Maximising Information Coverage in Investment Research
By Wian Stipp
The problems with current investment research practices
When conducting research around a stock, it is common to end up with documents and information isolated and siloed across many folders. If this information could be connected, it would be much easier to identify relationships across documents. The end result being much quicker trade evaluations. Instead, analysts’ work is less efficient because they spend a lot of research hours piecing documents together, which is time that instead could be spent on using this information to generate profitable ideas.
Consistently formulating compelling investment hypotheses requires some form of deeper level of knowledge about the target company than the market. One important way to gather this information is to monitor news that affects the company. With so many news sources, just subscribing to more and more news services is likely to only lead you to becoming overloaded with information. In other words, actually maximising your news coverage has traditionally led to Information overload, increasing the chance important information is missed.
A final problem with the traditional approach to collecting company information is that most sources only explicitly report first order consequences of the company. Despite the information about higher order connections being present in most investment research, this information is difficult to transfer from the original researcher as the understanding of these intricate connections is rarely recorded in a meaningful way. This is despite the higher order relationships being more valuable, as they lead to insight into scarce information, forming the basis for more profitable trade ideas.
To summarise the problems analysts have when researching companies: documents and information is normally siloed, which leads to hidden relationships being missed. As time goes by, analysts become overwhelmed by information overload, and it becomes difficult to keep on top of the relevant information for a company. In short, access to information isn’t enough, if the volume of information overwhelms you, you can’t make sense of it. Simply ingesting more data will not necessarily lead to more insights Lastly, only the first order connections are reported in the news, and there is a lot of value left on the table from not understanding the higher order connections. Next we will look at a better way of structuring information before applying it to these problems.
Knowledge graphs are a better way to store research information
Knowledge graphs are a more effective way to store information as they address the information storage problems described above. A knowledge graph (KG) organises information into an ontology — a set of concepts and the relationships between them. From this ontology, one can perform interesting inference tasks. Below is a simple knowledge graph, where nodes (the circles and squares) represent concepts/entities and edges (the lines) represent relationships.
Instead of having pieces of information isolated and static, KGs are able to create a latticework of relationships between concepts across documents. Information is a pile of unstructured concepts, but knowledge is adding a layer of connections (relationships) between these concepts. Structuring information in this way comes with the ability to perform many different powerful functions. For example, in the example above, we have captured an understanding that Sarah is the CEO of A and that A is a competitor of B (and to what extent). This could, for example, allow us to automatically evaluate the relevance of a news article about Sarah on company B. This is not the same as a relational database, which cannot easily represent these heterogeneous complex relationships where a KG can.
For example, here we can see an unsorted snapshot of a knowledge graph of AstraZeneca and some of it’s connections, the connections’ connections and how everything is connected. As you can see, we can quite quickly build up a very rich understanding of a companies position in its ecosystem.
How Knowledge graphs can be used to improve investment research flows
As we have laid out above, the challenge today is not how do I go out and get more news, but how do I synthesise the news and data I already have to derive new insights. Next we will look at how they can be used in practice to address this challenge.
As a concrete example, we will look at Auquan’s Portfolio Activity Monitoring (PAM) tool (powered by a knowledge graph) but the same could be achieved by a large team maintaining an in-house solution. The function of the knowledge graph in PAM is to act as an alternative data engine, automatically surfacing hidden links between pieces of information. This allows you to get more information out of the news you have, rather than having to search for more. The most relevant news is identified by evaluating the strength of the connections between the company and news articles. It does by looking at the knowledge graph and evaluating which concepts, and how many of them, the company and the news article have in common.
To make this point more concrete, consider Apple, which has various suppliers, including some in South Korea. Using the links between Apple and their suppliers in South Korea allows us to infer disruption to Apple from changes in government policy or coronavirus responses in South Korea. Mapping the entirety of these relationships can be overwhelmingly complex and not easy to spot manually. Using an automated tool allows for more rapid discovery of thousands of non-obvious connections. From this starting point an analyst could formulate a hypothesis and do further research to prove, or disprove, the trade idea.
Knowledge graphs have a number of attractive features and benefits for investment research. The first is the ability to have a layer of abstraction and unification over information, which makes available the ability to reason from the information to draw new insights. Secondly, the analyst is provided with a form of alternative data, which is key in finding something everyone else has missed. Lastly, using 2nd and 3rd order connections, non-trivial conclusions can be drawn by understanding the consequence of news affecting related entities on the target company. News is no longer a standalone unit of information, but rather a piece of evidence that can be viewed in the context of the current state of a company.
You can check out more details about PAM and other Auquan tools here
