On Governance is a series of guest blog posts from corporate governance thought leaders. The series, which is curated by the ESG Center research team, is meant to serve to spark discussion on some of the most important corporate governance issues.

At the U.S. federal level cannabis is still classified as a Schedule 1 substance, along with heroin and LSD. At the state level, however, legalization is becoming increasingly popular. Because laws differ widely from state to state, the cannabis industry is relatively fragmented. In October 2018, Canada legalized cannabis at the federal level, sparking further discussion over the future of the industry in the U.S. The Canadian cannabis industry has attracted a lot of attention from investors, sparking high valuations but also attracting naysayers and short-sellers.

As debates about the state of the Canadian cannabis industry continue, we wondered what cannabis companies have to say about themselves. We analyzed the text of public disclosure documents and uncovered a number of surprising results. There are not only significant differences in what cannabis companies say in their public disclosure, as compared to companies in more traditional industries but also differences in how they say it.

These differences may be the result of variance in risk-profiles or uncertainty levels, the usage of promotional language, less-sophisticated drafting or a number of other factors.

We used computational and statistical methods to extract the above insights, in lieu of having a human read and synthesize 105 separate disclosure documents. We used machine learning to discover which attributes of speech were most important in discriminating between a cannabis and non-cannabis company. We analyzed the text of the annual Management Discussion & Analysis (MD&A)[1], a required disclosure document that explains the company's strategy, primary risk factors and analyses changes in financial performance year-over-year.

This piece is part of an ongoing series exploring the application of a variety of data science disciplines to the field of governance. Our first piece in the series explored the CEO letters of Warren Buffett and others.

Risk, uncertainty & conditional verbs

Cannabis companies are operating in a new environment with unproven business models and their disclosure will tell you so. In fact, "material adverse effect" was the most common three-word phrase found in the cannabis MD&As after "Management Discussion & Analysis" and "year-ended December.”

While the word “risk” is much more frequently found in veteran, large-cap disclosures (keep in mind that risk management discussions frequently use the term), uncertainty is reflected in cannabis MD&As in other ways. For instance, the cannabis companies studied used conditional verbs like “may” and “could” far more often than the more established companies, big and small. The median, normalized proportion of conditional verbs in cannabis vs. veteran company disclosures is about 2-1.

Word & sentence structure

Simple linguistic characteristics, such as the average length of words, can distinguish an MD&A published by an established, veteran company from that of a cannabis company. Unsurprisingly, there are more marked linguistic differences between large-cap MD&A disclosures and cannabis disclosures than between small-cap and cannabis disclosures. For instance, the large-cap, veteran companies we studied generally used more varied words and sentences (i.e. their word and sentence lengths had higher standard deviations) than both cannabis and small-cap companies.

Interestingly, cannabis MD&As differed from both large and small-cap, veteran companies in using a significantly higher proportion of “stop words.” (Stop word is a term used in computing; it refers to high-frequency words that do not convey a lot of meaning like “because,” “and,” “the,” “be,” “that,” etc.) Using just the proportion of stop words employed in the MD&A, we were able to predict whether or not the MD&A belonged to a cannabis or a large-cap company with about 85 percent accuracy and distinguish between a cannabis and small-cap company with 75 percent accuracy. The large proportion of stop words in cannabis MD&As may be attributable to the use of promotional language in cannabis disclosure or, perhaps, to less sophisticated drafting. It may also reflect that veteran companies have more complex concepts to disclose and may, therefore, be more economical in their choice of language.

Distinctive vocabulary

It will come as no surprise that cannabis companies use a vocabulary distinct from other industries. For instance, the words “cannabis” and “medical” appeared frequently in the cannabis disclosures and close to never in the disclosures of the 70 veteran companies studied. However, other differentiating words and phrases are less predictable.

We analyzed which words have the most power to distinguish between cannabis and veteran disclosures by looking at a statistic called chi-square (a test that measures how expectations compare to actual observed data). After “cannabis,” the most discriminating term was “company” which appears considerably more frequently in cannabis MD&As. Both large and small-cap cannabis companies are considerably more likely to refer to themselves as “the Company” throughout their MD&A while veteran companies are more likely to use words like “we” or “our.” Could this suggest that the board and management in the cannabis industry identify differently with the business than is the case in more established industries?

Words indicating that the disclosure belongs to a veteran company include “million,” “earnings,” and “tax.”

Compliance & related parties

The cannabis companies studied discussed compliance-related topics less frequently than large-cap, veteran companies, using terms like “risk management,” “internal control,” and “monitor” significantly less frequently, on average. This is interesting in light of the regulatory regime within which cannabis companies operate. The small-cap, veteran companies also mentioned internal controls markedly more frequently than cannabis companies. Cannabis companies used the terms “related party” and “conflict of interest” more frequently than both comparator groups.

Conclusion

Our analysis shows that qualitative aspects of public disclosure can communicate, either intentionally or unintentionally, important information about the business and management's outlook. Most corporate documents are dense and lengthy and therefore, rarely read in full by the company's stakeholders. Computational analysis of corporate documents provides a new opportunity to better understand management's underlying message.

As the cannabis industry grows and becomes more mainstream, we look forward to seeing how their disclosures evolve. Our next piece will explore differences and similarities between cannabis and “dot-com” companies, given the parallels that some have drawn between the industries.

The views presented on the ESG Blog are not the official views of The Conference Board or the ESG Center and are not necessarily endorsed by all members, sponsors, advisors, contributors, staff members, others associated with The Conference Board or the ESG Center.