top of page

The Evolution of Financial Trading From Algos to AI


Over the past few decades, computerized trading has made significant strides, and the advent of AI is poised to usher in another era of digital disruption in the financial markets. Investment banks and traders have consistently embraced technological advancements, and AI will be no exception. However, while AI has the potential to create winners, it can also lead to numerous losers, garnering public attention with its spectacular impact.

AI Algo Trading

Algorithms: The Building Blocks of AI


Algorithms have become ubiquitous in our tech-driven world, although their precise meaning often remains elusive. In simple terms, algorithms are sets of rules that guide the execution of tasks. They play a vital role in shaping our digital landscape, from solving basic mathematical problems to powering complex machine-learning systems. Algorithms are employed in various domains, including personalized advertising on platforms like Netflix, where they match users with relevant content based on data. Even online dating apps leverage algorithms to connect individuals based on their preferences and compatibility. In the financial markets, algo trading has gained popularity as a term referring to traders who utilize computers and predefined rules to execute trades.


The Rise of Algorithmic Trading


The 1990s marked a pivotal turning point for technology-based investing. Global deregulation and advancements in computing power opened up new possibilities. One prominent example during this period was Long-Term Capital Management (LTCM). Founded by John Meriwether, a former Solomon Brothers employee, LTCM attracted substantial funding and assembled a team of brilliant minds, including PhD holders and Nobel laureates. Equipped with sophisticated algorithms, the fund sought to exploit correlations between financial assets and secure risk-free profits.


Understanding Correlation in Finance


Correlation is a fundamental concept in finance, representing the relationship between different assets. For instance, stocks of oil companies tend to move in tandem with oil prices. By recognizing and capitalizing on these correlations, LTCM identified investment opportunities. The fund employed computers to analyze vast amounts of historical data and identify pairs of assets with expected correlations, such as US treasury bonds with slightly different maturity dates. This approach allowed LTCM to profit from small price discrepancies. Despite the seemingly negligible nature of these discrepancies, LTCM believed that through large-scale trades funded by substantial debt, they could guarantee profits based on statistical models. Their approach proved highly successful, with LTCM generating impressive returns that eclipsed market averages.


Limitations of Traditional Computers


While LTCM's algorithms excelled at capturing correlations and identifying profit opportunities, they faltered when it came to accounting for rare events or "tail risks" in finance. These events, which are difficult to predict, historically surpassed the computational capacity of traditional computers. Although the computers of the time were more efficient than humans at executing tasks and algorithms, they struggle to exhibit anything resembling true intelligence. LTCM's machines were no exception. While they flawlessly implemented trading rules and algorithms, they lacked the capability to adapt to unforeseen circumstances.


The Downfall of LTCM


In 1998, the Russian government's default on its domestic debts triggered a bond crisis that caused significant disruptions in financial markets. LTCM operated under the assumption that if Russian bonds lost value, the Ruble would follow suit, as the events were correlated. By hedging their losses on the bonds through selling Rubles, LTCM believed they had a risk-free investment. If the Russian bonds performed well, LTCM would make a profit, given the higher interest rates they offered compared to US bonds. However, if the Russian bonds faltered, signifying a government debt default, the Ruble's value would crash. LTCM had entered into contracts with Russian banks, which had agreed to buy Rubles from the fund at a predetermined rate if the fund demanded it. In their calculations, LTCM guaranteed a profit regardless of the outcome.

  • LTCM bought high-yield Russian bonds. If everything went to plan the fund would receive a significantly higher return than if it bought US bonds.

  • If something went wrong, for example, if the government defaulted, LTCM would lose money on the bonds. However, they hedged their bets by entering into a contract to sell Rubles (which were correlated to the performance of the bonds) at a predetermined rate to a domestic Russian bank, in theory ensuring LTCM could not lose.



Unfortunately, the real world did not conform to LTCM's statistical model. When the Russian government defaulted on its domestic debt, a bond crisis ensued, as expected. However, LTCM faced a problem. The bank that had promised to buy their Rubles had gone bankrupt due to the crisis, leaving the hedge fund no option but to accept substantial losses.


While a human investor might have intuitively grasped the link between the government defaulting, a currency crisis, and the vulnerability of domestic banks, it had never occurred before and was absent from the mathematical model. The machines remained oblivious to the vital connection between the real-world event of Russian banks collapsing and LTCM's theoretically fool-proof trades. The machines continued to trade aggressively despite the mounting losses, not realising the fund would be unable to sell Rubles at the expected price.


The Limits of Traditional Algorithms


LTCM's algorithms, built on certain assumptions, proved insufficient in the face of unforeseen events. Despite the disparity between predicted and observed outcomes, LTCM's algorithms continued trading as usual, unaware of the paradigm shift in progress. Consequently, the fund suffered catastrophic losses, ultimately necessitating a bailout from the New York Fed. LTCM's downfall serves as a cautionary tale for traders, underscoring the limitations of adapting traditional algorithms to complex real-world scenarios.


The Promise of Artificial Intelligence


Artificial Intelligence, in contrast to traditional algorithms, offers a new dimension to trading. While traditional algorithms adhere to predefined rules, the emergence of machine learning introduces a revolutionary concept. What sets AI apart is its ability to generate abstractions and comprehend intricate patterns. While machines may lack intelligence by any human measure, they possess the capacity to adapt and learn from data, leading to more dynamic decision-making processes.


Could AI Have Averted LTCM's Implosion?


The answer is a resounding "yes." The machines powered by today's AI technologies would likely have prevented LTCM's demise. The Artificial Intelligence of today is advanced enough to realize the risks associated with a bank going bust and grasp the broader ramifications of the government default preventing LTCM from hedging its losses. However, this does not imply that current or future AI-powered trading tools are immune to mistakes. On the contrary, errors are bound to occur. Nonetheless, the factors that led to LTCM's downfall were deeply ingrained in the software of the 1990s, and AI is steadily progressing to address and overcome these limitations. The next financial implosion, when it transpires, will likely differ significantly from previous ones.


AI: From Rule-Based to Outcome-Driven Trading


Traditional algo trading revolved around the application of rules, whereas AI is driven by defining outcomes. Modern AI enables machines to determine the most effective path to achieving specific objectives, such as maximizing profit. Even relatively basic AI systems possess the capability to identify flaws in models, correct course, or halt trading altogether when necessary.


Conclusion: The Future of AI Trading


Looking ahead, we anticipate an AI trading gold rush, with the 2020s paralleling the 1990s era of algo trading. We cannot know for sure what the AI impact on financial trading will be, only that there will be one. It is quite likely that an AI-powered hedge fund will emerge, armed with financial expertise and computational prowess, capable of trading independently. This fund will skillfully capture opportunities before the proverbial market upheaval, much like LTCM did. However, over time, the changing dynamics of the financial landscape will catch up, and even the most advanced AI systems will confront new challenges. This type of arbitrage trading is often referred to ask picking up nickels in front of a bulldozer. No matter how many nickels you pick up, eventually, the bulldozer will catch you.


Comments


bottom of page