Článek v češtině zde / click here for Czech translation.
My name is Jakub Machata. The attempts to understand the world through numbers were apparent from my early childhood. That boy, who during the PE soccer games made mental notes about who scored and assisted to how many goals? Who then produced cards with the stats for all the kids? Yeah, that was me. So it was fitting—even though it was just simple luck—I stepped into baseball shortly after turning eight: baseball, only the most stats-happy and data-driven sport in the world.
Even nowadays, baseball, much like football, is viewed as an obscure sport in the Czech Republic. Mostly a boring spectacle that’s maybe good for Americans but doesn’t make sense to most Europeans. But back in the 90s? Forget about it—baseball was by and large considered a stand-off that gives a shot to the non-athletic kid. That is amusing because serious baseball players are damn studs with enough fast-twitch muscle fiber to knock your head off. Most spectators didn’t know what they were watching. For that matter, some of my teammates happened to be in that category as well. For teams, putting enough bodies together to fill the roster with was a challenge in itself. All you had to do to get some playing time was show up with tied shoelaces. Mentally, I was overprepared, knowing the game’s details before I even got to feel the baseball’s grass for the first time; before I stepped in the batter’s box to learn that my small frame can be my big advantage. My strike zone was minuscule and I knew how to work it. “Jakub learned baseball by playing a videogame” was a running joke in our team for a long time. Oh yeah, I played the hell out of Hardball II. That game looks like garbage when viewed through the optics of 2020. Yet it helped me learn many minor rules and tactics of the game like what’s infield fly or which players were responsible for which cutoffs after an outfield hit.
That served me well, so it was only natural I used the same strategy much later when I was in serious need of practice as a rookie-scorer for my local team in 2004-ish. This eventually led me to scoring several international tournaments. The first one was most memorable, especially the very first game. Slightly intimidated, I went to a dugout filled with English-speaking players to ask their coach about a lineup. In response, he gave me a sheet of paper with his team’s batting order, along with a good ol’ American smile and a souvenir—a little pin with a USA flag on it. Proud as hell, I’ve put that baby in my ballcap and kept it there for the duration of the tournament. I even held on to that through several times I changed addresses. As small as the pin was, it was a much-needed metaphor. It represented the link between work and results.
Anyway… I was in serious need of practice to become half-comfortable with recording the efforts of real-life people. And so I’ve spent my evenings scoring the games I’ve played in Triple Play 2001—what a nerd! Jumping from being a player to being a scorer was a logical step once I’ve noticed that I looked forward less to the actual games and more to the nights after. I went through the scorebooks and updated the Excel spreadsheets filled with our team’s stats. In retrospect, being in the actual competition was simply a means to an end. Like a junkie that needs his fix, I found a way.
I practiced scoring laboriously by doing the MLB games —if I was lucky to catch them on TV— or, more often, the games I played on PC. This is the scorecard of my Atlanta Braves beating the snot out of AI-controlled Toronto Blue Jays, 23-9.
I used my passion a little bit when my friends and I ran tournaments in Doom—yeah, that Doom. There I fully realized that it doesn’t matter what competition it is; a sport, a game, a mixture of both (I’m looking at you, poker), players love their stats. But broadly speaking, I pretty much went on hiatus after I ditched baseball in 2008, both in terms of my growth as a stat-head and as a person overall. Unless you count taking advantage of really horrible players in the golden days of online poker, or my epic franchises in sport manager games, or mastering the useless skill of chip tricks as growth, in which case, yes, I dominated on all fronts. Actually, maybe it wasn’t a total waste of time. Poker was a mechanical grind—playing twelve tables of micro stakes at once didn’t leave much room for art and great bluffs—but it paid several hundred bucks a month. The chip tricks were a good way to look skilled when playing live poker, and they came easy to me. Maybe it helped that I kicked ass at touch-typing — I was averaging 120+ words per minute — very early in my life, so I certainly have had some dexterity in my fingers. In the teeny-tiny world of chip tricks, I’ve finished top-3 in several online competitions. I was even writing how-tos for a Czech poker connected to Paradise Poker. They sent me a chipset to show their brand on camera in video tutorials. Nowadays, that’s just another day in the office for modern influencers, but it felt special to get a package like that. Anyway… It was more than ten years ago, and they are out of business for a while now, so now I can say it: those chips were slippery as hell, and I hated them with passion.
A few years later, my oldest brother told me about this book, The Signal and the Noise by Nate Silver which became my guidance. The book is focused on data: what kind of data we have at hand, how we should use it, and how we do use it. It was immediately apparent to me that this guy knows his stuff. Silver made a good name for himself for predicting election results and as an author of PECOTA, a model predicting baseball’s player’s performance. The first chapter I’ve read was regarding undersized, balding second baseman Dustin Pedroia. Most traditional scouts evaluated him rather poorly. He was drafted in the second round by the Red Sox and became an absolute allstar, winning MVP in his second year as a starter. I was hooked instantly. At the risk of making this sound too dramatic, I’m gonna go ahead and claim this book is the top-2 on the list of reasons why I turned my life around—the first one is meeting my wife. Silver’s knowledge and humor not only rebooted my knack for quantifying things, but they also made me appreciate reading. In the three years between 2016 and 2018, I’ve read 205 books.
We installed e-sport elements into competitions for the archaic game of Doom. This was well before e-sports were even a thing. Stats were so significant that I received two complaints via email within ten minutes of disconnecting parts of the site so that I could do small updates to our point system formula. True story.
In 2014, I was knee-deep in NFL. Football captivated me, and I was slightly frustrated with the state of ignorance regarding data in it. In a way, it was understandable. Football is one of the fiercest ways humans can compete with one another without killing themselves. Analytics would be in a tough position even if the traditional football guys didn’t notice what the wave of Harvard and Yale graduates did to baseball in the past ten years—and believe me, they noticed. The message to a stereotypical nerd hunched over his laptop was loud and clear: “You don’t belong here, you weakling“! The irony of the whole dispute was that FootballGuys™ were rejecting ideas that ultimately made the offenses play
more manly more aggressively. Either way, the contrast between the access to—and the level of—information between football and baseball teams was enormous. MLB had high-level cameras on every stadium. That let analysts know the speed, horizontal and vertical movement of the pitch. When the bat contacted, they could study the angle and speed with which the ball came off the bat—all of this in real-time. And yet, NFL teams were still judged by meaningless volume statistics. Enter Football Outsiders and Warren Sharp and his blog. These guys were smart and didn’t shy away from putting it on display. Each year, they invested hundreds of valuable hours and published dozens of articles to make sports fans smarter. All one needed to learn was right there for the taking.
I was preparing for the 2015 NFL season, which was my first (spoiler alert: it was my last, too) as a paid betting advisor on a Czech betting site KolemDvou. I’ve put together my first model. I called it Anthony, inspired by the football tout from Two For The Money, played by Matthew McConaughey. Initially, the purpose was Anthony to help me pick games to bet on. I quickly left this nice idea in the dust where it belonged. All the regressive analysis I did showed the harsh reality that numbers alone aren’t gonna be enough to win. Certainly not in the sport that combines the most efficient market with the smallest sample size. NFL teams play mere sixteen games of the regular season, compared to 82 for NHL and NBA, and whopping 162 for MLB. Still, the model served me beautifully as a way to comfortably filter stat sheets for each weekly match-up, saving me countless hours of research. On top of that, I probably got to an inevitable conclusion that I honest-to-god suck at betting maybe a year or two earlier than I would without the model. Even though my short-lived career as a betting advisor crashed and burned spectacularly, I decided to stick with the model, as well as with its silly name. After I licked my wounds and counted my losses, Anthony became my standard for rating teams.
The idea of building a machine that would pick games to bet never left me. Instead of further trying my luck with football, in 2016, I turned my attention to ice hockey. And it was once again Nate Silver
to inspire me whom I stole from. Silver’s website FiveThirtyEight compared teams across multiple sports using own adoption of Elo rating. Since 1959, Elo was a standard to rate chess players in USA and is used world-wide since 1970. It seemed fascinating that merely providing game results without any additional stats or details could help estimate how the teams stacked up. I spent the whole summer building a model with an Elo rating of my own. Putting together results from close to twenty hockey leagues since 2000 was reasonably quick, so was programming the formulas, but the manual collection of a few years’ worth of historical betting odds took a while. I looked at everything from the glamorous NHL to the lowest of the low: Ekstraliga w hokeju na lodzie, which is indeed a Polish league. As expected, I quickly learned there’s no way I can ever win in NHL, KHL, or Swedish league with my simple excel file. Oddsmakers were just too right at this level to get outsmarted by my simple excel sheet. In less popular European leagues, however, that was a different story. These markets weren’t even close to being efficient. The betting services try to balance their lack of knowledge by forcing bettors to pay more juice—they lower the odds—if they want to bet on these unknown, unpopular, unwatchable leagues. Still, my backtest suggested there was room to make money even with the increased vig. And so for the whole 2016 season, I bet real money on my model’s picks for +3.1% yield, which means I won three bucks for each hundred I risked. Normally, that wouldn’t be much of a reason to celebrate, but this wasn’t regular normal. A human bettor can make maybe 300 bets a season. But the computer model doesn’t get tired. It never wakes up with a crippling hangover, sore as hell from a 100K weekend hike, or simply feeling like not doing anything and watching Futurama instead. And so in 2016, when the model advised me to place precisely 867 bets, that’s exactly what I did. That’s why the fairly weak +3.1% yield meant a +28% in much more important column ROI—a return on investment. At that time, I considered it a success. These days, I’m thinking, “Good, not great,” which coincidentally is how I’m gonna name my biography.
In 2017, the model wasn’t going nearly as well. I pulled the plug on it after 194 plays with almost offensive +0.6% yield. I basically didn’t get paid for all the work I’ve put in—keeping up with the schedule, collecting scores and odds, betting the games model highlighted. I didn’t mind it. I was strangely having almost a perverse pleasure betting on these inferior leagues nobody but gamblers cared about. However, my presence with the Armchair Analysis project grew and was becoming a serious second job. When I was facing a decision of whether I should go with the sport I really loved, and where the real money was, or if I should keep grinding British hockey, I couldn’t close Excel fast enough.
The timing was amazing. When I found Armchair in the Autumn of 2016, its owner and my soon-to-be boss, Dennis Erny, was in the middle of scouting for people for his new charting team. The idea was, we would all collectively go through every play of every week to mark various details that interested clients: a mixture of betting public, fantasy players, and TV talking heads. How many yards did the receiver run after he caught the ball? Was the QB was under pressure? Was the pass incomplete because the receiver dropped the ball, or was the QB inaccurate? This would provide the layer of information I felt I’m missing tremendously in Anthony. Without thinking twice, I’ve typed a short motivational letter, added a bunch of my old articles filled with ugly charts, provided a link to my blog, and sent it over via the email form—only it wasn’t an email form. Oh, the horror when I realized my defacto CV ended up in public comments, just sitting there to be judged by anyone applying for the same job I do—awkward! We had a good laugh over that—Dennis surely more than me—and I worked the 2016 season for four hours a week as a volunteer, working the Carolina Panthers and Baltimore Ravens games.
I fell in love with the process and remained on the team. Before the 2017 season, my excitement went through the roof. So much that I couldn’t control myself. At times, I seriously tested the fine line between being a valuable, enthusiastic employee and being a downright annoying twat. But at the end of the day, that was for the best. After all, while Armchair had a long history of selling data, the charting project was brand new. It takes a while to build a dedicated, well-established team, especially when dealing with Joe and Jane from the internet if those are even your real names. To make things more interesting, several people quit within the first two weeks without as much as saying ‘this job sucks, bye‘. In that situation, without the benefit of an experienced group of reliable people with solid coaching and know-how, having a psycho on your staff can go a long way. We made it work without ever postponing the deadlines promised to clients, and I was promoted to be the lone senior analyst. And here’s where the stories connect. As much as I loved Polish and British hockey games, I was going all-in on this NFL thing. It paid off: I was stepping into the 2019 season as ‘Head of game charting’. But you know how people are with their fancy job-titles lately, right? In loose translation, this meant ‘You gonna do everything, and you’re gonna like it‘. And I did. The offseason in 2020 was pivotal. As the world was becoming a ball of chaos amidst the pandemic, we sensed a chance to transform our hobby project into a serious contender for the best source of data. For the first time, we didn’t take much of a break—not more than two weeks anyway. On the back of the NFL covid-season, we’ve put together the biggest downloadable data set on the web for about a third of the prize our competition asked.
There are more details in special articles about Football and Ice Hockey.