JOEL GREENBLATT MADE 34 PERCENT A YEAR FOR A DECADE AND THEN SHUT THE FUND DOWN HIMSELF.
Greenblatt ran Gotham Capital with six to eight stocks at a time and averaged roughly 34% a year for ten years. Then he handed every outside investor their money back and stopped taking new capital. Nothing had gone wrong. A portfolio that concentrated could drop 20 to 30 percent in a matter of days, and most people watching their account fall that hard sell at exactly the wrong moment.
He and his partner Rob Goldstein later ran the numbers across the whole fund industry over a full decade. Among managers who finished in the top tier for those ten years, 47 percent spent at least three years sitting in the bottom decile of their category along the way. The eventual winners looked like the biggest losers for stretches long enough to drive most clients out the door, and that's usually when the money actually left.
Even university endowments, sitting on what's supposed to be an infinite time horizon, judge fund managers on three year windows. So Greenblatt stopped building portfolios around his ten best ideas. He built Gotham Index Plus instead: long the cheap stocks inside the S&P 500, short the expensive ones, engineered so the swings stay within a fraction of a percent of the index instead of twenty or thirty. It picked up a five star Morningstar rating and ranked first in its category against more than a thousand competing funds.
The best strategy isn't the one with the highest number attached to it, Greenblatt says. It's the one an ordinary person can actually sit through without flinching.
BRIAN TRACY HAS WRITTEN 82 BOOKS AND SPOKEN IN 80 COUNTRIES.
In the video "Increasing Your Income 1000% Formula" he doesn't show a single secret trick. He shows math.
The idea is called the law of progressive improvement, and it rests on a second law, the law of accumulation. Improve something by 0.1 percent a day, and over ten years that compounds to roughly 140 percent total improvement. The number is tiny, almost invisible in the moment, which is exactly why Tracy calls it "small efforts and sacrifices that no one notices."
Translate that to a monthly pace and 2 percent improvement a month works out to 24 percent a year. Hold 24 percent a year for three straight years and income doubles. Not instantly, not in a quarter, but in exactly the time compounding needs to actually add up.
Then he adds the number people least want to hear. Reaching the top 10 percent of your field takes five to seven years of deliberate practice. Not a weekend motivational clip. Years.
To make the point land, Tracy tells the story of an employee named John who became a billionaire, not through one decision, but through a shift in mindset that then compounded daily, for years.
The 1000 percent in the title isn't a promise of overnight growth. It's what 0.1 percent a day eventually turns into, if you don't stop soon enough.
PETER BERNSTEIN DIED AT 90. NINE BOOKS IN HIS LIFETIME, FIVE OF THEM AFTER HE TURNED SEVENTY FIVE.
But none of that is where it started. Bernstein's grandfather was a leather tanner who also traded in sponges and chamois cloth. His father sold that business in 1929, right at the absolute top of the market, and put the entire proceeds into stocks. Within months the crash wiped out nearly everything.
The son drew the opposite lesson from "never invest again." In 1934 he opened a brokerage, Bernstein-Macaulay, with Frederick Macaulay. In 1941 he worked in the research department of the Federal Reserve, and during the war he was in London under V-bomb attacks.
Then he spent thirteen years managing client money, from 1951 to 1966, and the Dow roughly quintupled on his watch, climbing from around 200 to 1000. In 1967 he sold the firm. What came next was the real test. The market went nowhere for the following sixteen years. By 1976 the Dow had fallen back to 607, the exact same level it sat at in 1958. Twenty years, zero nominal return for anyone who just held stocks and waited.
In September 1958 something else happened that Bernstein himself treated as a historical turning point. For the first time on record, stock dividend yields fell below bond yields. Up to that moment, the assumption had always been that stocks had to pay more, because they carried more risk.
In 1981, with bonds yielding 15 percent, Bernstein told a major foundation to move its money into them.
Each of those numbers reads like ordinary luck or bad luck on its own. Put together, they're the reason a man who watched his father's ruin, then watched the market erase his own sixteen years of gains, then lived through a moment when a century of assumptions about risk simply broke, ended up writing not about how to predict markets, but about why trying to predict them always loses to the risk you didn't see coming.
NORWAY'S ECONOMY PRODUCES ABOUT 150 TIMES MORE PER PERSON THAN BURUNDI'S. MONEY EXPLAINS FAR LESS THAN THAT NUMBER SUGGESTS.
Esther Duflo, an MIT professor and Nobel laureate in economics, opens the first lecture of her Development Economics course not with a story about poverty but with a table of World Bank numbers.
Norway's income per person, adjusted for purchasing power, sits around 123,000 dollars a year. Burundi's sits around 790. That's nearly a hundredfold gap, and it's already adjusted for the fact that prices are lower in poor countries.
Then she puts up a second table, and the whole picture gets messier. Under five child mortality in Norway is 2 per 1,000 live births. The United States, roughly as rich as Norway by income, comes in at 7 per 1,000. Sri Lanka, a country far poorer than the US, also comes in at 7 per 1,000, the identical number. Guatemala, which is richer than Sri Lanka, comes in at 25 per 1,000, several times worse.
Income per person, in other words, barely predicts whether a child survives to age five. Something sits between the money and the outcome, a health system, specific policy choices, specific government decisions.
Duflo ties this to an older argument inside economic theory itself. A well known 1990s paper claimed that around 80 percent of the income gap between countries could be explained by differences in capital and human capital, meaning schooling and skills. Later work cut that number down. Klenow and Rodriguez Clare found closer to half. Caselli found about two thirds. Whatever's left over, economists just label it A, a productivity term nobody can measure directly, and it turns out to matter as much as machines and diplomas do.
The whole course, Duflo says, is an attempt to actually measure what's hiding inside that letter A, instead of writing off the gap between Norway and Burundi as simply a shortage of capital.
PATIENTS WHO FAITHFULLY TOOK A SUGAR PILL ON SCHEDULE LIVED JUST AS LONG AS PATIENTS WHO FAITHFULLY TOOK THE REAL HEART DRUG.
John Gabrieli, a professor in MIT's Brain and Cognitive Sciences department, tells this story in lecture two of Introduction to Psychology, right when he's explaining to students why correlation and cause are not the same thing.
The story is about a clinical trial for a drug meant to lower mortality from coronary artery disease. Patients who took the pill on schedule more than 80 percent of the time had a 15 percent mortality rate over five years. Patients who were less disciplined about it had 25 percent. On its own, that reads like solid proof the drug works.
Except the trial also had a placebo arm, and it produced the exact same split. People who faithfully took the placebo had that same low 15 percent mortality as the faithful drug takers. People who skipped their placebo doses had the same high 25 percent as the sloppy drug takers. What mattered wasn't what was inside the pill. What mattered was whether someone actually stuck to the schedule.
Gabrieli calls this self selection. Someone capable of taking a pill on schedule for years is usually also the kind of person who eats reasonably, exercises, and shows up for checkups. Taking the pill isn't the cause of their health. It's just a marker that this is someone who already takes care of themselves.
He then scales the same problem up to the whole field. By his estimate, a US college student is about four thousand times more likely to show up as a subject in a published psychology study than a random person anywhere else on the planet. A huge share of what gets called "the science of human behavior" is really the science of one narrow group of young, educated Americans.
And the researchers themselves aren't immune either. Gabrieli brings up a case where students were handed ordinary, randomly assigned lab rats and told half of them were bred to be maze bright and half maze dull. The supposedly smart rats ran the maze faster, not because they were actually smarter, but because the student experimenters unconsciously gave them an edge in a dozen tiny ways, how they were placed at the start, how a stopwatch reading got rounded.
Correlation only ever hands you a coincidence. Only a properly built experiment can hand you a cause, and even then the real question is what you actually measured the cause of.
A COMPANY OWNED 40 PERCENT OF A STEEL GIANT, 40 PERCENT OF AN ALUMINUM GIANT, AND A THIRD OF A CEMENT GIANT. THE WHOLE THING WAS PRICED LIKE ONE BOEING 747.
Peter Lynch told this story himself, at the National Press Club in October 1994, four years after he stepped down from running the Fidelity Magellan Fund, where he averaged 29 percent a year for thirteen straight years.
The story was about a company called Kaiser Industries. Lynch bought in at 14 dollars a share, already down from 26. Then it kept falling, all the way to 3. On paper he was sitting on close to an 80 percent loss.
The company wasn't the problem. Kaiser Industries held 40 percent of Kaiser Steel, 40 percent of Kaiser Aluminum, 32 percent of Kaiser Cement, plus a broadcasting arm and a handful of other pieces. The market was pricing the entire stack at 75 million dollars. Lynch's own comparison: that's what a single Boeing 747 cost.
The market wasn't wrong that those stakes existed. It just stopped pricing what they were actually worth while the stock kept sliding.
Lynch held on. Kaiser Industries eventually got broken up and sold off piece by piece, and shareholders walked away with 50 dollars a share, more than triple what he'd paid near the bottom.
In the same speech he admits to the other side of that bet. Over his career he bought roughly thirty long shot stocks, the kind pitched on a rumor or a pending contract, and never broke even on a single one of them.
He also drops a stat almost nobody keeps in mind while buying into a falling stock. Over ninety three years the market dropped 10 percent or more fifty separate times, about once every two years, and 25 percent or worse fifteen times, about once every six years.
That drop from 26 to 3 wasn't an exception to how markets behave. It was the schedule.
PEOPLE WILL WORK FOR HALF THE MONEY IF THEY THINK SOMEONE WILL ACTUALLY SEE WHAT THEY MADE.
Dan Ariely, a behavioral economics professor at Duke, ran a simple experiment on stage at TEDxMidwest that quietly wrecked the standard model of what motivates people.
He had people build Lego Bionicle robots for money, with each one paying a little less than the last. In one group, the finished robot got set on a shelf right next to the builder. In the other, the moment someone finished, the experimenter took it apart in front of them and handed back the same pile of pieces to build again. People in the first group built about eleven robots on average. People in the second, watching their work get undone over and over, built seven.
The gap itself isn't the interesting part. What's interesting is that outside observers, asked to guess the difference beforehand, predicted about one robot. The real effect ran roughly four times bigger than anyone expected.
The second experiment was rougher. People got paid per page for a dull task, finding matching letter pairs, with each page worth a little less than the one before. When the sheet got signed with their name and placed on a stack on the desk, people kept working down to about fifteen cents a page. When the sheet got torn in half right in front of them and dropped in a bin, the price they demanded to keep going roughly doubled. And when the sheet just got set on the pile silently, nobody glancing at it, nobody saying a word about it, the result came out almost as bad as shredding it.
Ignoring someone's work, in other words, demotivates almost as hard as destroying it in front of them.
Ariely sets this against two old models. Adam Smith figured efficiency, breaking work into small repeatable pieces, was progress. Karl Marx warned that the same breakdown alienates a person from what they make. In a knowledge economy, Ariely argues, Marx called it better than Smith did: people need to see meaning in what they're doing far more than most of us are willing to admit out loud.
His own summary of it is short. Everyone understands meaning matters. Everyone just badly underestimates by how much.
A COIN FLIP IS NOT RANDOM.
One MIT professor has been telling students that for twenty five years straight, and so far no physicist in the room has argued back.
According to Robert Gallager the outcome of a coin flip is fully determined: the initial velocity of the hand, the angle, the surface of the coin, the surface it lands on. Nobody's actually tracking all those variables at once, so it's just easier to call the result random and hand it a probability of one half. Convenient model. Deterministic reality underneath that nobody bothers to dig into.
This is lecture one of Discrete Stochastic Processes, the course Gallager has taught at MIT for exactly twenty five years, and the first thing he does is explain why anyone should bother listening at all.
He spots the same trick in data compression, where algorithms treat data as random when to whoever actually wrote it, none of it is random at all. The model holds up exactly until someone looks closer.
And when a model doesn't hold up, the cost can run a lot higher than an academic disagreement. Over roughly the last fifteen years the world's financial system has nearly collapsed more than once, and it wasn't wrecked by idiots, it was wrecked by very bright PhDs, most of them electrical engineers, people who calculate probability beautifully, who do well, then do well enough to start borrowing other people's money and risking it just as casually as their own. Then it all falls apart. Gallager's own words, more or less exactly: the models were no damn good, nothing wrong with the math.
The math doesn't break. The model underneath it does.
The rules meant to save that model showed up in 1933, when the Soviet mathematician Andrey Kolmogorov laid out three axioms across sixty pages: the probability of the whole sample space is one, the probability of any event is at least zero, the probability of a union of disjoint events is the sum of their probabilities. Every bit of probability math currently pricing insurance and running hedge fund risk models traces back to those three lines.
Then Gallager quotes the philosopher Alfred North Whitehead, and this might be the hardest line in the whole lecture: "seek simplicity, and distrust it." Not question it, he says, distrust it, because if you only question a simple model you're still psychologically attached to keeping it and it takes a mountain of evidence before you'll let it go.
He's in his eighties and says so openly to the room: his memory slips, he calls them senior moments, asks people to repeat questions if he loses the thread. He remembers the subject down to the axiom. Himself, not always.
And the best line in the whole lecture: the difference between someone who just solves textbook problems and someone who builds real systems isn't about how much they know. It's that one of them can take a pile of detail and pull out the one or two things that actually matter, and throw away the rest.
17.9 percent a year. Ten years straight. The best performing US stock mutual fund of that entire decade, according to the Wall Street Journal.
Anyone who held it from start to finish turned one dollar into $5.19.
The average investor in that same fund walked away with a completely different number.
Joel Greenblatt tells this story in a talk at Google. Founder of Gotham Capital, a manager whose own strategy beat the market for decades.
He uses this fund as the example of what happens when a strategy works and the people inside it do not.
The best strategy for you is not the one with the highest return. The best strategy for you is the one you can actually stay in.
Not performance. Endurance.
That fund, CGM Focus, returned 17.9 percent a year for ten years. The average investor in it lost 10.8 percent a year over that same decade. Same dollar. Same manager. One turned into $5.19. The other shrank to 32 cents.
The fund was never the problem. People bought in after the gains and sold after the drops, over and over, the whole market behaves this way, not just this fund's clients.
Finance has a name for it. The behavior gap. The distance between what a strategy earns and what an investor actually keeps.
Greenblatt admits the same thing about his own fund. Once every two or three years their strategy wiped out 20 to 30 percent of the capital. That is exactly why he eventually closed the fund to outside money. Outside investors simply could not sit through drawdowns like that.
Whoever understands the behavior gap picks a strategy they can survive psychologically, not the one sitting on top of last year's leaderboard. Whoever does not keeps buying at the peak of someone else's success and selling at the bottom of their own patience.
The talk is free and still sits on Google's channel. The gap between $5.19 and 32 cents never closed.
RABIN PROVED YOUR FEAR OF LOSING $10 SHOULD MATHEMATICALLY MAKE YOU REFUSE INFINITE MONEY.
In an MIT lecture, students meet Johnny. He turns down a 50-50 bet to lose $10 or win $11. Sounds like nothing.
→ If Johnny rejects that bet at every wealth level, expected utility theory forces you to calculate how fast the value of each next dollar collapses
→ At $42 richer, the value of a dollar drops to roughly (10/11)² of where it started
→ At $420 richer, it drops to roughly 3/20
→ At $900 richer, Johnny should value an extra dollar at less than 2 percent of where he started
This isn't about Johnny. It's about a model taught in first year economics breaking down over one small bet.
RABIN PROVED YOUR FEAR OF LOSING $10 SHOULD MATHEMATICALLY MAKE YOU REFUSE INFINITE MONEY.
In an MIT lecture, students meet Johnny. He turns down a 50-50 bet to lose $10 or win $11. Sounds like nothing.
→ If Johnny rejects that bet at every wealth level, expected utility theory forces you to calculate how fast the value of each next dollar collapses
→ At $42 richer, the value of a dollar drops to roughly (10/11)² of where it started
→ At $420 richer, it drops to roughly 3/20
→ At $900 richer, Johnny should value an extra dollar at less than 2 percent of where he started
This isn't about Johnny. It's about a model taught in first year economics breaking down over one small bet.
Take X squared. In ordinary calculus the derivative is 2X. In matrix calculus, that answer is wrong.
Matrices do not commute. X multiplied by an infinitesimal change dX is not the same as dX multiplied by X. So the derivative of X squared splits into two separate terms, not the familiar single 2X. The formula everyone learned in school breaks the moment the variable becomes a matrix.
This is not a footnote for mathematicians. Every neural network computes gradients through exactly this kind of derivative during training, just in dimensions with millions of parameters. Backpropagation is matrix calculus, executed billions of times in a single training run.
The course that breaks this down step by step is called Matrix Calculus for Machine Learning and Beyond, MIT, January 2023. It is taught by Steven Johnson, co-author of the FFTW library, together with Alan Edelman, co-creator of the Julia language.
Here is a direct quote from their syllabus, not a paraphrase: "Today's automatic differentiation is neither symbolic calculus nor finite differences. It is more in the field of the computer science topic of compiler technology than mathematics."
In other words, PyTorch and TensorFlow compute gradients in a way that matches neither what you learned in school nor what most engineers picture when they hear the word derivative.
The course had 40 students. Thirty five percent from computer science, twenty nine percent from math. The instructors warn on the very first page, before a single formula, that googling matrix calculus only shows a small slice of what the topic actually contains.
Someone who just calls .backward() and moves on will never trip over X squared. Someone who understands why 2X fails here understands what is actually happening inside every model they train.
The course sits fully in the open on MIT OpenCourseWare, no paywall anywhere in it.
January 2015. MIT Sloan School of Management. Eleven lectures in four weeks.
150 students signed up for a class where part of the grade comes from playing with real money.
Nothing about this class stayed a hobby. It produced a set of formulas that get carried into rooms where the stakes are no longer chips.
Kevin Desmond teaches it. A Sloan graduate student who paid his own way through poker before MIT.
In lecture three, Basic Strategy, he puts one formula on the board, and it flips the obvious read on every money decision in the room.
You do not need the best hand to take the pot.
Not the cards. The fear.
Bet $150 into a $350 pot. Get the other player to fold 30 times out of 100 and the bet has already paid for itself. Pot times fold rate, minus the bet times the rate people actually call.
The same math runs under every ultimatum in a price negotiation, under every order placed on an exchange not to get filled but to move someone else's behavior.
Poker has a name for it. Fold equity. The value of a bet that lives apart from the two cards in your hand.
Desmond's own notes warn his students about this. Most people who learn the formula still bluff too small, or aim it at opponents who physically cannot fold.
Whoever understands fold equity is betting on the other player's reaction. Whoever does not is paying for every bluff that should have cost nothing.
This is free. The notes and the video have sat in MIT's open archive for eleven years. A seat in that same Sloan classroom now runs $91,892 a year.
January 2015. MIT Sloan School of Management. Eleven lectures in four weeks.
150 students signed up for a class where part of the grade comes from playing with real money.
Nothing about this class stayed a hobby. It produced a set of formulas that get carried into rooms where the stakes are no longer chips.
Kevin Desmond teaches it. A Sloan graduate student who paid his own way through poker before MIT.
In lecture three, Basic Strategy, he puts one formula on the board, and it flips the obvious read on every money decision in the room.
You do not need the best hand to take the pot.
Not the cards. The fear.
Bet $150 into a $350 pot. Get the other player to fold 30 times out of 100 and the bet has already paid for itself. Pot times fold rate, minus the bet times the rate people actually call.
The same math runs under every ultimatum in a price negotiation, under every order placed on an exchange not to get filled but to move someone else's behavior.
Poker has a name for it. Fold equity. The value of a bet that lives apart from the two cards in your hand.
Desmond's own notes warn his students about this. Most people who learn the formula still bluff too small, or aim it at opponents who physically cannot fold.
Whoever understands fold equity is betting on the other player's reaction. Whoever does not is paying for every bluff that should have cost nothing.
This is free. The notes and the video have sat in MIT's open archive for eleven years. A seat in that same Sloan classroom now runs $91,892 a year.
In 2005 this man won a teaching medal. In 2013 he became provost of the entire university. In 2007, before any of that, he recorded a game theory course and gave it away to the whole world for free.
His name is Ben Polak. British, Yale professor. And in lecture 24 of his course he did something most economics teachers never do. He let his own students lose real money on camera, on purpose, to prove a theorem.
Two jars of coins were passed around the room. Students estimated how much was inside, then submitted sealed bids. Highest bid wins the jar and pays exactly what they bid.
The big jar held $2.07. The winning bid was $5.00. The student's name was Ashley. She paid $2.93 more than a jar of loose change was worth.
This was not a math error on her part. It is a law Polak states plainly: the winner of an auction is almost always the person who overestimated the most. Winning itself is the signal that you overpaid.
The term for this was not coined by economists in a lecture hall. It was named in 1971 by three petroleum engineers at Atlantic Richfield, who noticed something strange: companies that won bids on oil fields kept losing money on those exact fields. Winning the tender systematically meant overpaying for it.
The same mechanism later showed up in baseball, where teams with the most optimistic estimate of a player sign the most expensive contracts and regret them most often, and in IPOs, where shares bought on hype tend to fall once they actually start trading.
Polak also hands over the fix, in one line: bid not your estimate, but what your estimate would be if you already knew you had won. That forces you to cut your bid far more than feels natural.
The second jar, students estimated better. It held $1.48. Robert won it at $1.60. A twelve cent gap instead of a three dollar one. A lesson that usually costs a trading career took one repeat of the same experiment here.
Polak spends the rest of the lecture on four auction formats and lands on something counterintuitive: under the right conditions, the format barely matters, the seller earns roughly the same either way. The format is a detail. The selection mechanism for who wins is not.
Trading desks today compress the same idea into one sentence: if you got filled, you probably priced it wrong. A new hire spends months learning that the hard way. Polak proved it to a room of students with one jar of coins.
The lecture has been public since 2007, the last of 24 modules in a course that a future university provost decided to simply give away.
In 2005 this man won a teaching medal. In 2013 he became provost of the entire university. In 2007, before any of that, he recorded a game theory course and gave it away to the whole world for free.
His name is Ben Polak. British, Yale professor. And in lecture 24 of his course he did something most economics teachers never do. He let his own students lose real money on camera, on purpose, to prove a theorem.
Two jars of coins were passed around the room. Students estimated how much was inside, then submitted sealed bids. Highest bid wins the jar and pays exactly what they bid.
The big jar held $2.07. The winning bid was $5.00. The student's name was Ashley. She paid $2.93 more than a jar of loose change was worth.
This was not a math error on her part. It is a law Polak states plainly: the winner of an auction is almost always the person who overestimated the most. Winning itself is the signal that you overpaid.
The term for this was not coined by economists in a lecture hall. It was named in 1971 by three petroleum engineers at Atlantic Richfield, who noticed something strange: companies that won bids on oil fields kept losing money on those exact fields. Winning the tender systematically meant overpaying for it.
The same mechanism later showed up in baseball, where teams with the most optimistic estimate of a player sign the most expensive contracts and regret them most often, and in IPOs, where shares bought on hype tend to fall once they actually start trading.
Polak also hands over the fix, in one line: bid not your estimate, but what your estimate would be if you already knew you had won. That forces you to cut your bid far more than feels natural.
The second jar, students estimated better. It held $1.48. Robert won it at $1.60. A twelve cent gap instead of a three dollar one. A lesson that usually costs a trading career took one repeat of the same experiment here.
Polak spends the rest of the lecture on four auction formats and lands on something counterintuitive: under the right conditions, the format barely matters, the seller earns roughly the same either way. The format is a detail. The selection mechanism for who wins is not.
Trading desks today compress the same idea into one sentence: if you got filled, you probably priced it wrong. A new hire spends months learning that the hard way. Polak proved it to a room of students with one jar of coins.
The lecture has been public since 2007, the last of 24 modules in a course that a future university provost decided to simply give away.
HE IS 28. APPLE HAS NOT FIRED HIM FROM HIS OWN COMPANY YET
In 1983 Steve Jobs walked on stage at the International Design Conference in Aspen. Not a product launch. Not the keynote you have already watched ten times. Just a talk about design, two years before his own board pushes him out of the company he founded.
The recording sits on a channel called Steve Jobs Archive. Eleven thousand subscribers, no ads, no clickbait title, just an archive. The clip has pulled almost four hundred thousand views in a year and still lives in the shadow of every official Apple keynote.
Before NeXT. Before Pixar. Before the 1997 return and the iPhone that turned his name into a synonym for keynote. Here he has not proven anything to anyone yet, and the person on stage is a man, not a brand.
Whoever only watches the official keynotes sees a salesman. Whoever finds footage like this sees the person before he became the legend.