Using Alternative Data and Machine Learning in Alternative Asset Classes

Using Alternative Data and Machine Learning in Alternative Asset Classes

Keith Black, PhD, CFA, CAIA, FDP, Managing Director of Content Strategy, CAIA Association

Michael Oliver Weinberg and Peter Strikwerda work at the Dutch pension fund APG and serve as the head of hedge funds and alternative alpha and the global head of digital and innovation, respectively. CAIA Association and FDP Institute recently had a conversation with this APG team regarding their views on how alternative data and machine learning can be used in alternative investments.

There are generally three ways that an asset management firm can be designed. A traditional manager that uses discretionary or fundamental techniques typically has not explored the use of alternative data or machine learning approaches. Fully quantitative firms can be built specifically to facilitate machine learning and alternative data approaches. These teams work best when building a team from a variety of backgrounds including finance, computer science, engineering and physics. The key is to match data scientists with financial professionals with good intuitions, such as those who have studied for the Financial Data Professional program. In between the two are quantamental managers who may have started off as a fundamental or discretionary manager but are increasingly comfortable using quantitative methods and alternative data sources.

Before considering how machine learning and alternative data are applied to the investment process, teams must first evaluate how their organization has been designed. It is important that the goals of the team are clearly stated. Rather than running millions of models and simulations and seeing what the results might be, analysts need to understand why the models are doing what they are doing and have a testable hypothesis with an economic rationale before turning the computer loose. Ideally, your team will earn organizational alpha as you can improve efficiency by planning how you work with data, diversifying the skills of your team, and having a well-planned technology architecture. One of the key deliverables is to find lots of good quality data at reasonable costs that can be quickly loaded into your systems in just one day. The real strength of the team is in combining skills across disciplines, finance plus engineering plus computing is more multiplicative than additive. Senior management needs to support the alt data strategy and to empower the team to understand that failures are part of the research and investing process. Of course, it is hoped that failures lead to innovations and improved processes, but the research team needs to know that it is acceptable if not every model or trade is a winner.

In addition to placing controls on the modeling processes and working quickly and carefully, it is vital to employ risk management techniques to prevent the model from going rogue. Algorithmic risk management is similar to discretionary risk management, where limits are placed on position sizes, gross or net exposures, and sector weights. Portfolio managers and the model validation team both need to consider risk management in their processes.

Many large allocators do not manage money directly but invest in a number of external managers. Manager research processes should be able to carefully evaluate quantitative managers and strategies. It is important to evaluate each manager’s backtesting process and ensure that risk management has been carefully programmed into the algorithms. Emerging managers might offer transparency into their trading process after the investor signs a non-disclosure agreement. Data used to evaluate models and manager performance should be related to frequency of the model. Managers with weekly turnover can be easily evaluated on a six-month basis, but it takes a longer time to evaluate performance of managers with longer holding periods. It is important to have a portfolio diversified using multiple models and multiple managers, as you want to fly on a plane with more than one engine and not place your faith in a single airplane engine or a single factor model.

Investors need to understand these strategies before making a commitment to invest in funds based on machine learning and alternative data. The market for alternative data very inefficient and better alpha may come from cheaper data sets. The key is to look for more unique data that isn’t as widely distributed. Data on satellite images and credit card transactions is both expensive and widely used. As a result of this overexposure, the alpha of this data has been decaying rapidly. Given that 80% of the data ever created has been saved in the last two years, there is no shortage of new data sources. The key is finding this data, accessing it cheaply, and evaluating the alpha potential quickly in an efficient process.

Dutch investors are some of the global leaders of including environmental, social, and governance (ESG) factors into their investment and manager research processes. Machine learning and alternative data processes help APG better track impact of their ESG targeted investments. As part of their manager research process, it is imperative to investigate that alternative data sources used by managers have been obtained ethically and contain no personally identifiable information. Investors need to be careful with ESG data, as many companies don’t disclose this data or do so on an inconsistent basis. ESG data is of variable quality and needs to be closely evaluated.

Concluding remarks were that everyone should learn python, R, machine learning, and data science. These tools will be integral to success in financial markets and will be tested on the next FDP exam. Registration opens May 10 for the next FDP exam, which will be held from October 26 to November 8.

Listen to the full webinar.

 

 

 

Be Sociable, Share!

Leave A Reply

← Covid-19 Alpha Wedding ESG with Modern Portfolio Theory →