Executive Summary
Company culture is widely recognized to be a key intangible asset, yet few investors haveattempted to formally measure it. We use natural language processing to buildmultidimensional culture profiles for each company. Firms with strong cultures haveoutperformed the stock market, while those with toxic cultures have lagged.
Read as PDF
The Ultimate Intangible
CultureWorship
“Our number one priority is company culture. Our whole belief is that if you get the culture right, most of the other stuff ... will just happen naturally on itsown.”
🥾
“I came to see, in my time at IBM, that culture isn't just one aspect of thegame, it is the game.”
🖥 Lou Gerstner, former CEO of IBM
“Culture eats strategy for breakfast.”
💼 Peter Drucker
The primacy of company culture is a sacred truth. Business leaders from diversebackgrounds glorify culture, ascribing to it their companies’ triumphs and filling ourbookshelves with bestsellers on the subject. Culture forms a central part of the narrative ofsuccessful companies.
There is just one problem. While everyone knows culture is important, nobody knows howto measure it. Culture is a fuzzy concept, whose many facets cannot be reduced to a singlenumber. Moreover, there is no consensus around what exactly is a “good” culture.Amazon’s culture is held up as a paragon of both innovation and brutality.
In the face of these challenges, few researchers have dared attempt to quantify culture and its impact on performance. Most of the evidence we do have is anecdotal or biased, derived from cherry-picked case studies or opt-in surveys with tiny sample sizes. Culture is anorthodoxy, revered by business leaders without empirical question.
MeasureWhatMatters📐
Investors love numbers. In fact, we would argue that they are overly obsessed with metricsthat are easy to quantify. Conversely, this leads them to undervalue crucial intangibleassets simply because they are harder to measure. Culture is one of these overlookedintangibles. In fact, culture may be the ultimate intangible -- beloved by all but valued bynone!
Culture fits perfectly into our research on intangible investing. This paper will show that,like other intangibles, culture can be quantified using machine learning and the resultingmetrics are undervalued by the stock market.
Myers-Briggs for Companies
FoolingNoneofthePeople
The first challenge with measuring culture is that business leaders love nothing more thanpumping their culture. Over 75% of CEOs interviewed for HBR in the past 30 years madesure to flaunt their company’s culture.
Cultural cheap talk has reached such an extreme that it has lost its information content.Sull et al (2020) found that over 80% of company websites touted nice-sounding corporatevalues such as integrity, respect, and innovation. However, they found no correlationbetween stated and actual values.
Netflix points out that Enron went bankrupt from fraud, despite proudly displaying“integrity” in its office lobby.
Exhibit 1
Culture Fail
Source:Netflix.
Corporate values are not imposed top-down by the CEO but are emergent. They are thevalues lived daily by actual rank-and-file employees. According to Netflix, “actual companyvalues, as opposed to nice-sounding values, are shown by who gets rewarded, promoted, or let go. Actual company values are the behaviors and skills that are valued in fellowemployees.”
To examine the values held by rank-and-file employees, we will use Glassdoor. Glassdoorprovides a forum for current and former employees to post reviews of their employers.Over the past decade, millions of reviews have been posted on the website.
Reviewers submit both numerical star ratings (1 to 5) and free-form text. While mostresearch using Glassdoor data focuses on the structured star ratings, we’ll show how theunstructured text can be used to quantify culture.
Exhibit 2
Glassdoor Example
Source:Glassdoor,Sparkline.
While there is no perfect dataset, Chamberlain (2017) shows that Glassdoor suffers lessfrom the polarization afflicting other review sites such as Yelp and Amazon. We are alsohelped by our focus on textual data, which are harder to manipulate than simple starratings.
TheManyFacetsofCulture
The second challenge with measuring culture is that it is multidimensional. Unlike earningsor sales, which operate along a single dimension ($), company culture is expressed in many ways. In other words, quantifying company culture requires mapping each firm into amultidimensional space.
Since we can’t measure every facet of culture, the first order of business is to determinewhich are the most important. We’ll start by reviewing the values most cited by companiesthemselves. While stated values do not necessarily reflect values in practice, at least theycapture aspirations.
Two studies, Guiso et al (2013) and Sull et al (2020), analyzed how often various culturalvalues were posted on company websites. Although performed almost a decade apart, they reached similar results, of which the more recent is below.
Exhibit 3
The Long Tail of Corporate Values
Source:Sull et al (2020),Sparkline.
The long-tailed distribution suggests there are a handful of values that are more universal,such as integrity and collaboration, which each appear in over half the websites. Incontrast, there is a long tail of idiosyncratic values, such as civility and loyalty, each heldby only a few companies.
The management professors O’Reilly, Chatman and Caldwell (1991 and 2012) reached asimilar conclusion by surveying employees directly. They created a list of 54 cultural values(e.g., autonomy, fairness) and had employees sort this list based on both their ownpreference and their opinion of how characteristic each was of their employer.
They then performed principal component analysis (PCA) to cluster correlated values intobroader categories and assign an importance to each. They found that the top seven broadcategories captured 44% of the variance in their data. The resulting Organizational CultureProfile (OCP) has seven factors: innovation, teamwork, results-orientation, integrity,customer-orientation, detail-orientation, and transparency.
Exhibit 4
Organizational Culture Profile
Source:O’Reilly et al (2012),Sparkline.The“innovation”and“teamwork”factorswereoriginallynamed“adaptability”and“collaborative”.
They found the OCP framework to be useful for predicting job satisfaction. Employees withOCP preferences matching the OCP profiles of their employers were happier. They alsofound an association between CEO personality type and company OCP profile. Open-minded CEOs tend to lead innovative firms, conscientious CEOs captain detail-orientedfirms, and disagreeable CEOs helm results-oriented cultures.
BuildingCultureProfiles
This provides a great transition to our next topic. Personality testing has become a vibrantcottage industry in the world of business. Hundreds of online personality tests promise tohelp job seekers find their perfect career. Popular tests include Myers-Briggs (e.g., ENTJ)and Big Five (OCEAN). Not to be outdone, hedge fund personality Ray Dalio, with help fromprofessor Adam Grant, has created an even bigger 12-factor test called PrinciplesYou.
Our goal is to create personality profiles but for companies rather than individuals. We willuse the OCP framework to define the seven dimensions of company “personality.” As withpersonality tests, there are other possible frameworks. We chose the OCP due to it being inthe public domain for three decades and its relatively rigorous underpinnings.
In order to measure a company’s strength on each of the seven OCP dimensions, we utilizethe company embeddings from Investment Management in the Machine Learning Age (June 2019). However, rather than train our embeddings on 10-K business descriptions, we usethe text component of the Glassdoor reviews.
These embeddings collapse the tens of thousands of words in the English language into asmaller number of fifty dimensions. In doing so, the embeddings capture the semanticsimilarity between words based on their distance to each other in embedding space.
Once the embeddings are trained, we can use them to compute the similarity of all wordsto a given seed word. We define seed words for each OCP dimension to create sevenclusters. The TSNE visualization below shows that the OCP factors form mostly distinctclusters. However, we do see that certain values are more similar to some than others. Forexample, transparency is adjacent to integrity and teamwork but less related to detail andcustomer.
Exhibit 5
Culture Embeddings
Source:Glassdoor,Sparkline.Asof7/31/2021.
We ran a search for all words within a certain predefined distance from the seed words. This produced lists of around 40 to 140 words per cluster, from which some examples are drawn below. We included bigrams, trigrams and four-grams to capture common idiomaticpatterns such as “corporate social responsibility” or “open door policy.”
Exhibit 6
OCP Keywords
Source: Glassdoor, Sparkline. As of 7/31/2021.
The topic clusters are extremely intuitive. Innovation is strongly associated with the valuesof creativity and adaptability; integrity is tied to empathy and compassion; detail-orientation is coupled with quality and reliability; and transparency relates to openness and communication.
The next step is to determine each company’s strength on each OCP dimension. Oneapproach is simply to count how frequently a given company’s reviews mention thekeywords for each value (e.g., Li et al (2020)). The key is that we look for not only the seedwords (e.g., innovative) but also the dozens of other cultural values associated with thefactor (e.g., open-mindedness and agility).
A second approach is to simultaneously train our model for words and companies so thatthey coexist in the same embedding space. This enables us to directly compute the vectordistance of each company to the keywords for each OCP factor. We use both approachesand blend the final result into a composite. This allows us to generate a culture score foreach of the seven dimensions for each company.
The next exhibit shows the culture profiles for a few large companies. Nvidia focuses oninnovation, Honeywell excels at details, and Starbucks cares about the customer. Given the nature of their respective businesses, this makes sense.
Exhibit 7
Culture Profile Examples
Source:Glassdoor,Sparkline.Asof7/31/2021.
This suggests that companies mold their culture based on the business they are in. Thenext exhibit shows the average culture profile for each sector in the Russell 1000.
Exhibit 8
Industry and Culture
Source:S&P,Glassdoor,Sparkline.Asof7/31/2021.
Each industry has its own culture. Technology and health care companies tend to haveinnovative cultures. Consumer companies prioritize the customer. Material and industrialfirms have detail-oriented cultures suited to the rigors of industrial production. Of course,these averages may mask significant intra-industry variance. As with personality, culturevaries greatly across companies, even among those in the same industry.
YouCan’tForceCulture
“Many of the traits that make Amazon unusual are now deeply ingrained in theculture. In fact, if I wanted to change them, I couldn't. The cultures are self-reinforcing, and that's a good thing.”
📦 Jeff Bezos, Founder of Amazon
Culture is deeply rooted in organizations. As a result, culture tends to be slow-moving andpersistent. While founders typically establish a firm’s initial culture, once the casting hasset, culture can last for generations. CEOs brought in to turn around a company’s cultureface an uphill battle, which, even if successful, can take many years.
We can measure how company culture evolves over time. Instead of training ourembeddings once on the full sample, we re-train each month using a rolling two-yearwindow based on each review’s timestamp. The next exhibit shows the serial correlation ofOCP culture scores with varying lags.
Exhibit 9
Long Live Culture
Source:Glassdoor,Sparkline.Asof7/31/2021.
Culture decays extremely slowly, maintaining a strong 45% correlation to its original setting even a decade later. However, while culture is very stable on average, it is not predestined. For example, we find that smaller companies tend to have more malleable cultures, asexpected.
Culture is a very long-duration intangible asset. Companies with good cultures enjoy a long-lived advantage, while those with bad cultures can be dogged by this liability for manyyears. Business leaders would be well served to get culture right the first time!
LipService
We now return to our earlier point that investors cannot rely on CEOs to tell us theircompanies’ cultural values. We repeat the previous analysis, but instead of Glassdoorreviews we use earnings call transcripts. These calls, which consist of a presentation bymanagement followed by Q&A, are an important channel companies use to communicatewith the investing public.
Building culture profiles from both employee reviews and earnings calls for the samecompany allows us to contrast what management says is their culture with the beliefs ofrank-and-file employees. The next exhibit shows the correlation between the scoresgenerated from employee reviews and earnings calls for each OCP dimension.
Exhibit 10
Tale of Two Cultures
Source:S&P,Glassdoor,Sparkline.Asof7/31/2021.
With the exception of innovation, we find the correlation is quite weak. This corroboratesthe earlier finding that the values on company websites are mainly for public relationspurposes. Investors need to look beyond the official company line to the values held byactual employees.
Culture and Performance
WhyCultureMatters
“Why is culture so important to a business? Here is a simple way to frame it.The stronger the culture, the less corporate process a company needs. Whenthe culture is strong, you can trust everyone to do the right thing.”
🏡Brian Chesky, CEO of Airbnb
We just showed that culture can indeed be measured. But just because something can bemeasured does not necessarily mean that it matters. In fact, most company characteristics(e.g., logo color) have no material impact on performance. However, there is goodtheoretical reason to believe that culture matters.
As Chesky says, the advantage of strong culture is that it reduces the need for explicitlydefined rules and corporate bloat. Culture constitutes the “first principles” that underpindecision making. Employees can rely on principles such as “bias towards action” or “default to open” to guide the hundreds of micro-decisions required of them each day.
Relatedly, culture helps promote consistency across all parts of an organization. Firms wantto avoid employees or teams rowing in opposite directions, which can happen when theyare not aligned around a common set of principles.
Strong culture is especially valuable in complex, dynamic, and ambiguous environments,where fully monitoring and directing employee behavior is impossible. It is also particularlyvaluable in crises, such as the current pandemic, when there is no time to create a newplaybook.
Another view of culture is that it is a force multiplier on an underlying pool of talent. Aninnovative culture can help get the most out of employees, even those who would nototherwise be natural innovators. Conversely, a rigid culture can stifle innovation, driving acompany’s most creative employees to quit in frustration.
Finally, as we will later discuss, great company culture can improve employee satisfaction.Happy employees are more loyal and motivated and are more likely to recommend theircompany to potential hires and customers.
InnovativeCultures
Innovative cultures are extremely coveted. In a recent survey of over 700 CEOs, “creatingan innovative culture” ranked as the third most important priority. CEOs are increasinglyrecognizing that stale corporate cultures are holding back their employees' innovativepotential.
However, as with culture more generally, we cannot simply rely on firms to tell us if theyare innovative. In A Human View of Disruption (Feb 2021), we tracked investment intechnical human capital (e.g. PhDs, STEM degrees, AI jobs) to provide a “ground truth”indicator of firms truly embracing the digital age. Similarly, we can use employee reviewsto identify firms with truly innovative cultures.
The next exhibit shows how large-cap companies stack up on innovative culture score.High-scoring firms have employees who frequently cite their orientation toward values such as agility and creativity. The most innovative cultures tend to be found in the technologyand health care sectors, while the least innovative cultures are in banking. We also observe a strong correlation between innovative culture score and technical human capital.
Exhibit 11
Innovative Cultures
Source: S&P, Glassdoor, Sparkline. Units are percentages. As of 7/31/2021.
We can run a point-in-time backtest of the long-short performance of the most compared to least innovative firms. Each month, we compute the innovative culture score using datafrom the past two years and calculate the returns for the next month. We also control forindustry membership so that we don’t end up simply long tech and short financials.
Exhibit 12
Innovation Outperformance
Source:S&P,Glassdoor,Sparkline.Bluelineshowstheperformanceoftheinnovativeculturefactorinauniverseconsistingofthetop1000largestU.S.stocksbymarketcap.Redlineshowsthesamebutneutralizesnetexposuretothe11GICSsectors.Weexcludetransactionandfinancingcosts.From12/31/2009to7/31/2021.Seeimportantbacktestdisclosurebelow.
Companies with innovative cultures have outperformed. Interestingly, innovative cultureshave not only won in the technology sector, but in many other industries. This is consistentwith our argument that “all companies are tech companies.” While some sectors leadothers on innovation, disruption is ultimately impacting all parts of the economy.
TheRestoftheSeven
Innovation is the most important dimension of the OCP and is also the most coveted bymanagers and investors today. However, the other six factors together explain four timesthe variance of innovation alone and should not be ignored.
We ran the previous analysis for each of the seven dimensions of culture. We againconsider both market- and industry-neutral versions of the strategy.
Exhibit 13
The Primacy of Culture
Source:S&P,Glassdoor,Sparkline.Bluebarsshowtheperformanceofeachculturefactorinauniverseconsistingofthetop1000largestU.S.stocksbymarketcap.Redbarsarethesamebutneutralizenetexposuretothe11GICSsectors.Weexcludetransactionandfinancingcosts.From12/31/2009to7/31/2021.Seeimportantbacktestdisclosurebelow.
All seven factors have produced outperformance, both market- and sector-neutral. Theresults are strongest for innovation, results-orientation and customer-orientation andweakest for integrity and detail-orientation.
We next examine the correlations between the OCP factors. In practice, they are not fullyindependent. Integrity and transparency are correlated, which perhaps results fromunethical behavior being hard to disguise in transparent organizations. Teamwork andinnovation are also linked, which may reflect the benefit of recombinant innovation indiverse teams.
Exhibit 14
Culture Correlation Matrix
Source:S&P,Glassdoor,Sparkline.Asof7/31/2021.
Finally, we create a combined culture score by blending all seven factors. Given their strong individual performance and low correlation, it should be no surprise that the overall resultsare pretty decent.
An interesting feature of this performance is the huge gain at the start of the Covid-19pandemic. This is not simply driven by the windfall that innovative firms received from thepull-forward of digital adoption. All seven OCP factors had abnormally positive performanceat the onset of the crisis. As mentioned, culture can be an even greater asset in times ofuncertainty and stress.
Exhibit 15
Cultural Strength
Source:S&P,Glassdoor,Sparkline.Bluelineshowstheperformanceofthecompositeculturefactorinauniverseconsistingofthetop1000largestU.S.stocksbymarketcap.Redlineshowsthesamebutneutralizesnetexposuretothe11GICSsectors.Weexcludetransactionandfinancingcosts.From12/31/2009to7/31/2021.Seeimportantbacktestdisclosurebelow.
ContextualCultures
We were a bit surprised to find that all seven OCP factors produced positive results. We had initially expected only a few values to be universally beneficial, with others helpful only incertain contexts. For example, while integrity may be a universal good, stability is likelyonly helpful under certain conditions, as it generally involves a tradeoff with agility.
We believe that our consistently positive results are mainly due to our specific choice ofvalues. The seven OCP values are by construction the most universal cultural values. Recall that they were deliberately chosen from a set of 54 possible options due to being the topseven principal components.
Furthermore, as noted earlier, the OCP values have a very high overlap with theaspirational values stated on company websites. These are the values that companiesthemselves value, or at least want people to associate with them. There is a very goodreason that 65% of companies cite their deep commitment to integrity but only 2% cite the primacy of risk management (and an equivalent 2% even declare allegiance to the quasi-opposing value of risk-taking)!
We can test this theory by creating investible factors for some of the less universally heldcultural values, such as hierarchy, stability, and empowerment. Indeed, we find that allthree dimensions fail to produce excess returns, at least on average across all companies.However, it is possible the average masks some interesting heterogeneity.
We (crudely) split our investment universe into “procedural” and “creative” industries. Wefind that a culture of stability does confer some benefits to firms in procedural industries,but is a liability for creative firms. As a robustness check, we similarly found that stabilityhelps firms in industries with low growth expectations (but hurts high-growth firms).
Exhibit 16
Double-Edged Sword ⚔️
Source:S&P,Glassdoor,Sparkline.Bluelineshowstheperformanceofthestabilityculturefactorinauniverseconsistingofcompaniesinthetop1000U.S.stocksbymarketcapandtheGICSIT,healthcare,andcommunicationssectors.Redlineshowsthesamebutinauniverseconsistingoftheremaining8GICSsectors.Weexcludetransactionandfinancingcosts.From12/31/2009to7/31/2021.Seeimportantbacktestdisclosurebelow.
It stands to reason that all values, even the “universal” ones, are at least somewhatcontextual. Even integrity likely matters more for some businesses than others (e.g.,banking, medicine). Although a topic for further research, we expect we can furtherimprove the performance of our culture factors by taking context into account.
Strong Culture Theory
MasterofNone
“Innovation means saying no to a thousand things.”
📱 Steve Jobs, former CEO of Apple
So far, we have found that strength on all seven dimensions is good, allelseequal.However, what if all else is not equal? Suppose that firms only have a finite number of chips to spend on culture. In this case, one has to prioritize which values to spend their culturalcapital on.
We define “focused” company cultures as those that invest in one or two cultural factors atthe expense of all others. Amazon is obsessed with customer service and innovation, butless concerned with fostering a nurturing work environment. As Jeff Bezos said, “our culture is friendly and intense, but if push comes to shove we'll settle for intense.”
Similarly, Facebook’s motto “move fast and break things” implies an obsession withinnovation at the expense of details (and perhaps integrity). Apple is also focused oninnovation but notoriously non-transparent. Netflix’s culture of “freedom and responsibility” is also very opinionated; they are quite explicit that their culture is not for everyone.
Exhibit 17
Not Right for Everyone
Source:Netflix.
Ray Dalio is similarly unapologetic about his firm’s unique culture. Founded on “radicaltransparency,” Bridgewater is famous for recording workplace conversations and havingemployees rate each other on an app. Dalio explains why 25% of his employees leave intheir first 18 months:
“Some people love knowing about their weaknesses and mistakes and those of others because it helps them be so much better, while others can't stand it. Sowe end up with a lot of people who leave quickly and a lot of people whowouldn't want to work anywhere else.”
Turnover can actually be good if it improves firm-employee alignment. Zappos is anothercustomer-obsessed culture. Its founder Tony Hsieh went so far as to offer his employeesbuyouts after their first month of employment. The idea was to not trap those who felt they were not a good fit for Zappos’ unique culture.
While each of these firms has its own cultural focus, they all display a maniacal devotion totheir focal value above all others. They are not trying to be everything to everyone; theywould rather lose employees than compromise on culture. This is a form of investment. Inthe short run, they pay in employee turnover. However, the employees who do remain arebought in.
These companies also happen to all be wildly successful. But before we assume a linkbetween cultural focus and success, we need to consider the risk of survivorship bias. Thereason these firms’ unique cultures reached our notice in the first place is because of theirsuccess. It is even possible that focused cultures are high-risk / high-reward, with a muchgreater share imploding without our notice.
Thus, we need to create a metric of cultural focus that we can apply consistently and point-in-time across our entire investment universe. The next exhibit rearranges our radar chartsinto columns, allowing us to more clearly visualize the distribution of culture in fourhypothetical organizations.
Exhibit 18
Focused Culture Examples
Source:Sparkline.
Companies A and D have focused cultures; each prioritizes a single salient dimension.Company C is consistently high on all dimensions; while its culture is strong, it is notfocused. Company B has a bland culture; it is both weak and not focused.
While we will spare you the math, we define a battery of statistical metrics to identifydistributions similar to those of Companies A and D. We blend these metrics to create acomposite cultural focus factor, which we backtest below.
Exhibit 19
Focused Cultures
Source:S&P,Glassdoor,Sparkline.Bluelineshowstheperformanceofthefocusedculturefactorinauniverseconsistingofthetop1000largestU.S.stocksbymarketcap.Redlineshowsthesamebutneutralizesnetexposuretothe11GICSsectors.Weexcludetransactionandfinancingcosts.From12/31/2009to7/31/2021.Seeimportantbacktestdisclosurebelow.
We find that firms with focused cultures have outperformed. Netflix and Bridgewater arenot merely outliers. The market has indeed rewarded leaders with a strong vision for theircompany culture and the ability to stick to this vision even if it means not being able toplease everyone.
BreaktheMonoculture
Corporate culture can be extremely bland. Corporate leaders spout values such as integrity so frequently that they have become platitudes. Integrity should really be table stakes.Perhaps more information is revealed by companies that pride themselves on less obvious,more idiosyncratic values.
If nothing else, isn’t it so much more refreshing when Netflix makes opinionated statements like “we are a team, not a family”? Standing out can be good. Apple’s iconic 1984 SuperBowl ad successfully positioned the Mac as our savior from dystopian conformity.
Exhibit 20
Think Different
Source:Apple.
Our next study measures the distance of company cultures to each other in order toidentify interesting outliers. We compare companies both to their peers in the Russell 1000as well as their direct industry competitors.
Rather than calculate distance based on the seven OCP dimensions, we compute it directlyfrom the 50-dimensional company embeddings. By construction, the OCP framework onlyconsiders the seven most universal values, ignoring culture’s long tail. However, the wholepoint of this analysis is to identify firms with idiosyncratic cultures. For example, while theOCP might capture “radical transparency” in its transparency factor, “freedom andresponsibility” does not fit neatly into the framework.
Below are examples of companies that score highly on our measure of cultural uniqueness.In addition to Apple, Amazon and Netflix, companies such as Southwest Airlines, HubSpot,Chipotle, Slack and Costco are well known for their unique and occasionally quirky cultures.
Exhibit 21
Unique Culture Examples
Source:Glassdoor,Sparkline.Asof7/31/2021.
Finally, we can test the hypothesis that companies with unique cultures have outperformed.
Exhibit 22
I Want to Be the Minority
Source:S&P,Glassdoor,Sparkline.Bluelineshowstheperformanceoftheuniqueculturefactorinauniverseconsistingofthetop1000largestU.S.stocksbymarketcap.Weexcludetransactionandfinancingcosts.From12/31/2009to7/31/2021.Seeimportantbacktestdisclosurebelow.
Firms with unique cultures have outperformed. Perhaps cultural uniqueness is itself anintangible moat. Costco’s reputation for treating its workers well has become part of itsbrand, differentiating it from other retailers. Or perhaps the causality is reversed and it ismerely a trapping of success; only leaders who are already successful “earn the right” to be culturally idiosyncratic. In any case, investors should pay special attention to firms withstrong cultures.
Employee Goodwill
EmployeeSatisfaction
“Train people well enough so they can leave. Treat them well enough so theydon’t want to.”
🚀 Richard Branson, Founder of the Virgin Group
Relationships matter in the business world. And one of the most important relationshipsthat a company must manage is its relationship with its labor force. Human capital is notowned by the firm, but walks out the door each day. Firms must work hard to attract andretain top talent. Arguably, talent management is the only job of the modern firm.
Employee goodwill is an intangible asset that enables firms to achieve this goal. Treatingyour employees well builds up a reserve of goodwill, which is paid back in the form ofgreater motivation and loyalty. Furthermore, happy employees are more likely to spread the word, enhancing brand and recruiting efforts.
One way to build up employee goodwill is to pay well. Another way is to provide perks andother benefits. As an aside, accounting treats such human capital outlays as an expenserather than an investment. Unlike tangible capex, investment in intangible assets such asemployee goodwill, intellectual property and brand do not lead to the creation of a balancesheet asset. This inconsistency contributes to the undervaluation of intangibles.
But even if we did capitalize the cost of efficiency wages and dental insurance, we wouldstill be missing a significant component of employee goodwill. Contrary to classicaleconomic models, humans are motivated by more than just money. Intangibles likecompany culture, coworker relationships, and corporate mission play an important role inemployment choice.
MeasuringSatisfaction
Our next goal is to measure employee satisfaction and assess its relationship toperformance. As with culture in general, there are only a handful of rigorous papers on thistopic. First, Edmans (2011) links employee satisfaction to stock market returns by studyingthe winners of the “100 Best Companies to Work For in America” survey. More recently,Green et al (2019) uses Glassdoor reviews to measure employee satisfaction over a muchlarger swath of companies. However, they focus on structured star ratings.
We will instead use NLP on the text of Glassdoor reviews to identify firms with happyemployees. We can leverage the same embeddings we used to create company cultureprofiles. However, rather than use the seed words from the OCP, we create a new set ofseed words to measure key dimensions of employee satisfaction. The next exhibit showsthe keywords derived from our embeddings.
Exhibit 23
Employee Satisfaction Keywords
Source:Glassdoor,Sparkline.Asof7/31/2021.
These keywords are intuitively related to the seed words. For example, trust is related toterms such as “willing to listen” and “walk the talk.” Next, we score each company on eachof the factors above and combine these scores to produce a composite index of employeesatisfaction.
As a validation exercise, we cross-reference these scores against Glassdoor star ratings.The two metrics have a strong but imperfect 60% correlation. Companies such as Slack and Intuitive Surgical score highly on both NLP- and star-based metrics. However, firms such asEversource and Caesars Entertainment score highly on only one metric.
Exhibit 24
Two Satisfaction Metrics
Source:Glassdoor,Sparkline.Asof7/31/2021.
In order to determine which of the two metrics is more relevant, we test the relationship ofeach to subsequent stock performance. Assuming employee satisfaction is an intangibleasset not fully recognized by the market, firms with more satisfied employees shouldoutperform.
Exhibit 25
Pursuit of Happiness 😃
Source:S&P,Glassdoor,Sparkline.BluebarsshowthereturnsoftheNLPsatisfactionandstarratingfactorsinauniverseofthetop1000U.S.stocksbymarketcap.Redbarsarethesamebutneutralizenetexposuretothe11GICSsectors.Weexcludetransactionandfinancingcosts.From12/31/2009to7/31/2021.Seeimportantbacktestdisclosurebelow.
Employee satisfaction does seem to drive firm performance. Both NLP- and star-basedsatisfaction scores provide a useful signal, but NLP metrics appear stronger. This may bebecause NLP allows us to more finely measure the underlying facets driving employeesatisfaction. Or it may simply indicate that the market is already partly pricing in theinformation contained in structured star ratings.
ToxicCultures💀
“After we closed our Series C with Peter Thiel in 2012, we invited him to ouroffice. … I asked him what was the single most important piece of advice hehad for us. He replied, ‘Don’t f*ck up the culture.’”
🏡
So far, we have mainly focused on the benefits of good company culture. However, culturehas a dark underbelly. Just as good cultures can energize employees and engender loyalty,toxic cultures can de-motivate and drive away talent.
Leaders who cultivate an environment of fear, retribution, exclusion, and harassment areplaying with fire. While in the short run this may be tolerated or even help get things donefaster, toxic cultures inevitably create massive problems.
We define five types of toxicity. As before, we provide seed words and use our embeddingsto identify related words. The resulting words are worth examining as they are laden withinteresting idioms such as “dog eat dog,” “good ol’ boys club,” “gaslighting,” “two faced,”“power trip,” “double standard,” “unconscious bias,” and “bro (culture).”
Exhibit 26
Toxic Culture Keywords
Source:Glassdoor,Sparkline.Asof7/31/2021.
We also find that the clusters are quite strongly correlated. In particular, words related toracism and sexism frequently coexist. Sadly, companies that tolerate sexual harassmentare also more likely to enable racist behavior.
We can once again use our trained embeddings to identify companies with our desiredcultural characteristics, in this case toxicity. The next exhibit shows how these toxiccompanies perform compared to the broader stock market.
Exhibit 27
What Goes Around, Comes Around ⚖️
Source:S&P,Glassdoor,Sparkline.Bluelineshowstheperformanceofthetoxicculturefactorinauniverseconsistingofthetop1000largestU.S.stocksbymarketcap.Weexcludetransactionandfinancingcosts.From12/31/2009to7/31/2021.Seeimportantbacktestdisclosurebelow.
Firms with toxic cultures have dramatically underperformed. We also tested a sector-neutral version to remove the impact of some industries being more male-dominated and lessdiverse than others but found similar underperformance. This effect appears to haveaccelerated in recent years, which may be in part due to a business climate that isbecoming increasingly intolerant of bad behavior.
TheLongGame
Culture is an important type of intangible capital. Culture provides a moat around a firm’shuman capital. The rising importance of human capital in the knowledge economy impliesthat this moat is becoming increasingly key.
However, building cultural capital requires significant investment. This goes beyond themere cost of fancy gyms, holiday parties, and happy hours. Every time a company fires ordecides not to hire a “brilliant jerk,” it is trading a short-term tangible for a long-termintangible.
While such choices may in the short run disrupt productivity, they amount to smallinvestments in culture that compound over time. Conversely, ethical compromises may not be immediately noticeable but over time lead to irreversible cultural deterioration.
Culture is the ultimate long game. As a long-duration intangible, cultural investment isoften not only ignored by the market but can even be penalized by investors focused ontangible quarterly results. Building cultural capital requires the utmost conviction andpatience.
Our interest in measuring culture stems from our work on intangible investing.Interestingly, we share this goal with an unrelated group of investors. ESG investors selectfirms on the basis of perceived societal impact. Their work focuses on quantifying socialimpact with metrics such as CO2 intensity and board diversity. Their hope is that thesemetrics provide a yardstick for institutional investors to reward prosocial and punishantisocial corporate behavior.
Company culture is an important part of both the social and governance pillars. As a result,our framework for measuring company culture may be useful for ESG investors as well.Buying companies with great cultures may offer investors the potential to do both well andgood.
Conclusion
Company culture is the ultimate intangible asset. Everyone knows it is important butnobody knows how to measure it. While there is still a lot more work to be done, we hopethis paper provides some early evidence that culture both counts and can be counted.Furthermore, we hope it encourages business leaders to build tools to measure and thusimprove their cultures, to the benefit of both themselves and society.
Disclaimer
Thispaperissolelyforinformationalpurposesandisnotanofferorsolicitationforthepurchaseorsaleofanysecurity,norisittobeconstruedaslegalortaxadvice.Referencestosecuritiesandstrategiesareforillustrativepurposesonlyanddonotconstitutebuyorsellrecommendations.Theinformationinthisreportshouldnotbeusedasthebasisforanyinvestmentdecisions.
Wemakenorepresentationorwarrantyastotheaccuracyorcompletenessoftheinformationcontainedinthisreport,includingthird-partydatasources.Thispapermaycontainforward-lookingstatementsorprojectionsbasedonourcurrentbeliefsandinformationbelievedtobereasonableatthetime.However,suchstatementsnecessarilyinvolveriskanduncertaintyandshouldnotbeusedasthebasisforinvestmentdecisions.Theviewsexpressedareasofthepublicationdateandsubjecttochangeatanytime.
BacktestDisclosure
Theperformanceshownreflectsthesimulatedmodelperformanceaninvestormayhaveobtainedhaditinvestedinthemannershownbutdoesnotrepresentperformancethatanyinvestoractuallyattained.Thisperformanceisnotrepresentativeofanyactualinvestmentstrategyorproductandisprovidedsolelyforinformationalpurposes.
Hypotheticalperformancehasmanysignificantlimitationsandmaynotreflecttheimpactofmaterialeconomicandmarketfactorsiffundswereactuallymanagedinthemannershown.Actualperformancemaydiffersubstantiallyfromsimulatedmodelperformance.Simulatedperformancemaybepreparedwiththebenefitofhindsightandchangesinmethodologymayhaveamaterialimpactonthesimulatedreturnspresented.
Thesimulatedmodelperformanceisadjustedtoreflectthereinvestmentofdividendsandotherincome.Simulationsthatincludeestimatedtransactioncostsassumethepaymentofthehistoricalbid-askspreadand$0.01incommissions.Simulatedfees,expenses,andtransactioncostsdonotrepresentactualcostspaid.
Indexreturnsareshownforinformationalpurposesonlyand/orasabasisofcomparison.Indexesareunmanagedanddonotreflectmanagementortradingfees.Onecannotinvestdirectlyinanindex.TheS&P500isapopulargaugeoflarge-capU.S.equitiesthatincludes500leadingcompanies.TheRussell1000Indexconsistsoftheapproximatelytop1000U.S.stocksbymarketcap.TheRussell1000Value(Growth)IndexincludesthoseRussell1000companieswithlower(higher)price-to-bookratiosandexpectedandhistoricalgrowthrates.
Norepresentationorwarrantyismadeastothereasonablenessofthemethodologyusedorthatallmethodologiesusedinachievingthereturnshavebeenstatedorfullyconsidered.Therecanbenoassurancethatsuchhypotheticalperformanceisachievableinthefuture.Pastperformanceisnoguaranteeoffutureresults.