturing commercial transactions for certain public companies in quasi real-time (alt data) and forecasting revenues ahead of earnings using machine learning models.

Today, new enhanced predictive insights enable more informed and data-driven strategic decision making. Sophisticated machine learning and time series analysis also allow users to uncover deep relationships between financial metrics, market factors, and qualitative drivers that even the most seasoned industry veterans can miss. This augmented financial intelligence advances risk management and prepares institutions for a wider range of scenarios and products.

Financial modeling is another domain being optimized and reinvented by AI. Advanced algorithms have the capability to rapidly synthesize and analyze both structured and unstructured data to inform the assumptions and projections underlying models. AI can even generate models directly from natural language descriptions. This automates and enhances the flexibility of modeling processes versus tedious and static manual methods. Once created, AI enables continuous updating of financial models by monitoring and incorporating new data, research, and market conditions as they emerge. This allows models to remain relevant and optimize their predictive accuracy as changes occur.

In addition, AI can customize models for specific assets, geographies, risk scenarios and client needs. Coupled with some aspects of financial engineering, It can be even used for the evaluation and design of very sophisticated financial structures, such as valuation of slate of films yet to be created, for purposes of securitization (more about this in an upcoming article).

AI techniques are being applied across financial modeling processes - from options pricing and Monte Carlo risk analysis to forecasting, valuation of structured products, stress testing, and more. This expands modeling capabilities and frees up human analysts to focus on higher-value work. As mentioned before with the case of Medallion, systematic day trading is an area experiencing changes. By continually analyzing data and learning from performance outcomes, these algorithms adapt to evolving market conditions across various time horizons. This facilitates exploiting short-term inefficiencies and optimizing returns. AI autonomous trading also reduces the influence of human emotions and biases like loss aversion, overconfidence, and herding. This leads to increased discipline in trade execution. The scalability of AI further allows simultaneous trading across equities, futures, currencies, derivatives, and other asset classes while considering inter market dynamics. This expands capabilities dramatically beyond manual approaches. Machine learning models can greatly improve price transparency and valuation accuracy for complex securities like green bonds and parametric catastrophe bonds. During issuance, AI algorithms can analyze vast datasets - including past issuances, modeled climate projections, real-time sensor data. etc. to generate fair value estimates and transparent pricing models. These data-driven AI models account for the intricate relationships between interest rates, environmental factors, competing maturities and yields, and even catastrophic risk probabilities that influence the tail risk of expected returns. In secondary market trading, AI techniques could enable continuous re-calibration of pricing models as new data emerges. Advanced natural language processing could scan news, research, and regulatory changes to instantly update models. Meanwhile, deep neural networks could ingest streams of meteorological readings, seismic activity, and other data to recalculate catastrophe risk probabilities in real-time. This allows traders to mark green and parametric bonds to market with greater precision. As a side note, the green bond market is poised for major expansion as issuers globally embrace sustainable financing, and AI could play a key role in the success of this relatively new market. This is definitely at the top of the list where I see great potential, with new companies such as Hong Kong based Carbonbase, founded by Schwarzman Scholar, Forbes30U30, Climatetech Entrepreneur Max Song leapfrogging initiatives of even large players in the the ESG arena. Estimates suggest annual issuance could reach $1 trillion by 2025, almost tripling from 2021 levels. Growth will stem from surging corporate and sovereign supply. Over 140 governments have set net zero emissions targets, requiring green bonds to fund decarbonization. Major corporations are also committing to emissions-reducing projects financed through green bonds. Emerging markets will further boost issuance. Supply from emerging entities jumped five-fold from 2020 to 2022 as sustainable investing gains momentum there. EM corporates, municipalities and sovereigns will tap green bonds for climate-aligned growth. Meanwhile, global investors seek EM green bonds for diversification and yield.

Generative AI: A Game Changer for Risk Management

One of the most exciting emerging AI applications for insurance risk analytics is generative modeling. Generative algorithms can synthetically generate new, realistic data samples after learning patterns from real-world data.

For insurers, generative models open exciting new possibilities to enhance risk management:

The fledgling state of green bonds reminds me of the early days of insurance-linked securitization and litigation settlement fee trusts securitization, when I worked at Deutsche Bank (DB) in the early 2000s. Back then, rating agencies bundled catastrophic and insurance risks together, limiting investor appetite. My team at DB used bespoke models using generative AI (MCMC) to demonstrate to S&P that Insurance Liked Securities (ILS) and Litigation Settlements Liabilities were distinct categories with unique diversification benefits. In the case of CAT bonds, my team's research catalyzed market growth (Deutsche Bank’s ABS Report “Cat Bonds, opportunities for Issuers and Investors”, which you can find here) by attracting arbitrage CDO and ABS of ABS managers to these bonds due to their yield and correlation benefits. A similar thing is occurring with green bonds, where lack of understanding of the risks of the assets, and their true impact in the environment is limiting a great deal of mainstream adoption, although in a less dramatic fashion compared to the CAT market back in the early 2000s. But there is room for improvement in the regulatory & transparency aspects. I believe that with transparent AI-enabled analytics, the market can reach tremendous potential. As Elon Musk tweeted on May 2022:

Although I do not agree with everything Musk says, he has a point: ESG metrics may fail to capture true impact, similar to what S&P was failing to capture back with ILS as an emerging asset class. Just as our models and papers at Deutsche Bank ABS debunked myths about ILS and catalyzed growth back then, ML analytics and trustworthy AI can unlock the full potential of the green bond market, truly aligning environmental & social objectives with investors' profit objectives. Maybe the way to accomplish this is by a new entity in some sort of “financial sandbox” that transparently uses AI for risk assessment, pricing, distribution, and rating, allowing the "test" of some capital markets investors in rated securities, to new securities with low or zero correlations with other assets and with transparent and measurable links to ESG objectives, pricing and secondary market trading. Support of academia is crucial, to accelerate adoption and trust in ethical AI systems for the securities industries. In general, AI could deliver higher-fidelity valuations of green and catastrophe bonds by synthesizing more diverse, dynamic data than human analysts can handle. This increased transparency limits opportunities for arbitrage and manipulation that can occur with opaque, over-the-counter trading. Hence, responsible deployment of AI promotes fairer, more efficient markets aligned with the growth of sustainable finance.

In the area of risk management, AI is enabling more holistic, granular, and real-time assessments. By comprehensively analyzing vast, diverse structured and unstructured data sources, AI models can uncover emerging correlations, outliers, and early warning signals that conventional rule-based systems miss. This provides a more accurate view of risks across credit, market, liquidity, operational, cybersecurity, regulatory and other domains, facilitating proactive mitigation. In reference to trading and risk management, I can highlight 2 aspects where I have been personally involved: 1) The use of generative AI for scenario analysis to determine the risk of a fixed income transaction, and to help in the decision making process (you can read more about it here). Also below is a tweet about a meeting where I talked precisely about generative AI for ILS (but also for music) back in 2005, at the Bay Area (San Francisco) Machine Learning Group. Generative AI was not even a "thing" then, but "AI visionare" Shivon Zillis, head of Bloomberg VC arm at that time -Bloomberg Beta- and who had seen some of my generative AI work in finance, insurance and art, invited me to give a speech that included senior engineers and executives from some well-known tech firms in Silicon Valley. 2) A project I have been developing with some colleagues, where we use AI to analyze 3000 stocks of the Russell 3000 index on a daily basis, and 500 stocks of the S&P 500 on a tick-by-tick basis, over several decades of data and for decision making in the buy/hold/sell decisions in different time frames. Believe of not, some of those elements of generative AI for music described above have been incorporated into the money management project, since the proprietary models that generated those songs (including "regime" shifts and "correlations") are very applicable to the markets (you can sample some of this time series in the "Generative AI circa 2015" section towards the end of the page in this link) Below a table of some of the aspects we cover in our AI driven asset manager:

Without disclosing key performance metrics of our approach, below are some partial metrics of one of our first ML strategies, long-short, compared to the benchmark, the S&P500, from 2000-2017. Sudden regime changes (such as COVID) were not properly captured by our initial model, but we have made tremendous progress using techniques to assign probabilities to “regime changes” as a separate model, and generative AI to capture the effectiveness of the technique. (This early model consistently beat the S&P500 in backtests but experience drawdowns slightly larger than the benchmark due to regime changes like COVID, that presented unique challenges. The intraday model was designed out of this observation).

AI also enhances regulatory compliance and fraud prevention through advanced behavioral pattern recognition. By analyzing transactions, communications, customer data, and other behaviors, sophisticated AI algorithms can identify suspicious activities indicating financial crimes. This allows firms to preemptively detect and investigate money laundering, improper trading, identity theft, predatory sales practices, and other misconduct much earlier than traditional methods. Recently, I had the opportunity to attend an intensive class at the Artificial Intelligence Finance Institute (AIFI) at NYU. The AIFI mission is to be the world’s leading educator in the application of artificial intelligence to investment management, capital markets and risk. I got introduced to the excellent python Riskfolio-Lib library for quantitative, strategic asset allocation. Just this aspect of the AIFI class was worth the attendance. I highly recommend the AIFI bootcamp if you are interested in the topic, with an excellent program designed by its CEO, Miquel Noguer Alonso, AIFI Co-Founder.

Responsible AI Systems and Regulation

The development of AI systems for finance must adhere to principles of responsible and trustworthy AI. As discussed in "Connecting the Dots in Trustworthy AI", by Dr. Marcos Lopez de Prado and Dr. Natalia Díaz Rodríguez. Responsible AI demands accountability, auditability, and compliance with ethics and regulations. The EU's risk-based approach to AI regulation provides a framework to achieve this in high-risk sectors like insurance and finance. Under the EU's AI Act, high-risk AI systems must meet requirements around risk management, data quality, transparency, human oversight, and robustness. Responsible AI in these fields would undergo conformity assessments to demonstrate compliance. Regulatory sandboxes could allow financial firms to test AI systems per these specifications in a controlled environment. Overall, responsible AI ensures trustworthy outcomes in finance via rigorous auditing and alignment with EU regulations on ethical AI.

In summary, properly implemented and regulated, AI could bring game-changing improvements to risk analytics, trading, bond underwriting, mark to market pricing, liquidity, transparency and ultimately profitability. Leading financial firms are prioritizing AI adoption to remain competitive and maximize value in the 21st century landscape.