A few short years ago, a now-prominent investor in all things disruptive informed me that deep neural networks are "the most important software breakthrough of our time" and one that "could turbocharge every industry," yet they couldn't be used to build investment portfolios.
As iron sharpens iron, and as a fellow investment manager, I might agree that humans, in some cases, might be better at picking stocks. But machines have the upper hand when it comes to portfolio construction.
The use of end-to-end learning to solve complex investment problems is exactly what deep reinforcement learning provides and is the key to resuscitating active management. Whether fundamental or quantitative, long-only strategies are typically designed with portfolio construction as a means to conform to a benchmark and provide appropriate diversification. Superior stock selection is by definition diluted. Investment consultants like this behavior because it protects their businesses and their clients from the absence of skill. As long as everyone can explain the underperformance (e.g., "the manager's style is out of favor," or "it's difficult to create alpha when the market is so irrational"), the case for active management remains appealing because some day we will have that "comebacker." Meanwhile, we will hear more about how managers are "refining" their models in models in the wake of mediocre returns.
There's a better way for asset owners to achieve the returns they seek and for active managers to break the cycle of underperformance and deliver consistent value: use artificial intelligence — specifically, deep learning and deep reinforcement learning — to build portfolios.
These systems have disrupted health care, transportation, robotics and other verticals, demonstrating that machines can perform complex tasks previously thought to be only the domain of humans. The same disruption will certainly come to asset management. The convergence of computer power and neural networks provides a compelling pathway for developing investment strategies that are capable of superhuman pattern recognition and might be able to deliver the currently elusive but sustainable information edge to earn alpha.
Advanced AI can enhance the value of each investment decision through a dynamic system of reward and penalty that produces a prediction and an allocation in one holistic process. These models require significant research and development and success is uncertain. If properly designed, tested, and implemented, however, these models are scalable and could potentially provide alpha and beta more reliably without substantial additional cost. And while both traditional and AI models require humans for successful design and development, once fully trained, deep reinforcement learning models process complex datasets and improve autonomously through market cycles and do not require us to develop new parameters as they adapt through non-stationary market environments. The unique behavior of advanced AI can be beneficial during times of heightened market volatility.
When the volatility index signals instability in equity markets, deep reinforcement learning-generated signals will generally become more accurate as patterns driven by flight-to-quality behavior become more dominant in market pricing. A deep reinforcement learning model will modify trading decisions and adapt to changing market conditions more quickly than models designed with static factors that view risk/return relationships through a static lens.
A well-constructed deep reinforcement learning model has the potential of producing superior investment returns at risk levels that achieve better investment outcomes through more consistent diversification at the total fund level. The goal of deep reinforcement learning is to maximize the reward of each investment decision and self-moderate risk through the level of allocation to the underlying signal. This information edge could improve the path of returns produced. This behavior is already evident outside of asset management. Artificial intelligence models used in cancer diagnosis and prognosis know nothing about medicine. Yet by focusing entirely on the data, these models are at the forefront of cancer research.
Investment decisions produced by neural networks are similarly entirely data driven. The models are freed from statistical frameworks that force a solution through a prescribed set of investment characteristics. These models are agnostic about traditional quantitative drivers of investment return and are not programmed to mimic the decision-making of the greatest human investors. Instead, these algorithms hunt through the data, identifying patterns and use this knowledge to make both investment predictions and optimal allocations.
A growing body of evidence supports the case that advanced artificial intelligence could be used to build robust investment strategies that generate consistently attractive returns with low correlations to a range of asset classes. While some quant managers seek to adapt AI within their current investment strategies, in the not-too-distant future, more firms will likely use advanced AI exclusively as their investment engine to produce more consistent results. Surprisingly, today's new celebrated investors tout the benefits of transformative technologies for life and death decisions but lack the conviction to use them to make investment decisions. As Steve Jobs once said: "Innovation is the ability to see change as an opportunity — not a threat."
Julia Bonafede is co-founder of alternative money management firm Rosetta Analytics Inc. She is based in Minden, Nev. Previously, Ms. Bonafede was president of Wilshire Consulting and a member of Wilshire's board of directors and Wilshire Consulting's investment committee. This content represents the views of the author. It was submitted and edited under P&I guidelines but is not a product of P&I's editorial team.
