[The Institute for Innovation Development interview series invites innovation experts, innovative business leaders and emerging fintech companies to talk to our readers about their latest innovation activities. The series seeks to learn from innovative business creators, uncover innovation best practices, and discover how to apply these insights into a financial services business model.

We recently sat down with co-founders Andrew Dassori and Mark Landis of Wavelength Capital Managementan innovative money management firm leading the efforts in connecting academia and artificial intelligence to the investment process.]

Hortz: Your firm has allocated substantial resources to apply artificial intelligence to investing. How exactly are you executing this and what are you learning?

Dassori: Marvin Minsky, the founder of the artificial intelligence lab at MIT, breaks the field down into three basic approaches: connectionist reasoning and rule-based and case-based artificial intelligence. We focus on rule-based artificial intelligence, as it is the most applicable and most relevant to what we do when investing in the markets. Rule-based systems provide a way to test logic by looking back to see if investment rules have been return predictive. This can be done alongside an ongoing investment process. Artificial intelligence is thus a tool for us to look at the past as well as the future. It allows us to be more predictive.

Our research involves identifying investment logic that is fundamental in nature and actionable through liquid financial instruments. We test this logic across different time frames and economic environments using extensive amounts of empirical data. If the investment logic is consistent, significant and robust, we are then able to implement it live in parallel to our existing decision engine for investments. It can be added to the portfolio to enhance investment processes already in place.

We have also learned that as smart as humans are, computers simply have more capacity to keep track of rules and can process them more efficiently with discipline. Computers are able to analyze data using higher level statistics than what would otherwise be possible. We are not advocating for computers to take over the investment process entirely—there are meaningful risks to this. We believe having humans oversee the design and implementation of these processes is critical, but as markets and the set of factors driving them become more complex, incorporating artificial intelligence will be increasingly important to managing investment strategies successfully.

Hortz: How many different kinds or levels of algorithms are there being used in investing today? How are algorithms being improved or innovated?

Dassori: Algorithms can be incredibly sophisticated or designed as simple systematic ways to tackle any basic problem. We use a robust set of algorithms that are designed to work together, much like a doctor might prescribe a combination of medicines to fully address an ailment. Different algorithms are used to assess forward-looking risks as a base for our investment strategy, and others to actively gauge the primary investment characteristics we believe to be predictive of excess returns: carry, momentum and value. We measure each of these characteristics across multiple time frames and individual inputs. 

The different algorithms may work at different times or during different market cycles, but the combination of them, working over the long-term, is what makes our system robust. We combine their output efficiently to produce investment signals, and these signals determine buy and sell decisions within our investment process.

As to innovation or evolution of algorithms, know that the process described above improves each day through the ongoing reassessment and refinement of existing components of the overall system. In a rule-based system, you can build triggers that will remove an asset that is no longer responsive to changes in the economy and rebalance risk automatically to the assets that are responding. You don’t have to be sitting in front of a Bloomberg [terminal] waiting to spot disconnects from the market. We can automate that process as a risk management component to ensure that we address any issues in the portfolio when relationships change. This provides us more time to focus on building our systems by adding new pieces of investment logic and refining new algorithms that enhance the predictive return power of each of the signals.