top of page

A Deep Dive into Artificial Intelligence | Infinite Loops


Check out the full podcast show notes: Julia Bonafede — A Deep Dive into Deep Reinforcement Learning (EP.135)

This Week's Podcast Guest

Julia Bonafede is the co-founder of Rosetta Analytics Inc, “an alternative asset manager that is pioneering the use of advanced artificial intelligence to build and actively manage liquid investment strategies.”


Prior to co-founding Rosetta, Julia served as President of Wilshire Consulting and was a member of Wilshire’s Board of Directors and Consulting Investment Committee.



Intro

  • Rosetta Analytics uses artificial intelligence and machine learning to build and manage investment strategies. The company is a leader in the burgeoning use of AI in investing

  • People tend to get nervous about Artificial Intelligence and Machine Learning. The terms are even relatively new to us.

  • The term Artificial Intelligence was coined by Dartmouth Math professor John McCarthy, in the summer of 1956


Key Takeaways

  • Artificial intelligence will evolve from a nice-to-have to a must-have for asset managers. Artificial intelligence can provide a leg-up to asset managers - ultimately increasing returns.

  • Current applications of Artificial Intelligence include autonomous driving, medical imaging, robotics and more. The use-case for AI will continue to evolve as the technology improves.

  • Humans struggle with emotion when it comes to investing, especially when it comes to the risk of losing money. The best deal maker is the one who will walk away from an emotional trade.

  • The key to deep reinforcement learning is for the system to learn from mistakes, and optimize good decisions. The system can learn from mistakes and optimize the rewards.

  • When discussing Artificial Intelligence, people usually bring up SkyNet and The Terminator. Technology is neutral, it is humans that either use it for good or bad things.

  • If we succumb to fear over new technology, we will never be able to progress as a society - what if we never adopted computers, or advanced from horse and buggy to engine-powered automobiles? Where would our society be today?

  • "The treasure you seek is in the cave you fear to enter" - Joseph Campbell



What is the role of Artificial Intelligence in Investing?

  • Artificial Intelligence can help to dictate portfolio analytics and risk models.

  • Artificial Intelligence is particularly adept at discerning the relationships between different markets

  • Artificial Intelligence is a more sophisticated model than previous investment models. The Alpha and Beta style of investing reached its peak in the 1990's

  • Alpha and beta are two of the key measurements used to evaluate the performance of a stock, a fund, or an investment portfolio.

  • Alpha and beta are limited in their ability to identify relationships between data points


Performance of Active Management vs. Passive Investing
  • When looking at data on firm performance, active management has failed to consistently outperform index funds over time


Passive vs. Active Investing Performance

  • During some market cycles, active management performs better than passive management. However, at this time, there is no consistent outperformance generated by active management

  • Jack Bogle created the first Vanguard index fund to hedge against Beta

  • Both Jim and Julia see a strong use-case for using artificial intelligence in the investment process. However, asset management is a relatively slow field in adapting to new technology and change



What is Artificial Intelligence?

  • "Artificial intelligence is an umbrella term used to describe any model that simulates human intelligence"

  • Machine learning is a category within artificial intelligence

  • Neural Networks are mathematical functions that identify relationships between seemingly unrelated data. By nature, they are adept at finding relationships between non-linear datasets

  • Neural networks have the ability to "remember" relationships and the exhibited trends. It can use that "memory" to forecast future trends and relationship

  • When presented with a new decision or prediction, it will revert to its existing relationships and identify the closest pattern to the current scenario it is being presented with

  • Reinforcement learning (RL) is an area of machine learning concerned with how intelligent agents ought to take actions in an environment in order to maximize the notion of cumulative reward. - minimize a loss function.



Current Applications of Artificial Intelligence


Reinforced Learning

  • The old model in finance and investing was to reduce the variability between what you are trying to solve

  • Reinforced learning models are able to iterate in many steps.

  • Reinforced models are sequential, multi-step optimization frameworks used in making decisions



Applications of Reinforced Learning

  • Autonomous Vehicles - identify optimal distance and speed between two cars

  • Content moderation on social media



Jim's Thesis

  • Black swans, by their very definition, cannot be predicted. But can we use machine learning to identify when a black swan event has occurred?

  • Example: when the price of oil went negative (due to bad actors), Jim would have loved a confirmation that there was a black swam event occuring

  • Jim pitched this thesis to his friend, a deep learning and machine learning expert. His response: do you have an extra $500 million lying around?

  • The depth of information required to "predict" or confirm a black swan event would be staggering



Non-Linear Thinking

  • Artificial Intelligence will replace repetitive functions and jobs currently held by humans

  • However, jobs that require intuitive thinking will still require human input. Sometimes humans just know an answer intuitively, but computers do not

  • Humans struggle with emotion when it comes to investing and losing money

  • The best deal maker is the one who will walk away from an emotional trade

  • If you have a good investment model, do not override it. No matter how much your emotions want you to

  • When markets de-stabilize, there are patterns in how they behave

  • Optimization frameworks learn and adapt to new information - they are infinite loops with a reward and penalty function

  • The key to deep reinforcement learning is for the system to learn from mistakes, and optimize good decisions. Learn from mistakes and optimize the rewards

  • Arbitraging human nature is the last sustainable edge

  • In order to move forward in his quest, Odysseus needed to hear the Siren's Song.

  • He set up a system (tying himself to the ship's mast) to ensure he would reap the rewards of the Siren's Song without falling prey to the dangers



Regime Change in Financial Markets

  • Jim recognized that throughout his investment career, he and his compatriots were all relying on the same data sets.

  • Moreover, they had all made similar assumptions using those same data sets

  • Going back to 1927, general momentum was a good metric of securities to buy. However, momentum inverts - top performers become the bottom performers, and bottom performers become the top performers

  • Rosetta Analytics' investment models are most active when there is distress in the markets

  • Leverage is the cause of most financial downturns

  • True alpha is generated by swinging for the fences

  • Jim believes that Artificial Intelligence for Asset Managers will go from a nice to have to a must-have



How to address the human need to know "why"

  • It is in human nature to justify of our actions and views - this is why stories are so important to humans

  • It is challenging to explain the complexities of financial models to clients. The decisions of models may seem irrational, when in fact they are the right decisions

  • Ultimately, asset management will have to adopt artificial intelligence given outsized expected returns



How to pitch Artificial Intelligence to clients

  • Deep learning will produce uncorrelated returns, which are highly coveted in the current market regime

  • Deep reinforcement learning will product efficient compounding of returns

  • The important question is how the different strategies within your portfolio work together to produce a positive financial outcome?

  • Rosetta Analytic's portfolio structure will change over time as the models identify new regimes

  • The portfolio will be flexible and adaptive. For example, in March 2020, old models fell apart. However, Rosetta's models kept adapting to new information



Science Fiction has mischaracterized Artificial Intelligence

  • When discussing Artificial Intelligence, people usually bring up SkyNet and The Terminator

  • Technology in itself is neutral - it is humans that either use it for positively or negatively. People have the responsibility to use technology responsibly

  • Fire is a neutral technology. Fire can cook food and warm our homes, or it can burn down fields.

  • We developed fire departments, fire alarms, fire hoses and other systems to mitigate damage from fire

  • Do not suppress dialogue and opinions, we need to have active discourse. Open dialogue will lead to the optimal outcome

  • Julia's Parking Lot Theory: We all have a parking lot in our brain that can only fit so many cars - limited space for thought

  • If we succumb to fear over new technology, we will never be able to progress as a society

  • What if we never adopted computers, or advanced from horse and buggy to engine-powered automobiles? Where would our society be today?



Julia's Recommendations on Enjoying Life

  • Wake up and see the joy in everything

  • If we only pay attention to the negative things, negative is all we will see

  • Fear not, we have to take risks in order to advance - overcoming fear is part of the human experience

  • "The treasure you seek is in the cave you fear to enter" - Joseph Campbell

32 views0 comments
bottom of page