Although access to “alternative data” was once enough to generate alpha, in today’s digital landscape even the savviest investors are finding it difficult to extract predictive insights from the barrage of “alternative data” that has flooded the market. To find alpha in today’s markets, investors need to identify increasingly complex patterns buried in the digitized, nuanced footprints of human behavior. These patterns are invisible to those relying solely on traditional intelligence and tools.
Cognitive automation of specialized tasks traditionally performed by analysts can help investors gain an edge in the digital age of markets. Machines trained to reason, understand and learn in specialized ways can help investors overcome information overload, make better decisions and generate alpha.
By automating specialized cognitive tasks, investors can accomplish in minutes what would otherwise take an army of analysts weeks or months to achieve. Since capturing market inefficiencies is the key to generating alpha, understanding data more efficiently than competitors, no matter how novel the data, will lead to alpha.
Accrete had the pleasure of participating in the Hedge Fund Association Panel in San Francisco to discuss how specialized AI could help generate alpha in the digital age. Andrés Diana, Head of Product at Accrete, participated on a panel with Joe Rothermich, Sr. Director of Quant Research and Data Science at Thomson Reuters Labs and Daniel Ahn, Chief Economist and Head of Data Analytics at US Department of State.
1. What are some of the main ways that machine learning and alternative data have impacted the way investors make decisions?
Investors are drowning in information. The problem faced by investors these days is that the edge isn’t in access to alternative data. Rather, the edge is in understanding vast volumes of unstructured, alternative data at scale. In fact it’s impossible for human investors to continuously ingest and process all relevant digital information. It’s equally futile to discern the myriad of factors that could affect asset prices using biological intelligence.
Savvy investors are leveraging specialized AI to extract differentiated insights from ostensibly commoditized and mundane data. The reconciliation of the innately human ability to spot semantic nuance together with the machine’s ability to quickly identify statistical patterns and learn yields results that in effect scale domain specific human expertise. Specialized AI can give traditional human analysts super-intelligence and help them generate alpha in novel ways.
2. If someone is considering utilizing ML/Alt Data in their investment process, what are some of the most important factors for them to consider?
This depends on the type of investing. Having a robust understanding of the specific problem you’re trying to solve is very important. Your first step should be to formulate a clear hypotheses to test. An alternative approach may be to explore the data looking for patterns, but such an approach may lead to spurious insights.
It takes humans extensive periods of time, education and experience to gain specialized knowledge. Machines, on the other hand, struggle with generalization, not specialization. Accordingly, to create ‘super-intelligence’, it’s important for investors to realize there is no ‘super-intelligence’ without human expertise.
3. How will machine learning, AI, and alternative data impact the future of investment research? Can hedge funds stay competitive without integrating them into their investment strategy?
ML, AI and alternative data will enable investors to base decisions on data coming from ground truths vs derivative values (such as fundamental data) which are only compiled periodically and only summarize the internal workings of companies. Using high-fidelity, near real time feeds of alternative and unstructured data refined through ML and AI researchers will spot opportunities, risks, and important economic shifts much sooner and more accurately than competitors using traditional methods. The time and information advantage between funds utilizing these techniques and those that don’t will grow more apparent over time, ultimately making these tools an integral part of every competitive hedge fund’s research process.
4. Where does Accrete find alpha in unstructured data?
Accrete has found alpha in various domains with a primary focus on searching through public, widely available unstructured data for untapped sources of alpha. We’ve proven that by using specialized AI to monitor M&A chatter for actionable rumors, our Rumor Hound tool can source M&A rumors of increasing linguistic complexity that yield excess returns and also precede deal announcements with lead times as long as 245 days and growing. In fact, as Rumor Hound learns, it’s learning how to spot ‘super stars’ that post in lesser known chat rooms and forums but are highly accurate in terms of predicting deals before major news sources report on them.
We’ve also found that our specialized AI tuned to understand industry-specific topical nuances in earnings transcripts generates novel, feature rich data with high alpha content that could be used to create stand alone high Sharpe strategies or be used to augment Sharpe ratios of existing models.
5. How can investors utilize AI to augment their knowledge rather than replace it completely?
AI is a tool like no other. It helps investors transcend biological limitations to improve awareness, understanding, and synthesis of complex information at scale. By utilizing specialized AI that is transparent in reasoning, investors can combine machine derived insights with their own expertise and intuition. The end outcome will be better decisions that correct investor bias and the ability to take an objective view derived from a much broader spectrum of information.
It’s important to understand that specialized AI continuously learns- even when you are not. As such, tools like Rumor Hound and Topic Deltas actually provide a non-linear return on cognitive effort. Additionally, the insights derived by such tools compound to surface higher order patterns that would be impossible for even an army of skilled analysts to detect.
6. How do you use specific AI to create factors to backtest to keep ahead of the competition?
Utilizing backtests and historical benchmarks helps us determine the true value of our cognitive tools. Ultimately accurate predictive insight is the key differentiator between hyped up AI and true AI driven value. At Accrete, we keep on top of the competition by creating specialized AI tools that generate alpha and the reproducible research and attribution to back-up our findings. Because our tools produce predictive insights, we are able to pass that value on to our customers, including banks and hedge funds, so that they can process information more efficiently.
For example, if you are a quant, you can use the data produced by our specialized AI tool Topic Deltas to construct strategies with high Sharpe ratios. If you are a banker, you could use our Rumor Hound specialized AI tool to identify and contain deal leaks (leaked deals command higher premiums which is bad for banks). Bankers could also use Rumor Hound by generating leads before the competition is even aware a potential target is in play.
We first generate hypotheses about the the impact of the historical data upon company fundamentals. We then express the hypotheses using the features extracted from the historical data. These features are the dependent variables which can be used to build predictive models using machine learning techniques. When features are hidden and not obvious, we use deep learning system and to uncover these features. The dependent output is then predicted to accept or reject the hypothesis.
7. What is an example of a problem that cannot be solved using AI?
Since there is no such thing as general artificial intelligence at the moment, any problem that in effect tries to “boil the ocean” can’t be solved using “AI”.
An example of that would be trying to analyze the sentiment of earnings transcripts that simply measures the statistical distribution of key words to determine if management is positive or not. If an industry analyst analyzed the sentiment mappings of such an approach, the outputs would likely not endure expert validation nor would it endure rigorous backtesting.
The problem is that the language contained in earnings transcripts and analyst notes contain hidden nuances that would be difficult to detect using statistical methods. Unless the machine truly understands the industry and subject matter specific context like a human, there is no way to measure the sentiment of management of analysts accurately.
In order to train the machine how to reason, understand and learn like a domain expert, actual human experts with industry specific knowledge must seed the AI with specialized, topical context. The AI, comprised of a combination proprietary, patent pending deep learning and NLP methods, would then scale the human expertise and produce outputs that human experts could validate semantically. As the AI learns from mistakes pointed out by the human expert, accuracy of semantically derived linguistic classification and sentiment improves and human interaction wanes. Accordingly, rigorous backtesting will likely produce predictive insights and alpha because those using either human or machine to analyze management sentiment in earnings transcripts will not be able to properly account for semantic nuances buried in the language of earnings transcripts.
To learn more about how AI and unstructured data are reshaping the investment landscape, visit us at www.accrete.ai.
Attending SIBOS 2018?
Our senior leadership will be in attendance and sharing their insights on how financial enterprises can automate specific cognitive tasks performed by analyst's achieve dramatic results. Leveraging Accrete's platform, financial professionals are spending less time searching for information and more time to make well-informed decisions. We invite you to come speak with us or book a meeting.