Using heuristics and Symbolic Artificial Intelligence to solve a difficult problem in financial analysis

Published Date: 2019/10/23

Prepared by Sing Koo

The problem:

As much as quantitative analytics has been the popular approach to financial analysis, it falls short in providing answers to financial conditions as a result of new and unknown variables. For example, the climate change, the trade war effect on supply chain, negative interest rate, forever QE, evolution of digital payment, block chains, ecommerce free shipping, online mega merchants etc.; businesses respond to competition by devising new business models. Statistical models and stochastic models cannot function when there is a vacuum in critical metrics.

These missing financial metrics are actually available in the form of natural language discussions during business earnings conference call.

The following is an example used to illustrate the difficulties in analyzing earnings conference calls:

In an earnings conference call conducted by a biopharmaceutical company, the following was said:

“I'll highlight a few subtle differences in the patient dynamics specific to atypical aHUS. First, duration of therapy varies depending on etiology and is in general shorter than that in PNH given the silent nature of the disease between TMA events.”

How would a quantitative analysis be used to analyze the above excerpt?

The above excerpt is likely outside the reach of quantitative analytics.

It is also important to point out that Sentiment Analytics will equally fail to analyze this statement. The importance of analysis in this case is not the quantity or the sentiment, but the qualitative aspect of facts.

It is important for analytics tools to decode this semantics, and to make assessments with regard to risk and reward. For this reason, Artificial Intelligence based on heuristics and symbolic logic is used to understand the context and semantics without the benefit of knowing the domain knowledge of biopharmaceutical ahead of time.

The Solution:

To demonstrate the work of a system built with heuristics and symbolic logic, we turn to ELAINE. ELAINE stands for “English Language Artificial Intelligence NLU Enablement”. It is a project sponsored by SiteFocus since 2016 using many of the principles that is not found in today’s Open Source frameworks. It is the main engine that drives the Earnings Intelligence SaaS – an earnings call analyzer.

Result from Earnings Intelligence analysis:

The followings are related High Level Abstractions of type momentum:

  • (ahus*tma*ultomiris)
  • (ahus*pnh*tma*ultomiris)
  • (ahus*pnh*tma)
  • (atypical*pnh*tma)
  • (ahus*atypical*pnh*tma)
  • (ahus*pnh*tma*ultomiris) [inferred]
  • (ahus*atypical*pnh*soliris) [inferred]
  • (ahus*atypical*pnh*ultomiris) [inferred]
  • (ahus*pnh*tma) [inferred]
  • (atypical*pnh*tma) [inferred]
  • (pnh*tma*ultomiris) [inferred]
  • With this high level abstraction, it brings out the relationship between subjects and concepts discussed in the conference call. Additional information not shown in this article is the reference to annotations that help the user to get the details.

    This article draws references from Earnings Intelligence, a service provided by SiteFocus