Alex Edmans found that companies on the 100 Best Companies to Work For list beat their peers by 2.3% to 3.8% a year over 28 years. The list is public, it comes out every year, and the effect was still there when other researchers replicated it a decade after he published.
The efficient-market answer is that a published, freely available signal should be arbitraged away. His answer is that people cannot act on information they do not know how to put into a spreadsheet.
"But if I know that employees are happy, how do I change cell C23 in my valuation model?"
Edmans spent five years getting that paper past referees who told him you cannot beat the market, and certainly not with a measure of worker wellbeing; he now teaches finance at London Business School and has written three books on how markets misprice what they cannot see.
The full interview is covered here so you can skip it. 67 minutes of audio, 22 minutes of reading.
Here are the 13 insights that matter.
👤 Guest: Alex Edmans, Professor of Finance at London Business School, whose 2011 study linking employee satisfaction to stock returns is one of the earliest pieces of evidence that a company's most valuable assets are not on its balance sheet, and author of The Madness of Markets, Grow the Pie and May Contain Lies
🎙️ Host: Kai Wu, Founder and Chief Investment Officer of Sparkline Capital, who runs the Intangible Value Fund and does his own research on employee reviews and talent flows
📰 Published: 15 September 2026 on the Excess Returns YouTube channel and the show's own feed
🔴 YouTube | 🟣 Apple Podcasts | ⏱️ 1 hr 7 min | ✅ Time saved: 45 min
Key Takeaways
A public, annually updated list of good employers beat peers by 2.3% to 3.8% a year for 28 years and still worked a decade after publication
His explanation is not secrecy but usability: nobody knows which cell of the model to change
Errors cancel out when nobody cares about the answer, which is the opposite of a financial market
The county-fair crowd guesses an ox's weight well because no one is emotionally attached to an ox
The market overreacts to what is easy to see and underreacts to what is not
A sneaker company that put AI in its name went up 582%; adding .com to a name was worth about 74% in the tech bubble
Published strategies lose about a third of their alpha, not all of it
Across 97 strategies from top journals, two-thirds of the excess return survived publication
He does not think artificial intelligence closes the intangible gap, because the hard part is judgment rather than data
Value investing is not dead, but the denominator has to include brand, talent, innovation and network effects
A dollar in 1995 became $29 by 2021 on that definition, against 13 for the S&P 500
Demographic diversity does not predict returns; inclusion and psychological safety do
Pick your edge before you pick your strategy — knowledge, endurance or independence
1. Emotion on the Trading Floor
Wu opened on the scene the book starts with: Edmans on the Morgan Stanley trading floor in 2004, watching 10 billion euros leave the French market after a football defeat.
"So this showed me that markets are affected by emotions, not just fundamentals." He was doing his PhD at MIT, where the teaching was that markets are rational, dispassionate and efficient, driven by profits, dividends and sales growth incorporated without bias.
His point is that the trading floor should have been the hardest test of that claim — sophisticated investors, billions at stake — and it failed. The European Soccer Championship moved behavior even among American traders, who he said would root for the country their surname came from.
The human detail he gives is one trader who stormed off after a defeat and did not come back for a few days.
The question he took back to MIT sounded absurd: "And so this led me to think, is this just something I experienced at Morgan Stanley during that one summer, or is this something more systematic?" He found it was systematic.
He is explicit that the soccer paper was never a trading strategy. The goal was to establish that markets are inefficient, so that other, more exploitable mistakes could be looked for.
2. Why Errors Don't Cancel
Wu raised the wisdom-of-crowds defense: errors are random, they cancel, and the aggregate is efficient. Edmans said that argument works on an ox and not on a stock.
His summary of the efficient-market position: mistakes are real but they point in different directions. An Elon Musk fan overbuys Tesla, someone who cannot stand his politics oversells, and the price lands near the truth.
The source is James Surowiecki's book and its county-fair example, where random members of the public with no farming expertise averaged out to almost exactly the weight of an ox.
"But why this doesn't apply to financial markets is nobody's really emotionally attached to the weight of an ox." There are no strong biases about oxen.
In markets the biases line up. When England lose, everyone in England is upset at once — which is why he studied international tournaments rather than domestic league results.
His examples of one-directional enthusiasm are the tech bubble, electric vehicles at the start of the 2020s, and, he allowed, possibly artificial intelligence now. Being the lone skeptic in any of those is socially expensive.
"So when there's hype about something, human emotion causes many people to act in the same way, so we actually reinforce each other's biases rather than neutralizing them."
3. Six Months vs Three Years
Wu pointed out that two of the book's three sections look contradictory — overreaction and underreaction — and asked how both can be true at once.
Edmans said this was the other MIT objection: if you cannot tell which is happening, you cannot trade either, so hold the index.
The first thing that separates them is the measurement window. Richard Thaler's work found that on three years of past performance you buy the losers and sell the winners; on six months you buy the winners and sell the losers.
"So in the short term, the market underreacts, and in the long term, the market overreacts." He said both have been documented statistically outside the United States and across asset classes.
The second separator is the kind of information. Salient information gets overreacted to because everyone sees it in the news.
4. Salient Beats Substantive
The examples he used for salience are the sharpest numbers in the interview.
"So when a few months ago, Allbirds rebranded to New Bird AI, the market went up by 582%, even though there was no evidence that a sneaker company would be great at artificial intelligence."
He said the pattern is not new: "So in the tech bubble, just adding .com to your name added about 74% to your stock price on average."
The mirror image is intangible quality, which generates no headline: "So if you have great corporate culture or great brand strength or great network effects, that is not something as press release worthy, and therefore the market may well be ignoring it."
That asymmetry is what he calls the foundation of successful investment strategies — not that the information is secret, but that it is quiet.
5. The Costco Wage Puzzle
Wu asked him to retell the Costco story that opens the intangibles chapter. Edmans used it to explain why he picked employee satisfaction over any other intangible.
He wrote the soccer paper first at MIT and the employee satisfaction paper last, before starting at Wharton.
The first requirement was that the thing be important. Human capital is the most valuable asset in most companies, and most companies say so.
The second requirement is the one that makes it investable, and it is counter-intuitive: being good is not enough, because good and widely known is already priced. Buying the market leader in every sector, he said, is not a strategy.
What made employee satisfaction usable is that a large number of people believed it was bad for shareholders. His own experience at Morgan Stanley is the illustration: a vice president told him not to laugh, because the managing director would conclude he was not being worked hard enough.
The culture he describes treated harshness as a credential. A managing director in the UK was known as The Terrorist for how he treated analysts, a New York counterpart went by Badass Bill and liked the name, and Alex Ferguson's shouting was a management style people admired.
Costco was paying double the wages of Walmart, its closest rival, at the time of the study. For research scientists at a drug company that is easy to justify; for checkout staff, the standard view was that the labor is substitutable.
"There was an analyst quoted in Business Week saying Costco has focused too much on employees to the detriment of shareholders." The assumption underneath it is a fixed pie, where anything paid to workers is taken from shareholders.
The result: "I ran the numbers, and what I found was over a 28-year period, companies with high employee satisfaction, defined by being featured in the 100 Best Companies to Work For in America, delivered shareholder returns that beat their peers by 2.3% to 3.8% per year." That compounds, on his figures, to 89% to 184%.
6. Public, and Still Alpha
The paper came out in 2011, having been first circulated in 2006. Wu asked the obvious question: the list is public and the study is published, so why was the effect not arbitraged away?
Edmans agreed the persistence is strange and contradicts market efficiency. He put it against the size effect, published in 1981, where small-stock funds appeared and the effect faded reasonably quickly.
"Even 10 years later when it was independently replicated, you still have the alpha."
His first explanation is that the belief itself is a bias, and biases are hard to shake. "So some people will view employee-friendly companies as woke and tree-huggy, and employee satisfaction has been caught up in the ESG culture wars." He noted proposals to bar pension funds from considering these factors with public money, on the assumption that doing so must cost returns.
His second is that people who accept the idea often measure it badly, reaching for the share of ethnic minorities on the board, women in the workforce, or the chief-executive-to-worker pay ratio. Those are weaker predictors than the survey of employees that decides who makes the list.
His third is the one he thinks is deepest, and it is about what you do with the number once you have it. An earnings surprise of 5% tells you which line of the model to move. "But if I know that employees are happy, how do I change cell C23 in my valuation model?"
On the publication delay: it took five years of purgatory, because editors and referees held that you cannot beat the market and certainly not on worker wellbeing. He said the skepticism helped, because it forced him to rule out alternative explanations and settle the direction of causality.
Wu's restatement, which Edmans accepted, is that intangibles are both hard to assess and hard to process — two separate problems, either of which is enough.
Edmans generalized it to decisions outside investing. People say they value culture, hours and work-life balance, then take the job with the higher salary because salary is the comparable number, and burn out and leave. His literary version is Pride and Prejudice, where the question was whether Bingley's 10,000 a year or Darcy's 20,000 mattered more than the chemistry between the people.
7. R&D, Patents and Reviews
Wu asked him to widen the lens beyond employees. Edmans started with innovation, and the accounting that hides it.
The mechanical problem is that research and development is expensed rather than capitalized. "So if I was to buy a building, that is visible as a tangible asset. But with many intangibles, you have to expense them, and it just looks like money's gone out the door."
A 2001 paper did the crudest possible fix — add up R&D spending as though it had been capitalized — and buying the heavy spenders beat the market. Doing the same with advertising also beat the market.
The obvious objection is that what matters is effectiveness, not spending, and he says the research has followed. One paper links past R&D expense to future sales growth; that measure of efficiency also predicts outperformance.
Patents work the same way, and the refinements are about quality rather than count: highly cited patents, innovation efficiency, and patent novelty, where interdisciplinary patents predict higher returns than incremental ones in a single field.
What he likes about all of it is that it is common sense rather than statistical pyrotechnics. His counter-example from his own PhD is a paper on the elasticity of returns with respect to durable versus non-durable consumption — something no investor he knows actually considers when allocating money.
On brands, he cited Hamid Boustanifar and Young Dae Kang, the same researchers who replicated his employee study. They avoided branding expenditure because some firms build brands internally and because quality spending hides inside cost of goods sold.
Their two measures were Interbrand's valuation of company brands and the number of times a brand is mentioned in the annual report. The second sounds crude, he said, but a company that cares about its brand will tell shareholders what it is doing to protect it. Both predicted returns.
He added the caveat himself: this is a historical relationship, and if companies start stuffing the word into annual reports it will lose its power.
On customers: "So one paper indeed was looking at 14.5 million Amazon reviews, and buying companies with highly rated products and shorting the one-star disasters would've earned seven to nine percentage points per year over the 12-year period that he studied."
8. AI Erodes a Third
Wu asked whether better data and better tools eventually close the gap, given that the whole edge rests on intangibles being hard to measure.
Edmans conceded part of it: he expects artificial intelligence to erode some of the returns to basic intangible strategies, but not close to all of them.
His evidence is a Journal of Finance best-paper winner that took 97 strategies published in top academic journals and measured what happened afterward. "And what they found was that the alphas did go down, but they only went down by a third."
The first reason two-thirds survived is unflattering: "Number one is that many people actually just don't read academic research." They invest on instinct or on what he called, in quotation marks, lived experience.
The second is that some strategies are hard to run because they fight the investor's own psychology. Momentum is his example: the disposition effect makes people anchor on their purchase price, hold losers hoping for a bounce, and sell winners to book a gain — the exact opposite of what the strategy requires.
He expects the same protection for intangibles, because people still want to believe value sits in bricks and mortar, and because a rebrand to an AI name is more salient than the productivity of the spending behind it.
His route around the erosion is sophistication rather than secrecy, and he used Wu's own work as the example: not the Glassdoor rating, which anyone can pull, but textual analysis of what employees write, looking for specific cultures — innovative, egalitarian — in particular word clusters rather than a bag of words.
His own speculative next step, offered as thinking aloud, is scraping LinkedIn to see where highly qualified people move, and asking whether the flow reflects knowledge about a company's prospects or about its culture.
Wu said he has built that study, using a PageRank algorithm — the one behind Google's early search — to map where talent flows. A company losing people to its main competitor is a bad sign, because those people presumably know something. He also tracks the stock rather than the flow: which companies are dense in AI scientists or machine-learning PhDs, a measure that correlates with the patent data.
Wu's argument for why this edge lasts is dimensionality. There are only so many ways to define price-to-book, and a great many ways to measure talent flow.
Wu also offered two additions Edmans engaged with: a time arbitrage, because intangible spending pays off on a J curve — negative for one to three years while the money goes out, then materially higher earnings per share and return on equity ten years out, which he attributed to Erik Brynjolfsson's work — and a possible risk premium, because failed intellectual property has no salvage value the way an office chair does.
Edmans's answer to both is that anything exploiting a behavioral bias can keep working in an AI world. He added the institutional version: a portfolio manager has to justify a position to an investment committee, and a line about a company's revenues travels better than an account of walking its shop floor and seeing how staff are treated.
9. Go Where Others Won't
Wu asked about what the book calls level-two thinking: not spotting a bias, but working out where it runs deepest.
Edmans's answer is geographic and structural: go where the information is hardest to get. Employee satisfaction is more likely priced in the United States because Glassdoor and the Fortune list cover so much of the market.
In a country where you have to physically visit a retail store to see how staff are treated, the barrier to replication is the edge.
Inside the United States he points at small companies and those with little analyst coverage, on the reasoning that large caps already have equity analysts pulling patent-citation data because the institutional money is there.
"So in general, just go where others can't or won't. Go to the dusty corners of the market where you don't have the large sophisticated investors playing."
His analogy is scouting: look for the baseball player in Japan or Korea rather than the United States, where a language barrier and the effort involved keep him off other people's radar.
Wu put the counter-argument, which Edmans accepted and said he had omitted: sometimes you should go where the intangible matters most rather than where it is least observed. Large-cap technology has heavy analyst coverage but is, in Wu's words, "these companies are 99.9% intangible", while for a regional bank the balance sheet is more or less the whole story.
Edmans's own example of where it matters more is labor-market flexibility. His international extension found employee satisfaction predicts returns more strongly where workers can move between firms easily, because then keeping them happy is a competitive advantage. Innovation, similarly, matters more in technology than at a water utility.
10. Value Needs a New Engine
Wu turned to the book's argument that value investing is not dead but needs a new engine.
The traditional measure of cheapness is price to earnings or market to book, on the assumption that earnings or book value is what a company is really worth. For roughly the last decade, he said, value has significantly underperformed growth, and the Magnificent Seven in particular.
The first repair is accounting: a paper by Baruch Lev and Anup Srivastava capitalizes R&D into book value. That improves the value strategy, he said, but it still lags growth.
The reason it only half works is that some intangibles never appear as a line item at all. Money spent on product quality sits in cost of goods sold, not in a brand-building category, even though a better product builds a stronger brand.
The alternative he describes uses natural language processing and machine learning to estimate four sources of intangible value: innovation, human capital, brand and network effects.
His check on whether the method is a black box is whether the names it produces make sense: Nvidia and Moderna on innovation, Nike and Coca-Cola on brand, Google and Goldman Sachs on talent, Amazon and Uber on network effects.
The backtest is the number he ends on: "So if you invested a dollar in 1995, that would have swelled to $29 by 2021. That's double the S&P 500, which was 13, and a growth index would have got 14."
His framing is that this is still the Buffett strategy — buying a company cheap relative to its full value — with a fuller definition of what full value means. Wu added the biographical version: Buffett was taught Ben Graham's price-to-book approach and spent fifty years evolving toward intangible moats at Coca-Cola and Apple.
11. The End of ESG
Wu raised the paper titled "The End of ESG" and the line that ESG is both extremely important and nothing special.
Edmans said the employee satisfaction paper was never an ESG paper: "The phrase ESG is mentioned zero times in that paper." He chose the variable because human capital is material and mispriceable.
The adoption by the ESG world overreached, in his account: "But I didn't show that ESG pays off. I showed that one specific dimension of ESG does pay off, one where there is a lot of economic logic as to why it matters."
His two counter-examples are a Catholic values screen, where he sees no economic logic for higher returns, and carbon emissions, where he thinks the logic runs the other way — because the damage is an externality and there is no global carbon tax, a company can emit without bearing the cost.
The first reason for the title is that a label makes people think in black and white. One side treats everything under it as woke, which he says may be part of why the employee satisfaction alpha still exists; the other backs everything under it without granularity.
The second is the phrase itself. Extremely important, because a manager whose only goal is long-term client returns should still take these factors seriously — he pointed at Wu's own fund, which is not called ESG and treats this as a source of intangible value.
Nothing special, because these factors get put on a pedestal above other drivers of value: "So, as a fund manager, you are much more likely to be canceled if you're to be investing in a company with high carbon footprint or a low board diversity" than for owning a sluggish, un-innovative company, which he argues is worse for both shareholders and society.
His goal is to widen the lens and narrow it at the same time. Widen it to intangibles nobody files under ESG — his example is Vodafone launching M-PESA, which he says lifted hundreds of thousands of people out of poverty without improving a gender pay gap or a carbon number. Narrow it to factors that are financially material and likely to be mispriced.
On the trade-off he refuses to paper over: polluting companies do earn higher returns on average, and an investor is entitled to refuse that money. "It's not always the case that what's good for society is also good for the shareholder." What he objects to is pretending everything is a win-win.
12. Inclusion, Not Diversity
Wu asked about the related work on diversity, equity and inclusion, where Edmans finds the I predicts performance and the demographic D does not.
His first observation is that the acronym gets collapsed into headcount — the number of women or ethnic minorities on the board or in the workforce — and that measure does not predict long-term returns.
On the studies that say otherwise: "There are some studies by the likes of McKinsey that claim that, but these are very fragile and all the results go away if you just subject it to a little bit of scrutiny."
The economic logic is why he is not surprised. A board with more Asians does not automatically outperform, because reducing a person to their ethnicity throws away the differences that actually complement — big-picture against detail-focused, optimistic against pessimistic, quantitative against qualitative.
He added the market argument on top of the causal one: demographic data is easy to get and everyone is already looking at it, so any edge would be priced away quickly.
What his research and others' finds instead is that psychological safety and inclusion are more strongly linked to long-term returns and company performance.
"So again, what matters is not always what can be measured" — and however good the measures get, some things will stay beyond the machines.
On the gaming question Wu raised — whether companies will simply manage their Glassdoor reviews once everyone is watching — Edmans's answer is that the frontier moves. Spreadsheets did not end active investing; people put different things into the spreadsheet or looked at what could not be quantified. His example of something a computer still cannot read is whether people in a store are willing to disagree with each other, or are just following policy.
13. The Three Edges
The book closes on knowing your edge, split into the knowledge game, the endurance game and the independence game. Wu asked why that comes before any particular strategy.
"I think because if there was one behavioral bias that I was gonna say was more important than any other, it's overconfidence."
His illustration is a doctor who thinks understanding beta blockers makes him an expert on a pharmaceutical stock, while knowing nothing about the drug's pricing, cost of production, advertising spend, regulatory position or competitors. "So a little bit of knowledge is a dangerous thing because we think that we are experts, not realizing that we are competing against people who are studying this every single day."
The knowledge edge has three forms, and he rates the first as unlikely for an amateur: unique insight on a specific stock; knowledge of which investment strategies actually work; or knowledge of a region. For the second, "And this is why for many decades, one of my largest positions was the Parnassus Workplace Fund, which invests in employee-friendly companies." For the third, his example is buying Korea in 2024 on the Corporate Value-Up Program narrowing the Korea discount.
The endurance edge is a structural advantage over professionals, and he sets it in early 2009. A manager buying more into a falling market bleeds against peers, clients withdraw, and the position is liquidated before the thesis pays. An individual with a 35-year horizon has no such constraint.
The independence edge is freedom from a benchmark, and he gave his own portfolio as the example: he cut his technology positions six to nine months ago, not on a market call but because a mortgage comes up for refinancing next year and the gains were there to take. "I have a mortgage coming up for refinancing next year, and the tech portfolio's done pretty well." Benchmarked against other tech investors he would look wrong; benchmarked against the mortgage he does not.
Wu's institutional read on the three: professionals have more knowledge because they have more resources, less endurance depending on structure — a levered multi-manager pod has almost none, a closed-end fund has an infinite amount — and little independence. His view is that herding is pervasive because relative underperformance is the career risk, and that independence is the most underrated advantage an ordinary investor holds.
Bonus Insights
Asked what he believes that his peers do not, Edmans named market efficiency itself: "I believe markets are inefficient, that psychology and emotions drive the market." He says a significant share of finance academics still allow only for small deviations that cancel out.
He engaged with the as-if defense and rejected it. David Beckham can curl a free kick without knowing the physics of ball rotation, so the argument runs that investors can price assets as if they were running the model. Edmans does not think that is plausible even on a broad reading.
Wu's closing observation is that there should be more traffic between academic and practitioner work; Edmans said academics can only write about what practitioners put into effect.
Edmans's aside on data mining is a warning about his own field: run the numbers enough ways and something will predict returns, which is why he prefers measures with obvious economic meaning over statistical constructs.
Edmans's bottom line is that the mispricing of intangible assets is not an information problem that better data will solve, but a behavioral and institutional one — investors overreact to what is easy to see, cannot work out how to put culture or brand into a valuation model, and struggle to defend a position built on either to anyone else.
Products, Companies & Tools Mentioned
Costco and Walmart (The wage comparison behind his study: Costco paid double its closest rival's wages, and analysts called it a transfer from shareholders)
Morgan Stanley (Where he worked in 2004, and the source of both the soccer observation and the story about not laughing on the desk)
Glassdoor (The employee-review source he says a sophisticated investor should read textually rather than by rating)
Interbrand (The brand-valuation firm used in the study that linked brand strength to future returns)
Amazon (Both the source of the 14.5 million customer reviews in one study and one of his network-effect examples)
Nvidia and Moderna (The names his innovation measure surfaces, which he uses as the sense check on the method)
Google and Goldman Sachs (The talent examples)
Uber (Named with Amazon for network effects)
Tesla (His illustration of biases that are supposed to cancel out: the fanboy overbuys, the political opponent oversells)
Vodafone (M-PESA is his example of a company creating enormous social value that no ESG metric captures)
Parnassus Workplace Fund (For many decades one of his largest positions, because employee-friendly companies are its investment thesis)
LinkedIn (Where he speculates an investor could track talent moving between firms)
Sparkline Capital (Wu's firm, whose Intangible Value Fund is the practitioner version of the research)
Books & Resources Mentioned
The Madness of Markets – Alex Edmans (His new book, on why markets overreact, underreact and sometimes react to the wrong thing entirely)
Grow the Pie – Alex Edmans (His first book, and the source of the Vodafone M-PESA example)
May Contain Lies – Alex Edmans (His book on misleading evidence)
Principles of Corporate Finance (The textbook he co-authors)
The Wisdom of Crowds – James Surowiecki (The county-fair ox example he argues does not transfer to markets)
The End of ESG – Alex Edmans (The paper behind the line that ESG is extremely important and nothing special)
Pride and Prejudice – Jane Austen (His illustration that tangible facts crowded out intangible ones long before finance)
Watch the full episode:
If this was worth your time, send it to someone who has to have a view on this.
Get the latest market chatter as it happens: