Insights

Machine Learning

Machine Learning

Download Article as PDF

Blue Orbit Asset Management maintains an extensive quantitative research and development programme aimed at continuously improving our clients’ returns.

Machine learning – utilizing computer algorithms to provide insight and predictions – is one of the many tools we utilize in our research process. The main advantage of machine learning is that it can easily process huge amounts of data (commonly referred to as big data), to unlock insights beyond standard human processing capabilities and more traditional analytical techniques. Machine learning can be a very powerful tool. To maximize the benefits it is essential that application of the various techniques is undertaken in a thoughtful manner with a clear research question in mind. This issue is particularly relevant for Blue Orbit, where the underlying principle of our approach is to employ a highly transparent and clear/glass box approach to investing.

Machine learning – utilizing computer algorithms to provide insight and predictions – is one of the many tools we utilize in our research process.

The Blue Orbit research process is to first identify an investment thesis and rationale, based on our experience, market knowledge and academic research. The thesis can then be defined, tested, implemented and improved using our advanced quantitative tools and techniques available to us. Therefore, for Blue Orbit machine learning is a tool to inform and fine-tune our investment strategy rather being a blind driver of it.

This measured, scientific approach to investing avoids the temptation to data mine or let the black box drive the search for patterns that may appear to generate superior returns. Instead, our goal is to understand why and how we are doing what we do, so that our process is targeting real, sustainable outperformance rather than being fooled by temporary market patterns.

Machine learning in the research process

One of the first areas of focus within our machine learning research agenda was to seek enhancements to the fund’s Defensive Alpha signal. The intention of the signal is to provide downside protection, by overweighting those stocks that should outperform in a falling market but will not materially underperform in a rising market. While the existing signal construction has been performing as intended, it was anticipated that by employing machine learning further gains could be achievable.

In broad terms, machine learning algorithms fall into one of two categories; supervised and unsupervised learning, with the difference in simple terms relating to how much influence a researcher has in terms of specifying the learning model. Cluster analysis is an example of an unsupervised machine-learning tool. The intention of employing it was to separate stocks into meaningful clusters based on a host of variables, with these clusters showing meaningful return differences in down markets. Figure 1 provides an illustration of how cluster analysis can categorize stocks into clusters.

Figure 1:
Cluster Analysis Visualised

Cluster Analysis Visualised

The rationale for utilizing cluster analysis was twofold. The method allows Blue Orbit to undertake additional testing to confirm our signals and that the metrics we are using to construct them are performing as intended in given market environment.

blue orbit machine learning

Additionally, cluster analysis process has the capability to identify additional metrics that can improve Defensive signal’s performance. Importantly for Blue Orbit, it is possible to see how the clusters are defined, and assess the strength and explanatory power of the relationships. This is done through Dimension Reduction, a process which combines variables that contribute to the clusters. From Figure 2, one can see that dividend related metrics provide a significant influence on the formation of the clusters according to dimension 1. In addition, the influence of past returns and index weight is seen to have a lesser but a material influence, via their contribution to dimension 2.

Figure 2:
Blue Orbit: example of an Individual Factor Map – PCA

Factor Map - PCA

While Figure 2 provides a snap shot of the factors influencing the clustering in any given model, Blue Orbit can further utilise sophisticated visualisation tools to see the effect of the various contributors in a more granular, multi-dimensional fashion. Figure 3 provides an illustration of this work, which reinforces the existing construction of the Defensive signal.

Figure 3:
Blue Orbit: Metric Contribution to PC1

Metric Contribution to PC1

The second phase of the research agenda involved applying ‘supervised learning’- in this instance a classification model to identify the factor/s driving the specified outcome. In this process, rather than letting the algorithm naively form its own clusters, stocks were pre-labelled with a relative performance tag and the algorithm was asked to determine which metrics best explained the classifications. Figure 4 illustrates the resulting decision tree.

Figure 4:
Blue Orbit: Metric Contribution to PC1

Blue Orbit Metric Contribution to PC1

Using machine learning in this manner allows Blue Orbit to fully decompose our alpha signals to ensure they are working as intended, and to be able to visualise other interactions that are affecting the signal’s performance.

Through our machine learning analysis project into the Defensive signal, we have been able to identify an additional area of exposure within the signal that by controlling for, has the ability to improve signal returns in a falling market while minimally affecting its current upside capture. Machine learning is a powerful tool for us to gain a multidimensional insight into the exact drivers of our signal returns, to better understand the relationships between drivers, and to be able to control these relationships to gradually improve our signal outperformance. As part of our research and development program, we are committed to the continuous improvement of our existing signals and processes. We believe that the best and most consistent outperformance of a systematic strategy is achievable through continuously analysing, validating and improving our approach using the latest tools, data sets and technology available to us.

machine learning tool at blue orbit

Disclaimer

This document has been prepared for the general information of clients and professional associates of Blue Orbit Asset Management Pty. Ltd., ABN: 74 623 916 816 | AFSL: 513710 (Blue Orbit AM). This presentation has been prepared for use by wholesale clients only (within the meaning of the Corporations Act 2001 (Cth) and no other persons.

Information presented in this document is general information only, and is not intended to constitute personal advice or recommendations. This information has not taken into account your investment objectives, financial situation or needs. We strongly recommend that you seek your own professional financial and legal advice prior to any investment decision. While every effort has been made to ensure accuracy at the time of compilation, Blue Orbit AM makes no warranties or representations as to the accuracy, completeness or reliability of this information, nor that it is free from error. [You should read the information memorandum or other offer document for the fund and consider whether an investment is appropriate for you.] Unless otherwise stated, all returns shown in this document are simulated returns, and do not represent actual returns that
an investor received. Neither Blue Orbit AM nor any other party guarantees any income or capital return from an investment.

Past performance is not an indication of future returns. Any forward looking statements in this presentation are based upon Blue Orbit AM’s current views and assumptions and involve known and unknown risks and uncertainties, many of which are beyond Blue Orbit AM’s control and could cause actual results, performance or events to differ materially from those expressed or implied. These forward looking statements are not guarantees or representations of future performance and should not be relied upon as such. Blue Orbit AM undertakes no obligation to update any forward looking statements to reflect events or circumstances after the date of this document.

Download Article as PDF
Matthew Oldham Written by

Matthew Oldham

Matthew is an Assistant Portfolio Manager, responsible for assisting in all aspects of the investment process, including portfolio construction, investment research, trade execution and performance analysis. Matthew holds a Ph.D. in Computational Social Science (CSS) and is a CFA Charterholder. While completing his Ph.D. Matthew published various peer-reviewed journal articles researching the relevance of networks and feedback mechanisms in financial markets. Prior to his studies, and joining Blue Orbit, Matthew was an investment analyst at Equity Trustees (EQT) where he covered various sectors in the ASX200 and international equity funds. Preceding these roles Matthew performed corporate development activities at EQT and sales analytical and planning activities in the Australian automotive industry. Matt joined Blue Orbit Asset Management in January 2020.
Melbourne Head Office
Suite 409b, 480 Collins St,
Melbourne, VIC 3000
+61 3 5260 0929
+61 3 5260 0920

Sydney Office
Level 13, 333 George Street
Sydney, NSW, 2000
info@blueorbitam.com