Discover more from Lucky Maverick
The Time I Sold Furbies for Money
Or, more appropriately, why you should gamble to get better at estimating probabilities
I used to buy and sell Furbies for money.
For a few months in the late 90s, all I did was trade these furry little creatures on eBay. If you don’t know what a Furby is, it’s this hamster-looking thing:
Furbies were a collectible/toy kind of like Beanie Babies—except much cooler and not at all creepy if you ask me—and even though I was just a kid, I started a business trading them. The Furby stock market was hot back then man. Anyone who was anyone was involved. I got in after my grandpa—the same one who gave me the advice to “buy all the glass you can get your hands on, trust me”—opened a little trading desk himself.
I started slowly, and by that I mean I stayed up all day and night charting every sale to see which traits were most valuable. This was before Billy Beane brought his data-driven approach to baseball, and while I’m not saying my “Moneyball for Furbies” was the catalyst for his work, I’m not saying it wasn’t, either.
“The Oakland A’s have clinched the playoffs! We’re live with the most innovative GM of our time, Billy Beane…Billy, how did you come up with such a genius, outside-the-box foundation for building a championship baseball team?”
“Well, it all started when I was looking to purchase a Generation 1 Leopard Furby on eBay and noticed an intriguing analytical strategy being employed by a young child.”
Anyway, after a string of incredible trades in which I uncovered the most underpriced Furbies by targeting those with a high on-base percentage, I went busto. And not only did I lose all my money, but since I was buying Furbies in bulk and you didn’t need to actually pay right away after winning an auction, I went into debt.
Imagine this. The internet is basically new and I’m a child buying and selling ugly-ass furry owls in an online marketplace. I mean I didn’t even like these things. Look at them.
I didn’t tell my mom about the whole enterprise because not only was it illegal, but even worse, she would have yelled at me and made me stop. I had to hide these things all over my room—under my bed, in the closet. Imagine waking up in the middle of the night and seeing a line of Furbies staring at your half-asleep ass.
Think about the scene when I had to confess to my mom that the family was now in debt because I couldn’t pay for eBay auctions I won, then take her into my room to show her what we lost money on. I say “we” because this was our problem now.
So here’s what happened: I fucked up. Obviously I fucked up. I was so focused on identifying the value of each trait—more or less calculating the EV of each Furby—that I was overlooking the probability of the success for each trade, especially when buying in bulk. I had no downside protection—an unimaginably embarrassing oversight, even at that age.
What kind of fucking idiot assigns a static median projection to Furby value with no concern for the probability of profit—its range of outcomes—and the correlations between each Furby in his collection? My mom was left to pick up the pieces.
I learned a painful-but-valuable lesson that day, which was that as long as you have a safety net, you don’t need to be smart at all because someone else will be left to deal with the consequences. Just look at our entire economy.
But the real thing I learned was that you can’t think about EV in a static way. From a player projection in a sporting event to a company projection in business, the future must be represented by a probability distribution.
The inevitable ramification: your ability to assign probabilities to various outcomes is unimaginably important to your success in almost every area of life. You don’t need to build sophisticated models to do this (you mostly just need to continually refine your gut instincts about the chances of something happening), but if you cannot consistently determine if something is around 1% to happen versus 10% versus a coin flip, you’re never going to make it as a Furby salesman. There’s just no way.
These days, I’d never buy and sell such a ridiculous collectible…because everything is digital now. And you better believe I’m projecting the probabilities for different scenarios when buying and selling pixelated images or free highlight videos.
One funny side story: when he introduced me to this wonderful world, my grandpa told me that Furbies—which talked to you with canned lines, as if they weren’t creepy enough—utilized a really advanced technology and could see you and pick up things you were saying and repeat them back to you.
So I bought some of them for myself, taking them out of the box to talk to them. This was a giant mistake, by the way. Totally cut into my profits. Never get high on your own supply.
But I had a few in my room and every time I walked in, I’d say “Hello Jonathan.” Imagine what was going through my mom’s mind at that time as her child walked into his room 100 times a day and repeated “Hello Jonathan” like he had an imaginary pet parrot or something.
And that’s actually how she found out about the whole deal. She went into my room to clean one day and noticed a Generation 7 Red Wolf in the corner.
Startled, she fell to the ground as the furry toy opened his mechanical eyelids and began to speak, malfunctioning as he repeated what he’d heard so many times from his owner:
“Hello Jonathan. Hello Jonathan. You are a cute little guy aren’t you? Hello Jonathan. Oh no. I’m so screwed. How am I going to tell my mom I lost all our money on these big-eyed fucks?”
Estimating the Percentages: Start with a Baseline + Adjust
Most of you are likely familiar with Bill Belichick’s famous fourth-down decision in a 2009 game against the Colts. The Patriots were leading 34-28 with two minutes left to play and faced a 4th-and-2 at their own 28-yard line. Pretty much every other coach in the NFL would have punted the ball away in this situation, but Belichick (correctly) chose to go for it. New England was stopped short and the Patriots lost the game.
So how do we know that Belichick made the right decision even though it didn’t work? Well, there’s a whole lot of NFL data on breakeven percentages on fourth down in a variety of game situations. As a whole, coaches are still way too conservative on fourth down, even after a recent uptick in aggressiveness due to more widespread belief in the power of analytics.
Based on leaguewide data, an average NFL offense in an otherwise “neutral” game situation—say, tied the first quarter—should go for it on 4th-and-2 from their own 28-yard line. They should also go for it on 4th-and-3, and it becomes close on 4th-and-4.
That’s just overarching data, though. The Patriots scored the sixth-most points in the league that season. They were also facing maybe the greatest quarterback in NFL history in Peyton Manning, who had led the Colts to two touchdowns in the team’s prior three possessions. Indianapolis also had three timeouts, meaning there was plenty of time remaining for Manning to methodically move the Colts down the field had New England punted. If the Patriots had converted on fourth down, they would have either won the game with another first down or forced the Colts to use all three timeouts before getting the ball back with maybe 90 seconds to play.
On top of all that, one of the perhaps overlooked aspects of this decision was that a Colts touchdown didn’t necessitate a Patriots loss; if they scored “too quickly” from the New England 28-yard line, the Pats would get the ball back with time left to drive for a game-winning field goal.
You can bet Belichick knew the baseline percentages and understood all the game-specific factors that went into making the correct decision. And he made the right call—it wasn’t even close, despite all the backlash he received—with the actual result being effectively meaningless in dictating the merits of his choice.
In most situations, you should start your estimate of the percentages with any baseline data you have. I began my estimate of the Patriots’ fourth-down call with baseline go-for-it data, but Belichick likely used team-adjusted data, i.e. they should go for it more often on fourth down than the typical NFL offense. That baseline can then be modified with consideration given to all the major factors that would affect the quality of the decision.
Many times, this process doesn’t need to be incredibly rigorous. In the fourth-down example, our “all else being equal” starting point was to go for it, and with the specifics of the situation also leaning toward going for it, the choice very quickly became an easy one. Belichick didn’t have time to crunch the numbers to come up with a very specific forecast of the offense’s probability of converting the fourth down, but he didn’t need to; with a close enough starting point for a baseline probability and with the right questions being asked as to what could significantly alter that percentage, you usually don’t need highly accurate projections in order to make the right call.
In the fourth down-example, we’re aided by the plethora of NFL fourth down data. In many other areas, we can’t have the same degree of confidence in the baseline projection.
Let’s say you’re flipping a quarter for money because I mean come on you’re a degenerate, why not? Unless the coin isn’t legit, you can act with total confidence that the baseline for heads/tails is 50%; you should take any bet that’s better than even money on the flip of a fair coin. If the quarter comes up heads 10 straight times—which will actually happen one out of every 1,024 series of 10 flips—it means next to nothing because you have near 100% confidence it’s just variance.
I say “near 100%” because there’s always a chance the coin is weighted toward heads. Let’s say that rather than you supplying the coin, which would truly mean a near 100% chance of it being fair, the quarter and subsequent bets on it were offered by someone else you know to be a shady degenerate gambler with money problems. Now all of a sudden losing 10 straight bets on tails is a different story, right?
This line of reasoning is known as Bayes’ theorem, which is basically just the chances of an event occurring based on prior things known that might affect the probability.
So in the coin-flip example, the odds of the next five flips being heads if you supplied the quarter are 1-in-32, whereas we can’t just rely on that statistical baseline anymore for a crooked gambler who just took money off of you on 10 consecutive heads flips.
One of my favorite examples of the sometimes counterintuitive nature of Bayesian inference is with testing. Let’s say there’s a screen for a disease that correctly identifies those who have a disease 99% of the time, but also comes with a false positive rate of 1%. What are the chances you’d test positive for this disease but not have it?
Most people’s first guess is 1%. The real answer: you couldn’t possibly have any idea unless you know the overall rate of infection. As an example, let’s say just 1-in-10,000 people in the world has this disease, and you have no reason to believe you’re any more or less likely than anyone else to be infected. In this case, if a random sample of 10,000 people take the test, just one (on average) will be infected. But with a false positive rate of 1%, the test will identify 100 people as having the disease who actually do not. Thus, for a test that has a 1% false positive rate for a rare 1-in-10,000 disease, your odds of testing positive and actually possessing the disease would actually be 1-in-101, or just under 1%.
Now, let’s say you test positive, take the test again, and test positive again. Now what’s the likelihood you’re infected? Well, you have to revise your “prior” estimate of infection probability from 1-in-10,000 to 1-in-101. Now, a 1% false positive rate would happen just about one time in such a sample, meaning your odds of having the disease with a second positive test are right around a coin flip.
The Easiest Way to Become a Better Probabilistic Thinker
Sometimes, you can use data as a baseline for your probability estimates. If you were to independently forecast the odds of rain next week, where would you begin? A decent starting point might be how often it rains this time of year in the location you’re predicting.
Mostly, though, your probability estimates are going to be glorified guesses based on some combination of data and experience. In most areas that offer an edge, the data you use might actually be mainly anecdotal and experiential—mostly the result of pattern recognition. The ability to recognize patterns is enhanced via smart game play.
And the best way to make better probabilistic decisions—to improve your predictions—is to do it. You can study all the probability theory in the world, but the hallmark of true understanding, in my opinion, is the ability to do. People ask me all the time to recommend the best book on game theory, and I always tell them that the best teacher is to simply play games.
I’ve found the top gamblers, traders, entrepreneurs, and the like to have the keenest understanding of probability (and, thus, make the best predictions). Most of these people probably wouldn’t even be able to teach you what they know, but their understanding is pure, and that’s evident because they win. Unlike a statistics professor, who might know how to calculate all the odds of all kinds of hypotheticals but couldn’t win games against toddlers or gamble his/her way out of a paper bag, those who actually do don’t need to say anything to substantiate their competence; the proof is in the pudding.
You can know all the numbers in the world, but becoming an exceptional probability-based decision-maker in the real world is different from being one in theory; the economist knows all the financial data, but the entrepreneur knows how to make the money.
This should make intuitive sense, as gamblers, entrepreneurs, traders and anyone else who deals with risk and doesn’t quickly figure out how to win will die off; if you’re trying to become a world-class poker player, it makes sense to study the strategies of top poker players because they couldn’t possibly be there unless what they were doing actually worked. Anyone can try to tell you how you should play poker; few can actually pull it off at the highest level.
I’ve discussed the survivorship bias of those with downside in a previous Lucky Maverick post:
The primary reason I think it’s important to learn from those with skin in the game isn’t because it’s honorable to take on risk (although it is) or because doing > saying (although it is) or because I’m biased toward placing downside on beliefs (although I am). The main reason is simply because looking to those with skin in the game is a useful heuristic to finding what works—a quick way to get closer toward the “truth”—as there’s a selection bias in terms of who you’re analyzing.
As Nassim Nicholas Taleb proposes in Skin in the Game, those most worthy of our trust and praise—and our attention if trying to learn—are those who take on real risk; they’re forced to learn—and learn quickly—because they have downside. If they don’t figure it out, they go busto. And if someone doesn’t have downside from their opinions or actions—if they aren’t harmed when they’re wrong—then they’re probably full of shit. Not always. But probably.
Thus, my proposal for how to become the best possible probabilistic thinker in the fastest possible time is simple: play games, take on downside, and learn from others doing the same at the highest level.
In short: gamble.
This might seem like the most degenerate fucking thing of all-time—saying that if you want to succeed in life that you need to start gambling—and maybe it is. But I mean it.
This doesn’t mean you should be frequenting the nearest casino and dusting off your savings. But it does mean you should play strategic games—play poker or chess or dice with your friends—and have some sort of repercussions if you lose.
It doesn’t need to be money-related downside. The way I learned to effectively play Liar’s Dice was by competing with a group of friends in restaurants and bars, with the loser being forced to down shots and do the dumbest dares you can possibly think of.
Interested in traveling at warp speed from happy to miserable? Lose a bet with the consequence being you need to go hit on a group of attractive women (or men, or cattle, or whatever it is that you’re into), but only after you literally somersault into their feet. Or you need to talk to them from a perpetual squatting position. Or with six marshmallows in your mouth. Or after telling them you have to pee and asking where the bathroom is located, except you can’t actually speak and need to use only hand motions to do it. Because I’ve needed to do all of those things, and contrary to what you might expect, ladies were not at all throwing themselves at me. In fact, none of these things worked at all!
But I’ll tell ya what…I learned how to win at Liar’s Dice real fucking fast man. I figured out all the percentages. I knew all the data based on the number of players. I could adjust in real-time given the strategies of new players or when I identified a shift from a common opponent. I learned the optimal numbers to nominate and when it was optimal to lie big. I knew which players to sit next to. I’d suggest new game variations like more dice or more wild numbers or whatever, which necessitated learning new stats and probabilities that I had already memorized.
I mean when I put this in writing, I kind of sound like a psychopath.
Nonetheless, I went from a newb to a Liar’s Dice hustler in no time, and my prize was exactly what I wanted: never being forced to speak to a real, live woman out in the wild ever again. I was kicking my friends’ asses and watching them act like total morons, laughing and sticking to what I knew best: picking up girls in the real world (online).
One of the most important parts of winning is picking the right games to play. If the game isn’t suited for you to win, change the rules. I don’t mean in an unethical way; I mean change the hidden rules that govern success (much of these come before the game begins, such as picking the right person to sit next to—huge edge).
Life is best approached as one big game—the skills you learn to become an elite game-player translate very well to areas of life most don’t think of as a game—and the most effective way to become outstanding at games is 1) to play them with purpose and 2) in lieu of participation, study only those who are already winning—not those who teach how to win.
If you’re going to trust anything I say—and you should be skeptical about all of it—trust me not because I’m writing about how I’ve approached games and life, but because I actually did it in a way that’s (perhaps) useful to you. Much of it will be wrong, so cultivate the skill of learning what to believe.
Playing Games with a Purpose
Your goal in any game is to win. That’s it. We can seek to increase EV or win probability as a means of getting there, but ultimately you just want to win the fucking game, right?
Part of understanding probabilities to improve at games is properly internalizing what happens to you. Every decision you make should be accompanied by a specific reason for doing it. Having a bad reason for doing something is better than no reason; a bad reason can be continually improved until it becomes good reasoning. Make a hypothesis, test it, internalize the feedback you receive, and gradually chip away at your beliefs until they’re less wrong than before.
No matter what happens, you must learn from it. I always loved this excerpt from Louie Helm on how some top poker players learn the game:
When (Phil) Laak gets his money in behind, instead of moaning about his bad luck, he often says things like, “Wow. This is what 5% feels like,” pointing out that he still has a 5% chance to win the hand. Why would he do this? Is he trying to stay positive and upbeat in the face of long odds? Maybe. But I have another theory. Whether he knows it or not, Laak is actually calibrating. As the cards are dealt, he keeps updating the verbal commentary to reinforce in his mind what different probabilities “feel” like. By the river, he might be saying things like, “This is what 2% feels like.” He’s calibrating his mind to instinctively know what it’s like to be a 50:1 shot.
This may sound ridiculous to you, especially if you’re already good with math. You might be saying, “I already know that if I’m 5% to win, I will win 1 in 20 times. What’s the big deal?” The big deal is that, by default, nearly all the modules of your mind either can’t handle probabilities or skews them in self-serving ways. Even if your deliberative mind knows the math, unless you’ve explicitly done calibration exercises where you’ve got something on the line, the rest of your mind will consistently overestimate how often high probability events will occur (such as 80% probabilities) and consistently underestimate the likelihood that lower probability events will occur.
If you haven’t played poker before, you may hear that some good outcome for you is 91% likely to happen and think, “That sounds fantastic!” Since your mind wants to believe it, it may even trick you into feeling like it’s almost a sure thing. But if you play poker, you’ll know that 91% is like trying to have your hand hold against a gut-shot with one card to come. You’ll know what that feels like. So you’ll know that 91% is far less automatic than it intuitively feels.
I highly recommend that as part of your rationality training, you calibrate like this while playing poker. When you get all in, immediately calculate your chances of winning the hand. Then say to yourself, “This is what XX% feels like.”
Over time, you’ll begin to intuitively sense probabilities instead of just hear a number and think you know what it means. This has ramifications for wider life rationality too. As with all biases, being mis-calibrated costs you utility. The more mis-calibrated you are, the more subject you are to the loss of time, money, and well-being. You’ll be more likely to be loss averse, risk averse, vulnerable to zero-risk bias, dutch-booking, circular preferences, or even neglect probability entirely when making decisions under uncertainty. If you’re mis-calibrated, the world will take advantage of you at every turn. Remember: just like the poker economy, our real economy and even society at large is designed with your mis-calibration in mind. It will supply as much exploitation as your poor thinking will support. So calibration is a big deal. Make sure you’re well-calibrated to avoid this constant loss of utility!
You can learn the science of riding a bike all day long, but you’re not going to be able to ride a bike until you just get on and try. Similarly, you can study stats all day long, but you’re not really going to be someone capable of implementing probabilistic thinking until you get out there and learn what 2% truly feels like.
Turn your goals into games, and accomplish them by learning to win, mostly through deliberate practice with downside when wrong. Seems simple enough, but I want to mention that certain games—especially the most interesting ones in life—are played in open systems with undefined rules and unbounded upside/downside. Whereas only so many things can happen in a game of Connect Four, the possibilities are endless if your game of choice is, say, business.
While I still believe the best way to learn probability, find edges, and make +EV decisions is to actually do it over and over, you should be cognizant that this experience isn’t as useful in grasping outlying low-probability events. Can you really understand the odds of having a royal flush (1-in-649,739) simply by playing poker? Of course not.
These outlying events are very important—one of the biggest exploitable edges I can think of—but that’s a topic for another post.
Right now, I need to estimate the chances of a Furby revival to see what the probability is that I can get in the black to pay back my mom.